You are on page 1of 17

Original Article

A fracture mechanics interpretation


of rolling bearing fatigue
MWJ Lewis
1
and B Tomkins
2
Abstract
In fatigue assessment of structures and components, it is now commonplace to use a defect-tolerant approach to
account for the inevitable presence of surface and internal flaws, both detectable and undetectable. This approach
has been applied to rolling bearings in an attempt to present contact stress limits as a function of defect size below
which fatigue failure would not be expected. In essence, it is an extension to the existing rating life methods developed by
Ioannides and Harris that incorporate a fatigue limit stress in the estimation of bearing life, a concept similar to that of
the endurance limit in structural fatigue. Such a fracture mechanics approach to sub-surface initiated fatigue in rolling
bearing steels containing non-metallic inclusions, based upon the work of Murakami, has led to an allowable contact
stress limit as a function of maximum inclusion size (as estimated by extreme value analysis). An alternative, less
conservative fracture mechanics based stress limit is the prevention of propagation of cracks formed on inclusions
(termed butterflies) by shear (Mode II) loading. The approach has been developed by examination of butterflies formed
in service in rolling element bearings manufactured in bainitic steel with maximum inclusion sizes in excess of 100 mm.
The observed lack of micro-crack formation on inclusions and non-propagation of butterflies support the concept of a
fatigue endurance limit that is related to the cleanliness of the bearing steel.
Keywords
Fracture mechanics, bearing steels, inclusions, rolling bearing, fatigue
Date received: 12 April 2011; accepted: 17 November 2011
Introduction
Rating life methods for rolling element fatigue were
developed by Lundberg and Palmgren
1,2
in the 1940s
and 50s by analysis of many hundreds of tests on small
bearings at contact stresses well in excess of 2500 MPa.
Under these conditions, signicant cyclic strain-induced
metallurgical changes occur that are the precursors to
crack initiation and growth.
3
These changes consist of
the formation of white etching areas (WEAs) and dark
etching areas (DEAs) close to the running surface that
are oriented at specic angles. In view of the distribu-
tion in time to appearance of the rst spall, Lundberg
and Palmgren expressed bearing fatigue in terms of L
10
life, that is, the life that 90% of bearings would achieve
under nominally identical loading and lubrication
regimes.
However, at more modest contact stresses that are
typical of industrial bearings (12001800 MPa), no such
metallurgical changes are seen, which leads to the ques-
tion whether Lundberg and Palmgrens methodology
would equally apply.
The UK National Centre of Tribology has investi-
gated the failure of rolling element bearings of many
types and sizes in dierent applications for over
40 years. Recently, it has become apparent that the
estimated L
10
life is not being achieved in some appli-
cations, despite the fact bearings have been tted cor-
rectly, supplied with lubricant of sucient viscosity and
cleanliness and not subject to misalignment loads. This
is particularly observed in larger bearings manufac-
tured from bainitic through-hardened steels, as
opposed to the more common martensitic steel AISI
52100.
1
The National Centre of Tribology, ESR Technology Ltd, Warrington, UK
2
UK Forum on Engineering Structural Integrity, Warrington, UK
Corresponding author:
MWJ Lewis, The National Centre of Tribology, ESR Technology Ltd,
Whittle House, 410 The Quadrant, Birchwood Park, Warrington WA3
6FW, UK.
Email: mike.lewis@esrtechnology.com
Proc IMechE Part J:
J Engineering Tribology
226(5) 389405
! ESR Technology Ltd, UK 2012
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1350650111435580
pij.sagepub.com
It is now widely agreed that at more modest contact
stresses, sub-surface initiated fatigue arises from the
presence of internal aws, principally non-metallic
inclusions. In the case of ultra-clean steels where inclu-
sions are small and sparsely distributed, sub-surface
crack initiation often occurs on discontinuities such as
microstructural inhomogeneities.
In fatigue assessment of structures and components,
it is now commonplace to use a defect-tolerant
approach to account for the inevitable presence of sur-
face and internal aws, both detectable and undetect-
able. This approach has been applied to rolling
bearings in an attempt to present contact stress limits
as a function of defect size below which fatigue failure
would not be expected. In essence, it is an extension to
the existing rating life methods that incorporate a fati-
gue limit stress in the estimation of bearing life, a con-
cept similar to that of the endurance limit in structural
fatigue.
This approach is intended to be applied to situations
in which sub-surface initiated fatigue failure is domi-
nant, that is, the contact surfaces are separated by a full
uid lm (i.e. high z) and the likelihood of surface-
initiated fatigue is minimal. In the case of low z or
surface damage, however, it could be applied where
there are surface-breaking defects.
Bearing life calculations
Conventional approach
Rolling bearings are usually selected on the basis of a
percentage of failures in an application within a pre-
scribed number of revolutions. Lundberg and
Palmgren
1,2
formulated a theory of fatigue based
upon Weibull probability
4
and thus proposed a
method to calculate rolling bearing life. Their life pre-
diction model was based upon the equation
ln
1
S
_ _
N
e
t
c
0
z
h
0
az
0
l 1
The stressed volume V is given by az
0
l.
Substituting for the applied bearing loads, the fol-
lowing equation was obtained for the basic rating life
L
10

