You are on page 1of 12

Mohsen Nakhaeinejad


Dynamic Modeling of Rolling Element Bearings With Surface Contact Defects Using Bond Graphs
Multibody dynamics of healthy and faulty rolling element bearings were modeled using vector bond graphs. A 33 degree of freedom (DOF) model was constructed for a bearing with nine balls and two rings (11 elements). The developed model can be extended to a rolling element bearing with n elements and 3 n DOF in planar and 6 n DOF in three dimensional motions. The model incorporates the gyroscopic and centrifugal effects, contact elastic deections and forces, contact slip, contact separations, and localized faults. Dents and pits on inner race and outer race and balls were modeled through surface prole changes. Bearing load zones under various radial loads and clearances were simulated. The effects of type, size, and shape of faults on the vibration response in rolling element bearings and dynamics of contacts in the presence of localized faults were studied. Experiments with healthy and faulty bearings were conducted to validate the model. The proposed model clearly mimics healthy and faulty rolling element bearings. DOI: 10.1115/1.4003088 Keywords: rolling element bearings, modeling, multibody dynamics, bond graphs, rolling contacts, faults, dents, pits

Michael D. Bryant
e-mail: Department of Mechanical Engineering, University of Texas at Austin, Austin, TX 78712


Under high load rates and speeds, bearings are susceptible to faults and failures. Bearing faults comprise 41% of failures in induction motors 1. Early detection and precise isolation of bearing faults can decrease maintenance costs and increase machine safety 2. Diagnostic techniques for bearings can be either signal based or model based. Signal-based methods key on specic features of waveforms or spectra; however, since designs, manufacture, process dynamics, and operating conditions vary with machines and change over time, features used in signal-based techniques can vary markedly, making these techniques unreliable 3. Model-based methods, which gather more information and sort via a model, can overcome disadvantages of signal-based methods. Studies on diagnostics of rolling element bearings and rotating machinery has shown model-based diagnostic techniques are superior to signal-based methods for early detection and more accurate isolation of faults. However, these techniques are not adequately developed for real applications and require the support of detailed modeling. Several modeling software and virtual design tools have been developed 4. However, at present, there is no universal and detailed platform for the whole system. Many available models represent dynamics of rolling element bearings, but a generic modular model, which can be adjusted based on the complexity of the system to represent dynamics of both normal and defective bearings, is not available yet. Despite the long history of dynamic modeling of rolling element bearings, relatively few models consider faults. Most simplify models or neglect the details of faults. In 1960, Jones 5 introduced bearing element stiffness, damping, constraint forces, and moments. In 1979, Gupta 6 introduced a six degree of freedom DOF model of a bearing and validated the model experiContributed by the Tribology Division of ASME for publication in the JOURNAL OF TRIBOLOGY. Manuscript received June 24, 2010; nal manuscript received November 17, 2010; published online December 15, 2010. Assoc. Editor: Ilya I. Kudish.

mentally. In the early 1980s, McFadden and Smith 7 incorporated localized defects of bearings using impulse functions. Assuming a rotor-bearing system with a stationary outer ring, Afshari and Loparo 8 introduced a one dimensional linear time invariant state space model. Adams 9 presented an analytical 29DOF model of a shaft supported by two rolling element bearings using Lagranges approach, assuming only the radial motion of the balls with no contact slip. Harsha 10 developed an analytical model using Lagranges approach, assuming no slip and no friction between balls and races. Incorporating gyroscopic effects and shaft bending, El-Saeidy et al. 11 formulated dynamics of rotor-bearing systems using nite elements and Lagranges equations. In 2002, Liew et al. 12 obtained a 5DOF analytical model of angular contact ball bearing, including bearing centrifugal loads and axial and tilting stiffness effects. In 2003, Sopanen and Mikkolas model of a deep-groove ball bearing included localized and distributed defects, Hertzian contact deformations, and elastohydrodynamic EHD effects 13,14, but neglected centrifugal effects and slip between components. Analytical models by Jang and Jeong 15 and a 5DOF model by Changqing and Qingyu 16 considered the waviness faults neglecting contact slip and contact separations. Patil et al. 17 studied the dynamics of defects in bearings using a planar model neglecting slip, centrifugal force, and gyroscopic moments. Wensing 18 developed a nite element model of rolling element bearings. Sassi and Badri 19 developed a simple 3DOF model of ball bearings and introduced a numerical model to predict damaged bearing vibrations. Karkkainen et al. 20 studied dynamics of rotor systems with misaligned retainer bearings. Ashtekar et al. 21 used a dry contact elastic model to modify the Hertzian relationships for contacts with small surface faults. Also, they used superposition to include the effects of dents or bumps smaller than the Hertzian ellipse on bearing dynamics 22, assuming the contact stiffness K is unaffected by the presence of faults. Little has been done in bond graph modeling of rotating systems 2,23,24, including rolling element bearings, multibody dynamics, tribological aspects, and localized defects. Bond graphs JANUARY 2011, Vol. 133 / 011102-1

Journal of Tribology

Copyright 2011 by ASME

Downloaded From: on 05/12/2013 Terms of Use:

are energy-based modeling techniques, with modularity that permits system growth. Different submodels can be integrated into a large system. Despite the apparent simplicity of rolling element bearings, the presence of faults complicates dynamics. Most existing models cannot fully explain the dynamic behavior of bearings with faults since simplied models do not fully contain details of the fault. Here, the tribological aspects of rolling element bearings, nonlinear contacts, and surface defects are modeled in detail separately, and integrated into a unied bond graph model. This article presents a detailed model of rolling element bearings REBs with a direct physical correspondence between parameters of the model and physical system components and faults. The goal is a generic model of rolling element bearings, which can be simple or detailed, based on the complexity of the system, needed to support model-based diagnostics. First, multibody dynamics of bearings, including elastic deections with nonlinear Hertzian contacts, tractions with contact slips, and surface separations, are modeled in a generic bond graph. Localized faults, such as dents and pits, are modeled as surface prole changes and incorporated into the bond graph. The model is validated with experiments, and then the effects of parameters, such as geometry of faults, clearance, and radial loads on vibration responses, are investigated. Finally, dynamics of contacts with faults are studied in detail.

