Sei sulla pagina 1di 63

Contact Stress Concentration due to Surface Irregularity

in Cylindrical Rolling Element Bearings


by
Jeffrey A. Scarcella
A Project Submitted to the Graduate
Faculty of Rensselaer Polytechnic Institute
in Partial Fulfillment of the
Requirements for the degree of
MASTER OF MECHANICAL ENGINEERING

Approved:

_________________________________________
Ernesto Gutierrez-Miravette, Project Advisor

Rensselaer Polytechnic Institute


Hartford, CT
December, 2008

i
Copyright 2008
by
Jeffrey A. Scarcella
All Rights Reserved

ii
CONTENTS
LIST OF TABLES .........................................................................................................v
LIST OF FIGURES.......................................................................................................vi
LIST OF SYMBOLS.................................................................................................. viii
ACKNOWLEDGMENTS ..............................................................................................x

ABSTRACT..................................................................................................................xi
1. Introduction ..............................................................................................................1
2. Background ..............................................................................................................2
2.1 Project Goal ....................................................................................................2
2.2 Contact Theory Basics.....................................................................................2
2.3 Bearing Loading and Motion ...........................................................................2

3. Analytical Solution ...................................................................................................5


3.1 Simple Case: Infinitesimal Line Loading of an Elastic Half Space...................5
3.2 Extension of Simple Case: Parallel Cylinders in Contact .................................7
3.3 Application to Cylinder-Plane Contact and Sample Problem .........................12
3.3.1 Analytical Model Results ...................................................................14
3.4 Relationship between Normal Load and Contact Pressure..............................16

3.5 Relevance of the Flat Race (Infinite Radius R2) Assumption..........................17


4. FE Model and Validation........................................................................................19
4.1 Method..........................................................................................................19
4.2 FEA Tool Validation Test Case .....................................................................19
4.3 Baseline Elastic Model ..................................................................................21
4.3.1 Elastic Results ...................................................................................24

4.4 FEA Model including plasticity effects..........................................................28


4.4.1 Plastic Results....................................................................................29
5. Extension to Non-smooth Surfaces .........................................................................31
5.1 Rough and Irregular Surfaces in Contact........................................................31

iii
5.2 Sawtooth Roughness of Contacting Surface...................................................32
5.3 Square Wave Roughness of Contacting Surface.............................................33
5.4 Probabilistically Distributed Roughness of the Contact Surface .....................37
5.4.1 Normally Distributed Roughness .......................................................38

5.4.2 Weibull Distributed Roughness..........................................................41


5.4.3 Solution Convergence Challenges with Irregular Surfaces..................46
6. Conclusions ............................................................................................................51
Bibliography ................................................................................................................52

iv
LIST OF TABLES
Table 1: Example Problem Material & Geometric Properties........................................14

v
LIST OF FIGURES
Figure 1: Typical Roller Bearing ....................................................................................3
Figure 2: Line Loading of an Elastic Half-Space.............................................................5
Figure 3: Contacting Parallel Cylinders ..........................................................................8
Figure 4: Cylindrical Roller Loaded by Opposing Races...............................................13
Figure 5: Contact Pressure Distribution, Uniformly Loaded Flat Blocks .......................15
Figure 6: X- and Z- Stress in the Roller vs. Height (Z)..................................................16
Figure 7: Peak Contact Pressure vs. Normal Load.........................................................17
Figure 8: Peak Contact Pressure vs. Race Radius (R2)..................................................18
Figure 9: Conforming Elastic Blocks FE Model with Contact.......................................20
Figure 10: Contact Pressure Distribution, Uniformly Loaded Flat Blocks .....................21
Figure 11: Elastic FE Model, Smooth Cylinder & Block...............................................23
Figure 12: Contact Pressure Distribution, Elastic Cylinder & Block .............................24
Figure 13: Elastic Results, X-stress...............................................................................25
Figure 14: Elastic Results, Y-stress...............................................................................25
Figure 15: Elastic Results, 3rd Principal Stress .............................................................26
Figure 16: Elastic Results, 3rd Principal Strain .............................................................27
Figure 17: AMS6490 Bearing Steel Stress-Strain Relationship .....................................28
Figure 18: Contact Pressure Distribution Including Plasticity Effects............................29
Figure 19: 3rd Principal Stress, Elasto-Plastic Model....................................................30
Figure 20: 3rd Principal Strain, Elasto-Plastic Model....................................................30
Figure 21: Fly Cutter Typical Surface Finish Pattern ....................................................31
Figure 22: Sawtooth Wave Roughness Profiles.............................................................32
Figure 23: Contact Pressure Distribution, Square Wave Surface ...................................34
Figure 24: Square Wave Roughness, 3rd Principal Strain .............................................35
Figure 25: Square Wave Roughness, 3rd Principal Stress (Wide View) .........................36
Figure 26: Square Wave Roughness, 3rd Principal Stress (Close View) ........................37
Figure 27: Typical Randomly Distributed Surface Roughness Profile ..........................38
Figure 28: Gaussian Probability Density Function, Normalized ....................................39
Figure 29: Normal Cumulative Probability Distribution................................................40
Figure 30: Normal Surface Profile ................................................................................41

vi
Figure 31: Illustrations of Skewness and Kurtosis.........................................................42
Figure 32: Weibull Distributions with Varying Parameters a & b..................................43
Figure 33: Roughness Characteristics of Common Machining Processes ......................44
Figure 34: Typical Lathe Turned Roughness Profile .....................................................45
Figure 35: Typical Abrasive Ground Roughness Profile ...............................................46
Figure 36: Gaussian Roughness Results, 10N Total Load .............................................48
Figure 37: Weibull (Turned) Roughness Results, 2N Total Load ..................................48
Figure 38: Weibull (Ground) Roughness Results, 10N Total Load................................49
Figure 39: Gaussian Results, Elastic Properties, 1000N Load .......................................50

vii
LIST OF SYMBOLS
The following terms will be used in section 3. Units are listed if applicable.
Symbol Definition (units)
x, y, z Rectangular Coordinate System Axes
r, , z Cylindrical Coordinate System Axes
() Deformed Shape Function
Tensile & Compressive Stress (MPa)
Shear Stress (MPa)
P Applied Normal Load (N)
u Local Deformation at a Surface (mm)
E Elastic Modulus (MPa)
E* Effective Elastic Modulus (MPa)
G Shear Modulus (MPa)
Strain
Poisson's Ratio
zi Unloaded Surface Height above Origin(mm)
Unloaded Separation Distance Beetween Contacting Bodies
h (mm)
R'i Primary Local Radius of Curvature of a Body (mm)
R''i Perpendicular Local Radius of Curvature of a Body (mm)
i Bulk Body Displacement Under Load (mm)
A, B, C, D Arbitrary Constants
p() Normal Surface Traction (N)
q() Tangential Surface Traction (N)
a Deformed Contact Area Half-Width (mm)
R1, R2 Radii of Cylinders 1, 2 (mm)

viii
The following terms will be used in section 5. Units are listed if applicable.
Symbol Definition (units)
x, y, z Rectangular Coordinate System Axes
Standard Deviation (mm)
2 Variance (defined as the square of standard deviation)
R Stress Ratio
L Length of a Sample Surface (mm)
m Mean (Average) Surface Height (mm)
p() Probability Density Function
z* Normalized Surface Height (mm)
F(z) Cumulative Probability Distribution Function
a Shape Parameter for Weibull Distribution
b Scale Parameter for Weibull Distribution
P Applied Normal Load (N)
E Elastic Modulus (MPa)
Strain
Poisson's Ratio

ix
ACKNOWLEDGMENTS
Thank you to my wonderful wife Kate for your love and your support of my educa-
tional endeavors. I also owe a debt to my parents for instilling in me the value of a good
education and the need for continuous self-improvement.
Thanks to Professor Ernesto Gutierrez-Miravette of the RPI campus at Hartford, CT.
I appreciate your guidance and wisdom through my program of study and during the
creation of this work.
Thanks to Professor Al George of Cornell University. I owe a great deal of my tech-
nical expertise to your empowering teaching style. Thank you for helping me learn to
see the larger picture at times when I was focused on the details.
Thank you to Ray Matheis and Roger Bemont of United Technologies Corporation.
Not only did you mentor me in my first years with the company and provide crucial
technical advice, but your constant wit and ability to poke at the bureaucracy and hassle
of our daily tasks created a comfortable atmosphere in which to develop my abilities.

