Journal of Sound and Vibration 364 (2016) 6776

Contents lists available at ScienceDirect

Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi

Vibration response mechanism of faulty outer race rolling

element bearings for quantitative analysis
Lingli Cui a, Yu Zhang a, Feibin Zhang b,n, Jianyu Zhang a, Seungchul Lee c
a
Beijing Engineering Research Center of Precision Measurement Technology and Instruments ,Beijing University of Technology, Chao Yang
District Beijing 100124, China
b
Engineering College of Jiangxi Agricultural University, Jiangxi, China
c
Department of Human and Systems Engineering, Ulsan National Institute of Science and Technology, 50 UNIST-gil, Ulsan 44919, Republic
of Korea

a r t i c l e i n f o abstract

Article history: For the quantitative fault diagnosis of rolling element bearings, a nonlinear vibration
Received 14 November 2014 model for fault severity assessment of rolling element bearings is established in this study.
Received in revised form The outer race defect size parameter is introduced into the dynamic model, and vibration
26 September 2015
response signals of rolling element bearings under different fault sizes are simulated. The
Accepted 17 October 2015
signals are analyzed quantitatively to observe the relationship between vibration
Handling Editor: I. Trendalova
Available online 8 December 2015 responses and fault sizes. The impact points when the ball rolls onto and away from the
defect are identied from the vibration response signals. Next, the impact characteristic
Keywords: that reects the fault severity in rolling element bearings is obtained from the time
Rolling element
interval between two impact points. When the width of the bearing fault is small, the
bearings
signals are presented as clear single impact. The signals gradually become double impacts
vibration response
quantitative analysis with increasing size of defects. The vibration signals of a rolling element bearings test rig
fault diagnosis are measured for different outer race fault sizes. The experimental results agree well with
the results from simulations. These results are useful for understanding the vibration
response mechanism of rolling element bearings under various degrees of fault severity.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction

As a type of widely used universal component in the mechanical equipment, rolling element bearings directly affect the
dynamic performance, running accuracy, reliability and service life of the whole equipment. Hence, it is necessary to
diagnose and detect the faults in rolling bearings, to prevent early failure. The fault development of rolling element bearings
is usually a dynamic process and the fault degree gradually increases from a low level. The early damage detection is in
demand not only for high precision machines but also for general engineering equipment because minor faults often can
lead to large faults or catastrophic results. However, the rolling element bearings should be replaced when the fault reaches
a certain degree. Therefore, the service life of mechanical equipment can be extended and production costs can be lowered
through the quantitative diagnosis of fault degree in rolling element bearings.

n
Corresponding author at: Engineering College of Jiangxi Agricultural University, 330045, China; Beijing Engineering Research Center of Precision
Technology and Instruments, Beijing University of Technology, Beijing,100124, China.
E-mail addresses: acuilingli@163.com (L. Cui), yuzh_2014@163.com (Y. Zhang), zhfb_2014@163.com (F. Zhang), zhjy_1999@bjut.edu.cn (J. Zhang),
seunglee@unist.ac.kr (S. Lee).

http://dx.doi.org/10.1016/j.jsv.2015.10.015
0022-460X/& 2015 Elsevier Ltd. All rights reserved.
68 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776

