Journal of Advanced Mechanical Design,

Systems, and

Manufacturing

Vol.4, No. 4, 2010
728

Tolerance Design of Logarithmic Roller Profiles
in Cylindrical Roller Bearings
*

Hiroki FUJIWARA
**, ***
and Kazuto YAMAUCHI
***

**NTN corporation,
5-105 Hidamarinooka, Kuwana, Mie, 511-0867, Japan
E-mail: hiroki_fujiwara@ntn.co.jp
***Osaka University,
2-1, Yamadaoka, Suita, Osaka, 565-0871, Japan

Abstract
A logarithmic profile is essentially optimal crowning geometry for rolling machine
elements such as bearing rollers and raceways. Although some design methods of
the profile have been proposed, they do not refer to the tolerance of the geometry
required in engineering applications and in production. This paper shows how to
define the tolerance range associated with the optimum roller profile, previously
suggested by one of the authors. The overall tolerance is reasonably defined by
giving the reduced amount of roller radius at three points so that the von Mises
equivalent stress will not exceed a specified limit, where the point locations are not
measured from the roller end but from the intersection of the crowning and the
chamfer. For the purpose of design usefulness, the tolerance expression is offered
in dimensionless form.
Key words: Rolling Bearing, Optimum Design, Contact Problem, Tolerance,
Crowning, Logarithmic

1. Introduction
When a cylindrical roller bearing is subjected to a load, the rollers of finite length
contact the mating raceways of greater length, and compressive stresses at the roller ends
tend to be substantially higher than those at the center of the contact. This phenomenon of
stress concentration is referred to as edge loading. In roller bearings, the radii of the
rollers and/or the raceways are reduced on the order of micrometers to avoid edge loading.
This modified geometry is called crowning. Typical crowning profiles include a single
straight line, circular curve, their combination, and so on. In some cases, crowning
extends over the full length of the rollers and/or raceways (full crowning), while in other
cases crowning is processed only near both ends (partial crowning).
Contact stress is correlated with rolling contact fatigue life, and lower stress yields
longer life. Hence, under constant loading, a uniform axial stress distribution along the
contact would minimize the maximum stress value, resulting in a longer life. Lundberg
(1)

found that a logarithmic profile gives such a stress distribution based on the theory of
semi-infinite elastic body approximation. According to Lundbergs formula, however, the
radius reduction, which is called the crown drop, must be theoretically infinite at the contact
ends, thus making the formula impossible to be applied in real design. Many
researchers
(2) ~ (6)
thereby have been attempting to improve the logarithmic profile function.
One of the authors modified the Johns-Gohar formula
(2)
, a well-known logarithmic
crowning formula, to improve the flexibility of crowning design. Contrary to the
*Received 25 Feb., 2010 (No. T2-08-7025)
Japanese Original : Trans. Jpn. Soc. Mech.
Eng., Vol.75, No.749, C (2009), pp.1-7
(Received 2 July, 2008)
[DOI: 10.1299/jamdsm.4.728]
Copyright 2010 by JSME




Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
729
Johns-Gohar formula, the modified formula is applicable even if the rollers are misaligned
relative to the raceways
(6)
. With this formula, we are able to optimize roller crowning
design through a mathematical optimization technique when a set of loading and
misalignment conditions with bearing dimensional parameters is given.
Although the optimal crowning design yields only one solution without any tolerance,
real manufacturing of rollers and raceways inevitably requires the concept of dimension
tolerance. Therefore, this paper will cover the study of logarithmic crowning tolerance that
is required in real manufacturing. Crowning can be effectively formed on either or both a
roller and a raceway as long as the total crown drop reaches a predetermined dimension at
any cross section. In what follows, we can assume that crowning is exclusively processed
on rollers without loss of generality.