This work is dedicated to the memory of Kevin F. Ryan (1983-2008).

x
ABSTRACT
The goal of this work is to produce a comparison between the peak contact stress re-
sulting from contact between simulated roller bearing elements with smooth surfaces and
various roughness profiles.
First, the theory and principles of elasticity was used to create a baseline contact
case. Thin line normal loading of an elastic half-space was solved using elasticity, and
the results were extended to a solution for contact between a cylinder and a semi-infinite
flat planar surface. All problems were solved considering only two-dimensional varia-
tions in geometry, i.e., the surface profiles of the solids were assumed to be constant in
the depth dimension.
The second theoretical case (smooth cylinder-plate contact) was then modeled and
solved using ANSYS 11.0 finite element analysis software. The FEA contact stress
results compared favorably with the theoretical predictions, differing by less than 2%.
Square wave roughness was introduced on the flat plate, and the FEA results showed a
large increase in the apparent (overall) contact area width, but minimal variation in the
real (discontinuous) area of contact. Peak contact pressure actually decreased from the
smooth case.
Empirical data for typical surface roughness characteristics of machining processes
was used in concert with a spreadsheet random number generator and the Gaussian and
Weibull probabilistic distributions to generate sample roughness profiles fitting the
properties of machining processes that could be used to produce bearing races. These
profiles were decomposed into a series of points that were input into the ANSYS model
and fit with spline curves and solved using the same contact algorithms as the previous
cases.
Solution of the probabilistic roughness cases proved difficult. A failure analysis is
presented at the end of this work with root causes of the computational issues. It is
recommended that readers who intend to conduct further finite element analysis of finely
rough surfaces consider alternative modeling techniques to mitigate the problems
encountered in the production of this writing.

xi
1. Introduction
Contact between the rolling elements and the races of rotating bearing assemblies
creates stress in the material as the radial load is transferred from the inner race to the
outer race (or vice versa). Despite modern manufacturing processes, the surfaces of the
bearing components are, on some level, irregular, causing the contact stress to concen-
trate at discrete locations along the contact area. Concentrated, and thus higher, stress
and strain levels in the material can impact useful service life, friction, and wear of the
interfacing materials. To make an accurate estimate of the cyclic fatigue life of a
bearing, it is necessary to understand the true internal stress experienced by the material,
and then make a comparison to laboratory or service-based fatigue life statistical data.
Heinrich Rudolf Hertz was the first to publish analytical methods for describing the
behavior of elastic bodies in loaded contact when he did so in 1895. The Hertzian
contact model is still in widespread use and will form the basis for my initial investiga-
tions into elastic behavior at an example interface between a bearing roller and its
mating race. The Hertzian model can be extended to include material plasticity and
simple surface shapes, provided they can be described by a single continuous function.
The Finite Element Method (FEM), developed in the 1940s, allows complex geometry to
be decomposed into a series of many smaller, simple shapes that can be represented as a
series of simultaneous partial differential equations that are then solved numerically to
give an estimate of the closed-form solution for the complex geometry. This method
will greatly facilitate the incorporation of rough surfaces into the contact model, includ-
ing surfaces that have profiles based on the statistical characteristics of common
manufacturing processes. I intend to explore contact stress concentrations due to surface
irregularity through the use of Finite Element Analysis (FEA) tools, in order to assess
the importance of the roughness on contact pressure and stress, and ultimately, the
fatigue life of the product.

1
2. Background

2.1 Project Goal

The contact between the rounded surface of a cylindrical roller and the bearing race,
which has a much larger radius, can be approximated by the contact between a cylinder
and a flat plane parallel to the cylinder axis. An analytical model of the cylinder ap-
proximation will be solved theoretically to determine the stress and deformation as a
function of the force closing the contact (Section 3). Finite element analysis (FEA)
using ANSYS will be used to duplicate and verify those results (Section 4). Following
validation of the FE model accuracy, additional FE models will be created to examine
deformation and stress concentration that results if one of the surfaces is uneven (Section
5).

2.2 Contact Theory Basics

Contact between two solids can be broken into to categories. Conforming contact
occurs when the mating surfaces of the two bodies have identical (or nearly identical)
contours. Under normal load, the contact area is large and the pressure is distributed.
Non-conforming contact occurs when two solids are brought together that will initially
have an infinitesimal contact area, such as a point (sphere contacting a plane or another
sphere), or a line (cylinder contacting a plane or another cylinder). The non-conforming
bodies will deform under pressure and a relatively small area of contact will be created.
Contact stress will be concentrated in this small area and will thus be high compared to
stress in conforming contact. In practice, no surface is truly planar, cylindrical, or
spherical, due to manufacturing irregularities. Therefore, non-conforming contact is of
the greatest interest, especially when rough surfaces are involved.

2.3 Bearing Loading and Motion

The reader is assumed to have at least a basic familiarity with engineering design
principles and bearing types and uses. For completeness, a short description of the
bearings being studied is included below. Fundamentals of Machine Component De-

2
sign1 is recommended reading for more background on the design and development of
bearing technology.
Rolling element bearings are used in commercial and industrial applications to sup-
port and locate rotating shafts, and greatly reduce friction. This prolongs equipment life
and can drastically cut the energy input required to keep the mechanism moving.
Cylindrical roller bearings are used where high load carrying capability is required in a
compact space. Cylindrical roller bearings are intended to carry radial loads (perpen-
dicular to the shaft axis) only. The roller has a greater contact area with the race than a
traditional ball bearing, to better distribute the applied load.

Roller

Cage

Inner Race Outer Race


Figure 1: Typical Roller Bearing2

Figure 1 shows the basic components of a typical bearing. The inner and outer races
contain the assembly and transfer load from the rollers to the rotating and static mount-
ing points. The rollers maintain the separation of the races, and transfer radial load
while keeping friction to a minimum. The cage separates and evenly distributes the balls
around the assembly and prevents them from contacting each other. Grease or oil fills
the voids to provide lubrication and aid heat dissipation. Typically, seals or shields
cover the separation between the races to contain the lubricant and resist contamination.

1
Fundamentals of Machine Component Design, Juvinall, R. C. and Marshek, K. M., USA, 2000
2
Illustration courtesy of the Timken Company (bearing manufacturer)

3
Loads can be transferred at the interface point between bearing and race in three ori-
entations Normal, Tangential, and Axial. In a standard roller bearing (not a tapered or
thrust bearing), the axial force will be small. The tangential force is caused by a combi-
nation of friction and the tractive force driving the bearings rotation. The normal force
is due to the radial load in the bearing. In a well-lubricated bearing under low drive
torque, the tangential force will also be small in relation to the normal force. For the
purposes of this discussion the smaller axial and tangential forces will be neglected,
leaving only the predominant normal force. This is the so-called free rolling assump-
tion. Another important simplifying assumption is that the rolling elements in the
bearing are purely rolling. There is no slip at the bearing-race interface, such that the
surface speeds are identical. For a low rolling speed, there is not a significant difference
in the stress and load-carrying capability between rolling and static bearings.3 Cyclic
application of stress in a rolling bearing has a negative impact on useful life (compared
to static loading), but that is beyond the scope of this writing.

3
Rolling Tests of Plates, Wilson, W. M., United States, 1929

4
3. Analytical Solution

A solution based on the theory of elasticity can be applied to the case of a roller bear-
ing, with a few assumptions. Firstly, all strains must be small and linear (below the
proportional limit strain). This is generally a valid assumption as rolling element
bearings typically must be designed for billions of rotational cycles without failure. If
the ball was plastically deformed (to a significant degree) on every load cycle, it would
fail much sooner. Also, the geometry of the roller and its mating surface are assumed to
be invariant along the axis of the cylinder. The pressure, stress, and deformation will
also be constant, so that the interface can be modeled as semi-infinite, i.e. infinite and
invariant in one direction. This removes a dimension of variability and further simplifies
the calculations involved. Alternatively, these quantities can simply be defined in terms
of unit per length of cylinder, which is more practical and broadly applicable.

3.1 Simple Case: Infinitesimal Line Loading of an Elastic Half Space

The simplest semi-infinite geometry is the case of a planar semi-infinite (infinite in


one axis) solid body loaded in the normal direction along a line on the surface. The x-y
plane defines the surface, the loading line is along the y-axis, and the load is along the z-
direction. Figure 2 shows the axis and dimension definitions.

Figure 2: Line Loading of an Elastic Half-Space4

4
From Contact Mechanics, Johnson, K.L., United Kingdom, 1985

5
As described in Johnson5, the solution in the polar space is:

(r , ) Ar sin 3.1

From the equations of elasticity we know that the three components of stress in the
material will be

1 1 2
r 3.2
r r r 2 2
2
2 3.3
r
1
r 3.4
r r

Therefore, = r =0, so that the stresses are entirely radial from the line of contact.

2 A cos
r 3.5
r
It is evident from 3.5 that the stresses are dependent on the distance r from the line of
load, such that, for very small r the stresses approach an infinite magnitude, and, for
large values of r (locations distant from the application of load) the stress approaches
zero. The displacement vs. location curve is logarithmic due to the 1/r relationship as
evident in Figure 2.

