Sie sind auf Seite 1von 9

Simon Kabus

Vestas Turbines Research


and Development, Motion Systems,
Vestas Wind Systems,
8200 Aarhus, Denmark
e-mail: SIKAB@Vestas.com
Claus B. W. Pedersen
FE-Design,
22765 Hamburg, Germany
e-mail: claus.pedersen@FE-design.com
Optimal Bearing Housing
Designing Using Topology
Optimization
The internal load distribution in rolling bearings has a high impact on the bearing fa-
tigue life. This study presents a method to optimize roller bearing housing design in order
to maximize the bearing fatigue life by ensuring an optimal internal load distribution. An
FE-model of a cylindrical roller bearing utilizing nonlinear springs in the roller model-
ing is presented, which is capable of simulating the bearing load distribution efciently.
The optimal load distribution is achieved by specifying the desired internal load distribu-
tion as design constraints in a topology optimization of the bearing housing design. The
superiority of the method is clearly demonstrated through case studies involving a cylin-
drical roller bearing, where it is shown that the fatigue life is increased and the bearing
housing mass and roller contact misalignment is reduced. [DOI: 10.1115/1.4005951]
1 Introduction
The Lundberg/Palmgren rolling bearing fatigue life theory
adopted in Ref. [1] gives well dened methods for evaluating the
rating life of roller bearings based on the internal load distribution
among the rollers. Given a specic bearing design, the explicit
relationship allows for optimization of the internal load distribu-
tion with the purpose of maximizing the bearing fatigue life. This
has been investigated in Ref. [2] where the effect on bearing fa-
tigue life of the outer ring section and clearance in a planet gear
bearing is examined, in Ref. [3] where the maximum roller load is
reduced by making the inner ring elliptical, and in Ref. [4] where
the ring dimensions are optimized to maximize the fatigue life. In
these studies, the bearing ring deections are calculated by analyt-
ical relationships and do not support an arbitrary stiffness of the
supporting structure, which can have a signicant inuence on the
load distribution, as it will be demonstrated in this study.
In Refs. [5,6], simplied FE-models of a cylindrical roller bear-
ing are presented, where each rolling element is represented by a
number of nonlinear springs representing the Hertzian contact
stiffness between the roller and the raceways. The current paper
employs an equivalent model where the spring load from each
roller is used for bearing fatigue life evaluation. Topology optimi-
zation is employed on the bearing housing design in order to max-
imize the bearing life while reducing the housing mass. This is
achieved by formulating design constraints for each roller load to
ensure an optimal load distribution in the bearing.
There are several reasons for applying the design force con-
straints on each roller load instead of directly optimizing using the
bearing fatigue life as an objective function. Firstly, the force con-
straints on each roller should not yield different optimization
results compared to directly applying the bearing life function.
Secondly, to the authors knowledge no optimization tools exists
for optimizing directly on the bearing life function using rigor-
ously calculated sensitivities in topology optimization. Thirdly,
using force constraints on each roller also allows one to apply a
modied or different function for calculating the bearing life with-
out considerable reprogramming.
It is the aim of this paper to present a new methodology for
bearing housing designing. For ease of communication, the stud-
ied cases involves a simple single bearing housing setup with one
load case. Except for an increased computational demand, nothing
keeps the method from being implemented on an actual bearing
pedestal in a complete system taking into account more load
cases. The necessary steps in order to include multiple load cases
are included.
This paper is organized with a summary of the bearing life cal-
culation according to Ref. [1] followed by a subsequent deduction
of the optimal load distribution in a cylindrical roller bearing with
respect to maximizing the fatigue life. Subsequently, the cylindri-
cal roller bearing FE-model is presented with a description of the
implementation of load constraints in the model. A summary of
the basic concepts of topology optimization, with emphasis on
bearing housing optimization, follows before the optimal load dis-
tribution and the bearing model are utilized in a topology optimi-
zation where an optimal bearing housing is designed. Different
housing designs are nally compared in two case studies, one with
aligned contact and one misaligned. These studies are followed by
the concluding remarks.
2 Cylindrical Roller Bearing Fatigue Evaluation
By assuming aligned line contact in the roller/raceway contacts,
the bearing fatigue calculations follow the procedure in Ref. [7]
which is, given the assumptions, in accordance with Ref. [1]. It is
assumed that the outer ring is stationary. The basic dynamic
capacity of a raceway contact, i.e., the load, which a roller/raceway
contact, for a 90% probability of survival, will endure for 1 10
6
revolutions of a bearing ring, in a cylindrical roller bearing is
Q
c;m
b
m
552
1 c
29
27
16c
1
4
c
2
9
2r
r

29
27
l
7
9
Z

1
4
(1)
where m1 for the upper and m2 for the lower sign, which
refers to an inner and outer raceway contact, respectively. The
constant b
m
is a modication factor to accommodate for contem-
porary improvements in bearing quality and is 1.1 for cylindrical
roller bearings according to Refs. [7, 8].
For line contact in roller bearings, Ref. [9] determined that a
quartic mean roller load is appropriate for the rotating raceway,
Q
e;1