Fig. 1 Schematic drawing of a crowned roller

2. Axial position of crowning boundary
Figure 1 displays the main features of a crowned roller. Both ends of the roller are
chamfered, and each crown is processed to blend into the chamfer. If the chamfer length
exceeds the nominal design dimension, the crown drop at the intersection of the crown and
the chamfer becomes insufficient, although no problems are found in the reverse case from
the viewpoint of edge stress prevention. For example, suppose there is a logarithmically
crowned roller which is optimally designed so that the maximum equivalent stress predicted
by the von Mises distortion energy theory can be minimized. Table 1 shows a typical test
condition of a two-roller line contact test rig in which roller 1 is logarithmically crowned
while the profile of roller 2 is entirely flat without crowning. If the chamfer length is
longer than the nominal value by 0.1 mm, then edge loading occurs at the contact ends, as
depicted in Figs 2 (a) and (b).
As measures against such edge loading, we can propose two methods of crowning
design
1. Make the nominal chamfer length become the upper limit of its tolerance.
2. Make the axial boundary position between the crowning and the chamfer be
coincident with the position of the maximum crown drop.
Design 2 leads to a dimensional variation of the central flat part according to the
chamfer length error. A longer chamfer theoretically results in a shorter flat part with the
overall roller length kept constant, which raises the stress level. However, a small variation
of the flat part dimension has little effect on the optimized crown drop and in turn on the
stress distribution, since the chamfer length tolerance is less than a tenth of the flat part
length. On the other hand, in the case of Design 1, when the chamfer length is close to the
Straight part Crowned part
Chamfer
C
r
o
w
n

d
r
o
p





Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
730
lower limit of the tolerance zone, the crown drop at the intersection increases, and the
processing cost will therefore be higher. Hence, Design 2 is a better engineering choice to
set the axial position of the intersection.

Table 1 Optimizing condition of logarithmic roller profile in two roller contacts
(a) Two alternatives of chamfer design

(b) Equivalent stress distribution under the condition of Table 1 (Case )
Fig. 2 Chamfer length and von Mises equivalent stress


Crowned part
Chamfer
Crowned part
Chamfer
The chamfer length is nominal dimension
The chamfer length is nominal dimension + 0.1 mm
Maximum crown drop
Maximum crown drop
Depth
mm
S
t
r
e
s
s



G
P
a
Position of axial direction
mm
3.0
2.5
2.0
1.5
1.0
0.5
0
Roller 1 radius 6 mm
Roller 1 width 12 mm
Roller 1 chamfer width 0.5 mm
Straight part length 6 mm
Roller 2 radius 10 mm
Roller 2 length 20 mm
Roller load 13.7 kN





Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
731
3. Design optimization for logarithmically crowned rollers
Given a condition of roller bearing dimensions and loading, Johns and Gohars formula
yields a unique logarithmic crowning profile. Their formula, however, inevitably
generates edge loading unless the contacting roller and raceway are aligned perfectly
parallel, because the formula takes no account of misalignment. Hence, we have proposed
a new logarithmic formula with three unique design parameters, such that
( )
2
m
2
1
ln
2
1 1 exp 1
2
z y A
l
y
z
l
A
K
=
| |

|
| |
+
| ` |
\ . )
|
|
\ .
(1)
where,

1
2K Q
A
lE
=

E : Equivalent Youngs modulus
2
1
E

| |
=
|

\ .

E : Youngs modulus
: Poissons ratio
l : Effective length
Q : Load
y : Axial position
z(y) : Crown drop at axial position y

As shown in Fig. 3, K
1
, K
2
and z
m
correspond to the crowning curvature, the ratio of the
crowning region length to the effective roller length, and the crown drop at the crowning
region ends, respectively. We take a NU304E cylindrical roller bearing as an example of a
logarithmic crowning design under the condition given in Table 2. Cylindrical bearing
rollers often have a central flat land over half of the total roller length to assure the
machining accuracy. The flat zone length is accordingly set to be 5 mm, which results in
fixing the parameter K
2
, while leaving K
1
and z
m
as unknowns for mathematical
optimization. When K
1
and z
m
are designed in order to minimize the maximum von Mises
stress,
max
, the stress is distributed as a function of K
1
and z
m
as illustrated in Fig. 4.
At the same time, many theories on rolling bearing fatigue life have been propounded,
and one of them by Tanaka et al.
(7)
showed the following symbolic formula (2) to explain
that fatigue life of a local control volume, L
i
, is inversely proportional to the von Mises
stress,
i
, to the power of c/e.
( )
1
1
i i i i
e
c h
L V z
- -
D D (2)
where,
L
i
: Fatigue life of the stressed control volume

i
: von Mises equivalent stress of the control volume
z
i
: Depth of the control volume below the load-carrying surface
c, h, e : Constants to be empirically determined

Although Tanaka et al. did not specifically refer to the value of c/e, it is supposed to be
9.2 for line contact according to Lundberg and Palmgren
(8)
. With this value adopted,
fatigue life of the local material volume, L
i
, will decrease as much as 50 percent if the
stress,
i
, becomes 1.1 times greater. It is judged that a set of endurance test data of
dozens or less test pieces will be significantly different from another set of data if there
exists a margin of 1.5 to 2.0 times or more in fatigue life based on Fujitas statistical
simulation approach
(9)
. Consequently, we will allow the von Mises stress,
i
, to increase




Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
732
by up to 10 percent beyond the optimized design value. The maximum stress,
max
, does
not exceed 1.1 times that of the optimized value as long as it falls within the white heavy
line drawn in Fig. 4.
With this stress tolerance, the crowning profile yielding the maximum crown drop and
that yielding the minimum one are sketched in Fig. 5, respectively. Here it is to be noted
that a crowning curve lying between these two curves is not always accepted. That is
because a profile between these two curves in the tolerance zone with respect to z
m
may
exceed the tolerance with respect to K
1
. As an example, in some cases of a small z
m
, the
profile exists between the two curves in Fig. 5. However, it does not fall within the
allowable area in Fig. 4 with increasing K
1
.

Table 2 Optimizing condition of logarithmic roller profile in NU304E

Fig. 3 Logarithmic crowning parameters


z
m

l/2
K
1
K
2
l/2
C
r
o
w
n

d
r
o
p

Position of axial direction
Roller radius 4.5 mm
Roller width 10 mm
Roller chamfer width 0.5 mm
Straight part length 5 mm
Inner race radius 13.5 mm
Roller load 3.2 kN
Bearing misalignment 2/1000





Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
733
Fig. 4 Distribution of maximum von Mises equivalent stress to crowning design
parameters K
1
and z
m


Fig. 5 Permissible logarithmic profiles of rollers in NU304E

4. Design of crown drop tolerance
4.1 Measurement method of crowning profile
It is convenient to measure a crowning profile in a mass production line as depicted in
Fig. 6. That is, with reference to the maximum radius at the central flat land of a roller, the
drop is measured at prescribed axial positions. While the more measurement points we
set, the more accurately we can evaluate the profile, the number of points should be the
minimum required considering the operating cost. The logarithmic crowning function (1)
is characterized by the three parameters K
1
, K
2
and z
m
. The parameter z
m
is measurable at
the intersection of the crowning and the chamfer. On the other hand, it is difficult to
distinguish the central flat region from the crowned region and to measure the length
corresponding to K
2
since the flat land blends into the crowning very smoothly and
gradually. Thereby, we measure the length between two points where the drop is a tenth of
z
m
as an alternative to measuring the flat region length itself. Also, in order to evaluate K
1
,
we choose an axial position where the drop is a half of the optimized value of z
m
.
0
10
20
30
40
50
0 1 2 3 4 5
Position of axial direction mm
C
r
o
w
n

d
r
o
p

m

10 20 30 40 50
30
25
20
15
10
5
2.0
1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
1.1
1.0
z
m

K1
max GPa




Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
734

Fig. 6 A measurement method of roller profiles on a production line
4.2 A design example of crowning tolerance
In this section we attempt to design a logarithmic crowning tolerance for a NU304E
cylindrical roller bearing. Again, the design condition in Table 2 is used. The applied
load is 25% of the basic dynamic capacity, and the misalignment between the inner and
outer rings is set to be 2/1000. Note that most bearings in applications are subject to less
load and misalignment. This design condition, that would be the heaviest in actual use,
gives an optimized solution of K
1
= 5.25 and z
m
= 7.95 m, as depicted in the solid line in
Fig. 2 where the origin is located at the roller center. In this optimized profile, the drop is
10% of z
m
at the axial position y = 3.15 mm and is 50% of z
m
at the axial position y = 4.05
mm.
Let z(y) at each position of y = 3.15, 4.05 and 4.5 mm (crowning end) have some error,
then we evaluate the resulting increase of the von Mises equivalent stress. Here we
assume that a crowning profile plot drawn through three points can be approximated with a
smooth continuous function such as a spline function as long as the roller is normally
ground.
Table 3 summarizes a relationship between the drop error at each designated position
and the corresponding rate of increase of the maximum von Mises stress. Table 3 (a)
through (d), respectively, show the results when a drop error of 0.4 to +1.0 m is given at
the position of y = 3.15 mm in which the drop is 10% of the optimal z
m
. In Table 3, a
value in row a represents a drop error at y = 4.05 mm where the drop is 50% of the
optimal z
m
, while a value in column b represents that at y = 4.5 mm where the drop is z
m

