Progress in Engineering Science 1 (2024) 100030

Contents lists available at ScienceDirect

Progress in Engineering Science

journal homepage: www.sciencedirect.com/journal/progress-in-engineering-science

An improved approach on ball bearing characteristic frequencies

K.H. Munde a , Harsh H. Sawant b, M.J. Sable c , Kalpak J. Kamble d, R.N. Yerrawar b,
K.A. Mahajan b, P.D. Darade e , S.H. Gawande b,*
a
Department of Mechanical Engineering, K.E. Society’s Rajarambapu Institute of Technology polytechnic, Lohegaon, Pune 411047, India
b
Industrial Tribology Laboratory, Department of Mechanical Engineering, M.E.S. College of Engineering, Pune, S.P. Pune University, Pune 411047, India
c
Department of Mechanical Engineering, COEP Technological University, Wellesley Rd, Shivajinagar, Pune, India
d
Department of Mechanical Engineering, Maulana Azad National Institute of Technology, Bhopal, Madhya Pradesh 462003, India
e
Department of Mechanical Engineering, Sinhgad Institute of Technology and Science, Narhe, Pune, India

A R T I C L E I N F O A B S T R A C T

Keywords: Bearing Characteristic Frequencies (BCF) are crucial for diagnosing faults in rotary systems, as they depend on
BCF the number of balls, bearing geometry, and shaft speed. However, divergences often arise between theoretical
Fit factor and actual BCF values due to additional influencing factors. This paper addresses these divergences by incor-
Vibration analysis
porating two key factors into the BCF equation. First, the speed variation among the shaft, inner race, outer race,
Defect frequency
Bearing defect
and rolling elements is considered, as fault conditions can alter these speeds. Second, the fit factor, representing
the impact of bearing mounting and component fits, is introduced to account for variations in BCF. The initial
section of the paper reviews existing BCF equations proposed by various researchers, while the latter part pre-
sents modified equations that integrate these additional factors, offering a more accurate range of BCF values.

Determining the exact frequency location for a specific fault remains

challenging. In this study, the natural frequency of motor-driven ma-
1. Introduction chinery was found to reflect underlying time-frequency features, indi-
cating the ball bearing’s condition. Misinterpretation of fault symptoms
Vibration signatures provide crucial insights into the condition of often occurs due to improper representation of nonlinear and transient
machine components. These vibrations arise due to variations in signals in the spectrum. Chen et al. [1] proposed a method that is limited
fundamental parameters such as pressure, temperature, load, and tor- by its assumption of identical distributions between training and test
que. Premature failure in rolling bearings can occur if operated under samples, which fine-tuning cannot address. To tackle the issue of vary-
unsuitable loading conditions and at excessive speeds. However, vi- ing data distributions, they introduced adaptation layers. Yu et al. [2]
brations may also emerge as part of the normal wear process over the developed a new time-frequency analysis technique called the
bearing’s lifespan. Monitoring the condition of rolling bearings has transient-extracting transform, which is particularly effective at identi-
demonstrated cost savings exceeding 50 % compared to traditional fying and extracting transient components in fault signals. This tech-
maintenance methods. The prevailing trend is to continuously monitor nique, an extension of the short-time Fourier transform, requires neither
rolling element bearings, allowing for early detection of wear and timely extensive parameters nor prior information. They compared the per-
replacement before failure occurs. The most common approach involves formance of this method with classical and advanced techniques using
measuring machine vibrations at the bearing at regular intervals using a quantized indicators such as Rényi entropy and kurtosis. Patil et al. [3]
velocity transducer or an accelerometer mounted on the machine casing. conducted a review on vibration signature analysis for bearing fault
Bearing characteristic frequencies can reveal defects within the bearing detection and diagnosis, highlighting the unexpected relationship be-
components, and more detailed analysis can pinpoint the defect’s tween the ease of fault detection and its importance to users. Their study
location and size. underscores the extensive use of rolling element bearings in both do-
Frequency pattern analysis is essential for fault detection. A fault mestic and industrial applications and the potential for undetected de-
diagnosis procedure utilizing wavelet and statistical methods to extract fects to cause machinery failures.
features of bearing faults derived from time-frequency analysis, suc- Ibrahim et al. [4] applied the discrete wavelet transform (DWT) to
cessfully identifying ball bearing faults was developed [16,17].

* Corresponding author.
E-mail address: shgawande@gmail.com (S.H. Gawande).

https://doi.org/10.1016/j.pes.2024.100030
Received 25 July 2024; Received in revised form 3 November 2024; Accepted 20 November 2024
Available online 22 November 2024
2950-4252/© 2024 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