C
P
_ _
p
2
It can be seen that equation (2) makes no provision
for material structure or cleanliness, depending solely
upon the basic dynamic rating C and the equivalent
load P.
This equation was adopted by ANSI/AFBMA
in 1978 and by ISO in 1990.
5
Revisions to
ANSI/AFBMA Std-9 and Std-11 in 1990,
6,7
as well as
ISO 281-1
5
included factors a
1
, a
2
and a
3
to account for
dierent levels of reliability, material fatigue properties
and lubrication quality. Factors a
2
and a
3
are usually
combined to give a factor a
23
that is a function of mate-
rials and lubrication through the parameter k v/v
1
,
the ratio of actual to required viscosity. It is limited
to a maximum value of 2.5.
Using the basic rating life methodology of Lundberg
and Palmgren, the Society of Tribologists and
Lubrication Engineers have adopted dierent indepen-
dent factors, as summarised by Zaretsky.
8
These factors
cover bearing material, melting practice, metalworking,
oil ltration level and misalignment, amongst others.
Whilst the cleanliness of the steels used by Lundberg
and Palmgren is not stated, historical trends suggest
that the maximum inclusion size may have been
about 25 mm. Over the years, this has improved to
about 13 mm for good quality mass-produced AISI
52100 steel. Ultra-clean steels typically have a maxi-
mum inclusion size of 5 mm.
Fatigue limit approach
As a development of bearing life methodology,
Ioannides and Harris
9
proposed a fatigue strength
approach, similar to that in structural fatigue. Thus, a
fatigue limit was introduced according to the following
equation
ln
1
S
_ _
N
e
t t
u

