Meccanica (2012) 47:335–354

DOI 10.1007/s11012-011-9434-0

An experimental testing of a simplified model of a ball

bearing: stiffness calculation and defect simulation
M. Dougdag · M. Ouali · H. Boucherit ·
N.E. Titouche · M. Djaoui

Received: 28 December 2008 / Accepted: 13 May 2011 / Published online: 17 June 2011
© Springer Science+Business Media B.V. 2011

Abstract This paper presents an experimental verifi- To verify experimentally the developed model of
cation of a simplified model of a nonlinear stiffness rigidity in both static and dynamic conditions, a num-
ball bearing in both static and dynamic modes and test- ber of compression tests were done on the ball bear-
ing its capabilities to simulate accurately fault’ effects. ings. This requires the specimen grips system adap-
Analytical model was developed using a differ- tation of a mechanical universal testing machine to
ent method comparatively to classical ones; the ball’s receive the non-usual specimen. A number of exper-
deformation is obtained without using Palmgren’s imental simulations of the main faults are done on a
method. Modeling considers the balls scrolling in the testing bench to verify the defect model.
cage and the effect of the load-rotating vector. Ob- Results of defects simulation and model behavior
tained formula allows introduction of defects charac- in statics and dynamics are compared to experimental
teristics under parametric forms. results. The developed model gives an acceptable sim-
These modifications were done in order to realize ilarity and it proves its simplicity and robustness. Both
two objectives. The first one is to ameliorate ball bear- results were acceptable.
ings stiffness computing and to improve a more realis-
tic simulation of dynamic behavior of a defective ball Keywords Experimental testing · Modeling · Fault’s
simulation · Radial stiffness · Mechanical testing ·
bearing. The second is to use obtained results in design
Testing bench · Design · Maintenance
and maintenance domain.
Abbreviations
α: angle, fault position,
M. Dougdag () · N.E. Titouche · M. Djaoui
COMENA/CRNB/DTN/Laboratoire des Etudes β: contact angle of ball with ring,
Mécaniques, CRNB, B.P. 180, Ain-Oussera, 17200, δi : deflection/deformation,
Willaya de Djelfa, Algeria δT : total deformation,
e-mail: [email protected] γ: train (cage) angular rotation,
M. Ouali
ε: radius of curve of ball’s housing,
Département de Mécanique, Faculté des Sciences Pour ϕ: angle /x, ball’s position /x
l’Ingénieur, Université Saad Dahlab Blida, B.P. 27, ϕn : gap angle between two balls,
Route de Soumâa, Blida, Algeria ν: Poisson’s ratio,
e-mail: [email protected]
θ: inner ring tilt angle /z,
H. Boucherit ω: shaft angular or spin speed,
COMENA/CRND, Draria, Algeria ωca : cage rotational speed,
336 Meccanica (2012) 47:335–354

: fault size, The ball bearing took importance since Palmgren

φ: load angular position, [22] work, Harris [14] and Eschmann [13]. Their pur-
ψ: angle, inner ring tilt angle /y pose was to describe the vibration transfer towards the
a: half contact width, structure. In fact, the ball bearings introduce several
b: width, sources of nonlinearities associated with the mechani-
c: damping coefficient, cal gaps and the Hertzians contacts between the rolling
db : ball diameter balls and balls raceways.
dou : outer ring diameter in ball housing, A simplified method consists to introduce radial
din : inner ring diameter in ball housing, stiffness springs, which are often isotropic; it is a con-
di : displacement /R, X, y and Z, centrated parameters model but this approach does not
dlog : ball housing diameter, consider the coupling between bending and tension-
dp : bearing pitch diameter, compression. Rajab [25] introduced the coupling and
h: gap between rings, Young [33] has included the axial load effect. Lim [18]
hr : local deformation depth, and De Mull [9] validated this trend.
Lc : contact width of the (ball/ring), Afterward, works have been oriented to take into
n: ball number, account the geometrical defects; the models stay lin-
m: loaded balls number, mass, ear Yhlan [32]. Choi [8] and Bourdon [2] adapted the
p: ball bearing perimeter, ball bearing model to the finite element formulation by
q0 : maximum distributed loading, introducing the ball elements and considering the ring
q(p): distributed loading along perimeter p, deformation.
q(r): distributed loading along r, The fault is simulated by a pulse signal or by vary-
q(z): distributed loading along z, ing mechanical properties. Adams [1] simulates fault
rm : bearing pitch radius, by adjusting the inner and/or outer raceway diameter
Ar : elementary area according to the axis r, for each element rolling on fault. Hoonbin Hong [15]
introduces a signal model that contains several sim-
Ei : Young’s module for (ball, ring),
plified terms modeled as a sinusoidal function, like:
Fi : force /r, t, x, y & z,
impulse series produced by the fault, the modulation
Fm : maximum force,
effect caused by the non-uniform load distribution,
Fr : radial force,
the bearing-induced vibration including decay, and the
Krr : radial stiffness,
mixture of unwanted signal components yield by other
M: mass, moment,
parts of the machine and by noise. Tkachuk [31] in-
NR: shaft speed
troduces a fault element formed by a field of contact
NT: cage speed
forces applied on outer/inner raceway in order to sim-
NB: ball speed
ulate fault in a 3D finite element model.
R, Rb : ball radius,
In this paper, the main outline of rigidity and faults
Ra : shaft radius,
modeling is presented. However, a detailed description
R1e : external inner ring radius, of deformation calculus and experimental verification
R2i : internal outer ring radius, of stiffness modeling and faults simulation are studied.
Rlog : ball’s housing radius, Analytical model describes the nonlinear dynamic
Rb : radius ball, behavior of the ball bearings by considering load level
T: cycle time, on balls, cage rotation (ball scrolling) and load rotating
W: load rotating vector. vector. This model uses:
(1) A technique of deformation calculus [10, 12],
1 Introduction (2) Rajab technique of stiffness matrix elements cal-
culus [25],
The mechanical behavior modeling of this device part (3) Load model, and
is mainly based on the stiffness calculation which is (4) Dynamic model.
obtained by the local [7, 29] and global [30] ball bear- The obtained model will lead to a formulation,
ing deformations under various loadings. which introduces the ball bearing defects characteris-
Meccanica (2012) 47:335–354 337