McInerny et al. [11] developed a machine condition monitoring

Nomenclature application focused on rolling element bearings, presenting their
fundamental operational characteristics and formulas for calculating
N Shaft speed (rpm) characteristic fault frequencies. They discussed a fault detection module
D Pitch diameter of bearing (m) and the limitations of conventional vibration spectral analysis, using
di Inner race diameter (mm) synthetic vibration signals generated in Matlab to illustrate key vibra-
Z Number of balls tion signature features and the connection between fault signatures and
SF Speed factor amplitude modulation/demodulation. Patel et al. [12] researched the
δr Radial deflection detection of local defects in deep groove ball bearing races amidst
K Stiffness coefficient external vibrations using envelope analysis and Duffing oscillators. They
DS Defect size conducted experiments with a test rig to capture vibration signals,
ω Angular velocity applying external vibrations via an electromechanical shaker. Spectral
c Damping coefficient kurtosis was used in the envelope analysis to select the center frequency
Fs Shaft rotation frequency (Hz) for filters of different lengths and bandwidths. Tu et al. [13] investigated
d Ball diameter (m) skidding in rolling element bearings during shaft rotational acceleration,
do Outer race diameter (mm) which leads to wear and incipient failure. They developed an analytical
θ Contact angle model considering contact force, friction, cage gravity, and centrifugal
FF Fit factor force, applying Hertzian contact theory for non-linear contact force
k Load-deflection factor calculation. This model aids in designing applications to estimate wear
HD Defect depth during skidding and assess methods to prevent skidding. Wu et al. [14]
R,W
R Raceway radius analyzed the vibration characteristics of air blowers, crucial for per-
m Mass formance and system longevity. They simplified the rolling
W Load (weight) bearing-rotor system as a particle on a shaft with radial stiffness and
damping, ignoring the bearing structure’s impact on rotor vibration. Lei
et al. [15] introduced the Kurtogram for detecting and characterizing
transients in signals, a powerful tool for machinery fault diagnosis.
the q-axis component of the stator current for induction motor fault However, the use of short-time Fourier transforms or FIR filters limits its
detection. They developed a fault detection method using an artificial accuracy. They suggested developing precise filters to enhance the
neural network (ANN) to identify the motor’s operational state. A dy- Kurtogram’s ability to extract transient characteristics and identify
namic model of a squirrel-cage induction motor, incorporating bearing faults from noisy signals. Zhang et al. [16] explored roller bearing fault
faults, was constructed using Simulink/Matlab. Simulation results location classification and performance degradation assessment to
demonstrated the superior performance of the proposed method. Harsha reduce costs and unscheduled downtime. They mapped original feature
et al. [5] derived the governing differential equations of motion using vectors into higher-dimensional space, using kernel principal compo-
Lagrange’s equations. To solve these non-linear differential equations, nents as inputs to a PSO-SVM model. Experimental results demonstrated
they employed the implicit Newmark-beta numerical integration tech- the method’s effectiveness in identifying fault locations and perfor-
nique in conjunction with the Newton–Raphson method. An analytical mance degradation, proposing an automated fault diagnosis approach
model was developed to predict non-linear dynamic responses in a rotor for non-stationary vibration signals.
bearing system caused by surface waviness. William et al. [6] repre- Sheen et al. [17] proposed an algorithm for bearing vibration signal
sented the composite vibration signature in the spectral domain, processing based on the mechanical system’s resonance modes, quan-
normalizing features by the observation window length and classifying tifying the exponential decay frequency of the envelope signal and
them using a multilayer feed-forward neural network. The model was estimating resonance frequencies within the corresponding modes. Cong
evaluated using vibration data recorded by an accelerometer mounted et al. [18] analyzed the optimal order selection for the autoregressive
on an induction motor subjected to various single-point defects with (AR) model in diagnosing rolling element bearings, enhancing noise
different severity levels. cancellation by defining the optimal order as the maximum length of the
Kankar et al. [7] emphasized the critical importance of fault diag- periodic damping part recognizable by the AR model. Yan et al. [19]
nosis in bearings with point defects for designing high-performance reviewed wavelet applications in rotary machine fault diagnosis,
rotor bearing systems. Traditional diagnostic methods rely on signal covering classical wavelet transform, second-generation wavelet trans-
processing techniques that assume statistically stationary signal fea- forms, and new research trends such as wavelet finite element methods
tures. They introduced a feature-recognition system to diagnose faults in and dual-tree complex wavelet transform. Tang et al. [20] developed a
rolling element bearings. Tomovic et al. [8] experimentally validated a method using compressed vibration signals within a compressive
vibration model and conducted a parametric analysis. Their study pre- sensing framework to detect roller bearing faults, improving detection
sented empirical results, demonstrating how the specific construction of efficiency from minimal samples without full signal reconstruction.
a bearing significantly influences its dynamic behavior. They analyzed Bhende et al. [21] conducted an experimental study using acoustic
the effects of internal radial clearance and the number of rolling ele- emission techniques and an algorithm processing various signals to di-
ments on the vibrations of a rigid rotor in an unloaded rolling element agnose bearing faults, showing effectiveness in detecting fault presence,
bearing. Kumar et al. [9] explored the use of an acoustic emission location, and severity. Attaran et al. [22] presented an optimized arti-
technique, employing Analytical Wavelet Transform (AWT) to detect ficial neural network technique using the Bees Algorithm for diagnosing
and assess the severity of defects in the inner race of bearings, particu- localized faults in rolling element bearings, achieving 100 % accuracy
larly in the form of an axial groove. Their research examined the nature with radial basis function neural networks. Hiremath et al. [23] evalu-
of acoustic emission (AE) bursts and the correlation between AE bursts ated incipient fault detection in lightly loaded ball bearings using
and bearing defects. Kumar et al. [10] employed the discrete wavelet acoustic emission methods, demonstrating acoustic analysis effective-
transform of vibration signals to measure the width of outer race defects ness over vibration measurement. Singh et al. [24] detected bearing
in taper roller bearings. They underscored the critical role of rolling faults using motor current signatures and acoustic signals, with
element bearings in rotary machines and highlighted the ongoing increasing sidebands around running speed indicating faulty bearings
challenge in bearing fault monitoring: accurately estimating defect size and validating findings through system assurance of electrical fault
in bearing elements. absence. Esfahani et al. [25] proposed a stand-alone multi-sensor

2
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Fig. 1. Geometry of SKF-6206 Bearing.