position vector Ri = Xi Yi ZiT and i, respectively. For each contact between two elements i and j, a moving contact frame xij y ijzij with origin at the contact is dened with orientation ij described in the global coordinate as

ij = arctan

Yi Yj Xi Xj

Any vector X dened in the global frame can be transformed to a in the moving frame via vector X = TX X 2

where transformation matrix T for the body coordinate systems is Ti = sin i cos i 0 0 0 1 and for the contact coordinate systems is T = sin ij cos ij 0 0 0 1

cos i

sin i 0

cos ij

sin ij 0



Contact Modeling

A deep-groove ball bearing consisting of balls, inner race, and outer race was modeled as a multibody system using vector bond graphs. Nonlinear contact stiffness, contact damping effects, contact slip, contact separations, bearing clearance, traction between elements, rotational friction torques, and localized faults were included. The axial motion, lubrication effects, structural exibility except at contacts, and dynamics of the cage were neglected. A 33DOF bond graph model was constructed for a bearing with nine balls and two rings 11 bodies. The model can be extended to any n-body rolling element bearing with 3 n-DOF in planar motions and 6 n-DOF in 3D motions. 2.1 Geometry Considerations. Assuming planar motions, each element of the rolling element bearing has 3DOFs. Figure 1 shows a three-element rolling system representing inner race a, ball b, and outer race c in contact, with global xed coordinate system XYZ and rotating body coordinates xiy izi dened for each moving element, and coincident with the principal axes of moving elements. Misalignments and eccentricity can be included by relocation of inner race and outer race centers. Position and orientation of each element can be described in the global coordinate system by a

2.2.1 Contact Stiffness and Damping. According to Hertzian theory 25, the load-deection relation in the x direction for the contact between the two bodies with different radii of curvature in a pair of principal planes, as shown in Fig. 2, is F = K 3/2 5 where the contact force F is in N, the deformation is in mm, and the contact stiffness K is in N / mm2/3. The contact stiffness K is dened as K= 3


1 2

where E is the modulus of elasticity in MPa, is the curvature in 1/mm dened as = 1 / r, and the curvature sum is

= r


1 1 1 + + r a2 r b1 r b2

Subscripts a and b refer to bodies a and b and subscripts 1 and 2 refer to the bodies principal planes, in which curvatures are dened for each body. In rolling element bearings, includes main curvatures of ball and raceway, as well as side groove curvatures,

c b a

Plane 1
2 ne P la

yab y

y b xab B


X y





Body b Body a
Plane 1




ra 2


ne Pla 2

Fig. 1 Geometry and coordinates of a three-body rolling element system

Fig. 2 Geometry of two curved bodies a and b in contact under the normal force F

011102-2 / Vol. 133, JANUARY 2011

Transactions of the ASME

Downloaded From: on 05/12/2013 Terms of Use:

considering the sign convention for convex and concave surfaces. In Eq. 6, is a dimensionless deection factor graphed by Harris 25 as a function of curvature difference F dened as

F =

1 1 1 1 + r a1 r a2 r b1 r b2

, s = f , F = Fy = bdK3/2 + b Fz 0


K 3/2 + b


2.2.3 Bearing Friction Torques. In rolling element bearings, contact deformation, viscous lubrication, and seal rubbing generate rotational resistive torque, T f = Tload + Tlub + Tseal 15 where Tload is the torque caused by bearing radial loads and contact deformations. For a ball bearing under radial load, Tload has been formulated 25 as Tload = 0.0003

Note that the order of planes 1 and 2 should be selected in such a way that F is positive. For computational applications, the curvature function F is formulated as 26

= 327.61 + 1883.34F 3798.11F2 + 3269.62F3

1026.96F4 Since contact damping in ball bearings is relatively small, a constant damping value, based on the average load, can be used in the model. Kramer 27 estimated bearing damping by frequency analysis of the equivalent linear spring-mass-damper system as b = k0.25 2.5 102 N s/mm 9

Fs Cs


F d p


where Cs is the bearing static load rating dened by the bearings manufacturer, Fs is the static equivalent load dened as Fs =

0.6Fr + 0.7Fa Fr

F r F a F r F a


where k is the equivalent linear stiffness of the bearing in N/mm. Here, b is selected as b = 102k N s/mm. Therefore, the normal contact force including the damping force along xab is F x = K 3/2 + b 10

and F depends on the magnitude and direction of the applied loads, which equals to Fr in radial ball bearings under pure radial loads. The lubricant friction torque, Tlub, is dened by empirical equations 30 as Tlub =

is the rst derivative of the contact deformation in mm/s. where Note that when faults are smaller than the Hertzian contact ellipse, the Hertzian contact relationships should be modied for more accurate results 21,22. Here, it is assumed that Hertzian contact relationship is valid for contacts with faults. 2.2.2 Traction Model. Based on the rolling or slip regions, the traction model follows different behaviors with different coefcients of friction. Kragelsky et al. 28 described the dry contact coefcient of the friction as a function of sliding speed as