To solve for A, a force balance is used. The applied force P must be equal to the
sum of the vertical (z) stresses in the material.
/2
P
/ 2
r cos rd 3.6

Substituting 3.5 and recognizing the symmetry between 0 and /2 gives

5
From Contact Mechanics, Johnson, K.L., United Kingdom, 1985

6
/2
P 2 A cos d
2
3.7
0


P A or A 3.8
P
Substituting back into 3.5 results in an expression for the radial component of stress
2 P cos
r 3.9
r
Converting from polar to rectangular coordinates gives expressions for the x-, y-, and
z-components of stress in the region of loading
2P x2 z
x r sin 2 3.10
( x 2 z 2 )2

2P z3
y r cos 2 3.11
( x2 z 2 )2

2P xz 2
z r sin cos 3.12
( x2 z 2 )2
Application of Hookes law (generally, = E for elastic isotropic materials, with
E being the elastic modulus) as demonstrated in Johnson (1985) gives the strains and
their relationship to the displacements of the loaded material.
ur (1 2 ) 2 P cos
r 3.13
r E r
u r 1 u (1 ) 2 P cos
3.14
r r E r
1 u u u
r r 0 3.15
r r r G

3.2 Extension of Simple Case: Parallel Cylinders in Contact

The section 3.1 describes contact at a line of infinitesimal width, which is in line
with the initial contact geometry between the cylinder and plane that we intend to
examine. However, as soon as any load is applied to press the two bodies together, the
elastic nature of the solids will cause local deformation at the interface, and the area of
contact will begin to have width. This local strain will cause stress to develop in the

7
material, and the characterization of that stress is the ultimate goal. Analytically, both
surfaces are assumed to be smooth on a small scale such that their profile is exactly as
described by the equations of the bodies boundary lines. This limits the contact to only
one continuous, rectangular area. The defining equations of surfaces must also have
continuous second derivatives, meaning that there are no discontinuities and no sharp
transitions between regions of the contacting surfaces.

Lets consider two cylinders of arbitrary radii with parallel axes, contacting each
other under no load, so that the contact area is simply a line. First, define a coordinate
system with the origin at one end of the line of contact. The x-axis is placed tangent to
both cylinders and normal to the axes of the cylinders. The y-axis is parallel to the
cylinder axes, along the line of contact. The z-axis will then be along a line running
through the centers of the cylinders circular cross sections.

Y
X
Z

Figure 3: Contacting Parallel Cylinders

Given these assumptions, the surface profile of the each cylinder may be represented
as:

8
zi Ai x 2 Bi y 2 Ci xy Di x 3 ...
, where i=1, 2, for each cylinder 3.16
The unloaded separation distance between the two points at the same x-coordinate on
the two cylinders may be defined as
h z1 z2 3.17
or the difference of the vertical (z) locations of the unloaded points.

Johnson shows that z1 and z 2 can be expressed in terms of the radii of curvature of the

solid in the principal axes. R'i is the local radius of curvature of the surface about an

axis normal to x and z for cylinder i, and R"i is the local radius about an axis normal to y
and z.
1 2 1 2 1 2 1
z1 x1 y1 and z2 x2 y22 3.18a,b
2 R'1 2 R"1 2 R'2 2 R"2
Of course, for two cylinders that have parallel axes along y, the surfaces are linear
along y, which is effectively an infinite radius, i.e. R=. This eliminates variation in z
due to y-position and leaves simplified expressions for z.
1 2 1 2
z1 x1 and z2 x2 3.19a,b
2 R'1 2 R'2

Recalling that h z1 z 2 , if the x1 and x2 axes are aligned, or if one radius is infi-

nite, a value R can be defined that combines the R'1 and R'2 quantities into a single
effective radius of curvature for interface between the two cylinders. Solving for R in
terms of the two cylinders radii gives an expression for effective radius.
1
1 1
R' such that 3.20
R1 R2

1 2 1 1 1
h x x 2 3.21
2 R' 2 R1 R2

Given this relation for the separation between surface contours in the undeformed
and unloaded state, the next step is to apply a normal compressive load and find the
resulting deformation in both solids.

9
Under a load P (in this case, P is per unit of cylinder length), each cylinder will trans-
late an amount i , measured as the distance that the bulk solid body moves towards the

origin of the coordinate system defined above (Figure 3). As the two bodies cannot
occupy the same space, both will deform locally by amounts u zi . The interface is not

necessarily symmetric and, in this general case, the materials of the two cylinders may
differ, so u zi does not necessarily equal i . The sum of the cylinders translations can

be equated to the total combined deformation of both bodies plus any initial separation
in the undeformed state.
u z1 u z 2 h 1 2 3.22
Introducing a displacement sum 1 2 and substituting the generic expression

h A x 2 B y 2 for h, we find the displacements in terms of location on the body.

u z1 u z 2 h Ax 2 By 2 3.23
As earlier, the y term falls out when limited to the 2D cylindrical case, and we can
combine equations 3.21 and 3.23 to give:
1 2
u z1 u z 2 h x 3.24
2 R'
It is necessary to note that this equation holds true within the area of contact under
the normal load. Outside the contact area, the displacements will be lower than at
contacting points, or:
1 2
u z1 u z 2 h x 3.25
2 R'
Equation 3.25 can be differentiated with respect to x to give the displacement gradi-
ent at the contact, or the slope of the body surface when deformed under load. The
equation is thus simplified as constants are eliminated.
u z1 u z 2 1
x 3.26
x x R'
From Johnson, the general expression for the displacement gradient in z, derived
from elasticity and the half-space model, is given as:

u z 2(1 2 ) p( s ) (1 2 )(1 )
a

x

E b x s
ds
E
q ( x) 3.27

10
Where p(s) and q(x) are the normal and tangential surface tractions, respectively.
We are assuming the case of no tangential forces, so q(x)=0. The effective elastic
modulus E* will be defined as follows.
1
1 v12 1 v22
E* 3.28
E1 E2

If materials 1 and 2 are the same, E* simplifies to the plane strain elastic modulus of
the material. Combining equations 3.26, 3.27, and 3.28 gives the following.

E *
a
p( s )
x s ds
a
2 R'
x 3.29

The solution to 3.29, as described in Johnson (1985) shows the following distribution
of pressure over the contact surface:
2P 2 1
p( x) (a x 2 ) 2 3.30
a 2

P R'
where a 2 3.31
E*
The maximum contact pressure can be found by differentiating p(x), and, as would
be expected, the max pressure occurs at the center of the interface such that x=0.
2P
pmax or, with eqn. 3.31, 3.32
a
1
PE * 2
pmax 3.33
R '
Extension of equations 3.10, 3.11, and 3.12 to a contact region of a finite width, as
shown in Johnson, gives the following expressions for the stress at the contact area:

p( s)( x s)2
a
2z
x

ds 3.34
a x s 2 z 2 2
a
2z3 p( s )
z
a x s 2 z 2 2
ds

3.35

p (s)( x s)
a
2z2
xz
a
x s 2
z 2 2
ds 3.36

11
Substituting equation 3.30 for p(x) and the expression for pmax gives the following
stress relations. For simplicity of the calculations, we will consider the stress variation
across the contact surface and along the line (plane) x=0.

Contact stress at the surface:


x p( x ), a x a 3.37

z p( x ),a x a 3.38

Stress within the material above the point of initial (unloaded) contact:

x
pmax 2
a

a 2z a z
2 2 2

1
2
2 z

3.39

z pmax a a 2 z 2
1
2
3.40

1 pmax a z z 2 a 2 z 2 2
1
3.41

Differentiating the principal shear stress expression with respect to z gives the loca-
tion of the maximum shear stress along the z-axis:
1 max 1 (0.78a) 0.30 pmax 3.42
Building off these relationships, we will now begin to relate the findings of Section
3.2 to the example bearing configuration.

3.3 Application to Cylinder-Plane Contact and Sample Problem

A roller bearing contains a series of small cylindrical rolling elements, which run
between two cylindrical races. The inner surface of the outer race and the outer surface
of the inner race contact each roller at points 180 apart (Figure 4).

12
Figure 4: Cylindrical Roller Loaded by Opposing Races

The races typically have radii greater than the rollers by an order of magnitude or
more. Based on equation 3.20, if R2 R1 , the effective radius R will be much more
sensitive to changes in roller radius than race radius. As such, the geometry that will be
analyzed throughout this writing will use a cylindrical roller in contact with a flat, planar
bearing race. This simplifies the modeling of the rough surface cases and has a negligi-
ble effect on the results for bearings with a large race radius to roller radius ratio.
In the analytical calculations, the planar surface will be considered as a cylinder hav-
ing a large (infinite) radius. The computer model will use a rectangular prismatic section
of material. It should also be noted that all analyses will consider only the cross-
sectional geometry of each example and model in two dimensions. The third dimension
(depth, or roller/race width into the page) is modeled as infinite. Therefore, all values
of loads and any quantities not already expressed in terms of per unit area are in terms
of per unit width. For example, the normal load would be 100N/mm for a roller of X
mm in width.