wireless system for continuous induction motor condition monitoring, 2.2. Vibration analysis procedure
achieving 95 % average accuracy in fault category and severity
discrimination using multiple sensor modalities to reduce the need for The vibration analysis procedure began with the installation and
precise instrumentation. calibration of the bearings on a test rig. Proper alignment was ensured to
The comprehensive analysis underscores the critical role of vibration achieve precise measurements. The accelerometers were meticulously
signatures in diagnosing and monitoring the condition of rolling element calibrated to guarantee accurate vibration readings, eliminating poten-
bearings. Advanced diagnostic techniques, such as wavelet transforms, tial discrepancies from misalignment or sensor errors. Data collection
the discrete wavelet transform (DWT) coupled with artificial neural involved operating the test rig at various speeds to simulate different
networks (ANN), and compressed sensing frameworks, have demon- load conditions, thus providing a comprehensive range of operational
strated significant benefits, including timely fault detection, cost sav- scenarios. Vibration data were collected at regular intervals across these
ings, and improved accuracy. However, these methods also pose different speeds and loading conditions, capturing the bearing’s per-
challenges due to their complexity, data dependency, and the risk of formance under realistic and varying operational states.
misinterpreting non-linear and transient signals. Despite these chal- Signal processing commenced with preprocessing, where noise
lenges, the potential for substantial maintenance cost reductions and filtering was applied to the raw vibration signals using a low-pass filter.
enhanced machinery reliability through continuous monitoring and This step was essential to remove unwanted noise that could obscure
sophisticated diagnostic approaches cannot be overlooked. Future relevant signal features. The filtered signals were then normalized to
research should focus on simplifying these techniques and enhancing account for variations in measurement conditions, ensuring consistency
their robustness for real-time industrial applications. and reliability in the subsequent analysis. Time-frequency analysis
involved applying the wavelet transform to extract transient features
2. Methodlogy and identify non-stationary components in the vibration signals. This
method provided a detailed analysis of the signal’s time-varying char-
2.1. Experimental setup acteristics, essential for identifying and diagnosing transient events and
faults in the bearings.
Bearings of the type SKF-6206 (Deep Groove Ball Bearing) were Finally, the estimation of BCF was performed using standard equa-
selected for this study. The bearing geometry is shown in Fig. 1, with tions and modified equations incorporating Speed Factor (SF), which
specifications including 9 balls, a ball diameter of 9.525 mm, a pitch adjusts for speed-related variations like centrifugal forces, and Fit Factor
diameter of 46 mm, an inner diameter of 30 mm, and an outer diameter (FF), which accounts for variations in bearing fits affecting load distri-
of 62 mm. The shaft speed was set to 1500 RPM, resulting in a shaft bution and stress concentration. The SF was derived by combining
rotation frequency of 25 Hz. Vibration sensors, specifically accelerom- empirical data from experimental observations on ball bearing perfor-
eters, were mounted on the bearing housing to measure vibrations, and a mance at various rotational speeds with theoretical models that account
high-speed data acquisition system was utilized to capture high- for centrifugal forces, lubrication behavior, and thermal effects. This
frequency vibrations, ensuring a sufficient sampling rate. Commonly, approach allowed for refining the mathematical models with real-world
the sampling rate should be at least twice the highest frequency of in- data. Similarly, the FF was developed by analyzing empirical data from
terest, but in practice, it is was set significantly higher (e.g., 10 times the different fits (interference and clearance) between the bearing and its
maximum frequency) to ensure accurate representation and to prevent housing and shaft, which highlighted their impact on load distribution,
aliasing. The machinery vibration signals were decomposed into spectral stress concentration, and wear. Theoretical considerations in contact
components to define the machinery signature, which is a sum of har- mechanics and material deformation were integrated to enhance the
monic functions of discrete amplitudes and frequencies. These adjust- accuracy of the Fit Factor. These factors were included to account for
ments were intended to provide a more accurate reflection of the variations in speed and fit conditions, offering a more precise repre-
characteristic frequencies under practical conditions. sentation of the bearing’s performance. By incorporating SF and FF into
the equations, the analysis reflected real-world operating conditions,
leading to more precise and reliable diagnostic results.
The modified BCF equations in this paper, incorporating SF and FF,

3
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

enhance the accuracy of bearing fault detection by accounting for real- calculation and serve as a standard model.
world variations in speed and fit. This accuracy can lead to earlier and
more reliable fault detection in bearings, reducing unexpected down- a) Ball Pass Frequency of the Outer Race (BPFO)
times and extending the lifespan of machinery. The modifications are [ ]
Fs d
particularly valuable in industries where machinery operates under BPFO = Z 1− cos θ (1)
2 D
fluctuating speeds and loads, such as manufacturing, automotive, and
aerospace sectors. By improving the precision of diagnostics, these
equations can contribute to more efficient maintenance schedules and
b) Ball Pass Frequency of the Inner Race (BPFI)
better overall machinery performance. Industry professionals can
[ ]
implement these modified equations by integrating them into existing Fs d
BPFI = Z 1 + cos θ (2)
condition monitoring systems. This would involve recalibrating diag- 2 D
nostic software to include the SF and FF adjustments, ensuring that the
calculations reflect real-world operational conditions. While not tailored
to specific machinery, these equations offer a universally applicable c) Ball Spin Frequency (BSF)
approach to accurate bearing diagnostics. Compared to time-domain [ ( )2 ]
D d
analysis and advanced signal processing techniques like wavelet trans- BSF = 1− cos θ (3)
forms, these equations provide more precise frequency predictions but 2dFs D
may be less effective in detecting complex, transient signals. Thus, they
are best used alongside advanced techniques for comprehensive
diagnostics. d) Fundamental Train Frequency (FTF)
The main cost implications of implementing these modifications lie Fs
[
d
]
in the need for recalibrating diagnostic tools and potentially upgrading FTF = 1 ± cos θ (4)
2 D
software to accommodate the new equations. However, the benefits of
improved accuracy in fault detection may offset these costs through Eqs. (5–8) [27] are summarized as,
reduced machinery downtime and maintenance expenses. Practical N
[
d
]
challenges include ensuring that empirical data used for SF and FF BPFO = Z 1− cos θ (5)
2.6 D
calibration is accurate and representative of actual operating conditions.
Additionally, there may be a learning curve for technicians and engi- N
[
d
]
neers who need to understand the new factors and how they affect BPFI = Z 1 + cos θ (6)
2.6 D
diagnostic results. Overcoming these challenges will require investment
in training and potentially in new measurement equipment to gather the [ ( )2 ]
2ND d
necessary data for calibration. BSF = 1− cos θ (7)
2.6d D

2.3. Estimation of BCF [ ]

N d
FTF = 1− cos θ (8)
2.6 D
Each element of a bearing vibrates at a specific frequency that re-
veals the operational condition of that element. For instance, a fault in These four characteristic bearing frequencies are fundamental in
the outer raceway of a bearing generates vibrations at the frequency their respective harmonic series. The harmonic series’ amplitudes
corresponding to the rolling elements passing over the defect. By measure the deviation from sinusoidal force variation. A non-defective
considering the kinematics and geometry of the bearing, it is possible to rolling element bearing exhibits four distinct sets of frequencies or sig-
estimate the frequencies associated with defects in various bearing natures. The assumption in machine monitoring is that a localized defect
components. The analysis of bearing characteristic frequencies relies on modifies the signature in a way that indicates the defect’s location.
several assumptions: all the balls or rollers have equal diameters, there is Eqs. (9–12) proposed by SKF [28] are used to estimate approximate
no slip between the bearing and the shaft, the inner race rotates while frequencies in the field. These equations helped in practical applications
the outer race remains stationary, the components (balls, inner race, and to identify bearing faults based on the component frequencies.
outer race) are in pure rolling motion, the fits are perfect during as- [ ]
Z
sembly and mounting, and there are no manufacturing defects in the BPFO = − 1.2 × N (9)
2
bearing components.
[ ]
Z
2.3.1. Mathematical model for rotor bearing system BPFI = + 1.2 × N (10)
2
A mathematical model is created to forecast the vibration signatures
of rotor-bearing systems with localized defects. The model incorporated [ ]
Z 1.2
a non-linear spring approach for ball-race contacts, based on Hertzian BSF = − ×N (11)
2 Z
contact deformation theory. A computer simulation was developed to
model defects on raceways, with results analyzed in both time and fre- [
1 1.2
]
quency domains. Additionally, the impact of defect size and location on FTF = − ×N (12)
2 Z
bearing vibrations was examined.
Many researchers have compared theoretical and practical values of
2.3.2. Modified frequency model characteristic frequencies and have developed simple thumb rules for
The aim was to refine the BCF equations to better match practical the BCF equations [29,30]. The equations are as follows:
values. To achieve this, SF and FF were introduced to account for vari-
BPFO = 0.45[Z × N] (13)
ations in speed and bearing fits. Standard BCF equations were modified
to include these factors, providing a range of fault frequencies. In the BPFI = 0.55 − 0.6[Z × N] (14)
context of faulty rolling element bearings, four frequencies correspond
to the four main components: inner race, outer race, ball, and cage. BSF = 3.5 × N (15)
These frequencies (Eqs. (1–4) [26]) are fundamental in fault frequency

