An Improvement of Stator Current Based Detection

of Bearing Fault in Induction Motors

Sun Liling, Xu Boqiang, Member

School of Electrical Engineering
North China Electric Power University (Baoding)
Baoding, CHINA
xbqsllxtz@yahoo.com.cn

AbstractüThis paper develops an improved stator current based based scheme to detect bearing fault, which perfectly blends
detection scheme for bearing fault in induction motors. Firstly, subdivision Fourier transform, self-adaptive filter and rotor slot
experiments of bearing fault are performed. Secondly, by harmonics based slip estimation techniques together, is
thoroughly analyzing the experiment results, a series of proposed in this paper. Fault detection instances in laboratory
interesting highlights, which could help improve the sensitivity demonstrate that the presented scheme is effective.
and reliability of stator current based detection of bearing fault
in actual field, are revealed clearly. Finally, an improved stator
current based detection scheme, which perfectly blends II. STATOR CURRENT BASED DETECTION OF BEARING
subdivision Fourier transform, self-adaptive filter and rotor slot FAULT
harmonics based slip estimation techniques together, is proposed.
Fault detection instances in laboratory demonstrate that the It has been established that once bearing fault occurred,
presented scheme is effective. some harmonics would be induced in stator current of
induction motors, at the specific frequencies given by
Keywords-bearing fault; detection; induction motor; Fourier
transform; filter; slip estimation
f bearing = f s ± kf v . (1)
I. INTRODUCTION Where, f s is the supply frequency, f v is the characteristic
Induction motors are presently in common use as the vibration frequency depending on the faulted part of bearings,
machine of choice in a wide variety of industrial applications. k = 1, 2䯸3䯸" . As example, for outer race defect,
From a number of surveys [1-3], it can be deduced that bearing
fault constitutes approximately 40% of all the induction motor
nb § BD ·
failures. Therefore, early detection of bearing fault in induction fv = f rm ¨1 − cos φ ¸ . (2)
motors is of great significance. 2 © PD ¹
While detecting bearing fault, vibration based methods are
Where, nb is the ball number, f rm is the mechanical rotor
typically preferred. However, the installation and maintenance
of vibration sensors is relatively difficult, and so, vibration speed in Hz, BD is the ball diameter, PD is the ball pitch
signal is not always available to be measured. In that case, diameter, and φ is the ball contact angle (typically equals 0嘙).
stator current signal, which is rather easy to be measured, can Referred to [4-6] for detailed information.
be used to detect bearing fault. Significant efforts have been
expended on these stator current based detection methods of Thus, the detection of bearing fault can be realized by
bearing fault [4-8]. Regrettably, such stator current based analyzing the stator current spectrum and investigating its
detection is less sensitive/reliable than that based on vibration. harmonics at the frequencies given by Equation (1).
To resolve this problem, excellent Digital Signal Processing
techniques, such as statistic and wavelet analysis, have been III. EXPERIMENT RESULTS
employed to improve the performance of stator current based Bearing fault experiments of a 3kW, 380V, 50Hz, 4-pole,
detection methods for bearing fault [7, 8]. typed Y100L-2 induction motor have been completed. The
Similarly with [7, 8], this paper also addresses itself to motor bearing is typed 6206 with the following parameters:
develop stator current based detection method, which is with nb =9, BD=10mm, PD=46mm and φ =0嘙. Here only cites the
improved sensitivity/reliability, of bearing fault in induction results for outer race defect. This type of bearing fault is
motors. Firstly, experiments of bearing fault are performed. artificially modeled with electro-erosion (1mm hole performed
Secondly, by thoroughly analyzing the experiment results, a in outer race), displayed in Fig. 1.
series of interesting highlights, which could help improve the
sensitivity and reliability of bearing fault detection in actual Fig. 2 provides the stator a-phase current and its spectra
field, are revealed clearly. Finally, an improved stator current under the condition that the test motor is idle, and healthy. Fig.

This work is supported by National Natural Science Foundation of China

(50407016).