With the development and improvement of the fault diagnosis technology for rolling element bearings, research on the
fault diagnosis of rolling bearings has attracted considerable attention, and various new methods are being proposed [1].
Sopanen [2,3] found that the diameter clearance has a signicant effect on the level of vibration as well as on the natural
frequencies using a dynamic model of a deep-groove ball bearing. Sassi [4] developed a new application called bearing
toolbox to simulate the vibratory response of bearings to the excitations produced by localized defects. Zimroz [5] developed
a technique based on adaptive schur lter for bearing local damage detection. The method is well-suited to analyze the non-
stationary signals and it is potentially to use in bearing and gearbox monitoring. Zimroz et al. [6] also proposed a novel
method of timefrequency map enhancement and further processing for feature extraction in local damage detection. It can
clearly detect the local damage. But detection of cyclic impulses related to damage is not possible for that original raw
vibration signal contains strong non-informative contribution. Tadina [7] improved a comprehensive model of a ball bearing
to obtain the vibration response due to different sizes of localized defects. Petersen [8] developed a method for calculating
and analyzing the quasi-static load distribution and varying stiffness of a radially loaded double row bearing with a raceway
defect of varying depth, length, and surface roughness. Moazen [9] proposed a bearing model that takes the nite rolling
element size into consideration. While, some progress in the quantitative faults diagnose of bearing has been achieved by
the researchers in China and other countries. Antoni et al. [10] proposed a method of modulation intensity distribution that
can help to estimate the size of bearing fault, while it requires a relative long times in the diagnose process. Dowling [11]
revealed a phenomenon similar to double impacts in fault signals of rolling bearings owing to phase changes. In the fol-
lowing year, Epps [12] discussed the phenomenon in more detail. They noted that the double impacts occurred when the
ball rolled in and out of the fault. The rst impact is considered the step response of low-frequency components, and the
second one is the impact response of high-frequency components. Sawalhi [13] used the method of minimum entropy
deconvolution (MED) and spectrum kurtosis (SK) to analyze the experimental signals of bearing faults with considerable
noise. The results showed that the double impact phenomenon can be clearly attributed to fault signals of the outer ring of
the bearing after denoising treatment. Randall [14] used the minimum entropy deconvolution method to separate the
impulses from entry into, and exit process from an individual fault, which give information on the fault size. Then they nd
that the entry procedure of the damage zone is step signal but the exit is impulse. Baydar and Ball [15] examined whether
acoustic signals could be used effectively to detect the various local faults in gearboxes using the smoothed pseudo-Wigner
Ville distribution with three types of progressing local faults. Result indicated that acoustic signals are very effective for the
early detection of faults. Recently, Jiang [16] used signal complexity and morphological ltering method to analyze the fault
signals of rolling bearing quantitatively. Shen [17] adopted a diagnostic method based on Support Vector Regression (SVR) to
diagnose bearing faults quantitatively. A bearing fault severity measurement method based on the LempelZiv complexity
and the continuous wavelet transform was proposed by Hong. The result indicated that the LempelZiv complexity was
proportional (for outer race faults) or inversely proportional (for inner race faults) to fault size (severity) for all rotational
speeds [18]. Several timefrequency analysis methods are compared and analyzed in the quantitative diagnosis of outer race
fault based on the LempelZiv complexity [19]. Zhao [20] combined the approximate entropy theory with the Empirical
Mode Decomposition (EMD) algorithm. The method to distinguish between the step components and impact components
was realized through empirical mode recombination of the components. The existence of double impact phenomenon was
further veried, and it was feasible to realize quantitative diagnosis through separation methods. Kong [21] extracted the
double impacts from bearing faults by using Ensemble Empirical Mode Decomposition (EEMD) and complex morlet wavelet
method. Wang [22] proposed a new adaptive dictionary based on matching pursuit algorithms to construct the atoms with
double impact components. Good results were obtained in the fault diagnosis using experimental and engineering signals,
with research on algorithms constituting the majority of these.
However, faults in rolling bearings still need to be quantitatively analyzed from the aspect of bearing defective
mechanism. Randall et al created a dynamic model of a gearbox that compared to lumped parameter models and reduced
nite element method. And the dynamic model was used to simulate the vibration signals in the presence of local inner and
outer race faults. While this paper just highlights the plausibility of fault simulation in machine condition monitoring [23].
Patil developed a mathematical model for the ball bearing vibrations due to defect on the bearing race. With this model the
effect of the defect size and its position have been simulated and also the spectral components have been predicted. But the
prediction of the actual amplitudes of vibration is not possible by the model [24]. Ahmadi established a double-row
spherical roller bearing model and applied it to the quantitative performance analysis of bearings and investigated and
explained the path of a rolling element in the defect zone and the nature of the entry and exit events of the two main
features that appear in the vibration signal of a defective bearing [9,25]. Singh analyzed the contact force of the rolling
element entry and exit of the defect with the explicit nite element method. The nite element model presented in the
paper can be used to investigate the vibration characteristics of bearings with more complex defect geometries [26]. To
diagnosis the bearing fault quantitatively, not only the dynamic model should be developed but also the experiment signal
have to be analyzed.
In this paper, we analyzed the bearing fault of outer race quantitatively from the aspect of vibration mechanism. First, we
establish the vibration model of outer race faults in rolling element bearings, and next, the vibration response of rolling
element bearings under different fault severity is quantitatively analyzed in this paper. The scale of outer race faults is
changed to simulate the vibration response for different defect sizes. The impact characteristics when the ball rolls in and
out of the defect with different fault sizes are analyzed. Furthermore, the time interval between the two impacts is extracted
from the simulation signals. Next, the results are compared with the theoretical calculated values to verify the correctness of
 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776 69