13
3.3.1 Analytical Model Results

For illustrative purposes, a specific set of dimensions and properties has been chosen as
shown in Table 1, to facilitate comparison between the mathematical and computer
analyses. These parameters are typical of bearings in use in modern industry.

Material Properties:
Material Specification AMS6490 Bearing Steel
Elastic Modulus (E) 100,000 MPa
Poissons ratio () 0.29 (dimensionless)
Bearing Geometry
Normal Load P 100N (per mm of width)
Bearing roller radius R1 10mm
Bearing race radius R2 infinite (flat surface)

Table 1: Example Problem Material & Geometric Properties

Inserting the example dimensions into the equations derived in section 3.2 provides
the following values:
1 1
1 1 1 1
R' 10mm (3.20)
R1 R2 10mm
1 1
1 v12 1 v22 1 0.29 2
E* 2 54,591MPa (3.28)
E1 E2 100,000MPa

P R' 100 N 10mm


a2 2 0.153mm (3.31)
E* 54,591MPa
Indicating a deformed contact rectangle of width of 0.306 mm.
1 1
PE * 2
100 N 54,591MPa 2
pmax 417 MPa (3.33)
R ' 10mm
The maximum elastic contact stress predicted will be 417 MPa at the point of initial
(undeformed) contact. From equation 3.30, Figure 5 shows contact pressure plotted
against the lateral (X) distance from the center of contact. Because the surfaces are

14
smooth, the curve is continuous and has continuous first and second derivatives. The
contact stress peaks at the center and tapers to zero at the boundary between contact and
separation. The second curve in the plot shows a cumulative integration of the contact
load over the contact area. The sum total load is equal to the normal load applied to the
system (100 N).

Contact Pressure vs. X-Position, Elastic Smooth Cylinder-Plane Contact

400

300
p(x) (MPa)
Load (N)

200

100

0
-0.200 -0.150 -0.100 -0.050 0.000 0.050 0.100 0.150 0.200
x-position

Contact Pressure Cumulative Load

Figure 5: Contact Pressure Distribution, Uniformly Loaded Flat Blocks


From equations 3.39 & 3.40, Figure 6 shows the relationship between stress in the
roller and vertical (Z) location. Transverse (X) stress in the roller is caused by compres-
sion of the material at the contact surface and is localized. In this example, the
transverse stress falls to essentially zero for material more than 0.5mm (1/20th the radius)
above the initial contact point.

15
X- and Z-Stress in Roller vs. Z-height above Interface Plane

0
0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 0.900 1.000

-50

-100

-150
Stress (MPa)

-200

-250

-300

-350

-400

-450
Z-position (height) (mm)

x z

Figure 6: X- and Z- Stress in the Roller vs. Height (Z)

Note that the magnitude of contact pressure and stress in the material are equal at the
surface, but the signs are opposite (pressure is positive, but causes a compressive stress
which carries a negative value). The normal (Z) stress is also highly concentrated near
the contact. The theoretical minimum normal stress is not zero, but the net-section stress
at the diameter due to normal loading, or, in this example, 100 N / 20mm 5MPa . Also
notice that the stress state at the center of contact is symmetrical, x z pmax .

3.4 Relationship between Normal Load and Contact Pressure

The relationship between the normal load pressing the bodies together and the resul-
tant contact pressure and stress is nonlinear. A proportional relation would indicate that
the contact area was constant, and independent of load. This would require a contacting
body with a constant cross-section along the vertical axis. From equation 3.33 we can

see that pmax P , which results in a logarithmic relationship, as evident in Figure 7.

16
Peak Contact Pressure vs. Normal Load

1,400

1,200

1,000
Pmax (MPa)

800

600

400

200

0
0 100 200 300 400 500 600 700 800 900 1000
Normal Load (N)

Figure 7: Peak Contact Pressure vs. Normal Load

In practical terms, what is shown above is that, as load increases, the roller deforms
and translates down slightly. Due to the curvature of the roller, this causes the contact
area to widen, such that the load is distributed over a greater area. For this reason, the
average and peak pressures do not increase in proportion to an increase in the normal
load.

3.5 Relevance of the Flat Race (Infinite Radius R2) Assumption

The preceding sections work under the assumption that the bearing race surface can
be assumed to be planar, with no curvature. Using equations 3.20 and 3.33, a relation-
ship between race radius R2 and the maximum contact pressure (which is also equal to
the peak stress at the undeformed contact point) can be developed. Figure 8 shows the
results of this analysis. Note that, for these calculations, the value used for R2 was
negative, representing an outer race with a concavity of curvature that matches the roller
(typical of almost any roller bearing). The value plotted is positive to allow the use of a
logarithmic scale for better detail representation.

17
Peak Contact Pressure vs. Race Radius

500

417 MPa
400

300
Pmax (MPa)

At R2/R1=100, 0.5% error


At R2/R1=10, 5% error
vs. R2= assumption
vs. R2= assumption

200

100

0
10 100 1000 10000 100000
R2 (mm)

Figure 8: Peak Contact Pressure vs. Race Radius (R2)

For a race radius similar to the roller radius, the contact pressure differs greatly from
the flat plate assumption. However, for the range of race radius ratios between 10 and
100, the predicted contact pressure is very similar to that calculated by assuming the
outer race to be flat, with a maximum error of 5% that decreases as the ratio grows.
Given that nearly all roller bearings in commercial use have an R1/R2 ratio in this range,
the assumption that the outer race is a flat plate involves an acceptably low level of error.

18
4. FE Model and Validation
4.1 Method

Following the mathematical modeling conducted in section 3, a computer-based fi-


nite element model of the example in section 3.3 was created and run using a contact
modeling algorithm. Demonstrably similar results between the mathematical and
computer models increases confidence in the accuracy of the mathematics and validates
that the results obtained using the computer can be used as the basis for extension and
exploration of contact behavior with elastoplastic material properties and rough surfaces.

The first model will, as stated, attempt to duplicate the results of section 3. Follow-
ing that, a proportional limit and plastic yielding behavior will be introduced into the
model with the original geometry for comparison. Finally, the smooth planar lower race
will be replaced by a series of rough surfaces having sawtooth, sinusoidal, and random
contours.

4.2 FEA Tool Validation Test Case

To show the applicability of the planned finite element analysis technique, two rec-
tangular 2D blocks were created using ANSYS Mechanical 11.0 (the same software
version was used throughout this work). The two-dimensional solid was meshed using
4-node (linear) quadrilateral PLANE42 elements. The plane strain formulation of this
element was selected. Planar finite elements can be based on plane stress assumptions
(thin section, no stress in the depth direction), or plane strain (no expansion in the depth
direction). As the contact will create stress in the depth direction at the contact surface,
but the strain in that direction will be minimal, the plane strain formulation is the most
representative of the physical situation, without resorting to a 3D model and the in-
creased complexity that adds.
The reader interested in further information on the finite element method and quadri-
lateral plane element formulation is referred to the text Applied Finite Element Analysis6.

6
Applied Finite Element Analysis, Segerlind, L. J., USA, 1984

19
A symmetrically paired contact surface was created using CONTACT172 and
TARGET169 elements along the entire interface surface, and elastic material properties
as described in Table 1 were applied. This model represents completely conforming
contact with the simplest possible geometry at the interface (paired flat surfaces).
It is important to note that the coordinate system used in Section 3 differs from that in
Section 4. For the finite element model, two-dimensional modeling uses only X- and Y-
degrees of freedom, so X is the horizontal (same as the mathematical model), but Y is
now the vertical axis (positive up). The finite element model with constraints is shown in
Figure 9. In this section, y will be the same as z was in the calculations of section 3.

Figure 9: Conforming Elastic Blocks FE Model with Contact

A uniform 100 MPa pressure was applied to upper surface of the top block and the
lower surface was constrained in all degrees of freedom (DOF). This scenario should of
course result in an even distribution of the initial 100 MPa pressure over the entire
surface. As can be seen in Figure 10, contact pressure at the interface is within 0.5% of
the theoretical value. Variation of contact pressure along the length of the contact
surface is attributable to the slightly different state of stress at the edges and center of the
part. Material at the center of contact is constrained from Poisson expansion in the
transverse direction by the surrounding material, somewhat analogous (but not identical)
to a hydrostatic stress state. Material at the edges of contact is free to expand in the X

20
direction. The exact (continuous) solution will be equivalent to the finite element
solution for a mesh that has an infinite number of elements of negligible size.

Figure 10: Contact Pressure Distribution, Uniformly Loaded Flat Blocks

The same basic philosophy used to create this test case will be continued throughout
the FEA investigations. Plastic material properties and varying surface roughness
profiles will be added to the model in the following sections.