Fig. 1 Diagram of maximum force applied to the ball bearing

tics under parametric forms. Defect simulation is ob- The concepts and details concerning the geometri-
tained by handling stiffness formula. It can be formu- cal characteristics, physics and kinematics are found
lated by two methods. The first one, partial contact in [14, 20, 29].
method is based on elimination of the ball stiffness
that passes through fault while ensuring contact of the 2.1 Stiffness calculation
others balls with rings. The second one is to impose
an equivalent displacement, of the defect size, to the The stiffness evaluation is based on the Rajab’s model
[25] (Fig. 1). Thus, deformation and the corresponding
system response.
loads are calculated, in order to deduce the stiffness
To verify experimentally the developed model, two
into quasi-static mode.
devices are used: a mechanical universal testing ma-
Deformation in Fig. 2 due to the ball contact with
chine and a testing bench. The first one is used to ac-
the higher and lower raceways is calculated according
complish a number of compression experiences tests
to the Hertz law [19].
on the bearings, in sight of measuring experimentally
the rigidity. The machine requires adaptation of the 2.1.1 Rajab’s model
specimen grips system in order to receive the ball bear-
ing specimen correctly, as shown in Fig. 10. Modifica- Rajab [25] introduces the coupling between radial
tion consists to incorporate new objects on upper and displacement and rotation according to the bending
lower grip: a pressure bar in order to apply a compres- plane. Stiffness is calculated by resolving the nonlin-
sive force on the bearing and a clevis pins with a com- ear equation of the system under quasi-static equilib-
pensation ring, as shown in Figs. 10, 14 and 15. The rium (Fig. 1). A radial force (Fr) and a bending mo-
second device is used to execute three types of fault ment (Mb) load the inner ring, where the external ring
experimentation, on outer raceway, on inner raceway is fixed. Obtained rigidity is a matrix of order two as
and on ball. The main purpose is to verify the capa- follows:
bility of theoretical model to reproduce correctly the
fault effects in vibration mode. Krr = Fr /δr Krθ = Fr /θ and
Kθr = Mb /δr Kθθ = Mb /θ

Young’ [33] experiments validated Rajab’ proce-

2 Theoretical model
dure.
In the proposed model, the considered situation is
Different rolling bearings technologies are used in the defined when an equilibrium of dynamic forces is es-
rotating machines depending on the application. Due tablished in a stationary system as showed in Fig. 3.
to their complexity and their diversity, this study fo-
cuses only on deep-groove ball bearing. 2.1.2 Deformation’s model
The theoretical development aims to calculate stiff-
ness, which requires knowledge of contact scheme, Based on the Hertz law (Fig. 2(a)), a model of defor-
elasticity laws and an adequate loading model. mation calculus is proposed. The radial deformation
338 Meccanica (2012) 47:335–354

Fig. 2 (a) Local deformation according to the Hertz law [19]. (b) Local (hr ) and global elastic (δ) deformation of a symmetric part of
ball for the proposed deformation model

• The use of Timoshenko beam [30], according to the

Hook Law, allows modeling the ball compression
behavior under the resulting load (F ), knowing that
the ball has a local deformation (hr ) with a width of
(a). This gives elastic and global radial deformation
(δr) equation:
dδ F
= (2)
dr EA(r)
• Resolution of this equation in symmetrical domain
of r[0..(R − hr )] (Fig. 2(b)) gives total deformation
as [12]:
Fig. 3 Rotating coordinate system at equilibrium position
   
F db
δT = 2 ln − 1 + 2hr (3)
(Fig. 2(b)) is estimated by considering local and global πEdb hr
deformation.
Formula (3) includes the local and the global elastic
Local deformation is due to the ball contact with
compression of both sides of ball, supposed similar,
the higher and lower raceways and the global one is
and the same thing on both rings (Fig. 2(b)).
due to the elastic deformation of ball.
Equation (3) permits to adopt a method to construct
Deformation calculation is done as follows:
the curve of force-displacement f (δT ). We have for
According to the Hertz law [7, 29], local deforma-
each applied force Fi a corresponding local deforma-
tion of one side is given by (hr ) for a contact between (i) (i)
ball and ring: tion (hr ) using formula (1). Then (hr ) is introduced
 in formula (3) to obtain the corresponding total defor-
(i)
hr = R − R 2 − a 2 (1) mation δT and so on.