c
z
h
V 3
Whilst shear stress is used in equation (3), an alter-
native such as von Mises stress is equally applicable.
This approach suggests that there is an applied stress
below which fatigue failure should not occur such that
bearing life becomes innite.
In support of the approach, Ioannides and
Harris analysed results from rotating beam, torsional
beam and reversed bending beam tests with good agree-
ment.
9
They also carried out a small number of tests
on 6309 deep groove ball bearings under two dier-
ent steel qualities and lubrication conditions (high
and low z).
They concluded that values of the shear stress fati-
gue limit t
u
300 and 350 MPa applied to the bad and
good quality steels, whilst values of von Mises stress
o
vM
519 and 606 MPa were correspondingly
obtained. Applying a conversion factor of 0.55 for
von Mises stress, this equates to maximum contact
stresses of 950 and 1100 MPa, respectively.
The work of Ioannides and Harris was extended by
Harris and Barnsby
10
in which a stress-life factor a
SL
390 Proc IMechE Part J: J Engineering Tribology 226(5)
was proposed to modify the basic rating life. This factor
includes the following parameters:
. fatigue endurance limit;
. support structure exibility;
. outer raceway failures;
. inner raceway failures;
. rolling element failures;
. sub-surface initiated failures;
. surface-initiated failures;
. residual stress;
. hoop stress;
. rolling;
. elasto-hydrodynamic lubrication (EHL) oil lm
thickness;
. asperity interactions;
. sliding and surface tractions; and
. lubricant contamination.
Harris and Barnsby
10
carried out a much more sig-
nicant test programme with a variety of bearing types
up to a pitch diameter of 388.5 mm in materials includ-
ing AISI 52100, VIMVAR M50, case-hardening and
induction hardening steels.
However, for AISI 52100 which is of interest here,
the maximum pitch diameter was 72.5 mm for deep
groove and angular contact ball bearings and 141 mm
for cylindrical roller, taper roller and spherical roller
bearings. Endurance testing showed a mean von
Mises fatigue limit stress of 684 MPa for AISI 52100,
stated to be a conservative value.
The approach adopted by Harris and Barnsby
11
has
been incorporated in the ASMELife software program.
Within this software, it is possible to select a value for
the fatigue limit stress for specic materials that reects
their metallurgy and cleanliness.
Recently, ISO 281
12
has adopted a similar approach
that considers the eects of lubricant cleanliness and
viscosity as well as the concept of an endurance limit.
The latter is introduced as a fatigue limit load P
u
(or
C
u
) that is calculated from, for example
C
u
ZQ
u
sin o
100
D
pw
_ _
0.3
4
for a roller thrust bearing with D
pw
>100 mm and
C
u
0.2453ZQ
u
i cos o
100
D
pw
_ _
0.3
5
for a radial roller bearing with D
pw
>100 mm.
Figure 1 summarises the methodologies of
LundbergPalmgren and IoannidesHarris in terms of
contact stress vs cycles to failure. The basic L
10
life
relationship has a slope of 0.1 and suggests that fati-
gue failure would occur at all contact stresses. The
life modication factors (a
1
, a
23
and a
SL
) move this
line to left or right depending upon the reliability
required, lubrication conditions and material type/
cleanliness. The IoannidesHarris methodology intro-
duces an endurance limit that is a function of steel
cleanliness.
Figure 1. Summary of LundbergPalmgren and IoannidesHarris methods (schematic).
Lewis and Tomkins 391
Recent developments
In recent years, reviews of the state-of-the-art of rolling
bearing fatigue have been undertaken. Zaretsky
13
has
challenged the validity of this approach, concluding
that bearing steels do not exhibit an endurance limit
so that fatigue failure would occur under all loadings.
In contrast, Takemura et al.
14
report a variation in
endurance limit with steel cleanliness, with values in
the range 11001700 MPa dependent upon maximum
inclusion size.
Sadeghi et al.
15
carried out a comprehensive review
of empirical and research bearing life models. They
concluded that, though empirical models such as
Lundberg and Palmgren are practical life prediction
tools, they do not take into account the material
microstructure and inhomogeneity. These models
assume a Weibull strength theory, rather than the
resultant life distribution being a consequence of the
inhomogeneous and random nature of the material
microstructure.
Sadeghi et al.
15
proposed a model that treated the
material as a non-homogeneous microstructure consist-
ing of randomly shaped, sized and oriented grains. An
alternative life model was produced whose results were
compared to Lundberg and Palmgrens model.
In a development, Jalalahmadi et al.
16
accounted for
the eect of inclusion stiness, depth, size and number
on fatigue life. Spherical inclusions were placed within
the non-homogeneous microstructure. As with the ear-
lier model, comparisons were made with Lundberg and
Palmgren, together with experimental results at 2.4 GPa
and above.
Donzella and Petrogalli
17
have developed a fatigue
limit approach for rolling contact based upon the frac-
ture mechanics concept that avoidance of fatigue can be
guaranteed if inherent or early nucleated cracks are not
able to grow. Recognising the importance of inclusions
in bearing steel, they formulated a failure assessment
diagram to indicate whether a component was operat-
ing in a safe region or not.
They adopted a short crack approach in which the
KitigawaTakahashi diagram has been translated into
a failure assessment diagram as a function of the max-
imum contact pressure. A mixed mode relationship was
used to estimate crack growth thresholds, whilst it was
assumed that Mode II thresholds increased with crack
length (as do Mode I ones).
The equivalent stress intensity factor range was cal-
culated using the positive part for Mode I and the
whole range for Mode II. In order to construct the
failure assessment diagram, it was necessary to estimate
the relationship between the maximum contact pressure
and applied stress range for a material without defect as
a function of surface friction coecient.
An example of application of the failure assessment
diagram is given for a roller cam application with both
surface and sub-surface defects. A long crack threshold
relationship that was independent of crack length was
assumed. However, short crack thresholds have been
estimated by assuming an intrinsic crack length that is
added to the actual crack length.
It is noted that the above work, whilst considering
the eects of microstructure and other discontinuities
such as inclusions on fatigue response, has not yet
included a detailed interpretation of the physical
processes at play.
The role of inclusions in bearing fatigue
Butterfly development
Failure by rolling contact fatigue at lower contact stres-
ses involves the development of sub-surface cracks that
mainly form at non-metallic inclusions. These are
micro-cracks which grow from one or both ends of an
inclusion in a direction typically parallel to the surface
until they diverge towards the surface. Alongside this
crack growth, WEAs are also formed, described as
butteries (typical examples are shown in Figure 2).
In some unfailed bearings, these micro-cracks can ini-
tiate but only grow to a limited length until they arrest.
Figure 2. Typical butterflies.
392 Proc IMechE Part J: J Engineering Tribology 226(5)
The development of butteries from inclusions can
be described by the following sequential process:
. initiation of micro-cracks within the matrix adjacent
to the inclusion;
. formation of a WEA adjacent to the inclusion as a
result of polygonisation to form nano-crystalline
ferrite;
. growth of microscopic cracks along the WEA
boundary;
. extension of the WEA;
. growth of cracks alongside the WEA; and
. continuing crack growth (or arrest) beyond the
WEA.
There is, however, some continuing discussion over
the exact sequence under which micro-cracks and WEA
form.
Four stages of crack development have been dened
by Lamagnere et al.
18
in relation to the applied contact
stress which provide useful classications:
. stage 1 onset of dislocation movement (microplas-
ticity) at the inclusion/matrix boundary (H1 stress);
. stage 2 formation of WEA (H2 stress);
. stage 3 micro-crack initiation (H3 stress); and
. stage 4 crack growth (H4 stress).
Signicant crack opening displacement at crack ini-
tiation sites adjacent to inclusions suggests that the
cracks appear to be relieving a tensile stress of the
order of the yield stress. For the buttery cracks
shown in Figure 3, an estimate of the strain that is
relieved may be obtained from
c
o

6
In this instance, o 1.2 mm and 33 mm, so c
3.6%. Other observations give strains of about 0.8%.
Rolling contact fatigue has been related to structural
fatigue, such that the concept of endurance limit
applies. For materials in which there is a stress concen-
trating feature (e.g. an inclusion), the endurance limit is
reduced from that without a stress concentrating
feature.
Crack growth under Mode I loading
In fracture mechanics, crack opening and development
is related to the applied loading mode. Tensile loading
across the crack results in Mode I crack growth, whilst
shear loading in the plane of the crack results in Mode
II/III growth.
For buttery crack formation, it is postulated that
crack initiation and an initial extent of short crack
growth is caused by Mode I loading of the inclusion.
This means that a tensile stress eld must exist around
the inclusion in order that the overall compressive stress
arising from rolling contact can develop positive stress
intensity factors at inclusion-initiated crack tips.
The residual tensile stress eld around the inclusion
reduces to zero with distance from the inclusion, result-
ing in the eventual arrest of any Mode I crack growth at
an overall crack size determined by the Mode I thresh-
old stress intensity factor. However, this does not mean
that further crack growth is impossible as cracks could
continue to grow under the general high cyclic shear
stress eld outside of the local inclusion areas. This
growth would occur under Mode II/III shear loading
if the Mode II/III threshold stress intensity factor is
exceeded.
The threshold stress intensity factor for a Mode I
crack is given by Murakami
19
and is appropriate for
short cracks up to about 5001000 mm long
K
Ith
3.3 10
3
HV 120

area
p _ _
1,3
7
in which the stress intensity factor for an embedded
crack is given by
K 0.5o
0

area
p
_
8
This is also consistent with the following equa-
tions for an embedded crack subjected to Mode I
loading
20
K
o

b
p
E
sin
2

b
2
a
2
_ _
cos
2

_ _
1,4
9
E
_
,2
0
1 1
b
2
a
2
_ _
sin
2

_ _
1,2
d 10
Figure 3. Scanning electron microscopic image of a butterfly
showing crack opening.
Lewis and Tomkins 393
which leads to the following relationship when a b
K 0.48o
0

area
p
_
11
This results in the following relationship for the
endurance limit
19
o
w1
1.56 HV 120
1 R
2
_ _
o
_

area
p _ _
1,6
o 0.226 HV 10
4
12
Derived from the analysis of Murakami,
19
the fol-
lowing relationship for the fatigue endurance limit o
w1
of a cracked material (and hence peak contact stress
o
max
equivalent to the H3 limit) is proposed
o
max

0.55
o
w1
13
where 0.55 is the conversion factor from von Mises
to maximum contact stress (applicable to a depth
range z/a of 0.51). For justication of this conversion
factor, see, for example, Zwirlein and Schlicht.
3
Figure 4 shows the relationship between maximum
contact stress for avoidance of micro-crack growth and
inclusion (or total micro-crack) size for an R ratio of
1, steel hardness of 650 HV and various aspect ratios
b/a. It can be seen that Mode I stress intensity factors
are largely independent of the crack aspect ratio b/a.
Crack growth under Mode II/III loading
In the case of Mode II and III loading (relevant to
continuing propagation), an analysis proposed by
Kassir and Sih
21
has been carried out for an embedded
elliptical crack (Figure 5). It can be assumed that but-
teries act as cracks with total dimension of the inclu-
sion buttery cracks. The embedded crack is assumed
to grow under Mode II and III loading (i.e. in-plane
Figure 4. Maximum contact stress vs inclusion size under Mode I (HV650).
394 Proc IMechE Part J: J Engineering Tribology 226(5)
and out-of-plane shear) in both longitudinal and trans-
verse directions.
The stress intensity factors are given by
K
II

b
3
a
_ _
1,2
k
2
t
k
2
vEk vk
02
Kkb
1,2

14
K
III
ab
1,2
k
2
t1 v
k
2
vEk vk
02
Kka
1,2

15
k 1
b
2
a
2
_ _
1,2
16
k
0

b
a
17
Ek
_
,2
0
1 k
2
sin
_ _
1,2
d 18
Kk
_
,2
0
d
1 k
2
sin
1,2
19
where K
II
and K
III
are the stress intensity factors in the
longitudinal and transverse directions, a and b the semi-
axes of the ellipse and v the Poissons ratio. E(k) and
K(k) are elliptic integrals of the rst and second orders.
These have been estimated numerically.
The aspect ratio for calculation of Mode II and III
thresholds may be determined by the crack growth
under Mode I loading. It is observed that elongated
inclusions with low b/a ratio (e.g. ASTM E45 type A
sulphides) tend to develop cracks with b/a & 0.6,
whereas those with higher b/a (e.g. type D or DS glob-
ular oxides) tend to develop cracks with b/a & 0.9.
However, Mode II and III growths will be determined
by an aspect ratio b/a 0.49, that is, when K
II
and K
III
are equal.
Mode II and III threshold stress intensity factors
vary with crack size (as do Mode I ones) and those
measured by Matsunaga et al.
22
for martensitic SUJ2
bearing steel have been taken. The steel tested had a
hardness of 750 HV. For steels of diering hardness, it
can be argued that a correction factor similar to that
for Mode I loading given in equation (7) is applicable.
A curve t of the form below is appropriate
K
IIth
1.6a
0.333
20
This is appropriate to low interference of the crack
faces, that is, an open crack.
Stress limits
The maximum contact stress for avoidance of growth
by Mode I loading according to equations (12) and (13)
is plotted on Figure 6 as a function of inclusion size for
a steel hardness of 750 HV, R ratios 0.5 and 1 and
an aspect ratio b/a 0.49.
If, for a given inclusion size, the Mode I stress limit is
exceeded, then crack growth may occur. However, as
the initial tensile stress is relieved by the crack, it can
only then propagate by Mode II and III loading. So,
the Mode II stress limits are plotted according to equa-
tions (14) to (20) as a function of total crack size for
aspect ratios 0.49, 0.6 and 0.9. This assumes that the
shear stress t 0.25o
max
.
Whilst Mode I stress limits show little eect of the
crack aspect ratio b/a, those for Mode II show a signif-
icant eect, but in this instance, an aspect ratio of 0.49
is relevant (as discussed above).
Observations
Service experience
In one particular application, no butteries were found
in rollers that were manufactured from martensitic steel
with a maximum inclusion size of 15 mm and hardness
750 HV, suggesting a maximum contact stress of less
than 1700 MPa, according to Figure 6. In rollers, inclu-
sions tend to be oriented along the axis such that they
are quite small in the direction of rolling. However, the
bainitic inner and outer races of hardness 650 HV were
found to have butteries on inclusions down to about
12 mm in size, suggesting a maximum contact stress of
1600 MPa, according to Figure 4. The Mode I stress
Figure 5. Embedded crack geometry for Mode II and III
loadings.
21
Lewis and Tomkins 395
appears to t with observations of micro-crack forma-
tion on inclusions (that develop to butteries).
It should be noted that for martensitic steel of high
cleanliness (less than 10 mm maximum inclusion size),
the predicted contact stress for micro-crack initiation is
about 1700 MPa, but the stress for propagation by
Mode II is much higher (about 19002300 MPa).
However, it is observed that stresses in excess of
2000 MPa are necessary to initiate butteries in steels
of high cleanliness (e.g. Takemura et al.
14
), but for poor
cleanliness, the stress may be as low as 1100 MPa.
One possible explanation is that sub-surface residual
compressive stresses are developed under high contact
stress (greater than 2500 MPa). According to Zwirlein
and Schlicht,
3
a residual compressive stress of 20% of
the contact stress lowers the von Mises factor from
about 0.550.45. Thus, the Mode I stress would be
increased by about 22%. No such residual stresses are
seen at modest contact stresses.
For super clean steels (maximum inclusion size
about 5 mm), fatigue failure occurs less at non-metallic
inclusions, with micro-structural discontinuities becom-
ing important. Note that for very small inclusion sizes,
the Mode I limit for martensitic steel approaches the
theoretical endurance limit of 2300 MPa, quoted by
Zwirlein and Schlicht.
3
Incubation and initiation
Since inclusions have variable size, shape and compo-
sition, the assumption that they behave as cracks
requires explanation. Whilst there may be local areas
of sharp radius which can provide rapid initiation by
acting as a local stress raiser, initiation of micro-cracks
in general may require an incubation period.
Figure 7 shows an example of a neighbouring inclu-
sion to that shown in Figure 3 of similar size and sub-
jected to an identical stress eld. In one case, a buttery
has formed and in the other case there is no micro-
cracking. This suggests that the micro-crack initiation
phase may not form an important fraction of
overall life.
Figure 6. Stress limits for martensitic steel (HV750).
396 Proc IMechE Part J: J Engineering Tribology 226(5)
Effect of R ratio
In one particular application, there was the opportunity
to observe the characteristics of micro-cracking on
adjacent inclusions of varying sizes for a bainitic steel
of hardness 650 HV. These inclusions were subjected to
the same stress elds but gave dierent crack responses.
These observations are plotted in Figure 8, suggesting
that R ratio 1 is a reasonable value to assume.
Butterfly characteristics
In the course of failure examinations on bearings from
industrial applications, butteries have shown the fol-
lowing characteristics.
1. In the majority of instances, the micro-crack ran
along the edge of the buttery wing (WEA).
2. There were some instances in which the micro-
crack grew within the buttery wing.
3. There were a few instances in which the micro-
crack grew beyond the buttery wing.
4. There were some instances in which the micro-
crack crossed the buttery wing to continue
growth along the other edge.
Figure 8. Mode I micro-crack propagation.
Figure 7. Inclusion without micro-cracking (adjacent to that
in Figure 3).
Lewis and Tomkins 397
These characteristics suggest that the buttery wing
(WEA) and micro-crack grow simultaneously, with a
strong preference for growth at the interface between
the matrix and buttery wing. Typical examples of but-
teries are shown in Figure 2, formed at type A sul-
phide inclusions with type D oxide encapsulation.
Crack growth at the edge of the buttery wing, within
the wing and across the wing is illustrated.
Since cracks generally grow along a path of least
resistance, it is postulated that the characteristics of
the interface result in a lower threshold stress intensity
factor than either the matrix or buttery wing. It has
been observed that the zone within the WEA surround-
ing the main crack had a larger ferrite grain size and
contained many micro-cracks.
23
These could provide a
line of weakness for growth of the main crack.
Estimates of the plastic zone size ahead of the crack
tip according to equation (21) yield values in the range
24 mm. The plastic strains developed at the crack tip
are the most likely driving force for nano-crystallisa-
tion, as is seen in shot peening and drilling.
24
Thus,
the WEA is formed ahead of the crack, assisting
advance of the crack in a simultaneous manner
R
p

