J Fail. Anal. and Preven.

(2019) 19:1144–1152
https://doi.org/10.1007/s11668-019-00712-z

TECHNICAL ARTICLE–PEER-REVIEWED

Fault Diagnosis of a Double-Row Spherical Roller Bearing

for Induction Motor Using Vibration Monitoring Technique
Devarabhotla Sai Chandra . Y. Seetharama Rao

Submitted: 4 September 2018 / Published online: 28 August 2019

Ó ASM International 2019

Abstract Vibration monitoring techniques employing Keywords Vibration monitoring 

sophisticated instruments are widely used in industries for Bearing element defects  Rolling element bearing 
regular inspection and diagnosis of the health condition of Fault identification  Vibration spectrum analysis
rotating machinery. This paper discusses fault diagnosis of
the spherical roller bearing for a 630-kW induction motor
in a local industry through vibration monitoring technique Introduction
using a fast Fourier transform analyzer. The main attraction
of this technique is that it can be performed even while the Rolling element bearings are the most precisely made
equipment is in normal working condition, thus saving devices; however, they could fail prematurely if disturbing
precious downtime and avoiding production loss. forces are large enough. Fortunately, such forces, or at least
Accelerometer sensors are kept on drive and non-drive their effects, are often visible through vibration. Therefore,
ends of the motor. Vibration data on motor housing are when a vibration analysis indicates symptoms of bearing
collected in the triaxial direction in the form of amplitude, problems, it can further be analyzed to find out the primary
spectrum and time waveform signals in full-load and no- cause of the trouble. In case of frequent premature bearing
load conditions. These signals are analyzed and faults failures, one must watch out for other difficulties, such as
identified from the vibration characteristics. From the data, unbalance and misalignment, especially after replacing the
it is found that the dominating amplitude peaks have bearing. Most of the inspections in industries are preferably
appeared in the axial direction at 3960 cpm which is done while the equipment is running.
matching with harmonic frequencies of motor RPM. The- It improves the effectiveness of inspection; on the other
oretical calculations were also performed to find the hand, it reduces the downtime which is a limiting factor in
bearing fault frequencies, and it is observed that the time-based preventive maintenance. Instruments which are
vibration readings are matching with sub-harmonics of used for inspection effectively help in tracking down some
bearing ball pass frequency of outer race. These investi- of the complex machinery problems. Vibration monitoring
gations and calculations help the maintenance personnel in is a reliable and time-saving technique and aids in assess-
locating bearing faults and also in detecting the exact cause ing the machinery condition. This technique uses a fast
that leads to bearing deterioration so that a suitable remedy Fourier transform (FFT) analyzer and is easily applied to
can be suggested, tested and implemented. rotating machinery. Before a vibration analysis is under-
taken, it is helpful and advisable to consider the machine
and the circumstances that lead to a requirement for
vibration analysis. Defects like unbalance, bends in shafts,
race wear, element pitting, bearing distortion, magnetic
D. S. Chandra (&)  Y. S. Rao pulls, etc., can be identified easily using this technique. The
Department of Mechanical Engineering, Gayatri Vidya Parishad
College of Engineering (Autonomous), Madhurawada, nature and magnitude of vibration of a particular item of
Vishakhapatnam, Andhra Pradesh, India machinery can provide valuable information about
e-mail: [email protected]