1
Z

Z
j1
Q
4
j
_ _1
4
(2)
Equivalently, the mean load on the stationary raceway is
Q
e;2

1
Z

Z
j1
Q
4e
j
_ _1
4e
(3)
Contributed by the Tribology Division of ASME for publication in the JOURNAL
OF TRIBOLOGY. Manuscript received October 28, 2011; nal manuscript received
January 27, 2012; published online April 11, 2012. Assoc. Editor: Xiaolan Ai.
Journal of Tribology APRIL 2012, Vol. 134 / 021102-1 Copyright VC
2012 by ASME
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
where e is the Weibull slope approximated as 9/8 according to
Ref. [9]. The statistical combined life of the two raceways is cal-
culated from
L
10

Q
c;1
Q
e;1
_ _
4e

Q
c;2
Q
e;2
_ _
4e
_ _

1
e
(4)
From a contact point of view, the validity of Eq. (4) requires that
the rollers are at line contact and that only negligible contact mis-
alignment is present.
2.1 Optimal Roller Load Distribution. The explicit rela-
tionship between roller loads and bearing fatigue life in Eq. (4)
allows for optimization of the internal load distribution in the
bearing with respect to bearing life. This means that using a spe-
cic bearing, the life is optimized by ensuring an optimal internal
load distribution among the rollers. The optimization is expressed
as
Maximize: L
10
Q
subject to:
(5)
Equilibrium:

Z
j1
cos h
j
_ _
Q
j
F (6)
Equilibrium:

Z
j1
sin h
j
_ _
Q
j
0 (7)
where h
j
is the angle from the radial load to the position of roller
j. Since a pure radial load is applied, only the in plane force com-
ponents are considered in the equilibrium conditions in Eqs. (6)
and (7). The optimal roller loads Q
opt
from the optimization in
Eq. (5) represent the load distribution yielding the highest possible
fatigue life according to Eq. (4). This load distribution is visual-
ized in Fig. 1 for F1 N and Z 52 along with the equivalent
load distribution of a rigid bearing support. The bearing fatigue
life of the optimal load distribution exceeds the fatigue life of the
rigid support by approximately 24% according to Eq. (4) despite
supporting an identical external load.
It should be noted that the optimal load distribution yields a
steep load increase when the rollers enter the loaded zone, which
can be detrimental to the bearing because of the risk of smearing.
Firstly, this solution might not be obtainable, but if the nal
design yields an undesirable steep load increase, this can be
changed by modifying the Q
opt
values. However, any change to
Q
opt
yields reduced fatigue life, but other bearing factors should
be considered, if necessary.
In Sec. 3.2.1, it is described how Q
opt
is used for specifying
design constraints in the topology optimization of the bearing
housing.
3 FE-Modeling of Bearing System
Rolling element bearings constitute a signicant FE-modeling
challenge due to the high number of contacts between the rollers
and raceways. In Ref. [10], a single roller/raceway contact is mod-
eled in FE using 50 200 10,000 contact elements in order to
model the contact pressure with acceptable precision. In a com-
plete bearing model with 50 rollers, this mesh density corresponds
to 1 10
6
contact elements in addition to the solid elements in the
model. Obviously, simplications are needed in order to allow for
efcient analysis of complete rolling element bearings in FE. This
is addressed in Ref. [11] where a cylindrical roller bearing is mod-
eled in 2D using contact elements. In many applications, the bear-
ing housing is inadequately approximated as a 2D structure.
References [5,6] present a more general approach, where each
roller is approximated as a number of axially distributed nonlinear
springs connecting the bearing inner and outer raceway. Apart
from the high generality of this method, it does not involve con-
tact elements, which again allows for relative coarse modeling
mesh, a fast solution time, and good convergence performance.
Obviously, this method does not allow for detailed contact evalua-
tion including roller crowning, cage interference, contact pressure
evaluation, and etc., but such level of details only have limited
inuence with respect to simulating the load supported by each
roller, necessary in this study in order to optimize the bearing fa-
tigue life according to Eq. (4). It is recognized that a nal fatigue
life evaluation must include the actual pressure prole on each
contact, including crowning, misalignment, etc. and comply with
Ref. [1], but for optimization purposes, the importance lies in the
fourth power load/life relationship.
Since the spring modeling allows for direct access to the con-
tact forces used when evaluating the bearing fatigue life and the
possibility of optimizing the load distribution, as discussed in this
paper, it is indeed an appropriate approximation.
3.1 Spring Modeling Approach. Many different theoretical
and empirical expressions exists for the roller/raceway contact
deection. See Ref. [12] for an overview of several commonly
used relationships. In a parameter study in Ref. [6], it was found
that for the spring modeling approach, the best suitable expression
is the one specied in Ref. [13], in which the roller/raceway con-
tact deection for steel/steel contact, identical for both raceways,
is
d
i
d
o