where f r is the rotational speed in rpm, o is the kinematic viscosity of the lubricant in cSt, and coefcient f o, which depends on the bearing type and lubrication, equals unity for deep-groove ball bearings lubricated with oil mist and is given in Ref. 25 for other types of bearings with different lubrications. The rubbing seal friction torque, Tseal, which depends on the geometry and design of the seal, is assumed constant. 2.3 Fault Modeling. Bearings fail by numerous fault modes: corrosion, wear, plastic deformation, fatigue, lubrication failure, electrical damage, fracture, and incorrect design, among others. Most common spalling fatigue leaves pits on races or rollers, as in Fig. 3a, because of periodic contact stress 31. Most models for rolling element bearing faults have introduced mathematical impulse functions based on fault frequencies. Here, a kinematicsbased fault modeling is introduced, where the faults are dened in the model based on the surface prole change. For each fault, the width w, depth h, and location of the fault f in the body coordinate system are dened, as shown in Fig. 3c. Parameters T f , w, h, r f , and f in the model dene the type, size, shape, and location of localized faults, as shown in the block diagram of Fig. 4. T f denes the type of the fault, w and h represent size, r f determines the shape of the fault, and the location of the fault on each element is determined by f in the model. As a fault passes through the contact, the prole changes induce deections that result in dynamic interactions between elements, which generate force impulses that induce fault vibrations. Multiple faults of different sizes and shapes, at any location of the bearing, can be modeled via the proposed scheme.
w h

107 f oo f r2/3d p3, 16 10 f od p ,

6 3

o f r 2000 o f r 2000


= C1 + C2s2es sign s + C2s

where C1, C2, and are constants obtained experimentally based on the materials and surfaces. Here the traction force is modeled using traction-slip behavior in elastohydrodynamic contacts 29. Traction forces in directions y ab and zab are functions of the normal force and the friction coefcient as F y = Fx 11

The friction coefcient can be calculated based on the contact lubrication modes, determined by the oil lm parameter a ratio of oil lm thickness and contact surface roughness 30 as

bd, bd hd 1.56 + hd, = 0.01 1.66 hd,

0.01 0.01 1.5 1.5


Friction coefcient hd under hydrodynamic lubrication mode is a function of EHD parameters and temperature. Under boundary lubrication mode, friction coefcient bd is a function of slide-roll ratio s as

bd = 0.1 + 22.28se181.46s + 0.1

(a) (b) (c)

Therefore, assuming planar motion Fz = 0, the contact force vector, consisting of stiffness, damping, and traction forces as a func , and slip s, is given by tion of contact , Journal of Tribology

Fig. 3 a Subsurface-initiated fatigue spall 31, b outer race fault model, and c cross section of an outer race fault

JANUARY 2011, Vol. 133 / 011102-3

Downloaded From: on 05/12/2013 Terms of Use:

Fault Type Tf =1 IRF 2 BF 3 ORF

Fault Size wh [mm2]

Fault Shape rf = w/h

Fault Position 0 < f <2 [rad]

junction term representing gyroscopic forces. A cross product C B can be represented in matrix form using the cross product operator A as C B = C y B y = Cz Cy Cz Bz

Fig. 4 Four parameters Tf, w, h, rf, and f represent the type, size, shape, and location of faults in the model

Cx Bx

Cz 0 Cx

Cy Cx 0


B y = C B By 21

Bond Graph Modeling

Applying this to the second term, Eulers equation becomes =+ J r F J

i i 1

Bond graphs, introduced in the 1950s by Paynter 32, map energy storage, energy dissipation, and power ow between components in a system through energy bonds and elements 33. Associated with each bond are power conjugates, effort, and ow, the product of which is the instantaneous power owing to or from physical elements. Efforts and ows in mechanical systems are forces and velocities. Energy dissipation is modeled by the resistive element R, energy storage devices such as springs and inertia are modeled by elements C and I, sources of efforts and ows are represented by Se and Sf, and power transformations are modeled by transformers TF or gyrators GY. Power bonds join at 0 junctions summing ows to zero with equal efforts or at 1 junctions summing efforts to zero with equal ows, which implements Newtons second law. With constitutive laws for each element such as Eqs. 5 and 10, state equations are extracted from bond graphs. Bond graphs have many advantages. Multiphysics dynamic systems, such as electrical, mechanical, magnetic, uid, chemical, and thermodynamic systems, can be modeled and linked together. Also, nite element model FEM can be embedded in model 11. Furthermore, the modularity characteristic of bond graphs permits system growth. 3.1 Multibody Dynamics With Bond Graphs. The dynamics of a 6DOF rigid body system with the global coordinate XYZ and the body coordinate xyz, as shown in Fig. 5a, can be represented with NewtonEulers equations, = Mx


External forces in the global coordinates can be transformed to the body coordinates as Fi = TGBFi 23

where TGB is the transformation matrix obtained from Euler angles or direction cosines 34. Rewriting the NewtonEuler equations 19 and 22, = Mx

1 i i

J =+ where



ri TGB Ai =


1 i


=+ J

r F J
i 1


expressed in the global and the body coordinate systems, respectively. , also The body is subjected to external forces Fi and moment expressed in the global and the body coordinates, respectively. The second term on the right side of Eulers equation 20 represents moments due to external forces. The last term is the Eulerian

In terms of vector bond graphs, as shown in Figs. 5b and 5c, Newtons and Eulers equations can be represented as 1 junctions that embed conservation of linear momentum and conservation of angular momentum. Transformers TF with modulus of Ai transform coordinates between Eulers and Newtons equations and also convert external forces to moments acting on the center of i are represented by sources of efforts mass. External moments Se. The gyroscopic term J is incorporated as a modulated RE element. The overall bond graph structure of the system in Fig. 5a with dynamic equations 24 is the diamond-shaped bond graph structure of Figs. 5b and 5c. A multibody system consisting of n rigid bodies can be modeled by connecting n diamond-shape bond graph structures together through constraint models. 3.2 Bond Graph Model of Rolling Element Bearings. A deep-groove ball bearing consisting of nine balls, inner race, and