4
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Table 1 { [ ( )2 ]}/
Comparison of BCF by different methods. 2ND d
BSF = 1− cos θ 1.5 (19)
Sr. Method for calculation BPFO BPFI BSF FTF
2.6d D
No. of BCF (Hz) (Hz) (Hz) (Hz)
{ [ ]}
1 Standard Equation (SE) 89 136 58 10
N d
FTF = 1− cos θ × 1.3 (20)
2 IIT Roorkee 69 104 89 7.6 2.6 D
3 SKF VAL− 2 83 143 109 9
4 Thumb Rule 101 135 88 10 These equations are based primarily on the geometrical construction
5 Modified IIT Roorkee 89 136 59 10 of the bearing and the number of balls. However, they do not account for
6 SE modified by FF 79 – 100 121– 153 51 – 65 11 real-world factors such as improper installation, lubrication issues,
7 SE modified by SF 81 – 97 124–149 53 – 63 11
contamination, or fatigue loading, all of which can significantly affect
the actual characteristic frequencies. As a result, practical values of these
frequencies often differ from the theoretical predictions. The SF and FF
FTF = 0.4 × N (16) have been introduced to address these discrepancies. SF, which adjusts
These thumb rules are based on the observation that when a shaft for variations in rotational speed, is calculated as below.
completes one rotation, the cage rotates 0.4 times that of the shaft ( )1/3
1
rotation. This relationship simplifies field measurements and aids in sf = 3.218 (21)
N
practical applications. Equations for Ball Bearing characteristic fre-
quencies have been adjusted to better align with practical values by ( )
sf
incorporating two key factors; SF and FF. The modified equations are SF = 1 + (22)
3
designed to account for the impact of operational conditions on the
accuracy of theoretical frequency estimation. Specifically, the equations FF, on the other hand, accounts for variations in bearing geometry
for BPFO, BPFI, BSF, and FTF are given as follows: and operational conditions. It is defined as:
{ [ ]}
N d di
BPFO = Z 1− cos θ × 1.3 (17) KM = (23)
2.6 D k(do − di ) + di
{ [ ]} ( )
N d KM
BPFI = Z 1 + cos θ × 1.3 (18) FF = 1 + (24)
2.6 D 6

where k (Load-deflection factor) = 0.3 for ball bearings.

Fig. 2. Variation of acceleration with frequency for outer race with defect size 0.5 mm, load 2 kg & speed 500–3000 RPM.

5
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Fig. 3. Variation of acceleration with frequency for outer race with defect size 0.5 mm, load 4 kg speed 500–3000 RPM.

The modified equations that incorporate SF and FF to determine the components, such as perfect alignment and constant thrust load, were
higher limit (HL) and lower limit (LL) of BCF as below. assumed. Hertzian contact theory was used to compute contact forces
and deflections. Defect size and location were incorporated into the
BCF − HL = BCF × SF (25)
model to predict the spectral components of bearing vibrations.
This model for analyzing defective bearings is based on several as-
BCF − LL = BCF/SF (26)
sumptions: the outer and inner bearing races are perfectly circular, all
BCF − HL = BCF × FF (27) balls are spherical and of equal diameter, and the races are perfectly
aligned. The system is modeled with three degrees of freedom—radial
BCF − LL = BCF/FF (28) vibrations in the x and y directions and axial vibrations in the z direc-
tion—while applying a constant thrust load without radial load. Balls
Table 1 was prepared by solving the equations discussed above using are treated as massless, and the rings are assumed to be flexurally rigid
MATLAB. It presents the characteristic frequencies obtained from the with local deformations due to contact stresses. Deformations follow
standard equations and those provided by various institutes. The Table1 Hertzian elasticity theory, resulting in elliptical contact areas, and balls
highlights a significant discrepancy between the frequencies calculated are equally spaced around the inner ring without interactions. Damping
using the IIT Roorkee [27] equations and the standard values. To sources, such as elasto-hydrodynamic film and friction, are considered
address this, the IIT Roorkee [27] equations were modified to produce negligible, and the cage’s angular velocity is constant. Both bearings are
frequencies that are closer to the standard values. The differences symmetrically positioned, rotating simultaneously with constant
observed between the standard and calculated frequencies can often be angular separation maintained by the cage. Small elastic motions are
attributed to variations in speed or fit during operation. The modified considered, and all components, including the rolling elements, inner
equations, therefore, offer a range of fault frequencies that account for and outer races, and shaft, are treated as rigid with no bending. The
these variations, considering both speed and fit factors to provide more rollers have no angular rotation about their axes, and interactions with
realistic frequency predictions. the cage or races are disregarded.

2.3.3. Theoretical model for defective bearing

The goal was to develop a theoretical model to analyze the impact of 2.4. Mathematical formulation
defect size on bearing vibration. Ideal conditions for bearing
1. Load-Deflection Relation:

6
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Fig. 4. Variation acceleration with frequency for outer race with defect size 0.5 mm, load 6 kg& speed 500–3000 RPM.

Fig. 5. Variation acceleration with frequency for inner race with defect size 2 mm, load 8 kg speed 500–3000 RPM.

7
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Fig. 6. Variation of acceleration with frequency for inner race with defect size 2 mm, load 12 kg speed 500–3000 RPM.