123
J Fail. Anal. and Preven. (2019) 19:1144–1152 1145

Fig. 1 Measuring points to

collect vibration data from
induction motor

machine condition. Vibration arises from cyclic excitation whether vibration levels are within acceptable limits at a
forces within the machine. These forces can be inherent in given speed. Siddiqui et al. [7] diagnosed bearing fault on
the design of the machine or can be due to the propagation squirrel cage induction motor in transient condition as well
of some defect. Rolling element bearing having flaws in the as steady-state condition by signal processing techniques,
raceways, rolling elements or cage will usually have a high like FFT by time domain as well as wavelet transform for
frequency of vibration. The defective bearing will not early fault detection. Nandi and Toliyat [8] demonstrated
generally cause a single discrete frequency of vibration, but the experiment on grinding spindle-type rotor-bearing
instead may generate several frequencies simultaneously. system and carried out tests to identify system character-
Fault diagnosis of rolling element bearings is performed by istics and operating range by using FFT, short-time Fourier
the use of signal processing methods. It unveils the various transformation (STFT) and Winger–Ville distribution
physical modulation mechanisms present in vibration (WVD). These techniques were applied by them in varying
response of rolling element bearings. speeds for vibration data of normal and shaft–cracked
Kankar et al. [1] investigated the feature recognition rotor. Many other researches [9–24] have contributed in
system for the detection of bearing faults by classifying this important area. Recently, the authors [25] have
those using soft computing techniques like artificial neural investigated induction motor and found the bearing dete-
networks (ANN) and support vector machine (SVM). The rioration and their likely causes through collecting data
performance of SVM is found to be better due to its from motor drive and non-drive ends. The purpose of the
inherent generalization capability. Wong et al. [2] studied present paper is to explore different issues of bearing
the vibration signals of roller element bearings in the time failure and further extend the investigation by theoretical
domain and resample with a random Bernoulli matrix to calculations. In order to validate the theoretical results of
emulate the compressed sensing mechanism and also bearing faults, vibration readings on base foot of motor are
demonstrated that it is possible to sample the vibration data taken for detecting the exact root cause of bearing failure
of roller element bearings at less than the Nyquist rate and (Fig. 1).
reconstruct the signal for fault classification. Amir et al. [3]
validated the experiment on roller bearing using SVM. Li
and Tao [4] investigated the rolling element bearing using Fault Diagnosis of Induction Motor
SVM classifier for initial classification. SVM classification
results show basic probability assignment (BPA). The Fast Fourier transform analyzer is useful to take readings,
experimental results indicated that since SVM method has record them and track the trend of the machine over a
high recognition accuracy, stability, and broad applicabil- period of time. An accelerometer is held on motor drive
ity, it can effectively identify roller bearing health status. and non-drive end bearing housing for diagnosing. The
Rafiee et al. [5] performed experiments in four different overall readings are taken in amplitude versus frequency in
speeds with two load intervals on bearings, by capturing the triaxial direction on both ends of the bearing housing.
vibration data and validation is composed of four different These are analyzed by the collected data depending on its
defect states of rolling element bearing—outer race defect, vibration characteristics, and the nature of the defect is
inner raceway defects, ball defect and the combination of identified. The recorded readings are dumped into a com-
the bearing elements defect—and one representing the puter where all data are stored by using CSI Master Trend
normal state. Saravanan et al. [6] conducted experiments software which will be helpful for comparing the present
on engine test bed in order to verify that it meets the par- data with past data trends and these are represented in the
ticular needs at various load accumulation. These are often form of a graph. By analyzing this graph, one can diagnose
carried out with troubleshooting in order to determine the mechanical condition of the motor. The overall and

123
1146 J Fail. Anal. and Preven. (2019) 19:1144–1152

Fig. 2 (a) Knurling pattern

defect on motor drive end
bearing outer race, (b) normal
wear on non-drive end outer
race

Table 1 Equipment details of induction motor Table 2 Overall vibration data in full-load condition in mm/s before
dismantling the motor for inspection
Equipment type Single-suction overhung rotor fan
Direction MNDE MDE
Pressure 800 mmwc
Impeller diameter 2600 mm Vertical 2.81 3.05
Motor type 3-phase induction motor Horizontal 1.71 2.72
Rating 630 kW Axial 3.63 5.49
Operating voltage 6600 V
Full-load current 69 A
Motor speed 1000 rpm Table 3 Overall vibration data in full-load condition in mm/s after
Handling medium Nitrogen replacing with new bearing for motor
Coupling type Pin bush
Direction MNDE MDE
Foundation type Concrete
Motor bearing Double-row spherical roller Vertical 0.57 1.5
bearing Horizontal 0.51 1.15
Axial 0.91 1.11

spectral information from the vibration amplitude of

 2 !
machine points is analyzed for finding the bearing defect in Bd Bd 2
the outer race. Bearing fault frequency calculations are also BSF ¼ S 1 cos ð;Þ
2Pd Pd
done to find at what exact harmonic bearing fault fre-
quencies were found (Fig. 2). where S is speed (revolutions per minute)—996, Bd ball or
As per ISO 2372, this equipment belongs to class-3 and roller diameter—28.72 mm, Nb number of balls or roll-
the maximum vibration limit for this equipment is 4.5 mm/ ers—18, Pd pitch diameter—217.39 mm, [ contact
s, and collected vibration readings should not exceed this angle—9.5°, FTF fundamental train frequency, BPFI ball
value (Table 1). pass frequency of inner ring, cpm, BPFO ball pass fre-
quency of outer race, cpm, BSF ball spin frequency, cpm.