Q
9
10
l
8
10
3:84 10
5
(8)
where Q is in N and l in mm. Equation (8) is reordered to be
expressed by a stiffness constant,
Q
l
8
10
3:84 10
5
_ _10
9
d
10
9
i
k
i
d
10
9
i
(9)
Interpreting the roller/inner race and roller/outer race contact
deection as two springs in series, each roller can be represented
by an effective spring stiffness substituting both contact stiffness.
Noticing that the inner and outer race contact stiffness are equal
according to Eq. (8) and that centrifugal forces are neglected, i.e.,
d/2 d
i
d
o
, the effective contact stiffness is
k
r
d
10
9
k
i
d
10
9
i
k
i

d
2
_ _10
9
) k
r

k
i
2
10
9
(10)
In this study, three springs are used to model each roller, as
sketched in Fig. 2(a), because it has been found sufcient to cap-
ture the load distribution in the bearing and allows for modeling
of misalignment, as shown in Fig. 2(b). To justify the use of three
springs, Fig. 3 shows a convergence study using the cylindrical
bearing and setup presented in Sec. 5 and the V
2
bearing housing
shown in Fig. 8(a). Figure 3(a) shows a load distribution
Fig. 1 Optimal and rigidly supported load distributions
021102-2 / Vol. 134, APRIL 2012 Transactions of the ASME
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
comparison using four different numbers of springs, where Q
1
is a
single spring and Q
30
is thirty springs, which is considered to be
fully converged and is the basis of the error calculation. The rela-
tive error is only calculated for Q
30
>1 kN. Fig. 3(b) shows two
different convergence plots for the maximum force error and
U
sum
, which is the total deection error of the load node in the
model. It is evident that even a single spring ensures a reasonable
approximation of the converged load distribution. Although the
three spring results are not converged, the force error is <1%, and
considering the high computational performance, it is considered
to be appropriate. The presented method is, however, valid for
any number of springs.
It is assumed that the rollers are at line contact, which will be
the case for the majority of the loaded rollers, and as each spring
represents an equal length of the contact, l/3, the spring stiffness
is the total roller stiffness k
r
in Eq. (10) divided by the number of
springs representing each roller,
k
jk

k
r
3
; j 2 1; Z ; k 2 1; 3 (11)
Since rollers only transfer forces in compression, the force deec-
tion relationship for roller j spring k is
q
jk

k
jk
d
jk
c=2
_ _10
9
if d
jk
> c=2
0 if d
jk
c=2
_ _
(12)
The diametral clearance c is assumed zero in the rest of the paper,
but the procedure is general and will not require modications for
c =0.
The relative raceway misalignment, see Fig. 2(b), at the roller
position is derived from the spring deections
a
j
tan
1
d
j3
d
j1
_ _
3
2 l
_ _
(13)
It should be noted that the analytical contact stiffness in Eq. (8)
takes the compression of the raceway material into consideration.
This introduces a potential redundancy since the stiffness of the
raceways is also included in the solid elements of the FE-model.
To eliminate this error, it is thus recommended to use a relative
coarse meshing of the raceways in order to limit the local raceway
compression at the spring attachment nodes and/or increase the
spring stiffness in Eq. (12) to compensate for the raceway exibil-
ity. This modication can be established by applying unit loads on
the spring attachment nodes of the raceways in an FE-model only
consisting of the bearing rings, and in this way nd the redundant
linear stiffness of the raceway attachment nodes. The nal spring
stiffness is now found by considering the FE-spring to be the mid-
dle of three springs in series: the inner raceway exibility, the FE-
spring, and the outer raceway exibility.
3.2 Spring Deflection Constraints. In order for the topology
optimization to yield a load distribution that complies with the op-
timum load distribution calculated in Eq. (5), the maximum
allowed load on each roller is implemented as design constraints
in Sec. 3.2.1. The optimum load distribution is derived for forces,
which are furthermore converted to spring deections in order to
comply with the requirements of the topology optimization. Addi-
tionally, misalignment constraints dened in Sec. 3.2.2 ensure that
the raceway misalignments are of acceptable size.
3.2.1 Force Constraints. In most cases, the optimal load dis-
tribution derived in Sec. 2.1 is impossible to match in practice,
and a scaling e is added to the optimal values
Q
con;j
Q
opt;j
1 e ; e ! 0 (14)
The load constraint equations are formulated as the sum of the
load on the springs

3
k1
q
jk
Q
con;j
Q
opt;j
1 e (15)
Since the rollers are assumed to be at line contact, the springs rep-
resenting each roller should be uniformly loaded, and Eq. (15) can
be expressed in deections using Eq. (12)
Fig. 2 Modeling roller j using three springs
Fig. 3 Convergence study using different number of springs
Journal of Tribology APRIL 2012, Vol. 134 / 021102-3
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms

3
k1
d
jk
3

Q
opt;j
k
r
1 e
_ _9
10

c
2
d

j
(16)
where d

j
is the reference deection value used to dene the load
constraints in Sec. 4.1. d

j
is thus the limit on the mean relative
approach of the end nodes of the three springs of roller j along the
respective spring axes.
3.2.2 Misalignment Constraints. To limit the relative race-
way misalignments an additional constraint is added for each
loaded roller, using the misalignment angle calculated in Eq. (13)
a
j

a
max;j
(17)
tan
1
d
j3
d
j1
_ _
3
2 l
_ _

a
max;j
(18)
d
j3
d
j1

tan a
max;j
_ _

2 l
3
d

a;j
(19)
The maximum allowed misalignment angle for roller j, a
max,j
,
can be chosen based on bearing manufacturer recommendations
and evaluation of the life impact according to Ref. [1] by simulat-
ing the contact pressure at different misalignment angles using
dedicated FE contact simulations or elastic half space simulations.
Both methods are demonstrated in Ref. [10], and the latter used
for post-processing in Sec. 5.2.1. In all cases, the robustness of the
design is improved if the allowed misalignment angle is chosen as
low as possible without making the design domain infeasible,
causing unacceptable convergence performance or yielding excess
material usage. The variable d

a;j
is used to formulate misalign-
ment constraints using spring deections.
In the nal case study, an identical misalignment constraint is
applied for all rollers, but as indicated using subscript j, this can
be dened individually. This option can be utilized to allow a
larger misalignment on low loaded contacts for which the impact
on the total bearing life is small.
4 Conceptual Bearing Housing Designing Using
Topology Optimization
Topology optimization is frequently used in the early design
phase producing new design suggestions where the optimization
targets could be to obtain desirable stiffness properties and fulll
mass constraints. Usually, topology optimization provides the ability
to yield correct decisions in the early mechanical design phase. The
aim of the present paper is not to give a full theoretical background
of topology optimization. A comprehensive theoretical survey of to-
pology optimization is found in Ref. [14] and the references therein.
The intention is to give a brief introduction to topology optimization
and, specically, address how topology optimization can be applied
for obtaining conceptual designs of bearing houses.
4.1 Topology Optimization for Bearing Housing
Designing. Figure 4 shows the concept behind topology optimi-
zation. The elemental relative density q
e
of each individual struc-
tural nite element in the CAE model is dened as design
variable, which can be solid (relative density at upper limit
q
e
1) or void (relative density at minimum limit q
e
q
min
% 0).
The so-called SIMP-model is applied for modeling the elemental
stiffness as being proportional to a power-law of the relative ele-
mental density {q}, a vector containing the design variables of all
design elements. Thereby, the optimization formulation for the
optimal bearing housing design yields
Minimize: Global compliance u q f g f g; q f g
subject to:
(20)
Equilibrium: R u q f g f g; q f g f g
l
0 f g
for l 2 1; max load case
(21)
Mass: m q f g c
m
m
ini
(22)
Deflections:

3
k1
d u q f g f g; q f g
j;k;l
3
d

j;l
for j 2 1; Z ; k 2 1; 3
(23)
Misalignments: d u q f g f g; q f g
j;3;l
d u q f g f g; q f g
j;1;l

a;j;l
(24)
As an objective in Eq. (20), we want to maximize the stiffness.
However, instead the total compliance is minimized, which is
equivalent to maximizing the total stiffness. The force equilibrium
{R}
l
in Eq. (21) of the entire bearing housing for the given load
case l is solved using an implicit nite element solver [15] solving
directly for the displacements {u} of the entire bearing housing.
The total mass of the bearing housing is constrained to fulll a
certain weight target described in Eq. (22), where m
ini
is the mass
of the initial design space. As earlier described, the aim is to
obtain an increased fatigue life of the bearing by designing against
an optimal roller deection distribution. Thus, the roller deec-
tions d
j,l
are constraint against the constraint roller deection dis-
tribution d