According to Hertzian contact deformation theory, the non-linear for the contact of a ball with the outer race is
load-deformation relationship is expressed as:
(∑ )− 1/2 ( )
− 3/2
F = kδnr (29) Ko = 3.587 × 107 ρ δ∗o (32)

where: 4. Curvature Sum and Difference:

k = Load-deflection factor • Curvature sum:

δr = Radial deflection ∑ 1 1 1 1
n = Load-deflection exponent (3/2 for ball bearings, 10/9 for roller ρ = ρI1 + ρI2 + ρII1 + ρII2 = + + + (33)
rI1 rI2 rII1 rII2
bearings)

2. Load-deflection factor: • Function F(ρ):

The load-deflection factor k for the bearing system is given by:
(ρI1 − ρI2 ) + (ρII1 − ρII2 )
F(ρ) = ∑ (34)
1 ρ
k=[ ]n (30)
1
1/n + 1
1/n
5. Contact Force:
Ki Ko
{ (∑ )1/2 }
3/2
Q = 3.587 × 107 (δ∗ )− ρ δ3/2 (35)
where Ki and Ko are the elastic moduli for the contact of a ball with the
inner and outer races respectively. Where, δ∗ is normalized contact deformation and δ represents contact
3. Elastic Modulus for Contact: deformation
The elastic modulus for the contact of a ball with the inner race is 6. Radial Deflection:
(∑ )− 1/2 ( )− 3/2 The radial deflection at the ith ball, at any angle θi is given by,
Ki = 3.587 × 107 ρ δ∗i (31)
δr = [(x cos θi + y sin θi ) − Rc ] (36)

8
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Acceleration versus Speed for 2 kg Acceleration versus Speed for 6 kg

700.00 700.00
600.00 600.00
Acceleration (mm/s2)

500.00 500.00

Acceleration (mm/s2)
400.00 400.00
300.00 300.00
200.00 200.00
100.00 100.00
0.00
0.00
500 1000 1500 2000 2500 3000
500 1000 1500 2000 2500 3000
Speed (rpm)
Speed (rpm)
0.5 1 1.5 2 0.5 1 1.5 2

(a) (a)

Acceleration versus Speed for 4 kg Acceleration versus Speed for 8 kg

700.00
700.00
600.00 600.00

Acceleration (mm/s2)
Acceleration (mm/s2)

500.00 500.00
400.00 400.00
300.00 300.00
200.00
200.00
100.00
100.00
0.00
500 1000 1500 2000 2500 3000 0.00
Speed (rpm) 500 1000 1500 2000 2500 3000
Speed (rpm)
0.5 1 1.5 2 0.5 1 1.5 2
(b) (b)
Fig. 7. Variation in amplitude and speed of the outer race for (a) 2 kg load and Fig. 8. Variation in amplitude and speed of the outer race for (a) 6 kg load and
(b) 4 kg load. (b) 8 kg load.

The deflection along X axis and Y axis are x and y and Rc is the in-
ternal radial clearance. θt = (ωc − ω)t + 2π/Z(Z − i) (43)
7. Restoring Force (No defect): 9. Inertia Force:
Components of the restoring force along X and Y axes are:
Z
F = mew2 (44)
∑ 3/2
FX = K[(x cos θi + y sin θi ) − Rc ] cos θi (37) The direction of the rotating inertia force shifts with the rotation of
i=1
the mass center.
Z
∑ 10. Bearing Load Components:
FY = K[(x cos θi + y sin θi ) − Rc ]3/2 sin θi (38) The bearing load includes:
i=1
Fy = W + Fm cos θ (45)
K is the stiffness coefficient.
8. Restoring Force (With Defect): Fx = Fm sin θ (46)
For a defect on the bearing race:
Z [( ) ( ( ))]3/2 where W = mg and θ = 2πfs t.
∑ π
FXD = K x cos θi + y sin θi − Rc + HD sin cos θi
γ(θt − θi )
i=1
2.5. Equations of motion
(39)
[( ) ( ( ))]3/2 1. For Bearing Race without Defect:
∑Z
π
FYD = K x cos θi + y sin θi − Rc + HD sin sin θi mẍ + cẋ + FX = Fm sin θi
i=1
γ(θt − θi )
∑Z
(40) mẍ + cẋ + K[(x cos θi + y sin θi ) − Rc ]3/2 cos θi = Fm sin θ
i=1
DS mÿ + cẏ + FY = W + Fm cos θ
γ= (41)
RWR ∑Z
mÿ + cẏ + K[(x cos θi + y sin θi ) − Rc ]3/2 sin θi = W + Fm cos θi
K is the stiffness coefficient, HD = Defect depth, DS = Defect Size & i=1
RWR = Raceway radius (47)
θt = ωc t + 2π/Z(Z − i) (42) 2. For Bearing Race with Defect:

Where i = Z to 1

9
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

Acceleration versus speed for 10 kg Amplitude versus speed for 2 kg

700.00 350.00
600.00 300.00
Acceleration (mm/s2)

Acceleration (mm/s2)
500.00 250.00
400.00 200.00
300.00 150.00

200.00 100.00

100.00 50.00
0.00
0.00
500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000
Speed (rpm) Speed (rpm)

0.5 1 1.5 2 0.5 1 1.5 2

(a) (a)

Amplitude versus speed for 4 kg

Acceleration versus speed for 12 kg
800.00 400.00
700.00 350.00

Acceleration (mm/s2)
600.00 300.00
Acceleration (mm/s2)

250.00
500.00
200.00
400.00
150.00
300.00 100.00
200.00 50.00
100.00 0.00
500 1000 1500 2000 2500 3000
0.00
500 1000 1500 2000 2500 3000 Speed (rpm)
Speed (rpm)
0.5 1 1.5 2
0.5 1 1.5 2
(b)
(b)
Fig. 10. Variation in amplitude and speed of the inner race for (a) 2 kg load
Fig. 9. Variation in amplitude and speed of the outer race for (a) 10 kg load and (b) 4 kg load.
and (b)12 kg load.

mẍ + cẋ + FXD = Fm sin θi

∑Z [ ( ( ))]3/2
π
mẍ + cẋ + K (x cos θi + y sin θi ) − Rc + HD sin cos θi = Fm sin θi
i=1
γ(θt − θi )

mÿ + cẏ + FYD = W + Fm cos θ

Z
∑ [ ( ( ))]3/2
π
mÿ = cẏ + K (x cos θt + y sin θt ) − Rc + HD sin sin θi
i=1
γ(θt − θi )
= W +Fm cos θ
(48) 0.5 OuterMath 0.5 OuterExp 0.5 InnerMath 0.5 InnerExp

Where c is the damping coefficient, and K is the stiffness coefficient. 140.00

The initial displacement conditions are as follows: 120.00

Acceleration (mm/s2)

100.00
6
X0 = Y0 =10− m, Velocity (X0) = (Y0) = 0
80.00

The proposed model aims to predict the impact of defect size and 60.00

location on the frequency components of bearing vibrations. It considers 40.00

both the presence and absence of defects to provide a comprehensive 20.00
analysis. The model combines impulse response signal modeling with 0.00
rotor dynamic forces to simulate rolling bearing fault signals and predict 500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000

fault characteristics. This methodology allows for detailed analysis and 2 Kg 4 kg

Speed (rpm)
prediction of defective bearing behavior, contributing valuable insights
into bearing diagnostics and maintenance strategies.
Fig. 11. Variation of acceleration with speed with defect size 0.5 mm for 2 kg
& 4 kg.