Y F1



Eulers Eq. in body coordinate I: J

:A TF 1



r1 z

TF: A 2 0










1 I: M

Newtons Eq. in global coordinate

Fig. 5 Dynamics of a rigid body under external loads are represented with a diamond-shape vector bond graph model

011102-4 / Vol. 133, JANUARY 2011

Transactions of the ASME

Downloaded From: on 05/12/2013 Terms of Use:

Kh Rh


Ft_ib Fk_ib Fb_ib


Ft_bo Fk_bo Fk_oh



F Tf




outer race

Fk_bo Fb_bo Ft_bo

inner race

inner race &i



balls contact b/o
Cc RE :Tbo
:A bF T

outer race
:A TF o

Fk_ov Fb_ov

contact i/b
RE :Tib Cc
:A TF b


& b
1 I:Jb I:Mb

& o

1 I:Jo


I:Ji I:Mi

TF 0 0

TF 0 0

Eulers Eqs. in rotating frame

:A i TF

I:Mo :Kv,h C 1 :Rv,h R



Newtons Eqs. in global frame

Fig. 6 Rolling element bearing with single roller: a bearing with housing, b system with external loads and constraints, c free body diagram, and d bond graph models of each element and contacts

outer race is modeled as a multibody system using vector bond graphs. The outer race is xed in a housing characterized by stiffness and damping in the vertical and horizontal directions. The inner race moves and rotates under external forces and torques Fig. 6a. Weights are applied as external loads on each body. Each element is modeled using the diamond-shaped vector bond graph structure of Fig. 5. Contacts are modeled as nonlinear C elements representing nonlinear stiffness, damping, and traction forces inherent in Eq. 14. The bond graph model of a bearing with races and one ball, in Fig. 6d, has three of the diamond-shape bond graph structures with two contact models in between. Each diamond structure applies NewtonEulers equations 24 to inner race, balls, and outer race via the 1 junction. Each has rotational inertia J, translational inertia M, transformer TF with modulus matrix A transforming body coordinate system to the global coordinate system Eqs. 2325, and Eulerian junction elements RE representing gyroscopic moments. External torque and external force vector F are applied to the 1 junctions through Se elements. Bearing rotational friction torques are modeled as resistive elements R in the inner race model, with the constitutive law described by Eqs. 1518. Housing stiffness and damping effects are modeled through C and R elements with structural stiffness and damping matrices Kv,h and Rv,h. Each contact model consists of a transformer and a Cc element. Transformers TF transform the contact coordinate systems to the global coordinate system, as expressed in Eq. 2. Nonlinear elements Cc with the constitutive law of Eq. 14 represent contact and traction forces. Here, the bond graph model of the bearing with nine balls and two races consists of 11 diamondshape models combined through 18 contact models. Faults are modeled by surface prole changes, as shown in Fig. 3c. During simulations, the geometry of elements changes based on locations and growth of faults. Fault impulses and forces are generated via kinematics and dynamics.

The effects of the clearance and radial loads on the load zone distribution were studied. Three cases of clearance and two cases of radial loads were simulated. Healthy and faulty bearings with localized faults on inner race, outer race, and balls were simulated. For each type of fault, 16 cases with different fault sizes Table 2 were simulated to study the effects of fault type, size, and shape on the vibration response. For each case, the vertical vibration signal of the housing in Y direction of Fig. 6c will be presented. Also, normal contact force and displacements in the presence of ball faults are presented. For a bearing with stationary outer race and rotating inner race under downward vertical load F, the loading zone is stationary. In the load zone, reaction forces transfer load F between housings and shaft. Based on the angular position of the outer race fault ORF, the vibration response can change. However, in the case of inner race fault IRF and ball fault BF, fault locations do not

Table 1 Bearing specications and simulation parameters Parameter Bearing model Number of balls Material density g / cm3 Elastic modulus MPa Ball/inner race/outer race diameter mm Pitch diameter mm Race groove radius mm Poissons ratio Radial load on inner race N Inner race rotational speed Hz Housing vertical/horizontal stiffness N/mm Value MB ER-16K 9 7.75 210 7.94/31.38/47.26 39.32 4.1 0.25 110 35 4 107

Table 2 Studied faults

Simulations and Experiment Setup

faults size: Fault Level I Fault Level II Fault Level III
w = 5h 2.250.45 3.150.63 6.301.26 w = 2h 1.000.50 1.420.71 2.821.41
w h [mm

A bond graph model of a deep-groove ball bearing with specications in Table 1 was built and numerically simulated with the software 20-SIM 4.0. The RungeKutta Dormand Prince integration method with variable step size was used in simulations. Initial velocities and accelerations were zero. A radial load was applied to the inner race vertically downward. Running torque and external rotational loads were applied to the inner race. Journal of Tribology

w=h 0.710.71 1.001.00 1.411.41

w = 0.5h 0.220.44 0.320.64 0.450.90

JANUARY 2011, Vol. 133 / 011102-5

Downloaded From: on 05/12/2013 Terms of Use:

Data acquisition board

vibrations & speed

balance disks


Signals are analyzed in VibraQuest and MATLAB

bearing loader studied bearing

Fig. 7 Test bed for bearing experiments

affect the response since faults are not stationary with respect to the load zone. Here, the model is simulated with ORF located down at the load zone Experiments were conducted on the test rig shown in Fig. 7 with three types of rolling element bearing faults IRF, BF, and ORF. The test rig consists of a rotating shaft supported by two rolling element bearings Table 1. A three-phase induction motor rotates the shaft through a beam coupling. Bearing loaders apply static radial loads to bearings, and a magnetic loader applies rotational loads to the shaft through a gearbox and belts. Accelerometers measure the vertical and horizontal vibrations of the outboard bearing housing. Eddy current proximity probes measure the vertical and horizontal displacements of the shaft close to the bearing. A tachometer and an encoder measure shaft speed. Also, voltages and currents of the induction motor can be measured. Localized faults were created on inner race, outer race, and balls by spot grindings. A vertical load was applied to the inner race via the bearing loader. Since excessive vibrations can invalidate results, waterfall plots and hammer tests found machine critical speeds and natural frequencies, which were avoided during tests. At the shaft speed of 35 Hz, bearing vibration signals and rotor speed were recorded at a sampling rate of 76,000 Hz and computer analyzed.

bearings can change signicantly with clearance of the bearing. The model should be detailed enough to include this behavior. Figure 8 shows the simulation results for a rolling element bearing with 0.08 mm clearance, 0 clearance, and 0.02 mm preloads. For each case, the load zone proles for radial load conditions of 100 N and 60 N were calculated. For a given clearance, the load zone widens and lengthens with increased radial load. For zero clearance, the load zone covers half the ring regardless of the radial load and expands with load. In the case of preloads usually demanded by the design of the bearing, the load zone extends completely around the ring with zero radial loads and shifts to one side when radial load is applied. The simulated results of the load zone show the ability of the detailed model to incorporate load zone effects in vibration results. 5.2 Model Verication. To validate the model, experiments were conducted and vibration signals of the bearing in the vertical direction, in time and frequency domains, were measured and compared with simulations. Results are presented in Fig. 9 for IRFs, in Fig. 10 for BFs, and in Fig. 11 for ORFs. For each fault, the upper plots are simulations and the bottom plots are experimental results. The left plots present vibration waveforms with the magnied portions in middle plots. The right plots are the power spectra of these vibration signals. To isolate fault signals, bearing fundamental frequencies, which depend on the bearing geometry and rotor speed, are as follows: inner race fault frequency

Results and Discussions

5.1 Load Zone Tests. Clearance in bearings changes the prole of contact tractions in the load zone. Defects within the load zone generate much bigger impulses than at other locations in a bearing. This behavior is observed as modulations of the vibration responses for IRF and BF, wherein faults are not stationary. Outer race faults, when located in the load zone, create a much larger response than at other locations. Vibration responses in faulty

n db f IRF = f r 1 + cos 2 dp


clearance: 0.08 mm

0 clearance

Preload: 0.02 mm

Radial Load: 100 N Radial Load: 60 N

Radial Load: 100 N Radial Load: 60 N

Radial Load: 100 N Radial Load: 60 N




Fig. 8 Simulation results: effects of clearance and radial loads on load distribution in rolling element bearings. a 0.08 mm clearance, b 0 clearance, and c 0.02 mm preload.

011102-6 / Vol. 133, JANUARY 2011

Transactions of the ASME

Downloaded From: on 05/12/2013 Terms of Use:

Vibration response of the bearing with Inner Race Faults (IRF) Time signal Time signal (magnified) Power spectrum of vibration




Fault Width

(c) fIRF







Fig. 9 Simulation versus experiment: Vertical vibration response of the bearing with inner race fault fault size: w h = 3 1 mm2. Fault is located at the load zone down. a and d Time response, b and e time response magnied, and c and f vibration power spectrum shaft speed: 35 Hz.

outer race fault frequency

n db f ORF = f r 1 cos 2 dp


ball fault frequency f BF =

d b2 dp f r 1 2 cos2 db dp


Fig. 10 Simulation versus experiment: vibration response of bearing with ball faults fault size: w h = 1.0 0.8 mm2. a and d Time response, b and e time response magnied, and c and f vibration power spectrum shaft speed: 35 Hz.

Journal of Tribology

JANUARY 2011, Vol. 133 / 011102-7

Downloaded From: on 05/12/2013 Terms of Use:

Fig. 11 Simulation versus experiment: vibration response of bearing with outer race faults fault size: w h = 2.7 1.0 mm2. a and d Time response, b and e time response magnied, and c and f vibration power spectrum shaft speed: 35 Hz

case frequency f cage =

fr db 1 cos 2 dp

f BF


Here n is the number of balls, f r is the rotor frequency, db is the ball diameter, d p is the bearing pitch diameter, and is the contact angle, which is zero for radial bearings. Given f r = 35 Hz, for the bearing of Table 1, the bearing frequencies are f IRF = 189.28 Hz, = 166.32 Hz, f ORF = 125.72 Hz, f cage = 13.96 Hz

Figure 9a contains simulated vibration signals of acceleration versus time for IRF. Here, fault impact signals are modulated with rotor frequency f r because the inner race fault, xed on the inner race, passes through the load zone at a rate of f r as the inner race rotates. Similar behavior with the same frequency is observed in the corresponding experimental signals shown in Fig. 9d. The magnied portion of vibration signal plotted in Fig. 9b shows the fundamental fault frequencies of the bearing, which is conrmed by the experiments in Fig. 9e. As the IRF passes through the contact, two impact responses appear in the vibration signal. The fault model of Fig. 3 suggests that the leading and trailing edges of the surface prole dent would cause impacts with the balls in the load zone. The distance between these two impact signals contains information on the size faults. Figures 9c and 9f present power spectrum of the vibration signals. Depending on the frequency range, harmonics of the fault frequencies indicated by dashed lines might appear in the power spectrum. Figure 10 shows the vibration signals of a bearing with a single BF. Fault impact signals spaced by the fault frequency timing 1 / f BF are clearly observed in both simulation and experimental data. Also, simulated signals are modulated by the cage frequency f cage. However, in real bearings, three dimensional rotations of balls can prevent engaging of ball faults with races, causing randomness in measured signals. This phenomenon generates random 011102-8 / Vol. 133, JANUARY 2011