j;l
from Eq. (23).
In general, more than one load case will be considered for a
bearing housing optimization, and each load case can be weighted
in the objective function, as a starting point according to the per-
centage of use at that condition. The roller load and misalignment
constraints are more complicated to dene since the optimum load
distribution cannot be achieved for all load cases. A general rec-
ommendation is to dene the closest to optimum constraint values
for load cases representing the majority of use. For the remaining
load cases, focus should initially be on reducing misalignments
since contact misalignments in general are more severe than a
non-optimum load distribution. After the topology optimization
the results can be evaluated and if necessary the constraint values
can be adjusted and the optimization rerun.
4.2 Adding Topology Optimization in Existing CAE and
CAD Design Processes. The optimization formulation in
Eqs. (20)(24) is solved using an iterative design process scheme
Fig. 4 Schematic overviewof topology optimization workow
021102-4 / Vol. 134, APRIL 2012 Transactions of the ASME
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
as shown in the workow of Fig. 5. In the nal case studies, the
iterative design process requires approximately 50 automated
optimization runs. The iterative design process scheme can be
implemented in a predened CAE and CAD workow as in Ref.
[16]. In this study, Refs. [15] and [17] have been applied.
A nite element design space model is created for the topology
optimization as shown in Fig. 4(a). Afterwards, the design space
model is applied in an iterative design process (see Fig. 5). For
each design iteration the nonlinear equilibrium of the nite ele-
ment model are determined using Ref. [15]. Afterwards, the sensi-
tivities with respect to the elemental relative densities are
rigorously computed for the objective function and for the con-
straints by means of the adjoint method (see, e.g., Refs.
[14,16,18]). Then, the elemental relative densities are updated
using mathematical programming, and a new nite element model
is generated. If the optimization has not converged, a new optimi-
zation cycle is started.
When the optimization has converged, then the new conceptual
design for the bearing housing is still a nite element with non-
smooth surfaces as show in Fig. 4(b). An automated smoothing
process is applied, which smoothes the surfaces but keeps the top-
ological layout as shown in Fig. 4(c). A new nite element model
is created for verication of the smoothed design. The smoothed
optimization result is transferred into a nal CAD model if the
verication model shows to fulll the desired design targets. All
the previous optimization steps illustrated in Fig. 5 are integrated
in the software package TOSCA Structure [16].
5 Applying Topology Optimization on Bearing
Housing Design
The bearing housing topology optimization is demonstrated in
an example using a type N28/1400 cylindrical roller bearing with
basic data shown in Table 1.
For ease of communication, the bearing is mounted in a simple
bearing housing with a uniform thickness equal to the bearing
width of 175 mm, shown in Fig. 6. For the nite element modeling
nonlinear spring elements are applied, and the continuum model
consists of eight-node hexahedral elements. The bearing outer
ring is connected to the bearing housing elements using bonded
contact formulation. Numerical tests have proven that the bonded
contact has no impact on the load distribution compared to match-
ing the meshes at the interface and merging the coincident nodes
or applying different types of frictional contacts. In all cases, it is
assumed that the bearing rings are properly mounted without
clearance, misalignment, etc.
In the following aligned study, the housing is meshed using
seven elements in the depth and 35,000 solid elements in total. In
the misalignment study 15 elements are used in the depth and
64,000 solid elements in total.
All FE-nodes on the bearing feet bottom surface are xed in all
degrees of freedom, and the bearing inner ring is loaded by a ver-
tical force, F1 MN, applied at a load node connected to the
inner ring in a rigid MPC element spider web. Except in the verti-
cal direction, the load node is xed in all degrees of freedom. The
shaft can easily be included, but this has no principal inuence on
the method discussed and only minor inuence on the overall load
distribution, and is thus excluded from the model. The used bear-
ing housing material properties are E
h
170 GPa,
h
0.29 and
q
h
7000 kg/m
3
corresponding to cast iron, and the bearing ring
steel has E
b
200 GPa and
b
0.3.
The design space of the topology optimization is highlighted
with light colored elements in the V
1
design while the upper sup-
porting structure and the feet are excluded. Additionally the mate-
rial located 4 r
r
beneath the bearing mounting surface is also
excluded in order to improve the robustness of the design by mak-
ing the load distribution less sensible to variations in the design
domain and ensure that the roller load at intermediate positions,
between the locations in the FE-model, are comparable.
5.1 Aligned Contact Study. A total of ve different results
are compared in a model with perfect alignment constrained
between the bearing and housing:

rigid support/analytical distribution

V
1
- solid housing (c
m
1)

V
2
- arch housing

V
3
- only mass constraint (c
m
0.3)

V
4
- mass and load constraints (c
m
0.3, e 0.05)
First, a rigid support is included to allow for comparison of
load distributions with the analytical one not taking the bearing
Fig. 5 Topological iterative design process in a present CAE
and CAD workow using TOSCA [16]
Table 1 Basic Data of the N28/1400 Cylindrical Roller Bearing
d[mm] D[mm] B[mm] Z r
r
[mm] r
p
[mm] l[mm]
1400 1700 175 52 40 778.5 105
Fig. 6 Initial V
1
design with general model descriptions
Journal of Tribology APRIL 2012, Vol. 134 / 021102-5
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
housing exibility into consideration. V
1
is an analysis of the com-
plete design volume, and V
2
is an intuitive, geometrically simple,
and lighter design, shown in Fig. 8(a). To underline the impor-
tance of including roller load constraints in the optimization two
optimizations are included: V
3
without load constraints and V
4
including. Due to the boundary conditions, only very limited mis-
alignment is present and the misalignment constraints in Eq. (19)
are thus not considered in the aligned study. A remaining mass
fraction of c
m
0.3 is used in both optimizations, and in V
4
a tol-
erance of e 0.05 is used since it has been found to be close to
the minimum obtainable limit.
The total roller load distributions of the considered designs are
shown in Fig. 7, where each marker represents one roller. Figures
8(b)8(e) show the optimized relative elemental stiffness distribu-
tion and smoothed V
3
and V
4
design, corresponding to Figs. 4(b)
and 4(c). The convergence of the objective function and con-
straints of the V
4
design are shown in Fig. 8(f) where the horizon-
tal lines indicate the constraint values. Due to symmetry only half
of the roller positions are included.
From the comparison, it is obvious that the supporting structure
for the studied cylindrical roller bearing has a signicant inuence
on the load distribution in the bearing. Even the solid V
1
design
has a signicant deviation from the rigid housing. The high exi-
bility in the center region of the V
2
and V
3
design clearly demon-
strate the potential of applying the presented topology
optimization method, where the soft spot across the feet gap can
be compensated, as shown for the V
4
design.
Besides the load distribution of each design, Fig. 7 additionally
shows three different load distributions, namely the optimal load
distribution Q
opt
according to Eq. (5), the force constraints Q
con
according to Eq. (15), the load distribution of the nal optimiza-
tion iteration Q
nal
, which complies with all constraints, and the
realized load distribution after the automated smooth process
Q
smooth
. It should be noted that different smooth options are avail-
able, and some iterations were needed in order to obtain the excel-
lent correlation shown. The lower plot shows a verication of the
nal smoothed design where the bearing is additionally simulated
with a rotation of 180