10
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

ranging from 2 kg to 12 kg, and speeds from 500 to 3000 RPM. The
1 OuterM 1 OuterE 1 InnerM 1 InnerE
results showed the relationship between frequency and acceleration for
350.00 different loads. For a defect size of 0.5 mm and a load of 2 kg, the fre-
300.00 quencies of 29.76 Hz, 59.51 Hz, 89.27 Hz, 119.02 Hz, 148.78 Hz, and
Acceleration (mm/s2)

250.00
178.53 Hz corresponded to acceleration amplitudes of 4.08 mm/s²,
8.76 mm/s², 19.42 mm/s², 41.05 mm/s², 78.34 mm/s², and 98.46 mm/
200.00
s², correspondingly. The trend indicated that as the speed increased, the
150.00 amplitude of vibration increased as well. Fig. 2 illustrates the frequency
100.00 and acceleration relationship for a defect size of 0.5 mm in the outer
50.00 race under a 2 kg load and speeds ranging from 500 to 3000 RPM.
When the load was increased to 4 kg, the frequencies remained
0.00
500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 consistent with those observed under a 2 kg load, but the corresponding
2 Kg 4 kg
amplitudes were higher: 8.34 mm/s², 12.24 mm/s², 24.29 mm/s²,
42.23 mm/s², 85.37 mm/s², and 106.25 mm/s². This further supported
Speed (rpm)
the observation that amplitude increased with both speed and load. For
a load of 6 kg, the amplitudes at the same frequencies were: 12.51 mm/
Fig. 12. Variation of acceleration with speed with defect size 1 mm for 2 kg s², 15.89 mm/s², 31.23 mm/s², 45.45 mm/s², 82.89 mm/s², and
& 4 kg.
112.89 mm/s². The increasing trend in vibration amplitude with speed
and load was consistent. Fig. 3 shows the frequency and acceleration
relationship for a defect size of 0.5 mm in the outer race under a 4 kg
1.5 OuterM 1.5 OuterE 1.5 InnerM 1.5 InnerE load and speeds ranging from 500 to 3000 RPM and Fig. 4 depicts the
600.00 frequency and acceleration relationship for a defect size of 0.5 mm in the
outer race under a 6 kg load and speeds ranging from 500 to 3000 RPM.
500.00
The analysis of the inner race defects for a defect size of 2 mm under
Acceleration (mm/s2)

400.00 an 8 kg load showed frequencies at 45.24 Hz, 90.49 Hz, 135.73 Hz,
180.98 Hz, 226.22 Hz, and 271.47 Hz, with corresponding amplitudes
300.00
of 21.67 mm/s², 42.63 mm/s², 89.46 mm/s², 161.23 mm/s²,
200.00 241.31 mm/s², and 356.45 mm/s². This illustrated a notable increase in
100.00 amplitude with both load and speed. For a 12 kg load, the amplitudes
were notably higher at the same frequencies: 31.53 mm/s², 55.47 mm/
0.00
500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000
s², 106.46 mm/s², 181.23 mm/s², 241.31 mm/s², and 356.45 mm/s².
This further demonstrated the proportionality between defect size, load,
2 Kg 4 kg
Speed (rpm) and vibration amplitude. Fig. 5 illustrates the frequency and accelera-
tion relationship for a defect size of 2 mm in the inner race under an 8 kg
Fig. 13. Variation of acceleration with speed with defect size 1.5 mm for 2 kg load and speeds between 500 and 3000 RPM, and Fig. 6 shows the
& 4 kg. frequency and acceleration relationship for a defect size of 2 mm in the
inner race under a 12 kg load and speeds ranging from 500–3000 RPM.
3. Results and discussion The experimental findings for the outer race defects showed that the
amplitude varied significantly between defect sizes of 1 mm and
3.1. Results 1.5 mm, while 1.5 mm and 2 mm defects showed closer amplitude
values. This indicated that larger defects produced higher amplitude
Various frequencies and their corresponding amplitude values were vibrations. Similar trends were observed for loads of 6 kg and 8 kg, with
obtained through vibration analysis. Significant frequencies associated larger differences in amplitude between 0.5 mm and 1 mm defects, and
with bearing defects and their corresponding amplitude levels were closer values between 1.5 mm and 2 mm defects. The data confirmed
highlighted, providing insights into the severity and characteristics of that both speed and defect size increased vibration amplitude. Fig. 7(a)
the defects. The mathematical analysis of the outer race defects was and (b) show the amplitude versus speed relationship for the outer race
conducted at a defect size of 0.5 mm under various load conditions with defect for loads of 2 kg and 4 kg, respectively and Fig. 8(a) and (b)
depict the amplitude versus speed relationship for the outer race with
defect for loads of 6 kg and 8 kg, respectively.
For loads of 10 kg and 12 kg, defect sizes of 0.5 mm and 1 mm
2 OuterM 2 OuterE 2 InnerM 2 InnerE exhibited significant differences in amplitude, whereas 1.5 mm and
700.00 2 mm defects had almost overlapping amplitudes. This suggested that
600.00 the defect size’s impact on amplitude became less distinguishable at
higher loads. Fig. 9(a) and (b) illustrate the amplitude versus speed
Acceleration (mm/s2)

500.00
relationship for the outer race with defect and 10 kg and 12 kg loads
400.00
respectively.
300.00
For the inner race defects under loads of 2 kg and 4 kg, the amplitude
200.00 intervals were consistent, with 500 and 1000 RPM showing overlapping
100.00 values, while 3000 RPM exhibited a significant rise. This confirmed that
0.00 increased speed resulted in higher amplitude vibrations. Fig. 10 (a) and
500 1000 1500 2000 2500 3000 500 1000 1500 2000 2500 3000 (b) show the amplitude versus speed relationship for the inner race with
2 Kg 4 kg defect for loads of 2 kg and 4 kg, respectively.
Speed (rpm) The results indicated that as speed and load increased, the amplitude
of vibrations increased, which could be used to identify and diagnose
Fig. 14. Variation of acceleration with speed with defect size 2 mm for 2 kg bearing defects more effectively. The experimental findings aligned with
& 4 kg. the mathematical results, validates the trends observed in the study.