with: 2.1.3 Load’s model


3 3F [(1 − ν12 )/E1 ] + [(1 − ν22 )/E2 ] • Load distribution slice [−R, +R] (red and black
a= (1*)
8 (1/d1 ) ± (1/d2 ) color) in Fig. 4(a) encloses loaded balls before and
after the contact, it is expressed by:
where F , ν, E, d1 and d2 , are, respectively, the result- √
ing force, Poisson’s ratio, Young’s module, d1 ball’s R 2 − z2
diameter d2 ring’s diameter. q(z) = q0 (4)
R
Meccanica (2012) 47:335–354 339

Fig. 4 (a) Load distribution model on balls before contact (q(z)) and after contact (q(r)); (b) Balls radial loading model

• Variable change of z is R cos ϕ. Formula (3) allows to obtain the curve f (δT )
• The distributed load for the section [ϕi , ϕi+1 ] en- (force-displacement) as illustrated in Fig. 20. Formula
closing one ball only as illustrated in Fig. 4(b), is (9) permits to determine ball stiffness with using the
expressed as: curve f (δT ) as follows [10, 11]:
 Fr
q(ϕ) = q0 1 − cos2 (ϕ) = q0 · sin ϕ (5)
(b)
krr = db
(10)
Fr
2 πEd b
ln hr − 1 + 4hr
• The total loading Fi applied to ith ball is:
To obtain the total radial rigidity of ball bearing,
balls are considered loaded gradually (Fig. 5) and their
zi+1 ϕi+1
F i = Lc q(z)dz = Lc Rq0 − sin2 ϕdϕ positioning in cage is modifies due to the ball bearing
zi ϕi kinematics (scrolling), consequently, stiffness is varies
(6) in a nonlinear form.

• After the resolution of (6) for ϕi+1 = ϕi +ϕn = ϕi + Difference with Palmgren’s method The two meth-
2π/n and q0 = RL 2F
, Fi becomes: ods use the Hertz law but in the proposed method,
cπ
local deformation (hr ) is obtained directly by using
F geometry from contact sketch. This allows a different
Fi = [sin 2(ϕi + π/n) approach where stiffness is obtained step by step from
2π
curve f (δT ) [12].
− sin 2(ϕi − π/n) − 4π/n] (7)

• The radial force is defined by: 2.1.4 Dynamic’s model

Fri = |Fi | sin ϕi , (8) Load level and balls scrolling Figure 5 shows the
main positioning state of the balls for one rotational
• The law which links the loads to the deformations cycle and Fig. 6 illustrates the level of load applied on
is expressed as follows: each ball. Darker shades imply higher load. Cage rota-
tion with balls causes change of balls distribution; two
dFr cases appear case of one ball and case of two balls in
(b)
Krr = (9)
dδr the loading line.
In order to consider the following factors: loading
(b)
where Krr is the radial ball stiffness. level applied on the balls, the load mode (which covers
340 Meccanica (2012) 47:335–354

one-half of balls localized in dark area) (Fig. 6), the initial position of each ball, φ = ω · t + φ0 , where φ0
balls scrolling (with cage) and the load-rotating vector. is the initial position of rotating load vector and m is
The following formula is suggested [10]: the number of loaded balls (m ≈ n2 ).
m
Krr (γ ) = |Krr
(bi)
· cos(γi − φ)| (11) 2.2 Dynamic model
i=1

with γi is the ith ball angular position (Fig. 5), Theoretical dynamic calculation of the ball bearing,
(bi) under normal running conditions, uses a half-part of a
γi = γi−1 + ϕn with ϕn = 2π/n. Krr is the ball stiff-
real model (Fig. 7(a)) which is modeled by using one
ness, it is different from zero where γi is in the lower
degree of freedom (SDOF), as showed in Fig. 7(b).
half of bearing and equal to zero in the upper half as
The differential equation of motion has the follow-
showed by the Fig. 6, γ = ωca · t + γ0 , where γ0 is the
ing shape:

[M]{Ẍ} + [C]{Ẋ} + [K(t)]{X} = {f (t)} (12)

The stiffness K(t) can be expressed by:

K(t) = K max g(t) + K min (13)

with: g(t) is the a no dimensional and a nonlinear

function of the K which is varied over time and C is
the ball damping factor.
Indeed, the rigidity variation is characterized by
two states as show in Fig. 5. That makes a curvilin-
ear stiffness variation between the maximum and the
Fig. 5 Balls scrolling with cage (train) rotation (a & c) two
minimum value.
balls in the loading line, (b) one ball in the loading line
When the dynamic model and the total rigidity are
obtained, the nonlinear differential equation (12) res-
olution can be done by several methods. The Wilson-
Teta [23] and/or Newmark [17] step-by-step method
are chosen. With these methods, the time step T is
estimated by subdividing time between two successive
balls positions in the range of angle [ϕi to ϕi+1 ] by 100
steps or more as showed by Fig. 8. During one rotation
of the bearing cage, there are cyclic variations of the
nonlinear stiffness, where each cycle corresponds to
Fig. 6 Level of load applied on balls by sectors one angular section.