/Z to verify the roller load at the intermediate


positions. This plot shows that the maximum error between the
roller loads and the constraints are <2%. Furthermore, it is seen
that the deviation on forces at intermediate positions is negligible,
which is due to the non-design space below the bearing mounting
surface distributing the forces from the supporting trusses.
Fig. 7 Load distribution evaluation
Fig. 8 Considered bearing housings in aligned study
021102-6 / Vol. 134, APRIL 2012 Transactions of the ASME
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
Table 2 gives an overview of the fatigue life according to
Eq. (4) calculated for the optimal load distribution as well as each
of the ve designs. d
r
is the radial deection measured at the load
node to give an impression of the overall stiffness inuence of the
bearing housing design.
This evaluation clearly demonstrates the potential of the pre-
sented method for automated topology optimization of bearing
housings. Despite the simplicity of the considered problem, it
would have required numerous manual iterations to get even close
to the desired load distribution, which ensures a bearing fatigue
life superior to any of the considered designs.
Simultaneously, an optimum material usage is ensured.
5.2 Misaligned Contact Study. In the second study, a mis-
alignment on the inner bearing ring is imposed by rotating the
inner ring about the horizontal axis causing maximum roller mis-
alignment angle at w = 180

. A misalignment of 1 mrad is applied


and the V
5
design is optimized using the same constraint values as
for V
4
but also including misalignment constraints:

V
5
- mass, load, and misalignment constraints (c
m
0.3,
e 0.05, a
max
0.02 mrad).
The chosen misalignment constraints ensure that only 1/50 of
the externally applied misalignment is allowed as relative mis-
alignment between the raceways at the roller positions.
Figure 9 shows the relative elemental stiffness distribution and
smoothed V
5
design.
Figures 10(a) and 10(b) show the convergence of the objective
and constraints of which all are complied with in the nal optimi-
zation iteration. Figure 10(c) shows the nal load distribution
including results of the V
4
design in order to show the impact of
Table 2 Bearing Housing Performance Comparison
Design L
10
L
10
/L
10
(Q
opt
) m
h
[kg] d
r
[mm]
Q
opt
4786 1
Rigid 3881 0.81 0.047
V
1
3325 0.69 2769 0.096
V
2
3517 0.73 1977 0.110
V
3
2376 0.50 1438 0.150
V
4
4295 0.90 1438 0.158
Fig. 9 V
5
bearing housing for misaligned study
Fig. 10 Optimization convergence and evaluation of load distributions and misalignments
Journal of Tribology APRIL 2012, Vol. 134 / 021102-7
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
the misalignment, which is quite signicant compared to the rela-
tively small misalignment imposed on the inner ring. Figure 10(d)
shows the contact misalignment of all the designs presented in
Sec. 5.1, the nal optimization iteration, and the realized results
after smoothening. Two different smoothed results are shown; one
using 1st order 4-node tetrahedral elements and one using 2nd
order 10-node elements in the remesh after smoothening. Com-
pared to the nal optimization iteration, the 4-node model is stiffer
and thus results in larger misalignments, and the 10-node model is
more exible and thus results in smaller misalignments. This rela-
tively big difference is due to the fact that the misalignment con-
straints are very tight and the realized misalignment is, seen from a
contact point of view, negligible in practice, and the improvements
compared to V
1
-V
4
are evident. The difference on the load distribu-
tion in the bearing is, however, not signicantly inuenced by the
choice of element type, and thus the results are not illustrated.
The main differences compared to the V
4
design are a change in
the truss design and that these are supporting the bearing in an off-
set from the housing center plane in order to ensure that the bear-
ing housing angular deections match the imposed misalignment
on the bearing inner ring. The impact of this change is highlighted
in Fig. 11 where the deection of the V
4
and the two V
5
housing
designs are compared, and the superiority of the V
5
design against
the imposed misalignment is evident.
5.2.1 Life Evaluation of Misaligned Contact Study. The va-
lidity of the fatigue life calculation in Eq. (4) requires all rollers to
be at line contact with zero misalignment. To allow for a fatigue
life evaluation of the misaligned contacts, the more detailed pro-
cedure in Ref. [1], taking the actual contact pressure along the
roller/raceway contacts into consideration, should be followed.
Since the micro geometry is unknown, a full logarithmic roller
crowning prole as proposed by Ref. [19] is used with an assumed
design load Q
d
150 kN, while the raceways are assumed cylin-
drical. The crowning prole is thus
C
p
x
2 Q
d
1 v
2