4.3 Baseline Elastic Model

ANSYS was used to create a two-dimensional solid model of the geometry from sec-
tion 3.3. To minimize the size of the model and the solution time, symmetry boundary
conditions about the plane x=0 were applied such that only half the geometry needed to
be explicitly modeled. Element types used were the same as the model in section 4.2,
and both the race and the roller used the same (linear elastic) material properties as the
calculations in section 3.3. The PLANE42 two-dimensional element offers several
formulations, including the Plane Stress and Plane Strain philosophies. The formula-
tions differ in how they relate the 2D model to the physical, three-dimensional case.
Plane stress is generally used for very thin bodies (sheets, shells, etc). It assumes that
the body is thin enough that there is negligible stress in the thickness dimension, i.e.,

21
zz zx zy 0
. The bearing model we are using is significantly thick in the Z (into
the page) direction, and the contact pressure induces stress that acts in all directions at
the center line of undeformed contact, so this is not a representative assumption for our
problem.
The plane strain assumption, which was used throughout the finite element modeling
described in this work, describes a situation where deformation in the thickness direction
is negligible (zero). All externally applied loads act only in the X- or Y-direction, and
displacements are functions of X- and Y-location only. As there is no Z-strain,
zz xz yz 0
. When modeling a 2D section representative of a prismatic (constant
cross-section) body, all loads are applied in the XY-plane, there is no Z-deformation,
and there is no variation of stress with Z. It can be easily justified that strains in Z, along
the axis of the roller, are small and not generally of interest. Therefore, the plane strain
element formulation accurately represents the problem at hand.
While the calculations of section 3 neglected the effects of friction, and involved
normal load transfer only at the contact point, if was necessary to add a small but negli-
gible amount of friction to prevent the roller from running away tiny forces close to
zero as artifacts of the model can cause an unrestrained body to have high rigid body
displacements. Friction at the contact area eliminates this. A friction coefficient at the
contact area of 10 was used. Test cases run with the coefficient set both an order
5

of magnitude higher and lower showed no measurable difference in results, demonstrat-


ing that the addition of this friction did not significantly affect the results.
Figure 11 shows the finite element model and the mesh. The symmetry boundary at
the YZ plane is expanded to show the full shape. 8142 total quadrilateral elements were
used to mesh the shapes, with element size as small as 0.03mm in the region of contact.

22
Figure 11: Elastic FE Model, Smooth Cylinder & Block

The lower surface of the block is constrained in all DOF and there is a 100N force
applied downward at the center node of the roller. A comparison was run with an
equivalent load applied as a distributed pressure totaling 100N across the horizontal
radius of the roller. This resulted in some rigid body rotation of the quarter of the roller,
artificially increasing the deformed contact width and creating a sharp spike in contact
pressure at the outer edge of the contact area. The single point load most accurately
recreated the predicted results, so this load application method is used subsequently.

23
4.3.1 Elastic Results

Figure 12: Contact Pressure Distribution, Elastic Cylinder & Block

Figure 12 shows the contact pressure overlaid on the deformed model. Actual defor-
mation is very small and difficult to see. The peak contact pressure is 410 MPa,
differing by less than 1.7% from the theoretical value of 417 MPa. The measured half-
width of the contact area, 0.176mm, differs slightly from the prediction (0.153mm), but
the nature of the contact algorithm is such that contact status at each node is binary
there either is or is not contact. The 0.026mm difference is smaller than the minimum
node spacing in the region (0.03mm), so the error is easily explainable within this
tolerance band. As a test of the mesh, both the nodal (averaged between adjoining
elements) and element, or non-averaged, contact solutions were compared. Within the
resolution of the ANSYS output, there was no difference in results. Convergence of the
elemental and nodal solutions is a good indication that there will be no accuracy benefit
from further mesh refinement, and the solution is approaching the theoretical (continu-
ous, non-finite element) value(s).
Figure 13 & Figure 14 below show the resulting stress fields in the region of contact
for the elastic model.

24
Figure 13: Elastic Results, X-stress

Figure 14: Elastic Results, Y-stress

The above plots readily agree with the predictions plotted in Figure 6. The peak
stress values for both axes occur at the center point of initial contact, and the transverse
X-stress dissipates much more quickly as the location moves upward into the material,

25
away from the contact interface. Both components of stress peak at values very near the
theoretical maximum of 417 MPa. The X- and Y-stresses are of interest in the elastic
model, for verification of the theoretical calculations. However, these quantities are of
less interest practically, as the potential for damage (due to static or cyclic fatigue) to
either of the contacting components depends on the peak stress regardless of orientation,
or the principal stress components. The 3rd principal stress value is the minimal tensile
or maximal compressive stress state in any orientation. Further comparisons will use the
3rd principal stress as a convenient single quantity that is representative of the potential
for harm to the components due to applied loads.

Figure 15: Elastic Results, 3rd Principal Stress

Comparing Figure 14 & Figure 15, it is evident that the contribution of the vertical Y-
stress dominates the third principal stress. In other words, away from the center of
contact, the orientation of the maximal compressive stress in the region of contact is
close to vertical.

26
Figure 16: Elastic Results, 3rd Principal Strain

The observed principal compressive strain in the contact region peaks at 0.00287 or
0.29%. This exceeds the defined proportional limit strain of 0.2%, such that some
plastic deformation will be observed. Plastic, or unrecovered, deformation permanently
modifies the shape of the body and, if it occurs repeatedly, can result in much shorter
fatigue life than would be expected for a body undergoing repeated loading and defor-
mation solely in the elastic range of behavior.
Typically, fatigue life at high cycle counts is limited by cracks which initiate at tiny
flaws inherent in the material. These microcracks have sharp roots which concentrate
stress and allow plastic deformation to occur on a very small scale. Over time, these
small cracks grow into detectable cracks and can ultimately lead to rupture or component
failure. Further development of these fatigue concerns is beyond the scope of this work,
but from this explanation it is clear that plastic deformation, occurring on the macro
scale (assuming homogeneous material properties) during each cycle of the roller
interface, could lead to failure within a relatively short period of time. A design goal for
a bearing would be to ensure no macroplasticity during loading within the design range.
For the purposes of this exercise, a load that caused plastic deformation was deliberately
chosen to allow comparison of the plastic deformation resulting from loading the smooth
and rough surfaces.

27
4.4 FEA Model including plasticity effects

ANSYS allows input of a curve to describe an elastoplastic materials stress-strain


relationship. For this section and those following, the curve in Figure 17 was applied to
the material model. It is a simple bilinear curve, showing behavior matching the defined
elastic modulus of 100 GPa up to the 0.2% proportional limit, and a second, plastic
linear behavior from the yield point out to the ultimate strength of 465 MPa at 5%
elongation to failure. The material properties used in this work are based on scientifi-
cally developed statistical properties for the material, but the numbers have been
modified to some degree to protect the proprietary source of the information. It is the
judgment of the author that the results and conclusions herein are somewhat independent
of the actual values given for the material properties, but every attempt was made to use
properties that are representative of typical roller bearing materials in use at the time of
this writing.
Bilinear Stress-Strain Relationship for AMS6490 Bearing Steel at 70F
Yield Point Modified to reflect E=24,400 MPa @ 0.2% Strain for FEA

500

450

400

350

300
Stress (MPa)

250

200

150

100

50

0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
Strain (in/in)

Figure 17: AMS6490 Bearing Steel Stress-Strain Relationship

Adding material plasticity more accurately takes into account the relaxation of load
that occurs when a material is pushed beyond its yield point. When an area of material
begins to yield and its stiffness decreases, typically this results in the forces being
redistributed to other sections of material that remain in the stiffer elastic region. Load
redistribution from plasticity will tend to decrease the observed peak stress as compared

28
to an analysis conducted using a purely elastic material model. Also, again due to the
reduced stiffness, higher total strain values should be expected in the region of yielding.

4.4.1 Plastic Results

Figure 18: Contact Pressure Distribution Including Plasticity Effects

Figure 18 shows the contact pressure distribution from the plastic model. Note that,
as predicted, the peak stress decreases. The observed contact half-width in the model
was identical to that of the elastic model at 0.176mm. Given the nodal spacing as
described in Section 4.3, we can conclude that the contact area did not increase by
enough to trigger contact at the next node out from the center, such that the measure-
ments were the same.
Figure 19 & Figure 20 show the maximum compressive stress and strain near the
interface for the plasticity model. Note that, as with the elastic model, the peak com-
pressive stress is identical (within 1.4% margin) to the peak contact pressure. The value
of the peak stress/pressure has dropped by almost 8% to 370 MPa as the load is distrib-
uted over a wider area. The value of the highest compressive strain in the material has
increased by almost 10% to 0.306%. With plasticity allowing the material to yield and
become less stiff, once a region becomes plastic, it will strain to a much greater degree.

29
The region of permanent yield is still small, however, as can be seen in Figure 20 (the
three darkest levels of blue are beyond the 0.2% strain offset).
This elasto-plastic smooth contact representation will be used as a baseline for com-
parison of the rough surface models that will be developed and analyzed in Chapter 5.