Fig. 7 Bearing dynamic model of SDOF

Meccanica (2012) 47:335–354 341

2.3 Defects modeling partial contact rupture parameter or switch (0, +1) and
additional stiffness switch (0, +1). These two param-
The ball bearings fault modeling is obtained by adding eters are activated when the alignment occurs between
some special parameters to (11). E.g., some defect the balls and the defect.
cases are presented in Fig. 9 as crack or spall, bump The example shown in Fig. 9(c) cannot be solved
(a small metal extrusion, impurity or scaling), or a as mentioned in formula (14). Therefore, it can be
hardness zone on the ring race. The size of defect is de- calculated by the following expression illustrates this
fined by , the time necessary to pass through the fault case [10].
is δτ and it occurs at each period τ (τ corresponds to
an angular position ατ ). (τ ) and (δτ ) are defined ac- Xi = Xi + ri (15)
cording to the bearing kinematics and shock’s law.
where ri simulates the imposed displacement, it de-
Most of these defects can be modeled by introduc-
pends on the defect position and size, as shown in
ing some parameters. E.g. to simulate crack, the rigid-
Fig. 9.
ity of the rolling ball over crack is removed. To simu-
In addition, (14) can be enriched and improved by
late the hardened zone, stiffness is added to the rolling
introducing corrector parameters on the geometrical,
ball over that zone.
mechanical and/or kinematics variables in basic for-
The parametric formula becomes as follow [10]:
mulas.
m
(D)
Krr (γ ) = δ1 (τ ) · [Krr
(b)
+ δ2 (τ ) · Kd (τ )] Comparison to the unreduced models
i=1
 Compared to other recent and unreduced models
× | cos(γi − φ)| (14)
[1, 15, 31], the suggested model is simple; it does not
with Kd (τ ) is a proposal additional stiffness resulting present a heavy and complex formulation. It is based
from hardened zone, δ1 (τ ) and δ2 (τ ) are respectively on a parametric approach that modifies the bearing’s
stiffness according to the simulated case of defect.
This model can also be integrated into a finite ele-
ments structure. Results are alike to those of reference
and transitively to the unreduced ones.

3 Experimental tools

3.1 Static and dynamic device testing

Fig. 8 Bearing rotation corresponding to one cycle of stiffness

The universal testing machine of type INSTRON 1185
calculation (Fig. 10) was used for static tests realization on the ball

Fig. 9 Characteristic ball bearing defects simulation examples ened zone, causes e.g. by an overheating, the stiffness becomes
[10]. (a) the ball rolls on a crack or spall inducing a partial con- more hard; (c) the ball rolls on a bump causing an imposed dis-
tact rupture followed by a shock [7]; (b) the ball rolls on hard- placement (ri ); (d) the roll of a ball having a spall or pit
342 Meccanica (2012) 47:335–354

Fig. 10 Diagram of a universal testing machine. Type INSTRON 1185 [12]

Fig. 11 Defect bench testing [3]

bearings (reference FLT ISKRA 6203Z). This bearing 3.2.1 Mechanical part
contains eight balls.
This device is composed of a 100 KN capabil-
The testing bench is made of an electric motor with
ity mechanical frame controlled by an electronic part
[12]. a power of 1 Kw, gear box offering a speed range
This installation allows a compression load acting [0 to 900, 3000 rpm], a shaft, a test ball bearing placed
on specimen between two grip heads and monitoring at the middle of the shaft to support the imposed load
of the speed when force is applied using electronic de- and an embedded support with a rubber sole.
vice [12].
3.2.2 Instrumental part
3.2 Defects simulation device testing
An analyser of type 2033 Brüel & Kjaer and ac-
For investigating defects simulation, a testing bench celerometer of type 4381 Brüel & Kjaer compose the
[3] is used (Fig. 11). This tool is composed of me- instrumental part. Measures of signals are obtained by
chanical and instrumental parts. the following instrumentation arrangement in Fig. 12.
Meccanica (2012) 47:335–354 343

Fig. 12 Instrumentation arrangement for measuring system response [3]

Fig. 13 Ball bearing total deformation measuring [12]

4 Experimental verification of model

4.1 Experimental conditions of static and dynamic

stiffness measuring

The stiffness measure of the ball bearing requires a

supported loading. Therefore, the loading speed must
be the weakest as possible. In case of these presents
test, the load is limited, according to SKF reference,
at 4000 N and the loading speed is taken equal to
0.05 mm/mn. This operation is executed by the mov-
able cross-head [12].
The ball bearing is gripped between the pressure
bar and the compensation ring. The test is launched Fig. 14 Loading cases [12]
when the contact between all parts is done. The to-
tal deformation is then measured on the ball bearing
(Fig. 13). line. The second one, the loading line passes between
two balls.
4.1.1 Testing with ball bearing
4.1.2 Testing without ball bearing
Figure 14 shows the assembly, which allows measure
of the whole system stiffness. Two cases are consid- In this case, the pressure bar directly comes to butt
ered. The first one, the ball is placed in the loading against the compensation ring. This assembly (Fig. 15)
344 Meccanica (2012) 47:335–354

Fig. 17 Defect on surface of outer raceway: small (a), medium

(b) and large (c) [27]

Fig. 15 Testing without ball bearing

Fig. 18 Defect on surface of the ball: small (a), medium (b)

and large (c) [27]

Fig. 16 Defect on surface of inner raceway: small (a), medium

(b) and large (c) [27]

allows the system stiffness measure without ball bear-

ing.