p l E
b
ln
1
1
2 x
l
_ _
2
_
_
_
_
_
_ (25)
Where C
p
is the crowning prole, x is the coordinate along the
roller length, zero at center, and E
b
200 GPa and 0.3 are
elastic properties of the bearing. Obviously the precision of the
pressure calculations and the calculated fatigue life greatly depend
on the geometry, but for the purpose of illustration it is sufcient.
From each FE-analysis, the roller load Q
j
and tilt angle a
j
are
used to evaluate each roller/raceway contact (a
j
being split on
inner and outer raceway contact) using an elastic half space simu-
lation as proposed in Ref. [1] and demonstrated in, e.g., Refs. [10,
20], which is efciently able to calculate non-Hertzian contact
pressures including roller edge effects, misalignment, etc. For an
illustration of the calculated pressures, Fig. 12(a) shows the high-
est loaded inner raceway contact of the V
3
and V
5
design, the for-
ward nearly symmetric distribution being V
5
. The highest pressure
along the roller/raceway contact is extracted and Fig. 12(b) shows
the distribution for the highest loaded roller in each design, where
the pressure peak is enlarged on the right picture. Figure 12(c)
shows the unwrapped raceway pressure of the loaded zone of two
designs, where the intermediate space between the rollers, visual-
ized with vertical lines, is lled using linear interpolation of the
contact pressures. For each design, the bearing fatigue life is now
calculated, see Ref. [21] for a comprehensive explanation of the
procedure, according to Ref. [1] with the results shown in Table 3.
From this comparison, the superiority of the V
5
design is evident
with a life ratio ranging from approximately 2-7 compared to the
other design. As discussed previously, the actual fatigue life has
high dependence on the crowning prole, but it should be noted
that the life in Table 3 is inversely dependent to the misalignments
shown in Fig. 12, except for V
2
/V
3
, which is due to the general
better performance of V
2
.
Fig. 11 Deection sum comparison of V
4
and V
5
bearing cross
sections at 3300 magnication with misalignment indications
at bottom roller position. Scale in mm
Fig. 12 Evaluation of non-Hertzian contact pressures for misaligned contact study
Table 3 Bearing Housing Performance Comparison
Design L
10
L
10
/L
10
(Q
opt
) m
h
[kg] d
r
[mm]
Q
opt
4786 1
Rigid 632 0.13 0.047
V
1
869 0.18 2769 0.096
V
2
1143 0.24 1977 0.110
V
3
1085 0.23 1438 0.150
V
4
2134 0.45 1438 0.158
V
5
4219 0.88 1438 0.165
021102-8 / Vol. 134, APRIL 2012 Transactions of the ASME
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms
6 Conclusions
The load distribution in rolling element bearings is highly inu-
enced by the stiffness of the supporting structure. As demonstrated
in this paper, this inuence can yield an inferior as well as supe-
rior bearing fatigue life compared to a perfectly rigid support,
depending on the actual load distribution. Optimizing a bearing
housing is a complex challenge due to several factors. First of all,
the bearing modeling is a complicated task, interpreting the FE-
results is difcult, and nally the conventional workow involv-
ing CAD, FE-meshing, bearing modeling, simulation, result inves-
tigation, and interpretation is highly inconvenient. This is further
complicated if different people are involved in the different steps.
This paper presents a new method for bearing housing design-
ing where topology optimization is utilized to ensure an optimal
housing material usage and optimal load distribution. It is shown
how the optimal roller loads are explicitly included in the topol-
ogy optimization since the optimization will otherwise yield an
unfavorable load distribution. Besides ensuring an optimal load
distribution, it is also demonstrated how contact misalignment can
be minimized using design constraints. This is a huge benet since
even small contact misalignments can cause roller edge loads,
which severely reduces the bearing fatigue life.
The nal studies show the potential of this new method where
it is demonstrated that the topology optimization can be used in a
bearing housing design to yield optimal material usage, load dis-
tribution, and minimized contact misalignment. That shaft angular
misalignments due to, e.g., shaft bending can be compensated for
through housing design can yield a signicant improvement in
bearing fatigue life or have the additional benet that it can poten-
tially allow for usage of a cylindrical roller bearing in applications
that initially are disqualied due to excessive shaft misalignments.