Analytical Calculations for Bearing Fault Frequencies

Results and Discussion
Bearing fault frequencies calculations as per formula [26]
 
S Bd The exact fault detection can be analyzed by knowing
FFT ¼  1  cos / overall vibration data that are collected, and spectrums are
2 Pd
  analyzed for diagnosis of motor. The dominating amplitude
Nb Bd
BPFI ¼  S 1 þ cosð/Þ direction is helpful in detecting the root cause of bearing
2 Pd failure. Tables 2 and 3 show overall readings of motor
 
Nb Bd drive and non-drive end before and after replacement of
BPFO ¼  S 1  cosð/Þ bearing in full-load condition.
2 Pd

123
J Fail. Anal. and Preven. (2019) 19:1144–1152 1147

Fig. 3 MDE vertical spectrum before replacement of bearing

Fig. 4 MDE horizontal direction spectrum before replacement of bearing

Fig. 5 MDE axial spectrum before replacement of bearing

Analysis of Spectrum Before Replacement of Bearing Depending on the nature of vibration characteristics
shown in the spectrum, one can determine the type of
Figure 3 shows the amplitude of 3.05 mm/s at 960 cpm in defect in the machine and these spectrums are helpful in
the vertical direction, Fig. 4 shows the amplitude 2.72 mm/ finding the exact cause for bearing failure. By analyzing
s at 960 cpm in the axial direction, and Fig. 5 shows the this spectrum, it can be observed that the bearing defect is
amplitude 5.49 mm/s at 3960 cpm in the axial direction on on the outer race of motor drive end side.
motor drive end before replacement of motor bearing.
Figure 6 shows an amplitude of 2.81 mm/s at 960 cpm Analyzing the Spectrum of MDE and NDE Before
in the vertical direction in MNDE. Figure 7 shows an Bearing Replacement
amplitude of 1.71 mm/s at 960 cpm in the horizontal
direction in MNDE, and Fig. 8 shows an amplitude In the FFT spectrums of motor bearing, dominating peaks
3.63 mm/s at 960 cpm in the axial direction in MNDE appeared at 3974 cpm which is not matching with any
before replacement of motor bearing. These peaks show the harmonic frequencies of rotating rpm of the motor. Sub-
severity of the problem in the machine since, as per ISO harmonic frequencies of rotating motor rpm are also not
2364 the permissible value of amplitude is 4.5 mm/s. matching with 3974 cpm at any order. The above

123
1148 J Fail. Anal. and Preven. (2019) 19:1144–1152

Fig. 6 MNDE vertical spectrum before replacement of bearing

Fig. 7 MNDE horizontal spectrum before replacement of bearing

Fig. 8 MNDE axial spectrum before replacement of bearing

 
observations, conditions and predominant frequencies Nb Bd 18
appeared in the spectrums directing the analyst toward BPFI ¼  S 1 þ cosð/Þ ¼
2 Pd 2
suspicious of bearing condition. Based on the above  
28:72
spectrums, calculations were performed for determining  996 1 þ cosð9:5Þ ¼ 10132:017 cpm
217:39
bearing fault frequencies for the motor bearing and vibra-
tion readings on the base plate of the motor. 10132:017
Sub - harmonic is ¼ ¼ 5066 cpm ðEq 2Þ
Motor base plate vibrations readings are also high in the 2
 
