Bearing Fault Indicator in IM Using Stator Current Spectral Analysis

8/12/2019 Bearing Fault Indicator in IM Using Stator Current Spectral Analysis

1/5

Bearing Fault indicator in Induction Machine Using Stator urrentSpectral AnalysisBaptiste Trajin*, Jeremi Regnier*, ean Faucher**Universite de Toulouse; LAPLACE; CNRS, INPT, UPS;

2 rue Charles Camichel BP 712231071 Toulouse cedex 07, France

Email: [email protected]@[email protected]

Keywords: Bearing fault, Vibration spectrum analysis, Current spectrum analysis, Automatic energy extraction.AbstractThis paper deals with the application of motor current spectral analysis for the detection of rolling bearings damagein asynchronous machines. Vibration measurement is widelyused to detect faulty operations. However, this approach isexpensive and cannot always be performed. An alternative isto base the monitoring on electrical quantities such as themachine stator current which is often already measured forcontrol and detection purposes. Firstly, to define an automaticdetector for bearing faults based on mechanical quantities,vibration spectra are analyzed. Afterwards, it is shown thatbearing faults induce mechanical load torque oscillations.Then, a theoretical stator current model in case of load torqueoscillations demonstrates the presence of phase modulation.Related sideband components on current spectrum can beused for detection. Experimental measurements show thata transfer function including resonance links amplitudes ofsideband components to the torque oscillation frequency. Thissingularity will be used to improve the detection efficiency.A fault detector using the energy of stator current in specificfrequency ranges is then proposed. The efficiency of the detector is studied for different operating conditions. The supplyfrequency values are particulary investigated to demonstratethat using the resonance previously described could improvedetection performances.1 IntroductionElectrical drives using induction motors are widely usedin many industrial applications because of their low costand high robustness. However, faulty operations could beinduced by bearing faults [8, 10] A condition monitoringallows improving the availability and reliability of the drive.Traditionally, bearing faults detection uses vibration analysisbut measuring such mechanical quantities to detect bearingfaults is often expensive. To reduce the cost of monitoring,available electrical quantities such as stator current could beused. A general review of monitoring and fault diagnosisschemes using stator current can found in [6]. In caseof bearing faults, several studies demonstrate that specific

signatures appear on stator current spectrum [7, 11] Therefore,few of them concern the definition of a detector performing anautomatic extraction of relevant information from the currentspectrum. The present work deals with these specific aspects.In section 2 a short overview of bearing faults is presented.The principles of an automatic extraction of energy in vibrationspectrum are defined and validated. In section 3 a particularapproach based on the assumption that bearing faults induceload torque oscillations is studied [2]. Using the magnetomotive force theory, an analytical stator current model is usedto demonstrate that sideband components due to load torqueoscillations exist in the current spectrum. An experimentalBode diagram is determined to study the amplitude of thesespecific sidebands related to torque oscillation frequencies.In section 4, an automatic detector based on current spectralenergy estimation in specific frequency ranges is presented.Finally, section 5 demonstrates the ability of the proposeddetection scheme to distinguish healthy from faulty bearingsfor different operating conditions.

2 Vibration analysis for faulty bearings

2.1 Bearing faults and characteristic frequencies

6208-type bearings are modified using electro-erosion to createan outer or inner race 3mm-Iong single point defect comparable to the effect of spalling due to a severe ineffectivelubrification [3]. Thus, faulty bearings are mounted in a5 5kW induction machine supplied by a variable frequencyinverter. An acquisition board is used to sample the torque,vibrations and stator currents. Characteristic frequencies offaulty bearings in the vibration spectrum are theoreticallywell known. Only the low frequency ranges 1kHz areanalyzed. In fact, low frequencies render defect which arerelated to the rotating speed like eccentricity or bearing faults.Moreover, low frequency harmonics due to defects couldappear as combinations of mechanical rotating frequency. Thischaracteristic frequencies can be expressed using 1 [4 10].