0197-2618/07/$25.00 © 2007 IEEE 2277

 0.06 0.12
3 provides that under the condition that the test motor is idle,
and with bearing outer race fault. 0.05 0.1

Here, f v ≈ 176.1 Hz, and f bearing ≈ 126.1, 226.1 Hz

0.04 0.08

Amplitude / A

Amplitude / A
0.03 0.06
for k=1.
0.02 0.04

0.01 0.02

0 0
90 100 110 120 130 140 150 160 190 200 210 220 230 240 250 260
Frequency / Hz Frequency / Hz

(c) (d)

Figure 3. Stator a-phase current and its spectra of faulty motor

IV. HIGHLIGHTS ON BEARING FAULT DETECTION

By thoroughly analyzing the experiment results, a series of
interesting highlights, which could help improve the sensitivity
and reliability of bearing fault detection in actual field, are
revealed, listed below in detail.
Figure 1. Modeled bearing fault (1) It is clear that once bearing fault occurred, some
harmonics would be induced in stator current, at the specific
3 3 frequencies given by Equation (1), by comparing Fig. 2(c) and
2 2.5
Fig. 3(c), Fig. 2(d) and Fig. 3(d). Thus, the stator current based
detection of bearing fault is indeed theoretically correct and has
Stator a-phase current / A

1 2
a certain practicability.
Amplitude / A

0 1.5
(2) Fig. 2(b, c, d) and Fig. 3(b, c, d) demonstrate that the
-1 1
f s component and its harmonic components in stator current
-2 0.5
severely interfere with the feature of bearing fault, i.e., the
-3 0
0 1 2 3 4 5
Time / s
6 7 8 9 10 0 50 100 150
Frequency / Hz
200 250
harmonics at the specific frequencies, f bearing given by
(a) (b) Equation (1). Therefore, those components should be regarded
0.06 0.12 as the noise and then filtered out while detecting bearing fault.
This can be achieved by using subdivision Fourier transform
0.05 0.1
and self-adaptive filter techniques [9, 10, 11].
0.04 0.08
Amplitude / A

Amplitude / A

(3) Fig. 3 shows clearly that the stator current spectrum

0.03 0.06
comprises a few other peaks besides those associated with
0.02 0.04
bearing fault, ascribing to the intrinsic asymmetry of the actual
0.01 0.02 motor. Therefore, it is necessary to estimate the concrete values
0
90 100 110 120 130 140 150 160
0
190 200 210 220 230 240 250 260
of f s and s in advance, and then purposely inquiry the
Frequency / Hz Frequency / Hz

(c) (d)
f bearing component information in the stator current spectrum,
and finally accomplish the bearing fault detection with
Figure 2. Stator a-phase current and its spectra of healthy motor sensitivity and reliability. A method, which blends subdivision
Fourier transform and rotor slot harmonics based slip
3 3
estimation techniques together, can realize the above objective
smoothly [11, 12].
2 2.5
Stator a-phase current / A

1 2
V. IMPROVED STATOR CURRENT BASED DETECTION
Amplitude / A

0 1.5 SCHEME
-1 1 According to the highlights revealed in Section IV, an
-2 0.5 improved stator current based detection scheme for bearing
-3 0
fault in induction motors, which comprehensively uses
0 1 2 3 4 5 6 7 8 9 10
Time / s
0 50 100 150
Frequency / Hz
200 250
subdivision Fourier transform, self-adaptive filter and rotor slot
(a) (b) harmonics based slip estimation techniques, is put forward.