Fig. 1. Dynamic model of rolling element bearings.

the algorithm. The experimental results agree well with the result of the simulation analysis and are more accurate than
those from earlier research. In our study, we nd that the bearing damages with different sizes share different impact
characteristic. When the fault size is relative small, the vibration signals performances as a single impact. While it double
impact when the damage size is relatively large.

2. Dynamic model of normal rolling element bearings

Based on the previous studies [13], a unit resonator was introduced to establish an efcient nonlinear vibration model for
faulty rolling bearing, as shown as Fig. 1. A 5 degrees-of-freedom (dof) nonlinear vibration model of the bearing is proposed
which consists of 4-dof inner race and outer race in the horizontal and vertical directions and a 1-dof unit resonator in
vertical direction. The model was established based on the following assumptions: rst, lumped mass method is introduced
here for every component, and the moment of inertia and effect of surface waviness of the bearing raceway are ignored;
second, it is assumed that there does not exist other geometric errors on the moving contact surface aside from partial
damage and elastic contact between the rolling element and raceway satisfying Hertz theory; nally, all damps are linearly
viscous.
Based on the relevant knowledge of kinematics and dynamics, dynamic differential equations are as follows:
8
> ms xs cs x_ s ks xs f x 0
>
>
>
> m y cs y_ s ks ys f y F r
>
< s s
mp x p cp x_ p kp xp  f x 0 (1)
>
>

> mp y p cp cr y_ p kp kr yp  kr yb cr y_ b f y 0
>
>
>
: m y c y_  y_ k y y 0
r b r b p r b p

where ms , ks , cs , mp , kp , cp , mr , kr , and cr respectively represent the mass, stiffness and damping of inner race, outer race and
unit resonators; f x , f y , and F r respectively represent the nonlinear contact force of the bearing in the vertical direction and
horizontal direction and the vertical force applied to the inner race of the bearing; xs ,ys , xp , yp , and yb respectively represent
the two inner race dof, two outer race dof and measured vibration response.

3. Model of rolling element bearing with local outer race fault

In the normal rolling element bearing system, the inner race is xed with and rotates together with the rotating shaft,
whereas the outer race is usually fastened to the bearing pedestal, and the balls undergo pure rolling movement on the
raceway. During the operation of the rolling bearing, the raceway is divided into a bearing zone and non-bearing zone based
on the range of radial load. The balls that enter the bearing zone are deformed, producing varying exibility vibration.
70 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776

The total deformation of the jth ball is a function of the relative displacement between inner- and outer races as well as
the angular position of the jth ball and bearing clearance c, as expressed by the following equation:

  
j xs xp cos j ys  yp sin j  c (2)

where j 1; 2; nb , nb is the number of balls. The effects of elasto-hydrodynamic lubrication (EHL) can be obtained in the
current model by setting the clearance value c in Eq. (2) to a negative value [13,27]. It is assumed that the effect of the EHL
on the rolling elements damping is small and negligible [28].
The ball is deformed only when the ball is located in the bearing zone, and thus contact force is produced. Therefore, a
switch function j is introduced as follows:
(
1 j 4 0
j (3)
0 otherwise

The angular position of the jth ball j is a function of the period dt, the initial angular position of the bearing cage 0 and
the angular speed of the bearing cage c . j and c are dened as follows:
2 j 1
j c dt 0 (4)
nb
 
Db s
c 1  (5)
Dp 2

where s is the angular speed of the shaft, Db is the diameter of ball, and Dp is the pitch diameter of the bearing.
From Hertz contact theory, the contact force between the jth ball and the raceway is given by

f j kb j
1:5
(6)

According to the equations above, the total nonlinear contact forces of the bearing in the x- and y-directions are,
respectively,
X
f x kb j 1:5
j cos j (7)

X
f y kb j 1:5
j sin j (8)

where kb is the stiffness of the load contact.

When the local fault occurs on the outer race, the depth of the local fault is dened as cd , angle spanned by the fault d ,
the width of the fault L and angular position of the fault d . Fig. 2 shows a diagram of the model of the outer race fault and
some of the geometric relations.
 
d
L D sin (9)
2

4 2

d
5 1

d
D
6 8

cd L

Fig. 2. Spall denition on the outer race.

 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776 71

When the ball moves over the site of the local fault, the ball releases deformation. At this time, the deformation from the
jth rolling element entering the site of the fault is expressed as

  
j xs  xp cos j ys  yp sin j  c  j cd (10)

Switch function value j in Eq. (9) leads to

(
1 d oj o d d
j (11)
0 otherwise

The nonlinear contact force with the fault is calculated and substituted into dynamic Eq. (1) of the bearing vibration
model. The solution of the differential equation is the vibration response of the faulty rolling bearing.

4. Quantitative analysis of outer race fault of rolling element bearings

The rolling element bearing 6307 is taken used for the current simulation. Bearing parameters and fault parameters are
shown in Table 1. The widths L of outer race faults are set at 0.5 mm, 2 mm, 3.5 mm and 5 mm for each simulation. Four
response signals when the ball rolls over the defects are shown in Fig. 3.

Table 1
Geometrical and physical properties used for the rolling-element bearing 6307.

Number of rolling elements, Nb 8

Diameter of rolling element, Db 14 mm
Pitch diameter, Db 57.5 mm
Contact angle, 0
Radial clearance, c 5 m
contact stiffness kb 1.8978  1010 N/m
Stiffness of outer race kp 15.1056  106 N/m
Stiffness of the shaft ks 4.241  104 N/m
Damping of the outer race cp 2210.7 N s/m
Damping of the shaft cs 1376.8 N s/m
Stiffness of unit resonators kr 8.8826  109 N/m
Damping of unit resonators cr 9424.8 N s/m
Mass of unit resonators, mr 1 kg
Mass of the inner race, ms 1.2638 kg
Mass of the outer race, mp 12.638 kg

150 0.01325
600 0.01325
100
400
a(m/s )

a(m/s )
2

50
200
0
0
-50
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
t(s) t(s)

0.01325 400 0.01325

400
300
300
a(m/s )

a(m/s )

200
2

200
100 100
0 0
-100 -100
0 0.01 0.02 0.03 0.04 0.05 0.01 0.02 0.03 0.04 0.05
t(s) t(s)
Fig. 3. Simulation response signals under different fault scales: (a) L 0.5 mm; (b) L 2 mm; (c) L 3.5 mm; and (d) L 5 mm.
72 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776

After observing and comparing each response signal in Fig. 3, it can be observed that a series of step impacts are trig-
gered when the ball rolls over the defect. When the defect is small, the impact is presented as a series of oscillatory decays of
a single impact with a xed period To 0.01325 s (fo 75.45 Hz), which agrees with the theoretical value of the characteristic
period of the outer race fault. As the defect size increases, the double step impact phenomenon occurs. The time interval
between the two step impacts increases as the defect size increases. The time interval between two adjacent impacts is still
the characteristic period of the outer race fault.
The second impacts of each group of response signals in Fig. 3 are selected and magnied as shown in Fig. 4. To analyze
the effect of the width of the fault, the rst and second step points on the impact curve are marked as A and B, respectively.
The rst maximum value point is marked as C. Point A occurs at x 0.013257 s. The time when the sixth ball begins to
contact the defect can be calculated by the initial position and rotating speed of the inner ring set during the simulation.
2
t 0:013254s (12)
c nb
The error relative to the abscissa value of point A is 0.023 percent. Hence it can be determined that point A is the critical
point when the ball rolls into the defect area.
During the rotating process of the rolling bearing, the ball rotates around the shaft together with the cage. Hence, the
angular velocity when the ball rolls over the defect is equal to the angular rotation speed c of the cage. Therefore, the
elapsed time between the moments when the ball rolls in and out of the defect can be expressed as follows:
d
T (13)
c
The time interval between sections AB and BC in the response signals under different fault widths (Fig. 4) as well as the
time required for the ball to roll over the defect (calculated by Eq. (13)) are listed in Table 2.
It can be observed from the data in Table 2 that the result calculated from section AB is consistent with that by the
formula; the error is smaller than 0.3 percent. However, the error found when comparing the result calculated from section
BC with that formula is too large, and the result is obviously not accurate. It can be veried that B is the critical point at
which the ball rolls out of the defect.

600 B
100 B C
400
a(m/s )
a(m/s )

50
2

200
0 C
0
-50 A A
0.012 0.014 0.016 0.018 0.02 0.022 0.024 0.012 0.014 0.016 0.018 0.02 0.022 0.024
t(s) t(s)

400 B B
300
200
a(m/s )

a(m/s )

200
2

C 100 C
0 0
A -100 A
0.012 0.014 0.016 0.018 0.02 0.022 0.024 0.012 0.014 0.016 0.018 0.02 0.022 0.024
t(s) t(s)
Fig. 4. Magnied single responses under different fault scales: (a) L 0.5 mm; (b) L 2 mm; (c) L 3.5 mm; and (d) L 5 mm.

Table 2
Comparison between AB and BC with theoretical results.

Defect scales (mm) Theoretical calculation results (s) Simulation results

AB (s) Error (%) BC (s) Error (%)

0.2 0.0000944 0.0000942 0.212 0.001119 1085

0.4 0.000189 0.000188 0.529 0.001001 429
0.5 0.000236 0.000235 0.423 0.001030 336
2 0.000944 0.000942 0.212 0.001855 96.6
3.5 0.001652 0.001649 0.181 0.000824 50.1
5 0.002360 0.002356 0.169 0.001531 35.1
 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776 73

L=0.5mm
C

100
B

50

a(m/s )
2
0

-50
A

0.0128 0.013 0.0132 0.0134 0.0136 0.0138 0.014 0.0142 0.0144 0.0146
t(s)

Fig. 5. Single impulse under minor fault condition.

L=5mm
400 B

300

200
a(m/s )
2

100 C

0
A
-100
0.012 0.014 0.016 0.018 0.02 0.022 0.024
t(s)

Fig. 6. Double impulses under large fault scale.

When L0.5 mm, the step impact signal is locally magnied, as shown in Fig. 5. It can be observed that the maximum
value point of impact signal is not at B but C, which is located on the right side of B. The reason can be analyzed by
combining the dynamic equation of the system with the fault model. When the width of the fault is small, the step of
nonlinear contact force on the bearing decreases when the ball just rolls onto the defect area. The vibration acceleration
signal of the system triggers the downward step impact (with the starting point A). The step of nonlinear contact force on
the bearing increases when the ball just rolls away from the defect. The upward step impact (with the starting point B) is
triggered. Because the width of the defect is small, point B is located in the rst rising stage of the rst decay signal of the
step impact. The two step impacts gain and overlap in the same vibration direction. Next, the acceleration amplitude
continues to increase until it reaches the peak point C.
When L 5 mm, the step impact signal is locally magnied, as shown in Fig. 6. The time interval between step points A
and B is large. In other words, when the ball rolls out of the defect area to trigger the step impact, the impact generated
when the ball rolled into the defect area has already been seriously attenuated. Hence the signals show signicant oscil-
latory decay typical of double impacts. In references [20,29] point C is often taken as the roll-in location incorrectly. That is,
the time interval of section CB in Fig. 6 was often selected to analyze and determine the size of the fault. However, the
calculated result using section AB is more accurate than that using section CB according to Table 2. It can also be observed
from the error analysis in Table 1 that the result with A as the starting point is more accurate than that with C as the
starting point.
In summary, a step impact will be triggered when the ball rolls in and out of the defect, but the directions of the impacts
are different. When the defect is small, the decay of the rst impact is submerged in that of the second one; the signal
resembles a single impact. When the defect is large, the time interval between the rst impact and the second impact is
long. The rst one is not submerged, and the signal indicates double impacts.

5. Quantitative experimental verication of outer ring defects in rolling bearings

To verify the accuracy of the simulation model of faulty outer race rolling element bearings more effectively, the 6307
bearing defect test rig was used to obtain experimental data in the laboratory. The bearing test rig (Fig. 7) has a three-phase
asynchronous motor that is connected to a spindle equipped with a rotor through a exible coupling . The spindle is
supported by two 6307 bearings; is a normal bearing; and is a bearing with different single-point pitting faults. The
74 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776

Fig. 7. Bearing test rig.

0.01363 10 0.01328
5
5

a(m/s )
a(m/s )

2
2

0 0

-5
-5
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
t(s) t(s)

40 0.01302 0.01296
40
20
20
a(m/s )

a(m/ )
2

0 0

-20 -20
-40
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
t(s) t(s)
Fig. 8. Experimental vibration signals under different fault scales: (a) L 0.5 mm; (b) L 2 mm; (c) L 3.5 mm; and (d) L 5 mm;

5 20 B

C
a(m/s )
2

a(m/s )
2

0 0
A

-5 -20
0.258 0.26 0.262 0.264 0.266 0.268 0.27 0.466 0.468 0.47 0.472
t(s) t(s)

B B
50
C 50
a(m/s )

C
a(m/s )

2
2

0 0
A A
-50 -50
0.08 0.082 0.084 0.086 0.088 0.08 0.082 0.084 0.086 0.088
t(s) t(s)
Fig. 9. Magnied experimental vibration signals under different fault scales: (a) L 0.5 mm; (b) L 2 mm; (c) L 3.5 mm; (d) L 5 mm.
 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776 75

Table 3
Comparison between simulation results and experiment results.

Defect scales (mm) Simulation signals (s) Experiment signals (s) Errors (%)

0.5 0.00029452
2 0.00094248 0.00097656 3.6
3.5 0.00164935 0.00175781 6.5
5 0.00235622 0.00208333 11

motor speed is 1500 rev/min, the larger diameter of the bearing D 80 mm, the smaller diameter of the bearing d 35 mm,
the number of rolling elements Z 8, and the contact angle is 0.
The widths of bearing faults were set at 0.5 mm, 2 mm, 3.5 mm and 5 mm, respectively. The experimental system is
composed of the bearing experiment table, an HG3528A data acquisition unit acceleration sensors and a laptop. The
experiment was performed and vibration signals were collected. In this system, the sampling frequency was 15,360 Hz and
the number of sampling points was 8192. The experimental signals are shown in Fig. 8.
The periodicity and shock characteristics are not as signicant as the simulation signal because of the noise. The fault
period in the time domain vibration signal can hardly be found especially when the fault size is small.
After amplifying the signal in the time domain as shown in Fig. 8, clear double impact phenomenon can be observed in
the 5 mm fault vibration signal. It can be observed that the features are similar to the simulated signal for a 5-mm width
fault in Fig. 4(d). The rst step impact corresponds to the ball on the fault location point, and the second step impact
corresponds to the critical point when the ball left the fault location.When the fault was small, it was difcult to obtain
similar features owing to the noise.
It can be observed from Fig. 9 that when the width of the bearing fault is small, the signals appear as clear single impact.
When the width of the fault is large, clear double impact phenomenon can be observed. The signals are subjected to
quantitative diagnosis according to the introduced method, and they are compared to the simulation results, as shown in
Table 3. It can be observed from the table that the error between the experimental signal and simulation signal is within an
acceptable range. Hence the accuracy of the quantitative mechanism analysis and quantitative fault diagnosis method for
the bearing is veried.

6. Conclusions

Rolling element bearing vibration response signals for different outer race fault severity are simulated to analyze the
relationship between vibration signal characteristics and the fault size through a nonlinear rolling bearing vibration model.
The analyses in this paper have provided the theoretical basis for the quantitative diagnosis of rolling bearing faults. The
conclusions are obtained as follows:
Theoretical analysis and simulation results show that step impacts will be excited when the ball enters or exits the defect area,
respectively. The vibration response of outer race faulty rolling element bearings is a series of impact signals. When the width of the
bearing fault is small, the signals are presented as clear single impact. The signals gradually become double impacts with the
increasing size of defects, but the time interval between peak values of double impacts in the same direction is not equal to the
elapsed time of the ball rolling over the defect because the starting directions of the two impacts are opposite.
In the analysis of the experimental data, we found that the energy of the large scale fault impact signal is considerably
larger and may be free from the inuence of laboratory noise. Thus, it can be used to diagnose the size of the defect based on
the time interval of the double impacts from the original experimental signals directly. When the width of the fault is small
and the resulting impact signal energy is severely reduced, it is difcult to extract the quantitative information through the
original experimental signals. Effective denoising algorithms and diagnosis methods to extract early fault quantitative
features are worth investigating in the future.

Acknowledgments

This work is supported by the National Natural Science Foundation of China (Grant nos. 51175007 and 51575007), the Science
and technology Innovation Project (PXM2015_014204_500002). The authors would like to thank Professor R.B. Randall and
Dr. N. Sawalhi of UNSW for their help in theoretical guidance. Finally, the authors are grateful to the editors and anonymous
reviewers for their helpful comments and constructive suggestions.

References

[1] Z. Feng, M. Liang, F. Chu, Recent advances in timefrequency analysis methods for machinery fault diagnosis: a review with application examples,
Mechanical Systems and Signal Processing 38 (1) (2013) 165205.
76 L. Cui et al. / Journal of Sound and Vibration 364 (2016) 6776

[2] J. Sopanen, A. Mikkola, Dynamic model of a deep-groove ball bearing including localized and distributed defects. Part 1: theory, Proceedings of the
Institution of Mechanical Engineers, Part K: Journal of Multi-Body Dynamics 217 (3) (2003) 201211.
[3] J. Sopanen, A. Mikkola, Dynamic model of a deep-groove ball bearing including localized and distributed defects. Part 2: implementation and results,
Proceedings of the Institution of Mechanical Engineers, Part K: Journal of Multi-Body Dynamics 217 (3) (2003) 213223.
[4] S. Sassi, B. Badri, M. Thomas, A numerical model to predict damaged bearing vibrations, Journal of Vibration and Control 13 (11) (2007) 16031628.
[5] R. Makowski, R. Zimroz, New techniques of local damage detection in machinery based on stochastic modelling using adaptive schur lter, Applied
Acoustics 77 (2014) 130137.
[6] J. Obuchowski, A. Wyomaska, R. Zimroz, The local maxima method for enhancement of timefrequency map and its application to local damage
detection in rotating machines, Mechanical Systems and Signal Processing 46 (2014) 389405.
[7] M. Tadina, M. Bolte, V.Z. Ar, Improved model of a ball bearing for the simulation of vibration signals due to faults during run-up, Journal of Sound and
Vibration 330 (17) (2011) 42874301.
[8] D. Petersen, C. Howard, N. Sawalhi, A.M. Ahmadi, S. Singh, Analysis of bearing stiffness variations, contact forces and vibrations in radially loaded
double row rolling element bearings with raceway defects, Mechanical Systems and Signal Processing 50 (2015) 139160.
[9] A.M. Ahmadi, D. Petersen, C. Howard, A nonlinear dynamic vibration model of defective bearingsThe importance of modelling the nite size of rolling
elements, Mechanical Systems and Signal Processing 52 (2015) 309326.
[10] J. Urbanek, J. Antoni, T. Barszcz, Detection of signal component modulations using modulation intensity distribution, Mechanical Systems and Signal
Processing 28 (2012) 399413.
[11] M.J. Dowling, Application of non-stationary analysis to machinery monitoring, IEEE International Conference on Acoustics, Speech, and Signal Processing
1 (1993) 5962.
[12] I.K. Epps, H. McCallion, An investigation into the characteristics of vibration excited by discrete faults in rolling element bearings, Proceedings of the
Annual Conference of the Vibration Association of New Zealand, Christchurch, 1994.
[13] N. Sawalhi, R.B. Randall, Simulating gear and bearing interactions in the presence of faults: part I. The combined gear bearing dynamic model and the
simulation of localised bearing faults, Mechanical Systems and Signal Processing 22 (8) (2008) 19241951.
[14] R.B. Randall, N. Sawalhi, Signal Processing Tools for Tracking the Size of a Spall in a Rolling Element Bearing, IUTAM Symposium on Emerging Trends in
Rotor Dynamics 25 (2011) 429440.
[15] N. Baydar, A. Ball, A comparative study of acoustic and vibration signals in detection of gear failures using WignerVille distribution, Mechanical
Systems and Signal Processing 15 (6) (2001) 10911107.
[16] K. Jiang, G. Xu, L. Liang, G. Zhao, T. Tao, A quantitative diagnosis method for rolling element bearing using signal complexity and morphology ltering,
Journal of Vibroengineering 14 (4) (2012) 18621875.
[17] C. Shen, F. Hu, F. Liu, Quantitative recognition of rolling element bearing fault through an intelligent model based on support vector regression,
Proceedings of the Fourth International Conference on Intelligent Control and Information Processing, Beijing, 2013, pp. 842847.
[18] H. Hong, M. Liang, Fault severity assessment for rolling element bearings using the LempelZiv complexity and continuous wavelet transform, Journal
of Sound and Vibration 320 (1) (2009) 452468.
[19] L. Cui, J. Wang, H. Wang, P. Chen, Improved complexity based on timefrequency analysis in bearing quantitative diagnosis, Advances in Mechanical
Engineering 258506 (2013) 11.
[20] S. Zhao, L. Liang, G. Xu, J. Wang, W. Zhang, Quantitative diagnosis of a spall-like fault of a rolling element bearing by empirical mode decomposition
and the approximate entropy method, Mechanical Systems and Signal Processing 40 (1) (2013) 154177.
[21] Y.B. Kong, Y. Guo, X. Wu, Double impulses extraction of faulty rolling element bearing based on EEMD and complex Morlet wavelet, Advanced Materials
Research 904 (2014) 437441.
[22] L. Cui, J. Wang, S. Lee, Matching pursuit of an adaptive impulse dictionary for bearing fault diagnosis, Journal of Sound and Vibration 333 (10) (2014)
28402862.
[23] L. Deshpande, N. Sawalhi, R.B. Randall, Gearbox bearing fault simulation using a nite element model reduction technique, Proceedings of the 25th
International Congress on Condition Monitoring and Diagnostic Engineering Conference Series, vol. 364, 2012, p. 012082.
[24] M.S. Patil, J. Mathew, P.K. Rajendrakumar, S. Desai, A theoretical model to predict the effect of localized defect on vibrations associated with ball
bearing, International Journal of Mechanical Sciences 52 (2010) 11931201.
[25] A.M. Ahmadi, D. Petersen, C. Howard, N. Sawalhi, Defect size estimation and analysis of the path of rolling elements in defective bearings with respect
to the operational speed, INTER-NOISE and NOISE-CON Congress and Conference Proceedings, vol. 249 (3) 2014, pp. 42424251.
[26] S. Singh, U.G. Kpke, C.Q. Howard, D. Petersen, Analyses of contact forces and vibration response for a defective rolling element bearing using an
explicit dynamics nite element model, Journal of Sound and Vibration 333 (2014) 53565377.
[27] Y.H. Wijnant, J.A. Wensing, G.V. Nijen, The inuence of lubrication on the dynamic behaviour of ball bearings, Journal of Sound and Vibration 222 (4)
(1999) 579596.
[28] A. Moazen Ahmadi, D. Petersen, C. Howard, A nonlinear dynamic model of the vibration response of defective rolling element bearings, Proceedings of
Australian Acoustics, Victor Harbor, 2013, pp. 17.
[29] N. Sawalhi, R.B. Randall, Vibration response of spalled rolling element bearings: Observations, simulations and signal processing techniques to track
the spall size, Mechanical Systems and Signal Processing 25 (3) (2011) 846870.