592


2/5

{2

Qfeu1.5

tQ:r1 2eQQ.. L ~ : i : : ~ ~ ~ = = = ~ ~ :1 0.5

Q)II:

r . = = ; ; : ; : : c : = = : : : c = ~ - - . - - - , - - - - - - - - , - - r - - - - ,.5

c

U.Qo ~ o o o .Q5o,22eQQ 1>QII:

modulations at e-95 L L _ _ . . . . . . . J . . _ L _ ~ _ _ _ _ _ _ _ _ _ _10 20 30 40 50 60 70 80Frequency Hz)

Figure 1: Vibration spectrum for outer race fault Figure 2: Mechanical indicator for outer race fault

f == fr 1 DbCOS })or 2 b Dpf - fr N DbCOS })ir - 2 b 1 + DIe = ; 1 _ D b ~ O S }

p

1

The chosen frequency ranges are given in 2 . The proposedindicator uses the relative energy error in the specified frequency ranges between a faulty and healthy reference vibrationspectrum. A cumulative sum is then applied on the energydifferences extracted from the frequency ranges related to thebearing fault investigated.

3.1 Load torque oscillations due to bearing faults3 Load torque oscillations

After demonstrating the effects of bearing faults on vibrationsin section 2, the effects of bearing faults on mechanical loadtorque are required. Experimental spectrum of load torquedemonstrates the presence of harmonics at frequencies relatedto bearing faults. Here, the mechanical speed is chosen equal tothe nominal one namely 25Hz. The outer race fault frequencyequals 89Hz and the inner race frequency equals 136Hz.

where n E [1; 3].Fig. 2 presents this indicator for fde == for Notice thatFig. 2 is double scaled with a factor of 1000 betweenthe two vertical axes. Using a common healthy reference,the cumulative sum of the energy difference for an outerrace fault and a healthy case are compared. The distinctionbetween healthy and faulty case can thus be clearly done.This approach certifies that bearing fault detection is possibleusing the energy in the spectrum of mechanical quantities.However, setting Ide == for or fde == f i r j does notguarantee the distinction between inner and outer race faults.In fact, numerous harmonics due to inner race defect existsin frequency ranges corresponding to outer race fault andreciprocally. Consequently, such an energy detector cannotbe used for diagnosis but only for detection and monitoringof bearing faults. This observation is not really a negativepoint because many applications do not require an accuratediagnosis of the bearing defect.

2[nfde - fe; nfde + f

[nfde - fr - fe; nfde - fr + f[nfdej + fr - fe; nfde + fr + f

where:

In case of outer race fault, vibration spectrum is computedusing an averaged periodogram [5]. For example, Fig. 1demonstrates that characteristic fault frequencies occur invibration spectrum. Here, the supply frequency is chosenequaled to 13.3Hz. Considering the slip of the inductionmachine, the mechanical rotating frequency is then equal to6.56Hz. Using 1 and the assumption that the contact angleis null no axial load applied on the rotor shaft , the cagefrequency equals fe == 2.61Hz and the outer race faultfrequency equals for == 23.5Hz. Obviously, combinationsof several fault frequencies appear in the vibration spectrum.Strong modulations at fe and ir can be noticed around themultiples of outer race fault frequency.These observations lead us to define a mechanical detectorbased on spectral energy in vibration spectrum. The faultdetector is defined by extracting energies on frequency rangesrelated to the frequency components at fdej where fde iseither the inner or the outer race theoretical fault frequency.Moreover, the frequency ranges are extended to include modulations linked to the mechanical speed and cage frequenciesunderlined by the vibration spectral analysis see Fig. 1 .

2.2 Vibration spectrum and spectral energy detector

for outer race fault frequency; fir inner race fault frequency; fe cage frequency; fr mechanical rotating frequency; b number of balls; Db ball diameter; p pitch diameter; contact angle.

593


3/5

10 0

11 0

12 0

13 0

2.

165 170 175 180Frequency Hz

a)

185 262 264.5 267 269.5 272 274.5 2n 279.5 282Frequency Hz

b)Figure 3: Spect rum of mechanical torque - Comparison between healthy and faulty cases

Hence, Fig. 3 a) and 3 b) show a part of the mechanical torquespectrum around twice the characteristic fault frequency forouter and inner race fault respectively compared to the healthycase. I t could be noticed that bearing defects cause someload torque oscillations which can be detect by a mechanicalmeasure. In the following part, the link between load torqueoscillations and stator current will be recalled.

3.2 Stator current modelPrevious studies on mechanical failures in induction motorshave shown that load torque oscillations induce phase modulations P M) on stato r cur ren t [1, 2, 9]. Considering that theload torque oscillation is composed of a single harmonic atthe pulsation Wosc , the load torque on the shaft of the machinecan b e expre ssed using 3) as an average torque equal to theelectromagnetic motor torque and a sinusoidal component ofamplitude r c :

i t = s cos wst + s + Ir sin wst + p Jr cos woscwo sc 4)In 4), the first term is related to the stator MMF contributionand the second one, including the phase modulation of statorcurrent, is related to the rotor MMF contribution. If theamplitude of the torque oscillation is pretty slight, the FourierTransform Ff of the stator current can be expressed using 5) along the frequency 1/

Moreover, notice that in case of faulty bearings, the frequencyof load torque oscillations losc can equal any of combinationsof characteristic fault frequencies und erlin ed by the loadtorque spectrum.

3) 3.3 Amplitude variation law of sideband components

Physically, considering the mechanical equation linking torqueto mechanical angular position, it is proved that load torqueoscillations cause angular position oscillations. Here, the mechanical transfer function is equal to a simple inertia plusan integrator. The mechanical position is used to calculatethe rotor magnetomotive forces MMF). Thus, magnetic fieldin airgap is obtained by the product between the airgap permeance Ao, considered as a constant along time and position),and the rotor and stator MMF. The magnetic field is integratedon the coil s surface of a stator winding in order to determinethe airgap flux density in the coil. Hence, time derivation ofairgap flux density in a stator winding leads to induced voltage.The stator current expression 4) for an arbitrary phase isthen obtained considering a linear relation between current andinduced voltage. In 4), Isis the amplitude of stator current,r the amplitude of rotor current, p the number of pole pairsof the asynchronous machine and the inertia.

The knowledge of the amplitude variation of stator currentsideband components related to the fault frequency is of stronginterest for detection purpose. The link between PM harmonicsamplitudes of the stator current and load torque oscillationfrequencies is then experimentally studied and compared tothe theoretical expression 5).In order to draw a B ode diagram, the asynchronous motor iscoupled to a DC motor. The DC machine is connected to aresistor through a DC/DC converter which controls the DCmotor armature current. The reference current is composed ofan oscillating component plus an offset in order to induce loadtorque oscillation around a mean torque value. Mechanicaltorque is measured with a torque sensor. Sideband componentamplitudes of stator current are measured by off-line spectralanalysis. The experimental gain Bode diagram shown in Fig.4 is obtained by varying frequency of load torque oscillations.T he main observation lies in the existence of a resonancepoint around Ires 2 Hz Assuming that a single-point

594


4/5

1 r ; === C= ================ I ;I Statorcurrentsideband modulationenergyI5 .

2

25

30to: .. .. : : ,- 3 5 ~ :-- : : . ; .. : .. : ..

Is Ide where Ide is either the inner or the outer racetheoretical fault frequency. Moreover, the frequency rangesare extended to include modulations linked to the mechanicalspeed and cage frequencies underlined by the vibration andmechanical load torque spectral analysis see Fig. 1 and 3 .The chosen frequency ranges are given in 6 . The proposedindicator uses the relative error of energy between the currentspectrum in faulty and healthy case in the specified frequencyranges.