4.2 Experimental conditions of defects simulation

In this part, three types of defects were studied. De-

fects are put on the outer and inner raceway and on the
ball. In discrete manner, load variation is done using
dedicated disks [3].
Fig. 19 Electric marker pen and ball bearing of type 6205
FAG [3]
4.2.1 Experimental procedure and data

The experience consists to turn the motor with the

speed of 3000 rpm and a load of 150 N. Defects are
The defect is done artificially using electric marker
manufactured by one spall on the ball, the outer and
inner surfaces. Three sizes are chosen: (1 × 1 mm) pen (220 Volt and an adjustable electric current) [3]
small, (2 × 1 mm) medium and (3 × 1 mm) large [3] Fig. 19.
illustrated respectively in Figs. 16, 17 and 18. Fault’s
dimensions are obtained according to the limits of the
Electric marker pen (Fig. 19). 4.2.2 Ball bearing characteristics
The analyser measures signal with frequency span
of 5.63 kHz and time step of 61 µs for 3604 values per
signal and for 11 rotations of shaft on 0.22 s. Experiments of dynamic tests are done using the ball
To reduce noise effect, signals are measured five bearing type 6205FAG is loaded on radial direction,
times. Averaging signal and instrumentation are used where their characteristics are listed as follow accord-
to reduce random noise effect. ing to SNR’ mark (see Table 1).
Meccanica (2012) 47:335–354 345

Table 1 Ball bearing characteristics [3] the specimen grips system without bearing from the
SNR 6205 case with bearing (Fig. 15) [12].
Ball bearing with one row Globally, both results give the similar curves.
of balls and radial contact In detail, it is noted: In experimental curves, a small
interval figures for low displacements. This is ex-
Inner diameter 25 mm (+0, −10)µ plained by an existing gap in the machine and the bear-
Outer diameter 52 mm (+0, −13)µ ing. This gap is caught up during the machine starting
Pitch diameter 39 mm when compression forces are applied on specimen and
all elements comes into contact. This operation takes
Width 15 mm
a laps time called the period of adaptation.
Element number 9
Element 7.938 mm • A similarity is observed between experimental
diameter curves into the ball bearing working range.
Weight 128 g • A relative divergence of values at extreme loads
Maximum load Dynamic: 1400 daN; is observed. This difference can be explained by
Static: 790 daN (ISO) a non-consideration in the developed model of the
Velocity limit 12000–14000 rpm ball bearing material surface treatments undergone
in reality.

5.1.2 Dynamic testing

For a detailed comparison between dynamic curves

of stiffness, the relation of angle-displacement re-
lationship in quasi-static mode is used. It is based
on the establishment of a correspondence between
the displacement, which occurred under an applied
force, and the angular position of balls, as showed in
Fig. 21. Theoretical curve is obtained by changing the
ball’s position angle (φi ) for one cycle in the range
[(− π2 − ϕ2n ), (− π2 + ϕ2n )] (Fig. 8). Formula (8) and
(3) are used and rings are supposed not deformable.
Experimental results are obtained by measuring force
and displacement at three states as illustrated in Figs. 5
and 14.
The comparison between smoothed curves (Fig. 21)
shows an acceptable similarity between theoretical
Fig. 20 Comparison of the theoretical and experimental curves and experimental results.
(smoothing with polynomial curve of order two) A slight difference occurs at borders. It is due to
smoothing and singularities. It may be due also due
5 Results to the hypothesis or possibility of no consideration of
ring’s deformation.
5.1 Static and dynamic testing
5.2 Defects simulation testing
5.1.1 Static testing
In order to highlight experimentally the appearance
The theoretical and experimental results of the ball of defects and to compare them with analytical ones,
bearing behavior model in static mode are presented some experiments are realized with the previous con-
in Fig. 20. The experimental curve is obtained from ditions. Procedure consists to put an accelerometer on
averaging of three experimental curves carried out on the defected ball bearing housing in the vertical direc-
the same bearing and by deducing the deformation of tion of the shaft load; (at the top-left of the picture as
346 Meccanica (2012) 47:335–354

Fig. 21 Displacement change in function of ball’s position angle φi (smoothed curves)

Table 2 Defect’s frequencies

Frequency [Hz] BPFO BPFI FTF BSF RPM

Theoretical 178.6 271.4 19.8 116.01 50.0

Analytical spectrum 181.2 Hz 281.8 20.0 118.7 50.0
Experimental spectrum 175.0 295.0 – 176.0 50.0