modulations of vibration signals see Fig. 10d, which can be used to separate ball faults from other types of bearing faults. Another vibration signal feature that agrees with experiments is shape of the impact signals, as shown in Figs. 10b and 10e. As shown in the magnied plots, vibration signals due to BF consist of a single impact response signal for each fault, compared with the IRF and ORF that generate two impacts. The power spectrum of simulated signals Fig. 10c reveals harmonics of the fault frequency, which are not very clear in measured signals because of three dimensional motions of balls. Simulated vibration signals of the bearing with ORF are compared with experimental results in Fig. 11. Simulated vibration signals shown in Fig. 11a reveal a regular pattern of fault impacts spaced at the fault frequency timing, which appears in the experimental data of Fig. 11d. Two impact signals generated in ORF by the two edges of fault are observed in simulations. Similar to the IRF case, the ORF impact signals contain information on the fault size, which indicate the fault severity. In ORF, since the fault is stationary not moving, the vibration signal has a regular pattern, and fault frequency harmonics can be clearly observed in the power spectrum. Comparing simulation signals and measured signals in Figs. 911, the natural frequencies of the model and the machine seem to be slightly different. To investigate this issue, critical speeds and natural frequencies of the machine were obtained using startup/coast down and hammer tests. First and second critical speeds were about 70 Hz and 85 Hz. Two main natural frequencies obtained from hammer tests were about 480 Hz and 2100 Hz. The most dominant natural frequencies observed in the measurement signals during bearing experiments are about 2000 Hz, very close to the main natural frequency of the machine. The structure of the machine, including rotor and other components, is not included in the model. Therefore, there are natural frequencies of the real machine in measured signals that do not appear in simulations. These unmodeled frequencies likely from the machine Transactions of the ASME

Downloaded From: on 05/12/2013 Terms of Use:

creases the contact force and bearing radial deections. Centrifugal effects mainly appear at high speeds and are not signicant at low speeds. 5.3 Fault Severity Test. A model-based diagnostic employs a model to detect and assess faults, by estimating parameters in the model associated with faults. Therefore, a parametric study is needed. Here the effects of type IRF, BF, and ORF, size w h, and shape of the fault on vibration responses are studied. Referring to Table 2, faults in three levels corresponding to different sizes of w h are simulated and shown in Figs. 1315 for IRF, BF, and ORF, respectively. In time waveform and frequency signals, fundamental fault frequencies f IRF , FBF , FORF for all types of faults, cage frequency f cage for BF, and rotor frequency f r for IRF are observed. More severe faults with bigger defect sizes generate impact with higher amplitudes, for all types of faults. In IRF and ORF cases, vibration signals have a fault-element impact waveform characterized by two distinct peaks, with distance between these peaks proportional to the fault width, as shown in Figs. 13b, 13e, and 13h and Figs. 15b, 15e, and 15h. However, in the case of BF as shown in Fig. 14, the fault size only affects impact amplitudes rather than the distance between two impact peaks. To investigate the effect of fault shapes on the vibration response, faults with different size ratios w / h = 5 , 2 , 1 were simulated and results are shown in Fig. 16. For IRF and ORF, increasing the fault size ratio w / h enlarges the distance between the two impact peaks but does not change the impact amplitude much. For BF, the vibration response is relatively insensitive to w / h. To understand dynamics of the contact with faults, detailed simulations of contacts in the presence of BFs are shown in Fig. 17. The contact is within the load zone, but once the fault passes through the contact, nonlinear dynamic interactions can induce transient surface separations. Figures 17a and 17b show contact displacements solid line and forces dashed line for the contacts between inner race balls and outer race balls, for BF between inner race and ball. Negative force and displacements denote compression. Positive displacements denote surface separations, leading to zero forces. As shown in Fig. 17a, the inner race does not touch the bottom surface of the fault, and ball-race separation occurs until the inner race surface hits the trailing corner of the fault, causing the main impact. During this time, as shown in Fig. 17b at about 45.1 ms, the contact between ball and outer race unloads, which momentarily separate the surfaces. Just before 45.2 ms, a fault impact shock reloads the contact. These

Fig. 12 Radial deection versus speed for a roller bearing. The centrifugal force on rollers increases contact loads and radial deections when shaft speed increases. Simulations are compared with the results of Harris 35.

frame, bearing housing, rotating shaft, couplings, etc., can be easily considered in the model by appended submodels of machine components to the bond graph model. Simulations suggest that information on type and size of faults exist in the vibration signals from the bearing. However, the real system has numerous natural frequencies not in the bearing model that might mask the original spikes from faults. Tools such as high frequency lters might remove masking signals and retrieve fault information in the real system. Also, factors such as imbalance and misalignments can alter the regular patterns of the faults in a real machine. Interpreting the pattern of damage in real faults needs more than one snapshot of the signal and some statistical analysis to determine the condence level of the results. This need is more critical in the case of ball faults and inner race faults, in which faults are moving. The model includes centrifugal forces on balls. To illustrate this facet, a roller bearing with zero clearance under radial load was simulated for different shaft speeds. Bearing radial deections almost overlay data of Harris 35 in Fig. 12. The quadratic relation between radial deections and shaft speed in Fig. 12 represents the centrifugal effects in the bearing. The centrifugal force increases quadratically with the orbital speed of balls, which in-