10 1 2Loadtorque oscillation frequency Hz

Figure 4: Gain Bode diagram of experimental transfer function between loadtorque oscillations frequency and sideband components on stator currents

lis [nldef - fe nldef + e] IIfs [nfdef - T - fe nfdef - iT + e] Ilis [nidef + T - Ie; nldef + iT + e] I 6

defect related to characteristic frequencies determined in 1creates slight load torque oscillations at these frequencies, theresonance point is used as a natural amplifier to obtain higherPM hannonics on stator current. Then, in order to use theresonance point, the supply frequency of the asynchronousmachine is tuned to ensure that the theoretical mechanicalfault frequency of the bearing under test equals the mechanicalresonance noticed on the Bode diagram. In fact, with thisapproach, the PM modulation index shown in 5 will increasebecause of the resonance.4 Definition of a current spectral detector forbearing faults4 1 Improvement of SNR of stator current for faul t harmonic detectionAs defined in section 3, single-point defects induce load torqueoscillations and consequently, phase modulations on statorcurrent. However, amplitude of this PM is quite slight andcould be buried in noise. Reducing the Signal to Noise RatioSNR of the current signal is then necessary. On current measurements, noises are considered to be not correlated neitheralong time nor on two stator phases. Dividing the currentsmeasurements into time blocks, the averaged periodogramallows reducing the SNR by correlating measurements. It canbe estimated that the variance of the full signal is dividedby the number of block used [5]. Moreover, the SNR can beimproved and the noise variance reduced by correlating twophase currents using a multiplication of their Fourier Transform. As a consequence, the evaluated SNR of the resultingsignal is twice the SNR of a unique current. This methodallows improving the efficiency of fault hannonic detection. Incommon applications, measuring two stator currents is oftenperformed for control purposes so this SNR reducing methoddoes not require more sensor.4.2 Definition of current spectral detectorThe fault detector is defined by extracting energies on frequency ranges corresponding to the sideband components at

where n E [1; 5] iT is the mechanical rotating speed and Iethe characteristic cage frequency.Then, as expressed in 7 , the relative errors of energy extracted from outer and inner race fault frequencies rangesb E oTf and b Eir f respectively are added in order to geta single energy difference ~ t o t

Finally, a cumulative sum of b E tot is used to buildthe indicator. Only the last value of the cumulative sum isconsidered as the detector value. As a consequence, the finaldetector value could not perform a distinction between innerand outer race fault but only provide a distinction betweenhealthy and faulty case. The detection of inner or outer racefault will only be done with the supply frequency which istuned to equal the resonance frequency Ires and one of thecharacteristic fault frequencies (lorf or firJ).

5 Experimental ResultsThe acquisitions of the stator currents are done during 80susing a sampling frequency of 6400Hz As the analysis areperformed off-line, the data size and computation time arenot considered as constraints. Three experimental conditionsare tested, corresponding to three different supply frequencies.Each case uses a reference of energy obtained with a healthybearing.5.1 Detection of outer race faultIn order to use the resonance point, the supply frequency Isis tuned to 13 3Hz to ensure for f = fres. The detection ofouter race fault is then favored by the amplification due to theresonance. From Fig. 5 a , the representation of the cumulativesum allows differentiating outer race faulty conditions fromthe other cases. As expected, properly tuning the frequencysupply leads to focus on outer race fault. When Is is set toguarantee for f = Ires the detection of the outer race fault isensured.