shown in Fig. 11). One defect is put on one ball bear- Figures 22, 23 and 24 illustrate the emergence of ef-
ing. fect of one spall fault in acceleration response for each
The analytical conditions are an equivalent loading occurrence of contact over a faulted surface (located
mass of 7.65 kg, a modal damping ratio taken as 0.05 on outer or inner ring or ball). Results are illustrated
(5%), a speed of 3000 rpm and δ1 = 0 and δ2 = 0. by time signal history and spectrum.
All tests are done with one defect on the outer race-
way at φ = −90◦ (or 270◦ ) which corresponds to the 5.2.2 Fault on the outer raceway
maximum loaded zone (Figure 3), on the inner race-
way and on the ball surface. Analytical calculation is Figure 22(a) shows a time signal characterized by pe-
done for 100 subdivisions per cycle. Calculation us- riodical pulses and damped form depending on bear-
ing 30000 iterations includes 33 rotations of bearing’s ing’s mechanical properties.
train or 83 rotation of shaft. This method performs the Figure 22(b) illustrates the spectrum due to a fault
time step choice since it is based on geometry sub- on outer raceway. It is characterized by multiple har-
division and thus goes over the problem of calcula- monics of BPFO. Theoretical result shows similari-
tions convergence. The size of spectrum signal data ties with various references [6, 16, 21, 24, 26]. On the
is limited to 801 values and for the temporal signal other hand, the frequency rating 1×BPFO is 181.2 Hz.
between 2048 and 3000 values. Results are as fol- The results are similar for the theoretically and exper-
lows. imentally cases.
Comparison of results shows that time envelope
5.2.1 Results summary gives good correspondence between the theoretical re-
sults and the experimental results. In fact, good simi-
Table 2 shows that the results obtained from the ana- larity is found close to each other on the time axis.
lytical spectrum were closer to both the theoretical and Envelope spectrum comparison gives a similar ap-
the experimental ones. pearance of multiples harmonics of BPFO. However,
Meccanica (2012) 47:335–354 347

Fig. 22 (a) Time signal of defect on outer raceway; (b) Spectrum defect on outer raceway

a palpable difference exists in the order of highest defects), and at the initial stages of wear on these sur-
peaks or harmonics. faces. Higher harmonics, with multiples more than 10
The presence of multiple harmonic components to 20, are usually the consequences of shock pulses [6]
with biggest magnitudes at high frequencies can be
explained by the contribution of random vibration, ex-
5.2.3 Fault on the inner raceway
cited by friction forces and natural vibration compo-
nents excited by shock pulses.
On the other hand, experience has demonstrated Figure 23(a) shows a signal modulation illustrated by
that excluding unbalance, the magnitude of harmonic the black curve. The modulation is due to position
components below the fourth order average about variation of defect relatively to the most loaded ball
3–6 dB lower. Magnitudes typically increase with in- (Fig. 6) and it is associated to the relative inner ring
creasing load applied to the rolling surfaces (mounting speed as compared to cage’s one.
348 Meccanica (2012) 47:335–354

Fig. 23 (a) Time signal of defect on inner raceway; (b) Spectrum of defect on inner raceway

Figure 23(b) illustrates the spectrum due to a fault The time envelope comparison, gives globally cor-
on the inner raceway. It is characterized by a multiple rect correspondence between theoretical curve and
harmonics BPFI associated to its sideband harmonics experimental one; but a small intermediates wave-
around BPFI with a band of BPFI×k×RPM. This is forms rise between main ones. The reason is that the
due to the frequency of rotation of shaft, where k is an
proposed model cannot provide vibrations when the
integer number. Theoretical result has good similari-
faulted zone is unloaded.
ties with various references [4, 16, 21, 24, 26]. On the
other hand, the frequency rating 1×BPFI is 281.8 Hz. The envelope spectrum comparison gives similar
This result is the nearest to the theoretical and experi- appearance of multiples harmonics of BPFI and side-
mental result. band harmonics. However, two palpable differences
Meccanica (2012) 47:335–354 349

Fig. 24 (a) Time signal of defect on ball; (b) Spectrum of defect on ball

exist. The first one is the order of the highest peaks. cage. It also depends on the alternative contact of the
The second one is a slight shifting between curves. ball with the outer and the inner raceway.
The same reasons can be explained with the same Figure 24(b) represents the spectrum response of a
case already illustrated. bearing having one spall defect on the ball. It is char-
acterized by the harmonics of the ball fault frequency
5.2.4 Fault on the ball BFF (2×BSF) associated to its sideband harmonics
around BFF with a band of BFF×k×FTF. This is due
Simulation of one spall defect on ball is illustrated in to the frequency of train rotation. The spectrum is con-
Fig. 24. formed to author’s results [5, 21, 24, 26]. On the other
Figure 24(a) shows modulation of a signal due to hand, the frequency rating 1×BFF is 237.5 Hz and
the crossing of the defected ball through the loaded BSF is 118.75 HZ. This result is the nearest of both
and unloaded zones during the rotation of the bearing theoretical and experimental results.
350 Meccanica (2012) 47:335–354

Fig. 25 Two pulses due to chocks at leading and trailing edges of the cracks with different length on outer raceway (high resolution
of 1000 steps/cycle). (a) Zoom on the first pulse, (b) three pulses, (c) experimental illustration from the two pulses [28]