2278
A. Subdivision Fourier Transform rotor slot harmonics frequency, P represents number of pole-
Subdivision Fourier transform (SFT) technique is of great pairs, and Z r represents number of rotor slot. Referred to [12]
significance with improving the sensitivity and reliability of for more information.
bearing fault detection in induction motors. The precise
frequency of the stator current fundamental component can be
P § f rsh ·
accurately determined by means of SFT, as well as the precise s = 1− ¨¨ ± ν ¸¸ , ν = 1, 3, 5, " (3)
amplitude and initial angle. Thus, the concrete value of the Zr © fs ¹
frequency f s is obtained with a satisfactory degree of
accuracy before bearing fault detection. Moreover, an Furthermore, the mechanical rotor speed in Hz, f rm in
appropriate noise signal, needed to self-adaptively filter out the
stator current, can be formed according to the precisely Equation (2), can be calculated by Equation (4).
determined frequency, amplitude and initial angle. fs
Referred to [9, 10, 11] for detailed information.
f rm = (1 − s ) (4)
P
B. Self-adaptive Filter Based Fault Indicator Extraction D. Detection Method
Similarly with [10, 11], the sensitivity and reliability of The novel detection method for bearing fault in induction
bearing fault detection can be improved, by self-adaptively motors, which perfectly blends subdivision Fourier transform,
filtering out the f s component and its harmonic components, self-adaptive filter and rotor slot harmonics based slip
estimation techniques, is outlined as the following.
and thus endowing the f bearing component with prominence in
(a) Sample instantaneously the stator current signal,
the stator current spectrum. Pay attention that the f bearing
identified as is .
component, i.e., the indicator of bearing fault, should be
extracted, while the f s component and its harmonic (b) Employ subdivision Fourier transform to determine the
component should be filtered out as the noise in this situation. frequency, f s , amplitude, and initial angle of the is
fundamental component, and form the noise signal to be used
The principle of self-adaptive filter technique is shown in
in the self-adaptive filter, uS .
Fig. 4. Here is , which is the stator current signal, consists of
the useful signal S T to be extracted and the noise signal nT to (c) Counteract the f s component of is by using self-
be counteracted. Another noise signal uS has the same adaptive filter according to f s and uS .
frequency as the noise signal nT . Surely, S T is the f bearing
(d) Employ subdivision Fourier transform to estimate the slip,
component and nT is the f s component, and its harmonic s , by analyzing the rotor slot harmonics of eT .
components respectively. By adjusting the filter parameters
self-adaptively, eT will approach S T under the meaning of (e) According to f s and s ˈ work out f bearing given by
Equation (1).
minimum square error [10]. Notice that the noise signal uS
can be obtained through subdivision Fourier transform above (f) Counteract those harmonic components of is whose
mentioned.
frequencies are close to f bearing by reiterating (b) and (c). The
i s = S T + nT
result is identified as eT′ .
+ eT
¦ − (g) Employ subdivision Fourier transform to obtain the
us Self-tuning Filter spectrum of eT′ .

Self-tuning Algorithm (h) Inquire the information of fault indicator, i.e., the f bearing

component in the spectrum of eT′ and determine whether
Figure 4. Scheme of self-adaptive filter
the fault is present or not.
C. Rotor Slot Harmonics Based Slip Estimation A great deal experiment results have demonstrated that the
While the motor is operating, rotor slot harmonics appear in presented detection method is with more sensitivity and
the stator current due to the interaction between the rotor slot reliability. Here cites an instance.
MMF and the fundamental air-gap flux. The slip, s , can be
Fig. 5 and Fig. 6 respectively provide the spectra of eT′
estimated according to Equation (3). Where, f rsh represents under the two different conditions that the idle motor is healthy

2279
and faulty with bearing outer race defect. The concrete results TABLE I. EXPERIMENT RESULTS UNDER THE TWO DIFFERENT
CONDITIONS THAT THE IDLE MOTOR IS HEALTHY AND FAULTY WITH BEARING
are listed in TABLE I. The other corresponding results have OUTER RACE DEFECT
been provided in Fig. 2 and Fig. 3.
Apparently, the outer race defect can be detected with Healthy Faulty
improved sensitivity and reliability.

0.16
fs (HZ) 50.00 50.03
0.14

0.12
s (%) 0.3 0.3
Amplitude / A

0.1

0.08

0.06

0.04
f bearing (HZ) 122.8 / 222.8 124.8 / 224.5
0.02

0
50 100 150 200 250
Frequency / Hz Amplitude (10-3A) 0.4 / 0.2 7.1 / 0.9
(a)

VI. CONCLUSIONS
0.025 4

3.5
This paper addresses itself to develop stator current based
0.02
3
detection method, which is with improved sensitivity and
reliability, of bearing fault in induction motors. Firstly,
Amplitude / 10 -3A
Amplitude / A

0.015 2.5

2
experiments of bearing fault are performed. Secondly, by
0.01 1.5
thoroughly analyzing the experiment results, a series of
1
interesting highlights, which could help improve the sensitivity
0.005
0.5
and reliability of bearing fault detection in actual field, are
0 0 revealed clearly. Finally, an improved stator current based
90 100 110 120 130 140 150 160 190 200 210 220 230 240 250 260
Frequency / Hz Frequency / Hz scheme to detect bearing fault, which perfectly blends
(b) (c) subdivision Fourier transform, self-adaptive filter and rotor slot
harmonics based slip estimation techniques together, is
Figure 5. Spectra of healthy motor with improved detection scheme
proposed in this paper. Fault detection instances in laboratory
demonstrate that the presented scheme is effective.