11
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

3.2. Validation Future research should explore integrating advanced diagnostic

techniques such as machine learning and artificial intelligence to
The validation of the results is crucial to ensure the accuracy and enhance defect detection and prediction. Additionally, developing more
reliability of the mathematical models used in the study. In this section, robust models that can account for a wider range of operational condi-
the mathematical predictions are compared with experimental results to tions and noise factors could improve the reliability of vibration analysis
confirm the effectiveness of the proposed models. Figs. 11 to 14 illus- in diverse settings. Further investigations into real-time data processing
trate the comparison between the mathematical and experimental re- and predictive maintenance strategies could also provide significant
sults for various defect sizes (0.5 mm, 1 mm, 1.5 mm, and 2 mm) under advancements in bearing fault diagnosis.
two different loads (2 kg and 4 kg) over a speed range of 500 RPM to
3000 RPM. Financial and ethical disclosures

• For a defect size of 0.5 mm: Fig. 11 shows that the amplitudes of the This work is not supported fully or partially by any funding organi-
mathematical model and experimental data are nearly identical at zation or agency.
lower speeds (500 RPM and 1000 RPM). As the speed increases from
1500–3000 RPM, slight discrepancies appear, but the overall trend List of abbreviations
remains consistent.
• For a defect size of 1 mm: The results (Fig. 12) indicate a similar Abbreviation Full form
pattern where the mathematical and experimental amplitudes BCF Bearing Characteristic Frequencies
closely match at lower speeds, with minor deviations occurring at SF Speed Factor
higher speeds. FF Fit Factor
FFT Fast Fourier Transform
• For a defect size of 1.5 mm: The comparison (Fig. 13) demonstrates DWT Discrete Wavelet Transform
that the agreement between the mathematical and experimental ANN Artificial Neural Network
results holds well at lower speeds, with the differences becoming BPFO Ball Pass Frequency of the Outer Race
slightly more pronounced as the speed increases. BPFI Ball Pass Frequency of the Inner Race
BSF Ball Spin Frequency
• For a defect size of 2 mm: As shown in Fig. 14, the mathematical
FTF Fundamental Train Frequency
and experimental results are in close agreement across all speeds, AE Acoustic Emission
with only marginal differences observed at higher speed. AR Autoregressive
PSO-SVM Particle Swarm Optimization-Support Vector Machine
The overall comparison indicates that the mathematical models are AWT Analytical Wavelet Transform

effective in predicting the characteristic defect frequencies of rolling

contact bearings across different defect sizes and load conditions. Speed
variations among the shaft, inner race, outer race, and rolling elements
can significantly impact the accuracy of BCF calculations, as differences CRediT authorship contribution statement
from assumed speeds may cause discrepancies between calculated and
observed fault frequencies. These discrepancies often stem from factors Kalpak J. Kamble: Visualization, Validation, Resources, Method-
not accounted for in simplified models, such as variations in load, ology, Formal analysis, Data curation. M.J. Sable: Writing – review &
lubrication, and environmental conditions, as well as wear, contami- editing, Supervision, Project administration, Methodology, Investiga-
nation, or fatigue in real-world bearings. Misalignments and tion, Data curation, Conceptualization. Harsh H. Sawant: Writing –
manufacturing defects further reduce the accuracy of the theoretical original draft, Visualization, Validation, Software, Resources. K. H.
model by introducing unaccounted forces and altering contact dy- Munde: Writing – original draft, Resources, Methodology, Investiga-
namics. To address these challenges, the modified BCF equations tion, Formal analysis, Data curation, Conceptualization. S. H.
incorporate the SF and FF, which are calibrated with empirical data to Gawande: Writing – review & editing, Validation, Supervision, Re-
reflect real-world conditions. Continuous updates to these factors, based sources, Methodology, Investigation. P. D. Darade: Writing – review &
on ongoing monitoring, help mitigate errors caused by operational editing, Supervision, Resources, Methodology, Data curation, Concep-
variations like fluctuating loads and temperature changes. Despite some tualization. K. A. Mahajan: Writing – review & editing, Methodology,
deviations at higher speeds, the validation confirms the reliability of the Investigation, Formal analysis. R. N. Yerrawar: Writing – original draft,
mathematical model for low-speed applications, providing confidence in Software, Methodology, Investigation, Formal analysis, Data curation,
its use for further analysis and design. Conceptualization.

4. Conclusion Declaration of Competing Interest

Condition monitoring of rotating machinery typically employs The authors declare that there is no conflict of interests regarding the
techniques such as vibration analysis to assess the health and perfor- publication of this paper.
mance of equipment. Vibration frequency generally increases with ma-
chine speed, and amplitude rises correspondingly, with peaks at specific Data Availability
race frequencies indicating defects. Although defect frequencies remain
constant regardless of defect size or load, the amplitude is directly No data was used for the research described in the article.
proportional to both factors. Experimental results are often more reli-
able than theoretical predictions, as they better reflect real-world References
conditions.
[1] Zhihao Chen, Jian Cen, And Jianbin Xiong, “Rolling Bearing Fault Diagnosis using
However, this approach has limitations, including dependency on
Time-Frequency Analysis and Deep Transfer Convolutional Neural Network” IEEE
accurate data collection and interpretation, which can be challenging in Access, https://doi.org/10.1109/ACCESS.2020.3016888.
noisy environments or with complex machinery. Additionally, while [2] Yu Gang, A concentrated time-frequency analysis tool for bearing fault diagnosis,
modified BCF equations improved frequency accuracy, they may not IEEE Trans. Instrum. Meas. (2019), https://doi.org/10.1109/TIM.2019.2901514.
[3] S. Patil, Jose Mathew, P.K. Rajendra Kumar, Bearing signature analysis as a
account for all operational variations, potentially leading to medium for fault detection: a review, ASME J. Tribol. 130 (2008) 1–7, https://doi.
inaccuracies. org/10.1115/1.2805445.