Figure 19: 3rd Principal Stress, Elasto-Plastic Model

Figure 20: 3rd Principal Strain, Elasto-Plastic Model

30
5. Extension to Non-smooth Surfaces
5.1 Rough and Irregular Surfaces in Contact

The contact theory described in section 3 assumes both contacting surfaces are nomi-
nally smooth, i.e., there is a continuous contact surface within and limited to the
theoretical area of contact. For a surface that is jagged or uneven on a small scale, the
apparent area of contact, based on the nominal shape, is not an accurate measure. The
real area of contact is discontinuous, concentrated at the high points or asperities of the
rough surface. This real contact area is significantly smaller than the theoretical area
would be, and, as such, leads to higher contact pressures at the interface. For signifi-
cantly stiff materials (including metals), elastic and plastic deformation of the asperities
at the interface is not sufficient to create conforming contact under load.
Machining and other manufacturing processes often lead to a surface finish that re-
sembles a one-dimensional sinusoid, or repeating pattern, overlaid with a random
variation of much smaller amplitude. For example, the cutting edge of a fly cutter
creates a slight ridge as it removes material, and the head traverses a short distance each
revolution of the tool, creating a series of ridges that are of similar (but not the same)
geometry. Vibration of the tool and work piece, material inhomogeneity, and other
factors contribute to the difference in material removal with each cutting edge revolu-
tion, producing the overlaid random variation in the asperity height. Milling heads,
rotary grinders, and lathe cutters create similar patterns of alternating ridges.

Figure 21: Fly Cutter Typical Surface Finish Pattern7

7
Photo credit Mini-lathe.com

31
In the sections that follow, the finite element model of the roller-race contact, includ-
ing plasticity effects, will be modified to include different artificially rough surface
profiles. In keeping with the above description, the 2-dimensional model geometry will
be retained, and the surface roughness will occur in one dimension (X). The surface
contour will not vary with Z-location (model thickness).

5.2 Sawtooth Roughness of Contacting Surface

The first non-smooth surface profile analyzed was a standard sawtooth shape. This
shape consists of lines connecting a series of alternately high and low points with equal
horizontal spacing. Here, the troughs would represent passes of the cutting tool as the
material advances, with peaks between each pass. It is a good, if simplified, representa-
tion of surface roughness due to machining processes.

Sawtooth Surface Roughness Profile

0.4

0.3

0.2

0.1
Z (mm)

0
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

-0.1

-0.2

-0.3

-0.4
X (mm)

Figure 22: Sawtooth Wave Roughness Profiles

However, the sawtooth curve presents many difficulties when modeling using finite
elements. When using the more robust quadrilateral element shape (triangular elements
are prone to greater stiffness variation with size and shape), only one quad element will
make up each sharp peak. A single node will serve as the initial point of contact. The
single contact node, in addition to the marked difference in surface slope between the

32
contact and target surfaces, greatly increases the likelihood that, as the two surfaces
come together under load, they pass through one another undetected by the contact
algorithm. Contact modeling works by assigning an interface stiffness per unit area to
the contact elements which are overlaid on the solid elements. Since the effective
interface area is zero, the stiffness is zero, and the surfaces pass through one another
uninhibited. A model with sharp contact points can only be made to converge on an
equilibrium solution by dividing up the load into many time-consuming partial load
steps, or reducing the contact stiffness value and increasing the radius within which the
algorithm searches for potential contacting bodies. These modifications reduce the
accuracy of the model and increase solution time by an order of magnitude or more.
A more practical solution is to truncate the sawtooth, or chop the top of the pointed
shapes. This allows an increased element count and a larger area at the interface, greatly
aiding convergence. After cutting off the point, the remaining shape greatly resembles a
square wave, with its flat contact surfaces. Under a reasonable range of loads and
deformations, only the flat tops of the asperities will contact the roller. For these rea-
sons, the second roughness model will use regularly spaced rectangular asperities.

5.3 Square Wave Roughness of Contacting Surface

For the first roughness case, a repeating series of rectangular shapes 0.05mm wide
and 0.10mm high replaced the original flat surface of the bearing race. The material
properties and element formulations used were consistent with earlier cases.

33
Figure 23: Contact Pressure Distribution, Square Wave Surface

The contact pressure distribution within the areas of contact still displays a roughly
parabolic shape that parallels the pattern for the smooth cases. The deformed roller
contacts three of the peaks for a total area of 0.150mm, approximately the same as the
smooth case. The peak contact pressure, at the center, is also very similar to the smooth
case (381 MPa vs. 369 MPa). However, the minimum pressures are different. The
trailing edge of the contact has a pressure greater than zero due to the discontinuous
contact surface. The integral of the contact pressure, or the area of the three plot seg-
ments in Figure 23, will be equal to the normal load, just as for the smooth cases.
While the total contact area for the square wave case remains roughly the same, it is
evident that, due to the round profile of the roller, the center asperity bears a larger share
of the initial load as the contact interface deforms. The vertical displacement of the
center of the roller due to deformation rises significantly when roughness is added to the
model. The smooth elastoplastic roller moved down 0.008 mm under the 100N load,
but, in the square wave case, the shift was 25% higher, or 0.01 mm. Movement is
quoted in terms of rigid body displacements, i.e., not accounting for local deformation at
the contact area. The roller must displace further (deforming the asperities in turn) to
engage more material and distribute the normal load.
Figure 24 shows a very different circumstance in terms of plasticity. In contrast to
the smooth surfaces, which had very limited local plastic yielding, all of the rectangular

34
asperities in contact with the roller are plastic through their entire section, while the base
material (the solid below the asperities) undergoes very little plastic yielding. In this
case, each rotational cycle of the bearing in the model would result in local yielding of
the rough race surface, with potentially drastic effects on the useful life of the product.

Figure 24: Square Wave Roughness, 3rd Principal Strain

It is important to note that the plasticity and stress concentration resulting from the
rough surface remains a local effect. Comparing the macro views of Figure 19 and
Figure 25, the stress distribution away from the local roughness is very similar in magni-
tude and pattern.

35
Figure 25: Square Wave Roughness, 3rd Principal Stress (Wide View)

Figure 26 shows an enlarged view of the contact region from Figure 25, the 3rd prin-
cipal stress in the region of contact. Notice that there are areas of the base material that
experience stress in excess of the value required to cause yielding in the asperities, yet
the roller does not become plastic, and there is very little plasticity in the base of the
race. The Poisson effect again comes into play here, such that the base material is
constrained from transverse expansion, reducing its propensity to yield.
It can also be seen that where roughness peaks meet the rollers surface, there is sig-
nificant compressive stress in the roller (equal to the stress in the race and the contact
pressure at the surface, as before). In between the punch loading of the square peaks, at
the unloaded regions, there is actually a net tensile stress in minute local areas of mate-
min
rial. Let us define a value R, the stress ratio, such that R . The stress ratio is an
max
important factor in the prediction of fatigue life, which is generally dependent on the
ratio between the mean and alternating stress in the material, as well as the magnitude of
stress. With a smooth surface, the model predicts compressive stress only, such that
R=0. As the roller rotates, the points of contact with the peaks will vary with each
rotation (especially with extension to a random or probabilistic surface roughness).
Interspersed tensile and compressive stresses will lead to a degree of stress reversal (an

36
R-value less than zero), which can accelerate damage due to cyclic stress, both on the
micro and macro scales.

Figure 26: Square Wave Roughness, 3rd Principal Stress (Close View)

The analysis thus far has considered geometry that, even when not locally smooth, is
still regular and very much a theoretical construct, not a case that would be encountered
in industry. Roughness profiles resulting from actual production processes are irregular
to varying degrees, and the height of the asperities is best represented using a probability
distribution which gives the chance of finding a peak of a certain size on the rough
surface.

5.4 Probabilistically Distributed Roughness of the Contact Surface

As mentioned above, real-world roughness on manufactured surfaces will not follow


a regular pattern as in the previous cases. Milled or turned surfaces will often show a
pronounced, dominant wavelength pattern that is determined by the machine feed rate
and workpiece/tool rotational speed, which in turn establishes the amount of material
removed with each pass of the cutting edge. However, while there is a wide variation in
the profiles that can be produced by common manufacturing methods, most if not all are
well described by a probability density function. This function describes the probability
of a given surface height at any randomly selected point on the rough surface.

37
Figure 27: Typical Randomly Distributed Surface Roughness Profile

The profile of the surface can be represented as a height, which is a function of lateral
position (x). The sign conventions for the calculations of section 5 will revert to that
used in the calculations of section 3. The coordinate system for the finite element
portion will also mirror the previous FEA work, due to the coordinate system limitations
of the 2D analysis. In practice, there is an upper and lower bound to the possible height

at a given location, i.e., z min z ( x) z max . The height of a ground surface cannot
exceed the height of the original surface before grinding, and there cannot be a greater
thickness of material removed than existed to begin with.