itself. The rates of increase of the maximum stress are tabulated for various combinations
of the three errors, and if the rate of increase for a particular combination does not exceed
10%, that cell is shaded. Extracting the shaded data yields Table 4 to specify the tolerance
range. The optimized logarithmic crowning profile and the tolerance range are depicted in
Fig. 7, which shows that the tolerance range does not always cover the optimized profile.
As is shown in Fig. 4, there is a possibility that a slight deviation of K
1
from an optimized
solution may cause edge loading, and vice versa. It is therefore necessary that the
tolerance range be shifted over the optimized solution so that edge loading will not occur
even though each of the three given drops changes independently.
Axial direction
Gage
C
r
o
w
n

d
r
o
p





Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
735
Table 3 Relationship between error of reduced amount of roller radius and rate of increase
of stress (%) on NU304E



(a) Error of crown drop 0.4 m at y = 3.15 mm
a
b
1 0 1 2 3
2 51.5 69.7 80.4 197.8 101.0
0 5.6 5.3 21.1 32.1 42.7
2 10.6 5.8 7.7 9.5 11.1
4 14.7 6.2 8.1 9.9 11.8
6 18.2 6.8 8.3 10.2 12.0
8 21.5 9.8 8.4 10.3 12.1

(b) Error of crown drop 0 m at y = 3.15 mm
a
b
1 0 1 2 3
2 49.8 70.9 83.8 94.2 104.5
0 10.0 2.3 22.9 35.7 46.4
2 14.8 2.9 3.0 5.0 6.8
4 19.0 7.3 3.4 5.5 7.4
6 22.6 10.8 3.6 5.7 7.6
8 26.0 14.1 3.8 5.8 7.8

(c) Error of crown drop 0.5 m at y = 3.15 mm
a
b
1 0 1 2 3
2 46.5 68.6 87.8 98.3 108.5
0 14.6 3.2 21.8 40.4 50.9
2 19.1 8.1 1.0 2.0 3.0
4 23.0 12.4 1.3 2.4 3.4
6 26.6 15.9 4.5 2.5 3.6
8 29.8 19.3 7.8 2.7 3.7

(d) Error of crown drop 1.0 m at y = 3.15 mm
a
b
1 0 1 2 3
2 67.6 64.4 86.4 102.1 112.5
0 19.1 7.5 19.1 40.8 55.2
2 23.2 11.9 1.9 3.3 4.3
4 26.7 15.8 4.9 3.6 4.6
6 30.1 19.3 8.5 3.8 4.8
8 33.3 22.5 11.7 3.9 4.9

aError of crown drop at y = 4.05 mm
bError of crown drop at y = 4.5 mm
Rate of increase of stress <10%




Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
736
Table 4 Tolerance of logarithmic crowning in NU304E
Fig. 7 A optimum logarithmic profile of rollers in NU304E, and the profile tolerances to
satisfy the specified maximum von Mises stress condition
4.3 Generalized crowning design
For general-purpose cylindrical roller bearings, we obtain Fig. 8 if the axial position is
normalized with reference to the half length of the effective roller length, and the drop with
reference to the optimal z
m
. Although the plotted curves of Fig. 8 include optimal
solutions of eight bearings that are distinct in dimensions, they almost coincide with each
other. This fact implies that we may unify the tolerance expression by introducing
dimensionless parameters.
Fig.8 Dimensionless optimum logarithmic profiles of some rollers

Position of axial direction
mm
Optimum crown drop
m
Tolerance
m
3.15 0.449
+1
0
4.05 3.31
+3
+1
4.5 7.95
+6
+2

0
2
4
6
8
10
12
14
16
0 1 2 3 4 5
Position of axial direction mm
C
r
o
w
n

d
r
o
p

m

Tolerance
Optimum
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
Dimensionless position of axial direction
(Position of axial direction / (l/2))
D
i
m
e
n
s
i
o
n
l
e
s
s

c
r
o
w
n

d
r
o
p

(
C
r
o
w
n

d
r
o
p

/

z
m
)





Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
737
The tolerance obtained in Section 4.2 is summarized in dimensionless form in Table 5.
Likewise, Tables 6 and 7 are obtained for NU312E with larger rollers and NU2312E with
even longer rollers compared to their diameter, respectively. It should be noted that, in
dimensionless form, the tolerances in Tables 5 to 7 are thought to be, in effect, identical
regardless of the difference in bearing size. For engineering design where a small
discrepancy among the tolerances can be neglected, the arithmetic average listed in Table 8
will be applicable. As a result, once the optimal value of z
m
, the maximum crown drop, is
computationally obtained, the tolerance of the logarithmic crowning profile can be given as
0.06 0.17 z
m
at y = 0.35l, 0.56 0.83 z
m
at y = 0.45l and 1.26 1.86 z
m
at y = 0.5l, where
l is the effective roller length.