The present study features a simple cylindrical roller bearing
housing due to simplicity of communication, but the presented
method can be implemented equivalently on complete systems
including more load cases and on other types of rolling element
bearings.
Nomenclature
b
m
rating factor
B bearing width
c diametral clearance
C
p
crowning prole
d bearing bore diameter
D bearing outer diameter
e Weibull slope
E Youngs modulus of elasticity
F radial force
k stiffness
l effective roller contact length
L
10
basic rating life, in million revolutions
m mass
q spring load
Q roller load
Q
c
basic dynamic capacity of a raceway contact
Q
d
crowning prole design load
Q
e
equivalent roller load
r radius
u displacements
x coordinate along roller
Z total number of rollers in a bearing
a relative misalignment between inner and outer raceway
c auxiliary parameter, r
r
/r
p
c
m
mass reduction fraction
d total deection of both raceway contacts of a roller
e optimization constraint tolerance
Poissons ratio
h
j
angle from resultant radial force to roller j
q density
q
e
relative elemental density
W bearing angular position
Subscripts
b bearing
h housing
i inner ring/raceway
j roller index
k spring index
l load case index
m raceway index
o outer ring/raceway
p pitch circle
r roller
con constraint value
ini initial value
max maximum value
opt optimal value
Superscript
* reference value for constraints
References
[1] International Organization for Standardization, 2008, ISO/TS 16281:2008
Rolling BearingsMethods for Calculating the Modied Reference Rating
Life for Universally Loaded Bearings, Standard 1, Geneva, Switzerland.
[2] Jones, A., and Harris, T., 1963, Analysis of Rolling Element Idler Gear Bear-
ing Having a Deformable Outer Race Structure, J. Basic Eng., 85(2), pp.
273277.
[3] Harris, T., and Broschard, J., 1964, Analysis of an Improved Planetary Gear
Transmission Bearing, J. Basic Eng., 86(3), pp. 457462.
[4] Harris, T., 1965, Optimizing the Fatigue Life of Flexibly Mounted, Rolling
Bearings, Lubr. Eng, 21(10), pp. 420428.
[5] Molnr, L., Vradi, K., Bdai, G., Zwierczyk, P., and Oroszvry, L., 2010,
Simplied Modeling for Needle Roller Bearings to Analyze Engineering
Structures by FEM, Mech. Eng., 54(1), pp. 2733.
[6] Golbach, H., 1999, Integrated Non-Linear FE Module for Rolling Bearing
Analysis, Proceedings of NAFEMS World Congress 99 on Effective Engi-
neering Analysis, NAFEMS, ed., Vol. 2.
[7] Harris, T. A., and Kotzales, M. N., 2007, Rolling Bearing Analysis - Essential
Concepts of Bearing Technology, Taylor & Francis Group, Boca Raton,
Florida.
[8] International Organization for Standardization, 2007, ISO 281:2007 Rolling
Bearings-Dynamic Load Ratings and Rating Life, Standard 2, Geneva,
Switzerland.
[9] Lundberg, G., and Palmgren, A., 1952, Dynamic Capacity of Rolling
Bearings, Acta Polytech. Mech. Eng., 2(4), pp. 532.
[10] Kabus, S., Hansen, M. R., and Mouritsen, O. ., 2011, A New High Perform-
ance Method for Evaluating Contacts in Time Domain Rolling Bearing Simu-
lations, J. Tribol., submitted.
[11] Demirhan, N., and Kanber, B., 2008, Stress and Displacement Distributions on
Cylindrical Roller Bearing Rings Using FEM, Mech. Based Des. Struct.
Mach., 36(1), pp. 86102.
[12] Teutsch, R., and Sauer, B., 2004, An Alternative Slicing Technique to Con-
sider Pressure Concentrations in Non-Hertzian Line Contacts, J. Tribol.,
126(3), pp. 436442.
[13] Palmgren, A., 1959, Grundlagen der Walzlagertechnik, Franckhsche Verlag-
shandlung, Stuttgart, Germany.
[14] Bendse, M. P., and Sigmund, O., 2003, Topology Optimization: Theory, Meth-
ods and Applications, Springer-Verlag, Berlin, Germany.
[15] ANSYSVR
, 13.0, ANSYS, Inc., www.ansys.com.
[16] TOSCAVR
, 7.1, FE-Design GmbH., www.fe-design.com.
[17] CreoVR
, 1.0, PTC, www.ptc.com.
[18] Buhl, T., Pedersen, C. B. W., and Sigmund, O., 2000, Stiffness Design of Geo-
metrically Non-Linear Structures Using Topology Optimization, Struct. Multi-
discip. Optim., 19(2), pp. 93104.
[19] Lundberg, G., 1939, Elastische Beruhrung zweier Halbraume, Forsch. Geb.
Ingenieurwes., 10(5), pp. 201211.
[20] Cretu, S., Antaluca, E., and Cretu, O., 2003, The Study of Non-Hertzian Con-
centrated Contacts by a GC-DFFT Technique, National Tribology Conference.
[21] Kabus, S., Hansen, M. R., and Mouritsen, O.., 2011, A Cylindrical Roller
Bearing Model to Accurately Consider Non-Hertzian Contact Pressure in Time
Domain Simulations, J. Tribol., submitted.
Journal of Tribology APRIL 2012, Vol. 134 / 021102-9
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 04/28/2013 Terms of Use: http://asme.org/terms

Das könnte Ihnen auch gefallen