Fig. 13 Vibration response of a bearing with IRF for different fault severities: w h = 1 0.5 level I, 1.42 0.71 level II, 2.82 1.41 level I mm2, and fr = 35 Hz

Journal of Tribology

JANUARY 2011, Vol. 133 / 011102-9

Downloaded From: on 05/12/2013 Terms of Use:

Fig. 14 Vibration response of a bearing with BFs for different fault severities: w h = 1 0.5 level I, 1.42 0.71 level II, 2.82 1.41 level I mm2, and fr = 35 Hz

Fig. 15 Vibration response of a bearing with ORF for different fault severities: w h = 1 0.5 level I, 1.42 0.71 level II, 2.82 1.41 level I mm2, and fr = 35 Hz

Fig. 16 Vibration responses of faulty bearings IRF, BF, and ORF with different fault shapes, fr = 35 Hz

011102-10 / Vol. 133, JANUARY 2011

Normalized Acceleration

Transactions of the ASME

Downloaded From: on 05/12/2013 Terms of Use:

Fig. 17 Faults in rolling element generate impacts, which can cause contact separations. a Force and displacement at inner race-ball contact, b force and displacement at ball-outer race contact, and c outer race vibration ball fault: w h = 1.42 0.71 mm2 engaged at inner raceball contact, fr = 35 Hz.

nonlinear contact dynamics generate fault impulses, which transferred through elements, and generate vibrations, as observed in Fig. 17c.

Summary and Conclusions

A detailed model of rolling element bearings developed in vector bond graphs incorporated multibody dynamics of elements, centrifugal effects, dynamics of contacts, and surface defects. NewtonEulers equations for each element were encoded into bond graphs, with dynamics of contacts, traction forces, and rotational frictions formulated as constitutive laws of elements. A kinematics-based fault model was introduced. Tribological faults were modeled as surface prole changes, which generate impulses via dynamic interactions of faults and bearing elements. Fault parameters dene type, size, shape, and locations of faults. Simulations for different clearances and radial loads show ability of the model to incorporate load zone effects in vibration signals. Experiments for healthy and faulty rolling element bearings validated the model. A parametric study investigated the effects of type, size, and shape of faults on vibration responses. Dynamics of contacts under faults demonstrated how fault impulses are physically generated. The modular and generic rolling element bearing bond graph model represent complex dynamics of both normal and defective bearings, for rolling element bearings with different geometry and specications. With physical and kinematic parameters assigned to faults, the model can predict bearing response for fault conditions, and thus be used in model-based diagnostics of rolling element bearings, for information processing necessary for predictive maintenance of machinery.

fr rf v0 K k n r r

s x x i y iz i x y z
ij ij ij

v w,h

ij f

bd , hd

CS E Fi FS Fx , F y , Fr , Fa F J M T TGB bearing static load rating modulus of elasticity external force vector acting on the body static equivalent load forces in x, y , radial, and axial directions, respectively curvature difference body rotational inertia matrix body mass matrix coordinate transformation matrix transformation matrix from the global to the body coordinate rotational resistive torque fault type indicators 1: IRF, 2: BF, 3:ORF resistive torques due to contact deformations, lubrications, and rubbing seal global xed coordinate contact damping factor balls diameter bearing pitch diameter

bearing rotational speed in rpm fault size ratio kinematic viscosity of the lubricant contact stiffness equivalent linear stiffness of the bearing number of balls in the rolling element bearing radius of curvature position vector with respect to the body coordinate contact slide-roll ratio body position vector with respect to the global coordinate moving coordinate associated with the body i moving contact coordinate between bodies i and j Poissons ratio width and depth of localized faults oil lm parameter which is a function of oil lm thickness orientation of the contact coordinate described in the global coordinate angular position of the fault dened in body coordinate dimensionless deection factor body angular position vector orientation of the body coordinate described in the global coordinate friction coefcient under boundary lubrication and hydrodynamic lubrication modes curvature external torque vector acting on the body bearing contact angle bar accent assigned to vectors described in the body coordinate system

1 Raison, B., Rostaing, G., Butscher, O., and Maroni, C. S., 2002, Investigations of Algorithms for Bearing Fault Detection in Induction Drives, IECON02, pp. 16961701. 2 Nakhaeinejad, M., and Bryant, M. D., 2010, Multibody Dynamics of Rolling Element Bearings With Faults, Proceedings of the ASME/STLE International Joint Tribology Conference 2009, pp. 249251. 3 Isermann, R., and Balle, P., 1997, Trends in the Application of Model-Based Fault Detection and Diagnosis of Technical Processes, Control Eng. Pract., 55, pp. 709719. 4 Purdue University, 2003, Virtual Machine Design WorkshopDening the Requirements for the Next Generation of Analytical Software for the Design of Rotating Components, Technical Report. 5 Jones, A. B., 1960, A General Theory for Elastically Constrained Ball and Roller Bearing Under Arbitrary Load and Speed Conditions, ASME J. Basic Eng., 8221, pp. 309320. 6 Gupta, P. K., 1979, Dynamics of Rolling-Element Bearings.1. Cylindrical Roller Bearing Analysis, Mech. Eng. Am. Soc. Mech. Eng., 1011, pp. 9596.