The envelope spectrum comparison gives a similar the two impulsions, the ball bearing stiffness, the ro-
appearance of multiples harmonics of BFS and side- tational speed and the ratio between ball diameter and
band harmonics but a palpable difference exists in the fault’s size.
order of highest peaks. Figure 25 shows that the relation between the mag-
The same reasons can be explained with the same nitude of signal and the size fault is not as simple as
case already illustrated. imagined. In fact, several factors influence this rela-
tion, as already described but it seems that the most
5.2.5 Fault evolution important parameter is the dephasing phase displace-
ment between the two impulsions. Because, when
Bearing fault is created experimentally. After several the two pulses are synchronized the shock magnitude
shaft rotations, the size of spall increases and its bor- increases. Globally, for a small fault width, magni-
der becomes rounded. The border round causes attenu- tude will increase as the defect develops. Hong [15]
ation of chocks and make their detection difficult. For gives some interpretations to this relation. Sawalhi
this cause, measures are taken after few rotations to [28] mentions that this second impact creates a new
prevent Choc attenuation [3]. Adams [1] and Rubini wave which interferes with the first, resulting in a
[27] also mentioned the rounding effect. phase shift (the gap) (Fig. 25(c)).
To study the fault’s evolution in the two models, Figures 26, 27 and 28 illustrate the effect of fault’
various sizes of spall are used (large, medium and size increasing by superimposing the spectra.
small). In addition, theoretical results show that the or-
Figure 25 shows that when the width of the crack der of the highest harmonics shifts from the high fre-
is small, the two impulsions at the ball crossing of its quency to the low frequency (decrease) when the fault
edges are confused. However, when the separation be- grows as illustrated in Fig. 26(a). This result is showed
tween the leading and trailing edges of the fault be- better in Fig 26(b). Also, the Figs. 27(a, b) of experi-
comes relatively significant, the input resembles two mental results confirmed this trend.
small separated impulses [1]. The presence of the sec- Comparison between analytical and experimen-
ond pulse modifies the original signal. This change de- tal results shows a similar evolution of magnitude,
pends on many parameters like the dephasing between when size increases. In fact, the largest fault size is
Meccanica (2012) 47:335–354 351

Fig. 26 (a) Analytical spectra superposition of various fault sizes on outer raceway. (b) Evolution of outer raceway fault’s size effect
on theoretical spectrum for three frequency ranges

characterized by increment of peaks at low frequen- cretization of ball bearing stiffness and reveals through
cies (green color). Faults with medium size appear at experimentation difficulties related to mechanical test-
medium frequencies (red color). Faults with smallest ing devices.
size appear at high frequencies (blue color). Nondi- During the static mode, the result obtained shows
mensional magnitude evolution gives the same report that the deformation’s model has been successfully
in Fig. 28. verified. Also on dynamic mode, stiffness has been
presented using curve of angle-displacement com-
pared to experimental one. Finally, the possibility to
6 Conclusion use the stiffness’ curve that was calculated in quasi-
static mode to emulate dynamic one, has showed that
The experimental study gives acceptable results. That the choice taken was acceptable.
permits to verify the developed model with a relative Stiffness modeling relatively reflects experimenta-
tolerance. It also allows experimental measure con- tion, particularly in the ball bearing working range.
352 Meccanica (2012) 47:335–354

Fig. 27 (a) Experimental spectra superposition of various defect sizes on the outer raceway; (b) Evolution of outer raceway fault’s
size effect on experimental spectrum for three frequency ranges

To study the case of the theoretical model capability fault effect and identify sources of experimental per-
to accurate simulate faults, the method of partial con- turbations.
tact was uses. It gives acceptable results. The minor In conclusion, the proposed model gives globally
problems that can occur are dues to different phenom- correct representation of the behavior of ball bear-
ena not modeled. The other suspected factor explain- ing in static and dynamic modes under several condi-
ing the disappearance of fault signatures in experimen- tions. It also proves its capability to accurately simu-
tal results is the attenuation of chocks due to roundness late faults, knowing the difficulties encountered to ac-
of spall’s border and size increase. complish experimentations.
These problems make difficult identifying defects An interesting relationship is observed between the
correctly. Complexity is that part of the bearing defect order of the highest harmonics and the fault size. In the
frequencies may be close to frequencies excited by other hand, other advantages can be mentioned, such
other components in the machine, hidden or distorted as possibilities to simulate different faults, types and
by various phenomena and therefore hard to identify. locations.
Here, these difficulties become advantage for the the- Although, the proposed model is smaller compar-
oretical model and can help signal analysis to isolate ing to unreduced ones, it is rather competitive for sim-
Meccanica (2012) 47:335–354 353

Fig. 28 Nondimensional magnitude evolution under fault’s size variation from the three ranges of modes (smoothing with polynomial
curve of order two)