vertical direction at DE and NDE base plates. These Nb Bd 18
readings indicate that the motor feet are not in a common BPFO ¼  S 1  cosð/Þ ¼
2 Pd 2
plane and there was a gap of 0.2 mm with the base. This  
28:72
can be corrected by adding shims of 0.2 mm thickness to  996 1  cosð9:5Þ ¼ 7795:98 cpm
217:39
fill the gap from base. This helps in correcting shaft
misalignment and distortion causing bearing misalignment. 7795:98
Sub - harmonic is ¼ 3897:99 cpm ðEq 3Þ
  2
S Bd 996 !
FFT ¼  1  cosð/Þ ¼  2
2 Pd 2 Bd Bd 2
  ðEq 1Þ BSF ¼ S 1 ðcos/Þ
28:72 2Pd Pd
 1 cosð9:5Þ ¼ 433:11 cpm
217:39 ¼ 3705:5

123
J Fail. Anal. and Preven. (2019) 19:1144–1152 1149

3705:5 From the spectrum in Fig. 11, it is seen that there is an

Sub - harmonic is ¼ ¼ 1852 cpm ðEq 4Þ
2 amplitude of 1.15 mm/s at 960 cpm in the horizontal
direction. From the spectrum in Fig. 12, it is seen that the
Using the numerical values for S, Bd, Pd, [ and Nb
amplitude is 1.11 mm/s at 960 cpm in the axial direction
shown in ‘‘Analytical Calculations for Bearing Fault
after replacement of bearing on motor drive end (Table 6).
Frequencies’’ section, one obtains the following results.
From the spectrums observed in Fig. 13, it is observed
From the calculations, bearing fault frequencies with
that the velocity amplitude is 0.57 mm/s at 960 cpm in the
different harmonics are shown in Tables 4 and 5. Side-
vertical direction in MNDE, in Fig. 14 the velocity
bands are also tabulated to find sidebands for different
amplitude is 0.51 mm/s at 960 cpm in the horizontal
harmonics from 0.5 9 to 5 9 (Fig. 9).
direction in MNDE, and in Fig. 15 the velocity amplitude
As per the above-calculated equations, (1)–(4) the pre-
is 0.91 mm/s at 960 cpm in the axial direction in MNDE
dominant frequency in the spectrums 3974 cpm is very
before replacement of motor bearing. These peaks show
close to the sub-harmonic of ball pass frequency of the
machine is running in healthy condition. The maximum
outer race (BPFO) of the bearing.
amplitudes are found in vertical and axial directions only,
whereas in the horizontal direction the amplitudes are with
Analysis of Spectrum After Replacement of Bearing
in acceptable limits as per ISO 2372 which specifies
maximum vibration limit 4.5 mm/s.
From the spectrum in Fig. 10, it is seen that there is an
amplitude 1.5 mm/s at 960 cpm on the vertical direction.

Table 4 Bearing frequencies with different harmonics

Train Spin Outer Inner

Shaft speed (FTF) (BSF) (BPFO) (BPFI)
Harmonics (RPM) (cpm) (cpm) (cpm) (cpm)

0.5 9 498 216 1852 3897 5066

19 996 433 3705 7794 10,132
1.5 9 1494 649 5558 11,693 15,198
29 1992 866 7411 15,591 20,264
2.5 9 2490 1082 9264 19,490 25,330
39 2988 1299 11,117 23,388 30,396
3.5 9 3486 1516 127,970 27,286 35,462
49 3984 1732 14,823 31,184 40,528
4.5 9 4482 1948 16,676 35,081 45,594
59 4980 2165 18,527 38,980 50,660 Fig. 9 Base plate vibration measurements for motor

Table 5 Sidebands for bearing frequencies for different harmonics

Side bands outer race pass (BPFO) Side bands inner race pass (BPFI)
Harmonics Mean Minimum Maximum Mean Minimum Maximum

0.5 9 3897 3399 4395 5066 4068 5564

19 7794 7296 8292 10,132 9634 10,630
1.5 9 11,693 11,195 12,191 15,198 14,700 15,696
29 15,591 15,093 16,089 20,264 19,766 20,762
2.5 9 19,490 18,992 19,988 25,330 24,832 25,828
39 23,388 22,890 23,886 30,396 29,400 31,392
3.5 9 27,286 26,788 27,784 35,462 34,466 36,458
49 31,184 30,686 31,682 40,528 39,532 41,524
4.5 9 35,081 34,583 35,579 45,594 45,096 46,092
59 38,980 38,482 39,478 50,660 50,162 41,158

123
1150 J Fail. Anal. and Preven. (2019) 19:1144–1152

Fig. 10 MDE vertical spectrum after replacement of bearing

Fig. 11 MDE horizontal spectrum after replacement of bearing

Fig. 12 MDE axial spectrum after replacement of bearing

Table 6 Comparison before and after replacement of bearing to motor

End Direction Before (mm/s) After (mm/s) Reduction (%)

Motor drive Vertical 3.05 1.5 50.81

Horizontal 2.72 1.15 57.70
Axial 5.49 1.11 79.70
Motor non-drive Vertical 2.81 0.57 79.71
Horizontal 1.71 0.51 70.17
Axial 3.63 0.91 74.90

From the above comparison graphs, in Fig. 16 it is side. From these graphs, the velocity amplitude peaks were
observed that reduction in velocity amplitude is from 3.05 greatly reduced after replacement with new bearings on
to 1.5 mm/s on the vertical direction and from Fig. 17 it is motor drive end vertical and axial directions, clearly indi-
observed that the reduction in velocity amplitude is from cating that motor is running in a healthy state.
5.49 to 1.11 mm/s in the axial direction on motor drive end

123
J Fail. Anal. and Preven. (2019) 19:1144–1152 1151

Fig. 13 MNDE vertical spectrum after replacement of bearing

Fig. 14 MNDE horizontal spectrum after replacement of bearing

Fig. 15 MNDE axial spectrum after replacement of bearing

Fig. 16 Reduction in velocity amplitude on MDE vertical direction

Fig. 17 Reduction in velocity amplitude on MDE axial direction

123
1152 J Fail. Anal. and Preven. (2019) 19:1144–1152

Conclusions PSVM extracted using Morlet wavelet for fault diagnosis. Expert
Syst. Appl. 35, 1351–1366 (2008)
7. K.M. Siddiqui, K. Sahay, V.K. Giri, Modeling and detection of
Early failure of bearing after taking into the operation is bearing fault in PWM inverter fed induction motor drives. J.
not the problem of bearing, but some external forces are Electr. Eng. 8(1), 11–24 (2014)
responsible for its failure. Early and efficient fault detec- 8. S. Nandi, H.A. Toliyat, Condition monitoring and fault diagnosis
tion is very important in process industries, and failure of of electrical machine, in Proceedings of IEEE IAS Annual
Meeting, vol 1 (1999), pp. 197–204
any machine leads to huge production loss. Vibration 9. A. Boudiaf, A. Djebala, H. Bendjma, A. Balaska, A. Dahane, A
analysis is the versatile tool in the identification of bearing summary of vibration analysis techniques for fault detection and
failures at an early stage and also helps in identification of diagnosis in bearing, in 8th International Conference on Model-
root cause like misalignment in the system, misalignment ling, Identification and Control, Alglers (2016)
10. P.K. Gupta, Dynamics of rolling element bearings (part I–part
of bearing with its shaft or in the housing, unbalance in the IV): cylindrical roller bearing analysis, ball bearing analysis, ball
rotors, and resonant conditions in the surrounding sup- bearing analysis, ball Bearing results. ASME J. Lubr. 101, 293–
porting structures. 326 (1979)
11. P.K. Gupta, Advanced Dynamics of Rolling Element Bearings
1. From the above calculation, it is observed and found (Springer, Berlin, 1984)
that the predominant frequency in the spectrum, i.e., 12. C.T. Walters, Dynamics of ball bearing. ASME J. Lubr. Technol.
93(1), 1–10 (1971)
3960 cpm, is very close to sub-harmonic of ball pass
13. T.A. Harris, M.H. Mindel, Rolling element bearing dynamics.
frequency of the outer race (BPFO), i.e., 3987 cpm, of Wear 23(3), 311–337 (1973)
a double-row spherical roller bearing. 14. E.A. Boesiger, A.D. Donley, S. Loewenthal, An analytical and
2. Shaft misalignment is the primary reason for bearing experimental investigation of ball bearing retainer instabilities.
ASME Trans. Tribol. 114, 530–539 (1992)
failure, and shaft-to-shaft misalignment was corrected
15. R. Tomovic, V. Miltenovic, M. Banic, M. Aleksandar, Vibration
by laser kit and attained zero-to-zero accuracy. response of rigid rotor in unloaded rolling element bearing. Int. J.
3. The vibration readings taken on the base plate of motor Mech. Sci. 52, 1176–1185 (2010)
found it is another reason for the failure of bearing and 16. P.Y. Kim, A review of rolling element bearing health monitoring
(II): preliminary test results on current technologies, in Machin-
that vibration was eliminated by tightening bolts and
ery Vibration Monitoring and Analysis Meeting (Vibration
soft foot is done by adding shims having thickness of Institute, New Orleans, 1984), pp. 127–137
0.2 mm at the base to avoid clearance that leads to 17. N. Tandon, A comparison of some vibration parameters for the
vibration in a motor. condition monitoring of rolling element bearings. Measurement
12, 285–289 (1994)
4. The vibration amplitudes were greatly reduced from
18. P.D. McFadden, J.D. Smith, Model for the vibration produced by
5.49 to 1.11 mm/s in the axial direction of the motor a single point defect in a rolling element bearing. J. Sound Vib.
drive end bearing after replacing with the new bearing. 96(1), 69–82 (1984)
5. The temperatures readings also came within the limits 19. P.D. McFadden, J.D. Smith, The vibration produced by multiple
point defects in a rolling element bearing. J Sound Vib. 98(2),
on shafts 50° and on bearing 53° centigrade on motor
263–273 (1985)
drive end bearing. 20. Y.T. Su, S.J. Lin, On initial fault detection of a tapered roller
bearing: frequency domain analysis. J. Sound Vib. 155(1), 75–84
(1992)
21. Y.T. Su, M.H. Lin, M.S. Lee, The effects of surface irregularities
References on roller bearing vibrations. J. Sound Vib. 163(3), 455–466
(1993)
1. P.K. Kankar, S. Sharma, S.P. Harsha, Fault diagnosis of rolling 22. A. Choudhury, N. Tandon, Vibration response of rolling element
element bearing using cyclic autocorrelation and wavelet trans- bearings in a rotor bearing system to a local defect under radial
form. Neurocomputing 110, 9–17 (2013) load. ASME Trans. Tribol. 128, 252–261 (2006)
2. M.L.D Wong, M. Zhang, A.K. Nandi, Health status identification 23. H. Arslan, N. Aktürk, An investigation of rolling element
of rolling bearing based on SVM and improved evidence theory, vibrations caused by local defects. ASME Trans. Tribol. 130,
in 23rd European Signal Processing Conference (2015), pp. 041101-1-9 (2008)
2256–2260 24. M.S. Patil, J. Mathew, P.K. Rajendra Kumar, S. Desai, A theo-
3. R.B. Amir, S.T. Gul, A.Q. Khan, A comparative analysis of retical model to predict the effect of localized defect on
classical and one class SVM classifiers for machine fault detec- vibrations associated with ball bearing. Int. J. Mech. Sci. 52,
tion using vibration signals (IEEE, 2016) 1193–1201 (2010)
4. M. Li, Z. Tao, Health status identification of rolling bearing based 25. S.C. Devarabhotla, Y.S. Rao, M. Sambaiah, Experimental
on SVM and improved evidence theory (IEEE, 2016), pp. 378– investigation and fault diagnosis through CBM techniques for
382 anti-friction bearing of Mill Fan. Int. J. Mech. Prod. Eng. Res.
5. J. Rafiee, P.W. Tse, A. Harifi, M.H. Sadeghi, A novel technique Dev. (IJMPERD) 8, 803–812 (2018)
for selecting mother wavelet function using an intelligent fault 26. NTN bearing frequencies NTN americans.com
diagnosis system. Expert Syst. Appl. 36, 4862–4875 (2009)
6. N. Saravanan, V.N.S. Kumar Siddabattuni, K.I. Ramachandran, Publisher’s Note Springer Nature remains neutral with regard to
A comparative study on the classification of features by SVM and jurisdictional claims in published maps and institutional affiliations.

123