Tf Tf Tload , Tlub , Tseal XYZ b db dp

Journal of Tribology

JANUARY 2011, Vol. 133 / 011102-11

Downloaded From: on 05/12/2013 Terms of Use:

7 McFadden, P. D., and Smith, J. D., 1984, Vibration Monitoring of Rolling Element Bearings by the High-Frequency Resonance TechniqueA Review, Tribol. Int., 171, pp. 310. 8 Afshari, N., and Loparo, K. A., 1998, A Model-Based Technique for the Fault Detection of Rolling Element Bearings Using Detection Filter Design and Sliding Mode Technique, Proceedings of the 37th IEEE Conference on Decision and Control, Vols. 14, pp. 25932598. 9 Adams, M. L., 2010, Rotating Machinery Vibration: From Analysis to Troubleshooting, 2nd ed., CRC Press, Boca Raton, FL. 10 Harsha, S. P., 2006, Nonlinear Dynamic Analysis of Rolling Element Bearings Due to Cage Run-Out and Number of Balls, J. Sound Vib., 28912, pp. 360381. 11 El-Saeidy, F. M. A., and Sticher, F., 2010, Dynamics of a Rigid Rotor Linear/ Nonlinear Bearings System Subject to Rotating Unbalance and Base Excitations, J. Vib. Control, 163, pp. 403438. 12 Liew, A., Feng, N., and Hahn, E. J., 2002, Transient Rotordynamic Modeling of Rolling Element Bearing Systems, ASME J. Eng. Gas Turbines Power, 1244, pp. 984991. 13 Sopanen, J., and Mikkola, A., 2003, Dynamic Model of a Deep-Groove Ball Bearing Including Localized and Distributed Defects. Part 1: Theory, Proceedings of the Institution of Mechanical Engineers Part K-Journal of MultiBody Dynamics, 2173, pp. 201211. 14 Sopanen, J., and Mikkola, A., 2003, Dynamic Model of a Deep-Groove Ball Bearing Including Localized and Distributed Defects. Part 2: Implementation and Results, Proceedings of the Institution of Mechanical Engineers Part K-Journal of Multi-Body Dynamics, 2173, pp. 213223. 15 Jang, G., and Jeong, S. W., 2004, Vibration Analysis of a Rotating System Due to the Effect of Ball Bearing Waviness, J. Sound Vib., 26935, pp. 709726. 16 Changqing, B., and Qingyu, X., 2006, Dynamic Model of Ball Bearings With Internal Clearance and Waviness, J. Sound Vib., 29412, pp. 2348. 17 Patil, M. S., Mathew, J., Rajendrakumar, P. K., and Desai, S., 2010, A Theoretical Model to Predict the Effect of Localized Defect on Vibrations Associated With Ball Bearing, Int. J. Mech. Sci., 529, pp. 11931201. 18 Wensing, J. A., 1998, Dynamic Behaviour of Ball Bearings on Vibration Test Spindles, IMAC, Proceedings of the 16th International Modal Analysis Conference, pp. 788794. 19 Sassi, S., Badri, B., and Thomas, M., 2007, A Numerical Model to Predict Damaged Bearing Vibrations, J. Vib. Control, 1311, pp. 16031628. 20 Karkkainen, A., Helfert, M., Aeschlimann, B., and Mikkola, A., 2008, Dy-

21 22 23 24

25 26 27 28 29 30 31 32 33 34 35

namic Analysis of Rotor System With Misaligned Retainer Bearings, ASME J. Tribol., 1302, p. 021102. Ashtekar, A., Sadeghi, F., and Stacke, L. E., 2008, A New Approach to Modeling Surface Defects in Bearing Dynamics Simulations, ASME J. Tribol., 1304, pp. 041103. Ashtekar, A., Sadeghi, F., and Stacke, L. E., 2010, Surface Defects Effects on Bearing Dynamics, Proc. Inst. Mech. Eng., Part J: J. Eng. Tribol., 224, pp. 2535. Bhattacharyya, K., and Mukherjee, A., 2006, Modeling and Simulation of Centerless Grinding of Ball Bearings, Simul. Model. Pract. Theory, 147, pp. 971988. Nakhaeinejad, M., Lee, S. H., and Bryant, M. D., 2010, Finite Element Bond Graphs of Rotors, Proceeding of Ninth International Conference on Bond Graph Modeling & Simulation ICBGM 2010, Vol. 42, Issue No. 2, pp. 164171. Harris, T., and Kotzalas, M., 2007, Essential Concepts of Bearing Technology, Taylor & Francis, Boca Raton, FL. Aktrk, N., Uneeb, M., and Gohar, R., 1997, The Effects of Number of Balls and Preload on Vibrations Associated With Ball Bearings, ASME J. Tribol., 1194, pp. 747753. Kramer, E., 1993, Dynamics of Rotors and Foundations, Springer, Berlin. Kragselsky, V., Dobychin, N., and Kombalov, S., 1982, Friction and Wear: Calculated Methods, Pergamon, Elmsford, NY. Gupta, P., 1984, Advanced Dynamic of Rolling Elements, Springer-Verlag, New York. Palmgren, A., 1959, Ball and Roller Bearing Engineering, SKF Industries, Inc., Philadelphia. Shin, H., 2004, Bearing Diagnosis Using Embedded Modeling Method and System Identication Technique, Ph.D. thesis, Rensselaer Polytechnic Institute, Troy, NY. Paynter, H. M., 1961, Analysis and Design of Engineering Systems, MIT, Cambridge, MA. Karnopp, D. C., Margolis, D. L., and Rosenberg, R. C., 2000, System Dynamics: Modeling and Simulation of Mechatronic System, Wiley, New York. Sinokrot, T., Nakhaeinejad, M., and Shabana, A. A., 2008, A Velocity Transformation Method for the Nonlinear Dynamic Simulation of Railroad Vehicle Systems, Nonlinear Dyn., 5112, pp. 289307. Harris, T., and Kotzalas, M., 2007, Advanced Concepts of Bearing Technology, Taylor & Francis, Boca Raton, FL.

011102-12 / Vol. 133, JANUARY 2011

Transactions of the ASME

Downloaded From: on 05/12/2013 Terms of Use: