bearing fault detection of induction motor by ann …

34

Int. J. Elec&Electr.Eng&Telecoms. 2014 S M Shashidhara and P Sangameswara Raju, 2014

BEARING FAULT DETECTION OF INDUCTIONMOTOR BY ANN METHOD

S M Shashidhara1* and P Sangameswara Raju2

*Corresponding Author: S M Shashidhara, [email protected]

Three-phase induction motors are the most-widely used electrical machines and are consideredas the workhorses of the industry. About 60% of the electrical power generated is used byinduction motors and hence have gained high importance. Their fault-free operation is desiredin industrial processes. For this reason, detection of incipient faults has gained significance.Motor current signature analysis is the technique popularly used for the detection of fault inmachines. It is non-intrusive and an online method, which analyses the health of the machinethrough the spectrum monitoring of the stator current. Bearing fault is the most common fault inIM and this paper deals with detection of Bearing fault using Artificial Neural Network (ANN).

Keywords: Induction motor, MCSA, Bearing fault, ANN

INTRODUCTIONInduction motors are workhorses of industrialprocesses and are frequently integrated incommercially available equipment andindustrial processes. There are manypublished methods and numerouscommercially available tools to monitorinduction motors to ascertain a high degreeof dependability uptime. Despite these tools,many companies are still faced withunforeseen system failures and decreasedmotor lifetime. The analysis of induction motorbehavior during abnormalities and the

ISSN 2319 – 2518 www.ijeetc.comVol. 3, No. 1, January 2014

© 2014 IJEETC. All Rights Reserved

Int. J. Elec&Electr.Eng&Telecoms. 2014

1 Department of E&CE, Proudhadevaraya Institute of Technology, Hospet, India.2 Department of EEE, SVU College of Engineering, Tirupati, India.

possibility to diagnose these circumstanceshave been a challenging issue for manyelectrical machine researchers.

The literature indicate that majority of thefailures in the three-phase induction motors aremechanical in nature such as bearing faults,eccentricity or misalignment faults and balanceassociated faults (Shashidhara andSangameswara, 2013; and Allbrecht et al.,1986).

The faults occurring in motor bearing iscommonly due to the excessive load, rise oftemperature within the bearing, employment

Research Paper

35


of defective lubricant and so on. The bearingconsists of primarily of the outer race, the innerrace way, the balls and cage which ensuresequidistance between the balls. The differentfaults that may occur in a bearing can becategorized according to the affectedcomponent:

• Outer raceway defect

• Inner raceway defect

• Ball defect

BEARING FAULTSThe bearing faults can be grouped into cyclicfaults and non-cyclic faults (Thomson andGilmore, 2003). Cyclic faults emerge when therolling component and the rolling element cageof the bearing passes through the point ofdefect. The deep scratches in a rolling elementare a case of cyclic fault. The materialabrasion, quality degradation of the lubricantdue to contaminants, slither, insufficientlubrication and skid amongst the movablebearing components induce mutilation of thecontact areas, which is a non-cyclic fault family.The bearing defects cause non-stationary andfault specific frequency constituents in thestator current and the generated vibrations(Randy et al., 1995; and Benbouzid, 2000).

Ball DefectThe bearing cage in a ball bearing bears onthe balls at evenly balanced berths and aidsthe confined rolling of the balls along theracetracks. While the motor shaft is rotating,the bearing cage rotates at a steady angularvelocity that is average of the inner and outerrace angular velocities. The cage angularvelocity can be exploited to work out the valueof dominant fault frequency due to cage defect,fCD as given below (Eschmann et al., 1958):

2cos

2cos

601

60

dDdD

DDrrf oiooii

CD

...(1)

where

i = Angular speed of the inner race in RPMo = Angular speed of the outer race in RPMD = Pitch Diameter

d = Ball Diameter

= Ball contact angle

ri = Inner race radius, ro = Outer race radius

The outer race is attached to the casing thatis stationary. The shaft and inner race aremounted together and both revolve at the same

Figure 1: Fault Percentages in InductionMotor

Figure 2: Roller Bearing Geometry

36


angular speed. Consequently, it can beassumed that:

o = 0 and i = r ...(2)

where r = Rotor angular speed in RPM

Incorporating the above mentionedassumption as shown in Equation (5) bringsEquation (6) as given below:

D

dF rCD

cos1120 ...(3)

Empirically, the fundamental frequency dueto cage defect for a ball bearing with six totwelve balls in it is given as:

fCD = 0.4 rs ...(4)

where

60rpminspeedangularRotor

rs

The ball defect is simulated in MATLABusing Artificial Neural Network algorithm. In thesimulation procedure, the harmonicfrequencies generated due to the bearing balldefect fed into the stator current of a healthy

machine. Thus the current of phase A obtainedis as shown in Figure 3.

Characteristic FrequenciesEach anomaly in the motor operation getsreflected in the stator current as acharacteristic frequency. Motor currentsignature Analysis is a nonintrusive and easymethod of finding the fault in the motor. It needsno other input, but, only the stator current. TheCharacteristic Frequencies can be identifiedby the frequency spectrum of the current.

In this paper only ball defect is beingconsidered for analysis. Hence, thecharacteristic frequency of the ball defect isgiven by

0,1 . ibearing fmff

where = 1, 2, 3, 4, …, and fi,0 is one of thecharacteristic frequency based upon thegeometry of the bearing shown in Figure 1.

EXPERIMENTAL SETUPThe experimental setup consists of two threephase induction motors 2.2 kw, 440 V, 4 pole,

Figure 3: Stator Current in Phase A

37


50 Hz coupled to a mechanical load. Onemotor is healthy and the observations wereconsidered as reference to compare with theother motor which has a bearing fault.

The data acquisition and conditioningsystem was developed and data visualizationdone on a PC using FFT and Power SpectralDensity.

Case 1: Healthy Motor

Experiment was conducted on a healthy motor;the measured current and speed are tabulatedin Table 1. Their corresponding waveforms andfrequency spectrum are shown in Figures 4and 5. Interpolation technique was employedto determine the waveforms for the (0.1, 0.2,0.3, 0.4, 0.6, 0.7, 0.8, 0.9) of the rated loadcases to furnish the entire data needed for theNeural Network (NN) to be trained for eachcase in order to increase its ability ofdiagnosis.

Case 2: Faulty Motor

Now, with the defective bearing test resultswere taken and the results are presented asbelow:

The above spectrum analysis diagramsshow the presence of harmonic componentsaround the fundamental frequency of 50 Hz.The above spectrums show the existence ofa harmonic components located around thefundamental l ine frequency. Theseconstituents are used to be called as lowerside-bands and upper sidebandscomponents. It is apparent that their distance

Figure 4: Experimental Setup

Load Speed

No Load 1440

Half Load 1380

Full Load 1295

Table 1: Load Applied on Motor and theCorresponding Speed

38


Figure 5: Healthy Motor at No Load (a) Phase Current Signal (b) FFT Current Spectrum

Figure 6: Healthy Motor at Half Load (a) Phase Current Signal (b) FFT Current Spectrum

39


Figure 7: Healthy Motor at Full Load (a) Phase Current Signal (b) FFT Current Spectrum

Figure 8: Bearing Fault at No Load (a) Phase Current Fault Signal (b) Fourier Analysis(c) Side Bands at No Load (d) Side Bands at Half Load (e) Side Bands at Full Load

40


Figure 8 (Cont.)

Figure 9: Sideband Frequencies in Stator Current Due to Fault in Motor

41


from the fundamental component in thespectrum is increasing with load. Also themagnitude of these sidebands increases asload increases. Figure 9 represents thesesidebands clearly in general.

Implementation of Neural NetworkNeural Network was simulated using MATLABNeural Network toolbox choosing DampedLeast-Squares (DLS) method, also knownas Levenberg-Marquardt Algorithm (LMA)

Figure 10: ANN Algorithm

42


method. It provides a numerical solution to theproblem of minimizing a function, generallynonlinear, over a space of parameters of thefunction.

CONCLUSIONThis paper dealt with the design and testing ofthe induction motor for the diagnosis of bearingfault. The results obtained experimentallymatch with the fault signature frequenciesderived analytically. It is proved that the MCSAis an effective tool in the detection of bearingfault of induction motor.

REFERENCES1. Allbrecht P F, Appiarius J C, McCoy R M

et al. (1986), “Assessment of theReliability of Motors in Utility Applications-Updated”, IEEE Transactions on EnergyConversion, Vol. 1, No. 1, pp. 39-46.

2. Benbouzid M E H (2000), “A Review ofInduction Motors Signature Analysis as aMedium for Faults Detection”, IEEETransactions on Industrial Electronics,Vol. 47, No. 5, pp. 984-993.

3. Eschmann P, Hasbargen L and WeigandK (1958), “Ball and Roller Bearings: Their

Theory, Design, and Application”, K GHeyden, London.

4. Randy R Schoen, Thomas G Habetler,Farrukh Kamran and Robert G Bartheld(1995), “Motor Bearing DamageDetection Using Stator CurrentMonitoring”, IEEE Transactions onIndustry Applications, Vol. 31, No. 6,pp. 1274-1279.

5. Riddle J (1955), “Ball BearingMaintenance”, OK University of OklohamaPress, Norman.

6. Shashidhara S M and SangameswaraRaju (2013), “FPGA Based EmbeddedSystem Development for Rolling BearingsFault Detection of Induction Motor”,International Journal of ElectricalEngineering and Technology(IJEET), Vol. 4, No. 5, pp. 78-86.

7. Thomson W T and Gilmore R J (2003),“Motor Current Signature Analysis toDetect Faults in Induction Motor Derives-Fundamentals, Data Interpretation, andIndustrial Case Histories”, Proceedings of32nd Turbo Machinery Symposium, Texas,A&M University, USA.

bearing fault detection of induction motor by ann …

Documents