Table 5 Dimensionless tolerance of logarithmic roller profile in NU304E
Table 6 Dimensionless tolerance of logarithmic roller profile in NU312E
Table 7 Dimensionless tolerance of logarithmic roller profile in NU2312E
Table 8 Dimensionless tolerance of logarithmic roller profile in cylindrical roller bearings
Position of axial direction Optimum crown drop Tolerance
0.7 0.06
0.13
0.00
0.9 0.42
+0.38
+0.13
1.0 1
0.75
+0.25

Position of axial direction Optimum crown drop Tolerance
0.7 0.06
0.11
0.00
0.9 0.42
+0.41
+0.14
1.0 1
0.86
+0.26
Position of axial direction Optimum crown drop Tolerance
0.7 0.05
0.10
0.00
0.9 0.40
+0.41
+0.14
1.0 1
0.82
+0.27

Position of axial direction Optimum crown drop Tolerance
0.7 0.07
0.10
0.00
0.9 0.43
+0.45
+0.15
1.0 1
1.00
+0.25





Journal of Advanced Mechanical Design,
Systems, and
Manufacturing
Vol. 4, No. 4, 2010
738
5. Conclusion
This paper defines the tolerance of logarithmic crowning applicable to cylindrical
bearing rollers. Considering chamfers formed at roller ends, it is proposed that the axial
boundary position between the crowning and the chamfer be coincident with the position of
the maximum crown drop.
Also, the crowning performance can be approximately assured by controlling the crown
drop tolerance at three distinct axial locations; the intersection of the crowning and the
chamfer where the optimal drop is equal to the optimized maximum drop z
m
, and the
locations where the optimal drop is 50% and 10% of z
m
, respectively.
For general-purpose cylindrical roller bearings that vary in geometry, when the applied
load is less than 25% of the basic dynamic capacity and the misalignment between the inner
and outer rings is less than 2/1000, the tolerance of the logarithmic crowning profile can be
uniquely given as 0.06 0.17 z
m
at y = 0.35l, 0.56 0.83 z
m
at y = 0.45l and 1.26 1.86 z
m

at y = 0.5l, where y and l are the roller axial position and the effective roller length,
respectively.
Design procedure of the logarithmic roller crowning tolerance requires considerable
computational efforts and is cumbersome, hence, the proposed simple approach will be
useful in engineering applications.
References
(1) Lundberg, G., Elastic Contact Between Two Semi-Infinite Bodies, Engineering research
(in German), Vol. 5 (1939), pp. 201-211.
(2) Johns, P. M. and Gohar, R., Roller bearings under radial and eccentric loads, Tribology
International, Vol. 14 (1981), pp. 131-136.
(3) Reusner, H., The logarithmic roller profile the key to superior performance of
cylindrical and taper roller bearings, Ball Bearing Journal, Vol. 230 (1987), pp. 2-10.
(4) Takata, H. et al., Experimental Study of Fatigue Life of Profiled Roller Bearings, NSK
Technical Journal, Vol. 653 (1992), pp.1-7.
(5) Kamamoto, S. et al., Research on Crowning Profile to Obtain The Maximum Load
Carrying Capacity for Roller Bearings, KOYO Engineering Journal, Vol. 159 (2001),
pp.44-51.
(6) Fujiwara, H. and Kawase, T., Logarithmic Profile of Rollers in Roller Bearing and
Optimization of the Profile, Transactions of the Japan Society of Mechanical Engineers,
Series C (in Japanese), Vol. 72, No.721 (2006), pp. 3022-3029.
(7) Tanaka, H. and Tsushima, N., Estimation of Rolling Bearing Life under Contaminated
Lubrication, ASTM STP1419, BEARING STEEL TECHNOLOGY, (2002), pp. 213-225.
(8) Lundberg, G. and Palmgren, A.,Dynamic Capacity of Roller Bearings, Acta Polytechnica,
Mechanical Engineering Series, 2,4(1952).
(9) Fujita, T., Life Test Designing Method to Balance Test Time with Reliability and
Valificating Method of Test Results, Material of Technical Commitie on Rolling Contact
Fatigue (in Japanese), 98(2007).