12
K.H. Munde et al. Progress in Engineering Science 1 (2024) 100030

[4] M.M. Ibrahim, H.J. Nekad, Broken bar fault detection based on the discrete wavelet [17] Y.T. Sheen, Y.H. Liu, A quantified index for bearing vibration analysis based on the
transform and artificial neural network, Asian Trans. Eng. (ATE ISSN: 2221-4267) resonance modes of mechanical system, J. Intell. Manuf. 23 (2012) 189–203,
3 (2013) 1–6. https://doi.org/10.1007/s10845-009-0344-0.
[5] S.P. Harsha, K. Sandeep, R. Prakash, Non-linear dynamic behaviors of rolling [18] F. Cong, J. Chen, G. Dong, Research on the order selection of the autoregressive
element bearings due to surface waviness, J, Sound Vibr. 272 (2004) 557–580, modeling for rolling bearing diagnosis”, Proc. Inst. Mech. Eng. Part C: J. Mech.
https://doi.org/10.1016/S0022-460X(03)00384-5. Eng. Sci. 224 (2010) 2289–2297, https://doi.org/10.1243/09544062JMES1958.
[6] P.E. William, M.W. Hoffman, Identification of bearing faults using time domain [19] R. Yan, R.X. Gao, & X. Chen (2014). Wavelets for fault diagnosis of rotary
zero-crossings, Mech. Syst. Signal Process. 25 (2011) 3078–3088, https://doi.org/ machines: a review with applications. https://doi.org/10.1016/j.sigpro.2013.04.0
10.1016/j.ymssp.2011.06.001. 15.
[7] P.K. Kankar, S.C. Sharma, S.P. Harsha, Fault diagnosis of rolling element bearing [20] G. Tang, W. Hou, H. Wang, G. Luo, J. Ma, Compressive sensing of roller bearing
using cyclic autocorrelation and wavelet transform, Elsevier Neurocomput. (2012) faults via harmonic detection from under-sampled vibration signals, Sensors 15
1–9, https://doi.org/10.1016/j.neucom.2012.11.012. (10) (2015) 25648–25662, https://doi.org/10.3390/s151025648.
[8] R. Tomovic, V. Miltenovic, M. Banic, A. Miltenovic, Vibration response of rigid [21] A.R. Bhende, G.K. Awari, S.P. Untawale, Comprehensive bearing condition
rotor in unloaded rolling element bearing, Int. J. Mech. Sci. 52 (2010) 1176–1185, monitoring algorithm for incipient fault detection using acoustic emission,
https://doi.org/10.1016/j.ijmecsci.2010.05.003. J. Tribol. 2 (2014) 1–30.
[9] Rajesh Kumar, Divya Prakash Jena, Manpreet Bains, Identification of inner race [22] B. Attaran, A. Ghanbarzadeh, Bearing fault detection based on maximum
defect in radial ball bearing using acoustic emission and wavelet analysis, Proc. likelihood estimation and optimized ANN using the bees algorithm, J. Appl.
Isma (2010) 2883–2892. Comput. Mech. 1 (1) (2014) 35–43, https://doi.org/10.22055/jacm.2014.10547.
[10] Rajesh Kumar, Manpreet Singh, Outer race defect width measurement in taper [23] N. Hiremath, D.M. Reddy, Bearing fault detection using acoustic emission signals
roller bearing using discrete wavelet transform of vibration signal, Measurement analyzed by empirical mode decomposition, Int. Res. Eng. Technol. 3 (3) (2014).
46 (2013) 537–545, https://doi.org/10.1016/j.measurement.2012.08.012. [24] S. Singh, A. Kumar, & N. Kumar (2014, July). Detection of bearing faults in
[11] S.A. McInerny, Y. Dai, Basic vibration signal processing for bearing fault detection, mechanical systems using motor current signature and acoustic signatures. 21st
IEEE Trans. Educ. 46 (1) (2003) 149–156, https://doi.org/10.1109/ International Congress on Sound and Vibration (pp. 1-8).
TE.2002.808234. [25] E.T. Esfahani, S. Wang, V. Sundararajan, Multisensor wireless system for
[12] V.N. Patel, N. Tandon, R.K. Pandey, Defect detection in deep groove ball bearing in eccentricity and bearing fault detection in induction motors, IEEE/ASME Trans.
presence of external vibration using envelope analysis and Duffing oscillator, Mech. 19 (3) (2013) 818–826, https://doi.org/10.1109/TMECH.2013.2260865.
Measurement 45 (2012) 960–970, https://doi.org/10.1016/j. [26] S. Kim, D. An, J.H. Choi, Diagnostics 101: a tutorial for fault diagnostics of rolling
measurement.2012.01.047. element bearing using envelope analysis in MATLAB, Appl. Sci. 10 (20) (2020)
[13] W. Tu, Y. Shao, C.K. Mechefske, An analytical model to investigate skidding in 7302, https://doi.org/10.3390/app10207302.
rolling element bearings during acceleration, J. Mech. Sci. Technol. 26 (8) (2012) [27] Vibrations of Gears and Bearings, Diagnostic Maintenance and Monitoring of
2451–2458, https://doi.org/10.1007/s12206-012-0627-9. Machines, (Chap. 3, IIT Roorke), http://www.iitr.ac.in/outreach/web/CIRCIS/
[14] H. Wu, Q. Zhou, Z. Zhang, A. An, Vibration analysis on the rolling element bearing- UG/DMM/ Diagnostic%20 Maintenance%20and%20Monitoring%20of%20Machines-
rotor system of an air blower, J. Mech. Sci. Technol. 26 (3) (2012) 653–659, 3.pdf.
https://doi.org/10.1007/s12206-011-1201-6. [28] S. Ganeriwala (2010, February). Review of techniques for bearings & gearbox
[15] Y. Lei, J. Lin, Z. He, Y. Zi, Application of an improved kurtogram method for fault diagnostics. In 28th International Modal Analysis Conference (IMAC), Jacksonville,
diagnosis of rolling element bearings, Mech. Syst. Signal Process. 25 (2011) FL, USA.
1738–1749, https://doi.org/10.1016/j.ymssp.2010.12.011. [29] V.N.A. Naikan et al., Bearing defects of various types: rotating machinery fault
[16] Y. Zhang, H. Zuo, F. Bai, Classification of fault location and performance simulation, (Sakshat Virtual Labs, IIT, Kharagpur.), 〈http://iitkgp.vlab.co.in/?
degradation of a roller bearing, Measurement 46 (2013) 1178–1189, https://doi. sub〉= 40&brch=213&sim=b 739&cnt=1.
org/10.1016/j.measurement.2012.11.025. [30] Vibration Analysis - Level 2, (SKF Reliability Maintenance Institute, 2012).

13