0.16 REFERENCES
0.14
[1] IAS Motor Reliability Working Group, “Report of large motor reliability
0.12
survey of industrial and commercial installations, Part I~II,” IEEE
Transactions on Industry Applications, vol. 21, No. 4, pp. 853-872, July-
Amplitude / A

0.1

0.08 August 1985.

0.06
[2] IAS Motor Reliability Working Group, “Report of large motor reliability
0.04 survey of industrial and commercial installations, Part III,” IEEE
0.02 Transactions on Industry Applications, vol. 23, No. 1, pp. 153-158,
0
January-February 1987.
50 100 150 200 250
Frequency / Hz
[3] Olav Vaag Thorsen, Magnus Dalva, “A Survey of faults on induction
(a) motors in offshore oil industry, petrochemical industry, gas terminals,
and oil refineries,” IEEE Transactions on Industry Applications, vol. 31,
No. 5, pp. 1186-1196, September-October 1995.
0.025 4
[4] Randy R. Schoen, Thomas G. Habetler, Farrukh Kamran, Robert G.
3.5 Bartheld, “Motor bearing damage detection using stator current
0.02
3 monitoring,” IEEE Transactions on Industry Applications, vol. 31, No. 6,
pp. 1274-1279, November-December 1995.
Amplitude / 10 -3A
Amplitude / A

2.5
0.015

2 [5] Caryn M. Riley, Brian K. Lin, Thomas G. Habetler, Randy R. Schoen,

0.01 1.5 “A method for sensorless on-line vibration monitoring of induction
1
machines,” IEEE Transactions on Industry Applications, vol. 34, No. 6,
0.005
0.5
pp. 1240-1245, November-December 1998.
0
90 100 110 120 130 140 150 160
0
190 200 210 220 230 240 250 260
[6] Ramzy R. Obaid, Thomas G. Habetler, Jason R. Stack, “Stator current
Frequency / Hz Frequency / Hz analysis for bearing damage detection in induction motors,” Symposium
on Dianostics for Electric Machines, Power Electronics and Drives, pp.
(b) (c)
182-187, Atlanta (USA), August 2003.
[7] T. Lindh, J. Ahola, J.-K. Kamarainen, V. Kyrki, J. Partanen, “Bearing
Figure 6. Spectra of faulty motor with improved detection scheme damage detection based on statistical discrimination of stator current,”

2280
 Symposium on Dianostics for Electric Machines, Power Electronics and squirrel cage induction motor with rotor defects,” Transactions of China
Drives, pp. 177-181, Atlanta (USA), August 2003. Electrotechnical Society, No. 4, pp. 176-179, November 1996.
[8] Michael J. Devaney, Levent Eren, “Detecting motor bearing faults,” [11] Xu Boqiang, Li Heming, Sun Liling, “Sensitive and reliable detection of
IEEE Instrumentation & Measurement Magzine, pp. 30-35, 50, broken rotor bar fault in induction motors,” 2004 IEEE Industry
December 2004. Applications Conference (39th IAS Annual Meeting), vol. 2, pp. 714-719,
Seattle (USA), October 2004.
[9] Liu Jin-ming, Ying Huai-qiao, “Fourier transform with FFT spectrum
continuous subdivision analysis,” Journal of Vibration Engineering, vol. [12] K. D. Hurst, T. G. Habetler, “Sensorless speed measurement using
8, No. 2, pp. 162-166, February 1995. current harmonic spectral estimation in induction machine drives,” IEEE
Transactions on Power Electronics, vol. 11, No. 1, pp. 66-73, January
[10] JIANG Jian-guo, WANG Qing-sheng, YANG Bing-shou, QIU A-rui, 1996.
“Applying the adaptive noise cancellation to extract the features of

2281