5.4.1 Normally Distributed Roughness

It is convenient to describe the collective heights of the asperities in terms of a popu-


lation mean, m, and a population standard deviation, . The standard deviation is
defined as the square root of the variance (2):
L
1
( z m) 2 dx
2
5.1
L0

38
where z is the height of the surface at a distinct point, and L is the x-dimension length of
the surface in question. Over the length, the normal, or Gaussian, density function is
given as:8

1 ( z m) 2
p( z ) exp 5.2
(2 )1/ 2 2 2

Figure 28 shows the probability density function. The area under the curve (or integral)
from - to a value of the standard deviation represents the cumulative probability of that
value, or the chance that a number equal or lower will occur. The normal distribution
has a finite (positive) value for all values of . The integral from - z is equal to
unity (1.0), as the total cumulative probability for all possible values must equal 1.0, or
100%. For standard deviations beyond 3, however, it can be seen that there is a very
low rate of occurrence. There is only about a 1 in 1000 chance of a value occurring that
is greater than three standard deviations away from the mean value. Practically, this is
acceptable, as we realize that there is a very low probability of a physical impossibility
occurring in the roughness model. An example would be a trough that was deeper than
the total material thickness. The actual probability of this is zero, or course, but if the
model shows a probability of, say, 1 in 109 occurrences, there is little danger of this issue
arising in our model.
Normal (Gaussian) Probability Density Function

0.4

0.3
Probability

0.2

0.1

0
-4 -3 -2 -1 0 1 2 3 4
Standard Deviation

Figure 28: Gaussian Probability Density Function, Normalized

8
From Introduction to Tribology, Bhusan, B., USA, 2002

39
For simplicity, it is helpful to normalize the density function to a curve representing a
Gaussian probability density function with a central mean at z=0 and a unit standard
deviation =1 (as in Figure 28). This is accomplished by defining a normalized inde-
pendent variable z*.
zm
z* , or, conversely, z z * m 5.3a,b

Substituting 5.3 into 5.2 gives:

1 ( z*)2
p( z*) exp 5.4
(2 )1 / 2 2

Integrating this expression symbolically to obtain the cumulative distribution is not


feasible using standard methods. Figure 29 was plotted using the trapezoidal rule to
integrate piece-wise. As described above, the total sum is 1.0 at the limit of +.
Cumulative Probabilty Distribution, Normal

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
-5 -4 -3 -2 -1 0 1 2 3 4 5

Figure 29: Normal Cumulative Probability Distribution

Numerical methods have facilitated the integration and subsequent inversion of equation
5.4, resulting in an expression that will produce normally distributed heights at random
to generate a surface profile.
R 0.135 (1 R) 0.135
zi 5.5
0.1975
where zi represents the profile height at a location, and each instance of R is a random

number generated from a normal distribution over the interval (0,1). Using Microsoft

40
Excel 2003 spreadsheet software, the sample profile shown in Figure 30 was generated.
The points generated in the spreadsheet were input to the FE model. Using a spline, or
piecewise polynomial curve fit, the plot was duplicated and meshed with contact ele-
ments.

Normally Distributed Surface Roughness Profile


Sk=0, K=3
0.04
Height (Z)

0
-0.04 0 0.2 0.4 0.6 0.8
-0.08
-0.12
Lateral Position (X)

Figure 30: Normal Surface Profile

5.4.2 Weibull Distributed Roughness

While a normal distribution offers a useful approximation of most surface profiles,9 it is


often necessary to distinguish between the characteristic roughness patterns created by
common material removal processes. To aid in the description of these non-normal
surfaces, we will add two additional parameters of the probability density function
(along with the mean and variance as described earlier). It will also be helpful to intro-
duce the Weibull probability distribution at this point.
The skewness (Sk) is a value representing the symmetrical nature of the density func-
tion, or the lack thereof. A symmetrical density function, such as that in Figure 28, will
have a skewness of zero, as it is perfectly balanced about the mean value. The mean of a
non-skewed density function is equal to its median. A positive skewness (Sk>0) indi-
cates that the function is skewed right, meaning the right tail is longer. For a positively
skewed function, the mean is larger than the median. Conversely, a density having

9
While very few manufacturing processes produce a surface roughness that truly fits the normal distribu-
tion, random processes such as air-propelled shot blasting have been used to create Gaussian roughness
profiles under laboratory conditions.

41
negative skewness (Sk<0) will have a longer left tail, and a mean less than its median.
Plot (a) of Figure 31 shows functions with positive, negative, and zero skewness.
Kurtosis (K) represents the degree of peakedness of the density function, or the
strength of the central tendency. A given level of variance could represent either a small
number of outliers located far from the mean, or, alternatively, a larger occurrence of
values that differ from the mean only slightly. A higher value of kurtosis describes a
function with a sharper peak and thinner tails, or a greater degree of clustering about
the mean. The normal density function, typically an appropriate baseline, has a kurtosis
of 3, leading some to define an excess kurtosis that differs from K by 3 and gives the
normal function a value of zero. In this work, the standard kurtosis representation
(normal has K=3) will be used exclusively. Plot (b) of Figure 31 shows functions with a
kurtosis greater than, less than, and equal to three.

Figure 31: Illustrations of Skewness and Kurtosis10

10
Introduction to Tribology, Bhusan

42
The Weibull probability curve is defined as having a cumulative distribution described
by:
b
z

F ( z) 1 e a
5.6
Where a is the scale parameter and b is the shape parameter. The Weibull distri-
bution differs from the normal distribution in that it is not defined for negative values
(see a) Probability Density Function b) Cumulative Probability Distribution
Figure 32). As such, it is often used for modeling and predicting quantities such as
the size of a particle or the useful life before failure of a product. Both of these quanti-
ties can never be less than zero, so the probability density fits the situations well. By
setting a limit to the depth of the roughness (a minimum Z value) and defining the
surface profile in terms of height above that minimum, the Weibull distribution can be
applied to the rough surface model.
The advantage of using this description of the surface is that, by varying the scale and
shape parameters, density functions with a wide range of Sk and K can be created,
broadening the type of roughness that may be described. A Gaussian distribution,
described in terms of only a mean and variance, will not have a nonzero skewness.

a=0.5 b=2
a=1.0 b=2
a=1.5 b=3
a=3.0 b=4

a=0.5 b=2
a=1.0 b=2
a=1.5 b=3
a=3.0 b=4

a) Probability Density Function b) Cumulative Probability Distribution

Figure 32: Weibull Distributions with Varying Parameters a & b11

11
Figure 32 plots courtesy of Philip Leitch, created using MathCAD

43
Each shaping process has its own, characteristic roughness patterns related to the method
of material removal. Two common processes used to create bodies of revolution are
lathe turning and grinding. Figure 33 shows typical values of skewness and kurtosis for
these processes as well as several others. Honing is typically used to smooth the surface
of bores. Honing can produce a very smooth surface, as evidenced by the high Kurtosis
values possible (high K indicates significant clustering of values about a central ten-
dency). Milling and electrical discharge machining (EDM) are very flexible in
application and could be used to produce almost any shape. Most bearing races will be
produced by either lathe turning followed by grinding or, alternatively, die forging
followed by grinding to produce a smooth surface for the rollers to contact.

Figure 33: Roughness Characteristics of Common Machining Processes12

12
Introduction to Tribology, Bhusan

44
Notice that there is a pronounced difference in skewness for turning and grinding. The
turning density is skewed towards lower heights, and the grinding density is skewed
towards higher heights. This is easily explained by the nature of these procedures.
In lathe turning, the workpiece is rotated and a stationary tool is brought into contact
with the surface. Often, the tool is roughly rectangular or triangular in shape and has a
sharp point on its leading cutting edge. Each pass removes a thin strip of material,
leaving a sharp valley with peaks on either side. The width of the peaks depends on the
distance between each pass of the tool, but the effect will be the same. Some lathe tools
have a rounded cutting edge to reduce the height of the resulting asperities, but the speed
of material removal is greatly reduced, tool life is decreased, and tool friction is in-
creased, limiting the tools applicability and cost-effectiveness.
Figure 34 shows a sample roughness profile that was generated using a spreadsheet
similar to that used in section 0, only with the Weibull function (equation 5.6) substi-
tuted. An additional advantage of the Weibull function is that it is symbolically
integrable and differentiable, eliminating the need for numerical solution as an interme-
diate step. Notice how the surface height appears skewed towards the lower range of
values, with occasional sharp peaks. The variance does not seem to be significantly
different from the normally distributed curve in Figure 30, and given that the kurtosis is
similar (3.66 vs. 3.0), this is to be expected.