ulation and can give better results when increasing its tion aux boites de transmission de puissance d’hélicoptères
degrees of freedom, including the housing structure Thèse de doctorat INSA de Lyon n◦ 93. ISAL 0034:224
and noise. 9. De Mul JM, Vree JM, Maas DA (1989) Equilibrium and as-
sociated load distribution in ball and roller bearings loaded
in five degrees of freedom while neglecting friction. Part I:
General theory and application to ball bearings. Trans of
References the ASME J Trib, 111,149–155
10. Dougdag M, Ouali M (2001) Introduction de défauts dans
1. Adams ML (2010) Analysis of rolling element bearing l’étude des machines tournantes, thèse de magister, soutenu
faults in rotating machinery: experiments, modelling, fault juin 2001, Université de Blida
detection and diagnosis. Thesis for Doctor of Philosophy 11. Dougdag M, Ouali M (2006) Modélisation d’un Roulement
in the case Western Reserve University, August 2001. pp à billes et de ses Défauts Etude Comparative, XV colloque
12–13, 25–28, 50–53, 74–108, 147–183 Vibrations Chocs & Bruit, Lyon France du 14, 15 et 16 Juin
2. Adelin B Modélisation dynamique globale des boites 2006
de vitesses automobiles, thèse de doctorat à INSA de Lyon, 12. Dougdag M, Ouali M (2008) The calculation of ball bearing
1997, n◦ 97 ISAL 0084, pp 23–42 nonlinear stiffness theoretical and experimental study with
3. Bouchrit H (2007) Analyse expérimentale et numérique comparisons. J Eng Appl Sci 3(11):872–883
des défauts de paliers à roulements pour le diagnostic des 13. Eschmann P, Hasbargen L, Weigand K (1985) Ball and
machines tournantes. Thèse de Magistère, Ecole Nationale roller bearings: Theory, design and application. Olden-
Polytechnique El Harrach, Alger 2007. pp 28, 54–59, 96 burg/Wiley, Munchen
4. Barkov V, Barkova NA (1996) Non-linear signal models in 14. Harris TA (1966) Rolling bearing analysis, 3rd edn.
vibroacoustic machine diagnostics Lavoisier, 1991
5. Barkov V, Barkova NA (1999) The artificial intelligence
15. Hong H, Liang M (2009) Fault severity assessment for
systems for machine condition monitoring and diagnostics
rolling element bearings using the Lempel–Ziv complex-
by vibration. In: Proceedings of the Saint Petersburg post-
ity and continuous wavelet transform. J Sound Vib 320:
graduate institute of the Russian Federation power industry
452–468
and vibration institute, USA, Saint Petersburg, vol 9
6. Barkov A, Barkova N Condition assessment and life pre- 16. Konstantin-Hansen H, Herlufsen H (2010) Envelope and
diction of rolling element bearings—Part 1, Vibroacousti- cepstrum analyses for machinery fault identification. Brüel
cal systems and technologies. St Petersburg, Russia, Edited & Kjaer, www.SandV.com, Sound & Vibration, May 2010,
by John S Mitchell, Contributing Editor for Sound & Vi- 1–10
bration 17. Ladevèze P, Pelle J-P (2005) Mastering calculations in
7. Bourgain L, Dart R, Bourgain J (1988) Machines tour- linear and nonlinear mechanics. Springer, Berlin, p 250
nantes et circuits pulsés. In: Applications Industrielles et (Translated by Theofanis Strouboulis, Mechanical Engi-
médicales de l’analyse spectrale, pp 392–411, Dunod, Edi- neering Series Frederick F Ling)
tion Bordas 18. Lim TC (1989) Vibration transmission through rolling ele-
8. Choi IS (1993) Simulation des mécanismes complexes en ment bearings in geared rotor systems. The Ohio State Uni-
CMAO étude des non linéarités de comportement applica- versity, pp. 12–60
354 Meccanica (2012) 47:335–354

19. Louf F (2003) Contact: théorie de Hertz. Site Internet: 27. Rubini R, Meneghetti U (2001) Application of the enve-
http://agregb1.dgm.enscachan.fr/Documents/TheorieTP/ lope and wavelet transform analyses for the diagnosis of
files/Contact.pdf, pp 1–7 incipient faults in ball bearings. Mech Syst Signal Pro-
20. Moret M (1995) Roulements et butés à billes et à rouleaux. cess 15(2):287–302. doi:10.1006/mssp.2000.1330. Avail-
Techniques de l’ingénieur, vol B5-I, Août-1995 able online at http://www.idealibrary.com
21. Muthukumarasamy A, Ganeriwala S (2009) Diagnosis 28. Sawalhi N, Randall RB (2008) Simulating gear and bearing
of rolling element bearing faults using envelope analy- interactions in the presence of faults Part I. The combined
sis http://www.spectraquest.com/. Tech Note, SpectraQuest gear bearing dynamic model and the simulation of localised
Inc pp 1–11 bearing faults. Mech Syst Signal Process 22:1924–1951
22. Palmgren A (1959) Ball and roller bearing engineering, 3rd 29. Shigley JE, Mitchel LD (1983) Mechanical engineering
edn. Burkank design, 4th edn. McGraw-Hill, New York, pp 85–88,
23. Paz M (1985) Structural dynamics. Theory and computa- pp 484–514
tion, 2nd edn. van Nostrand Reinhold Company, New York, 30. Timoshenko S (1968) Résistance des matériaux. Théorie
pp 389–397 élémentaire et problèmes, vol I, collection Dunod tech-
24. Piersol AG, Paez TL (2010) Harris’ shock and vibration nique, pp 2–19
handbook, 6th edn. McGraw-Hill, New York (2010, 2002, 31. Tkachuk P, Strackeljan J (2010) A 3D-ball bearing model
1995, 1988, 1976, 1961 by The McGraw-Hill, Printed for simulation of axial load variations. In: The seventh in-
in the United States of America, pp (16-8)–(16-24) and ternational conference on condition monitoring and ma-
(24.14)–(24.16)) chinery failure prevention technologies
25. Rajab MD (1982) Modelling of the transmissibility through
32. Yhland E (1992) A linear theory of vibrations caused by
rolling-element bearing under radial and moment loads.
ball bearings elasticity in gear at moderate speed. Trans
PhD thesis. The Ohio state university. Columbus. Ohio
ASME J Tribol 154:348–359
26. Roque AA, Silva TAN, Calado JMF, Dias JCQ (2008)
33. Young WB (1988) Dynamic modelling and experimental
Rolling bearing fault detection and isolation—a didactic
measurement of gear shaft and housing system. MS Thesis,
study. In: 4th WSEAS/IASME international conference on
The Ohio state university, Columbus, Ohio
educational technologies (EDUTE’08). Corfu, Greece, Oc-
tober 26–28, 2008