595


5/5

6 8 10 12Terms cumulative sum

(a) lor = Ir es 6 8 10 12

Terms cumulativesum

(b) lir = Ires 6 8 10 12

Terms of cumulativesum

(c) Is = 50Hz

Figure 5: Cumulative sums of relative error in

5.2 Detection of inner race faultIn order to favor the detection of inner race fault, the supplyfrequency Is is tuned to 6.7Hz to ensure lir IresFrom Fig. 5(b), the representation of the cumulative sumallows differentiating inner race faulty conditions from theother cases. Once again, properly tuning the frequency supplyleads to focus on the selected fault. Moreover, Fig. 5(a) 5(b) underline that if the characteristic fault frequency doesnot equal the resonance frequency, the associated fault is notdetected.5.3 Detection of faults at nominal frequencyThe supply frequency Is is tuned to 50Hz correspondingto the nominal supply frequency of the machine. No characteristic fault frequencies are equaled to the resonance point.In this case, lir 136Hz and lor 89.4Hz. The PMsignatures are located in the attenuation part of the electromechanical transfer function (see Fig.4). Fig. 5(c) underlinesthat distinction between healthy and faulty cases is not possiblewith this detector when the sideband components are stronglyattenuated. It emphasized the importance of properly tune thesupply frequency to ensure the detection of a possible bearingfault.6 ConclusionIn this paper, a novel method for an automatic detectionof bearing fault in induction motors using stator currentmonitoring has been presented. As inner and outer racecharacteristic fault frequencies are well known, bearings areartificially damaged to ensure the assumption of single-pointdefect. A vibration spectrum analysis has been proposed tovalidate the principles of an automatic spectral detector basedon the extraction of energy in frequency ranges related tobearing faults. Mechanical measurements have shown thatbearing defects induced load torque oscillations. Thus, a modelof stator current demonstrates that load torque oscillationslead to sideband components on stator current spectrum. Theamplitude variation law of these signatures with respect to faultfrequency has been determined by experimental measurements

in order to refine the proposed stator current model. The exploitation of the resonance point allows detecting preferentiallyinner or outer race fault. A detector based on extraction ofenergy in frequency ranges related to PM signature in thestator current spectrum has been built. The automatic detectorallows clearly distinguishing healthy and faulty cases with agood confidence rate while the supply frequency is tuned toensure the detection by using the resonance point. To reducethe computation complexity and demonstrate the efficiencyand reproducibility of the detector, short data length recordswill be studied in further work. Moreover, more realistic faultswill also be investigated.References[1] M. Blodt, J. Faucher, B. Dagues and M. Chabert. Mechanical load faultdetection in induction motors by stator current time-frequency analysis ,IEEE International Conference on Electric Machines and Drives pp.1881-1888, (May 2(05).[2] M. Blodt, P. Granjon, B. Raison and G. Rostaing. Models for bearingdamage detection in induction motors using stator current monitoring ,IEEE International Symposium on Industrial Electronics volume 1 pp.383-388, (May 2004).[3] R. A. Guyer. Rolling Bearings Handbook and Troubleshooting Guide ,Chilton Book COlnpagny Radnor, Pennsylvania, (1996).[4] T. A. Harris. Rolling bearing analysis , Wiley New-York, 3rd ed ..(1991).[5] S. M. Kay. Modem Spectral Estimation: Theory and Application ,Prentice Hall Englewood Cliffs. New Jersey, (1988).[6] S. Nandi and H. A. Toliyat. Condition monitoring and fault diagnosis

of electr ical machines - a review . IEEE Transactions on EnergyConversion volume 20, no. 4, pp. 719-729, (Dec. 2005).[7] R. R. Obaid, T. G. Habetler and J. R. Stack. Stator current analysisfor bearing damage detection in induction motors , Symposium onDiagnostics for Electric Machines Power Electronics and Drives pp.182-187, (Aug. 2003).

[8] B. Raison, G. Rostaing, O. Butscher and C. -So Maroni. Investigationsof algorithms for bearing fault detection in induction drives , IEEE28th Annual Conference of the Industrial Electronics Society volume2, pp.1696-1701, Nov. 2002).

[9] R. R. Schoen and T. G. Habetler. Effects of time-varying loads on rotorfault detection in induction machines , IEEE Transactions on IndustryApplications volume 31, no. 4, pp. 900-906, (Jul-Aug. 1995).[10] J. R. Stack, T. G. Habetler and R. G. Harley. Fault classificationand fault signature production for rolling element bearings in electricmachines , IEEE Transactions on Industry Applications volume 40,no. 3, pp. 735-739, (May-Jun. 2004).[II] J. R. Stack, R. G. Harley and T. G. Habetler. An amplitude modulationdetector for fault diagnosis in rolling element bearings , IEEE Transac-tions on Industry Electronics volume 51, no. 5, pp. 1097-1102, (Oct.2(04).

9

Bearing Fault Indicator in IM Using Stator Current Spectral Analysis

Documents

Transcript of Bearing Fault Indicator in IM Using Stator Current Spectral Analysis