Weibull Distributed Surface Roughness Profile


Turned Surface, Sk=1.05, K=3.66

0.04
0
-0.04 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-0.08

Figure 34: Typical Lathe Turned Roughness Profile

Grinding is typically accomplished by bringing a high-hardness, abrasive material into


contact with the workpiece. For round parts, often both the work and the abrasive rotate
in opposite directions, but at least one must be in motion for material to be removed.
The abrasive material has a much wider area of contact with the rough surface, and, as

45
such, has a tendency to chop the tops of the asperities, without altering or deepening
the valleys. If the surface begins with a profile similar to Figures Figure 30 or Figure 34,
following grinding it will have a flattened appearance with less sharp peaks protruding
above the mean surface. Figure 35 shows a randomly generated Weibull distributed
roughness profile again generated with the aid of a spreadsheet.

Weibull Distributed Surface Roughness Profile


Ground Surface, Sk=-0.94, K=4.86

0.04
0
-0.04 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
-0.08

Figure 35: Typical Abrasive Ground Roughness Profile

The shape and scale parameters were adjusted to provide the height density of a typical
ground surface. The negative skewness is readily apparent in comparisons with the
previous profiles. There is a pronounced trend towards the higher values of height (near
zero) as compared to the other plots. Also, as would be expected, given the higher value
of K, the variance appears lower, with the resulting surface smoother than the turned
feature.

5.4.3 Solution Convergence Challenges with Irregular Surfaces

Similar to the experiences of Section 5.2, many difficulties were encountered attempt-
ing to drive the probabilistic rough surfaces to converge on a solution. The surface
contours described in the previous sections have better, but by no means ideal, geometry
for contact analysis. The asperity tips are rounded and have a finite area, but there is still
a very, very small real area of contact during the initial stages as the bodies are pressed
together. This leads to high contact stress and quick onset of plastic yielding. As the
modeled material becomes plastic, its effective stiffness decreases dramatically, increas-
ing the speed at which the contact and target bodies approach each other.

46
Speed is a relative quantity here: contact FE models are solved by gradually ramping
up the load to the specified value, with each increment of load fully solved to satisfy the
equations of equilibrium. Therefore, even a static model such as this one is solved in
terms of an arbitrary time (if there are 10 time steps, each will have a value of 0.1 greater
than the previous interim solution point).
The results for the three probabilistic roughness cases echoed those of the sawtooth
case, although with marginally better results. It was possible, with some manipulation of
the contact element parameters, to force the model to converge at low normal loads (up
to about 16N total, or 8N per side with symmetry). Beyond that, the contact surfaces
tended to pass through each other, leading to unconstrained rigid body motion, which
prevents the solution process from reaching equilibrium.
There are several tactics available to aid convergence of nonlinear (iteratively
solved) contact models in ANSYS. Increasing the pinball region radius swells the
circle around each contact node within which a potentially nearby or contacting node is
searched for. This helps prevent pass-through but can dramatically increase the time
required for solution calculation. Decreasing the contact stiffness reduces the push-
back that the interfering surfaces experience, facilitating equilibrium. Increasing the
penetration tolerance allows more leeway in terms of position of the two surfaces to
still be considered a converged solution. Each of these choices will have the same direct
result of increasing the potential for error in the solution. The iterative solving process
repeats the same matrix algebra with increasing loads and increasingly tight criteria.
The forces and reactions must be equal at the end of an equilibrium calculation, within a
certain tolerance for error, for the solution to be considered complete. Increasing that
tolerance will aid solution time at the expense of confidence in the results. It is almost
always possible to get a difficult model to converge to a solution, but, if you have
manipulated the parameters to favor solution to such a degree that you no longer trust the
answer, there is little value to be gained from the analysis.
Increasing mesh density (smaller elements) can also help in some circumstances,
but, when there is only a small point of potential contact, it is not always useful, and that
proved to be the case here. Figure 36, Figure 37, & Figure 38 show the results (3rd
principal strain) using the highest normal loading that was able to converge without

47
excessive manipulation of the contact algorithm. Only one asperity was in contact and
bore the entire normal load in each of the three cases. The peak contact pressures in the
three load cases were 1372 MPa, 476 MPa, and 1242 MPa, respectively, showing orders
of magnitude increases in pressure per unit normal load, as compared to the smooth and
square wave cases. The real area of contact was, as would be expected, very small, on
the order of 10-2 mm. The slight gaps between contacting surfaces visible in the plots is
an illusion: Any existing gaps due to modeling were detached from the solution by using
a normal offset of the contact element surfaces to ensure the contact pairs were exactly at
the moment of contact (no gap, no penetration) at time zero for the analysis.

Figure 36: Gaussian Roughness Results, 10N Total Load

Figure 37: Weibull (Turned) Roughness Results, 2N Total Load

48
Figure 38: Weibull (Ground) Roughness Results, 10N Total Load
One other way to potentially eliminate the runaway bodies that occur because of
plastic material properties is, not surprisingly, to remove plasticity effects from the
model. This has the dubious effect of causing the bodies to behave linearly, even at very
high strain values, which is not generally realistic. A test case using a high load (1000N)
and elastic material properties identical to those of section 4.3 was the only case that
allowed enough deformation (without rigid body motion) to cause the second-highest
asperity to contact the roller surface and begin to share load with the tallest peak (Figure
39).
Another difficulty arose with this method, related to the element formulation. When
one or a very small number of elements, with a relatively small cross-sectional area, are
supporting the entire normal load of contact, there is a tendency for the elements at the
contact surface to be squashed disproportionately, especially if the section becomes
wider as one moves towards the bulk body, as in the case of the roughness models.

49
Figure 39: Gaussian Results, Elastic Properties, 1000N Load

In modern FEA codes, the actual geometry of each element is mapped onto an
ideally shaped element using a transformation matrix. The Jacobian ratio describes
how good of an approximation this mapping is, and is a useful measure of element
quality. If an element is compressed to the degree that edges pass over one another, the
Jacobian ratio becomes negative and the element solution, while definite, has no applica-
tion to a physical case and cannot be interpreted as actual results. High loads, tapered
asperities, and elastic material properties led to several cases where the elements at the
tips of the peaks suffered inversion.

50
6. Conclusions
Finite element analysis tools create many interesting opportunities for investigation
of the behavior of complex geometry under loading. As was demonstrated, the results
agree very well with universally accepted closed-form solutions for cases in which such
a calculation is possible. This allows the experienced user to have great confidence in
the results obtained using an FEA code to predict product performance, prior to the
fabrication of even a prototype. However, any analyst or researcher should maintain
measured caution as FEA, like any computer tool, subscribes to the programmers
adage: garbage in, garbage out. The computer tool is only performing a series of
repetitive calculations at the instructions of the operator. One must be careful not to give
instructions that cause erroneous output.
While the ANSYS methods employed in this work showed great promise for the in-
vestigation of smooth and relatively smooth surfaces, it has been shown that extending
this success to sharp, rough surfaces is not a simple task. If solutions for rough contact
are desired, it is suggested that the analyst look at a much smaller, more detailed section
of the geometry. Submodeling and substructuring techniques would be potentially
applicable, as well as simply modeling only the contact surfaces and ignoring the bulk
geometry outside the contact region.
Despite the difficulties encountered, it is clear from the results that were achieved
that, when rough surfaces are involved, real stress and strain values experienced on a
micro scale at the interface can be significantly higher than what might be predicted by
analysis using a smooth surface assumption. If long service life and an accurate predic-
tor of that life is required, a method should be devised to account for the roughness
effects and include them in the life predictions. This could extend from simply adding a
factor of safety to the analysis, to calibrated statistical life values with empirical data at
the complex extreme. It is not practical to actually measure and analyze surface rough-
ness in a production environment, as each piece will be different. The goal should be to
capture the overall (statistically driven) behavior that can be expected so that the predic-
tions and the experienced service life are aligned. This calibrated model will allow
bearing components to be properly designed to have adequate, but not excessive, life
capability to meet the needs of the products in which they are put to use.

51
Bibliography
Fundamentals of Machine Component Design, Juvinall, R. C. and Marshek, K. M.,
John Wiley & Sons, New York, USA, 2000

Rolling Tests of Plates, Wilson, W. M., University of Illinois at Urbana-Champaign,


United States, 1929

Contact Mechanics, Johnson, K.L., Cambridge University Press, Cambridge, United


Kingdom, 1985

Applied Finite Element Analysis, Segerlind, L. J., John Wiley & Sons, New York, USA,
1984

Mechanical Behavior of Materials, Dowling, N. E., Prentice Hall, New Jersey, USA,
1999

Introduction to Tribology, Bhusan, B., John Wiley & Sons, New York, USA, 2002

B46.1 2002: Surface Texture, Surface Roughness, Waviness and Lay, American
Society of Mechanical Engineers, New York, USA, 2002

Mechanical Metallurgy, Dieter, G. E., McGraw-Hill, Boston, USA, 1986

Roarks Formulas for Stress and Strain, Young, W. C. and Budynas, R. G., McGraw-
Hill, New York, USA, 2002

Engineering Mechanics, Timoshenko, S. and Young, D. H., McGraw-Hill, New York,


USA, 1956

52

Potrebbero piacerti anche