1
2
K
o
y
_ _
2
21
Examination of failed bearings also considered the
overall size of the buttery in relation to the inclusion.
Figures 9 and 10 suggest that small inclusions can yield
large buttery sizes as well as evidence that many but-
teries reach about 90 mm. This suggests that crack
arrest is occurring, in line with previous observations
of many butteries at an overall size of 90 mm. It was
not possible to distinguish between the behaviour of
type A and type D/DS inclusions.
Additionally, it can be seen that the larger inclusions
develop butteries about twice their size. Figure 11
shows an example of advanced cracking from one or
more butteries. This suggests that WEA may or may
not be formed alongside the cracking once it has
extended well beyond the initial buttery.
Mode II growth
Using equations (14), (15) and (20), Figure 12 shows an
estimated boundary for crack growth in Mode II based
upon observations made for bainitic steels of hardness
650 HV. This corresponds to threshold stress intensity
factor data according to Matsunaga et al.
22
but cor-
rected for hardness according to the method of
Murakami,
19
as shown in equation (7), namely
K
IIth
HV650 K
IIth
HV750
650 120
750 120
22
Figure 9. Relationship between butterfly size and initiating
inclusion.
Figure 10. Ratio of butterfly to inclusion size.
398 Proc IMechE Part J: J Engineering Tribology 226(5)
As presented in Figure 9, a large number of butter-
ies were seen of maximum size 90 mm. This implies
that the butteries are non-propagating (NP). So, the
correction applied to the Mode II threshold stress
intensity to account for the lower hardness gives agree-
ment with observations from service bearings.
A number of interesting observations may be made
concerning the Mode I and II stress limits.
1. The Mode I stress increases as the R ratio becomes
more negative this is because less of the Mode I
crack tip stress intensity factor (K
I
) is positive and
hence a higher stress is necessary to overcome the
threshold value.
2. If the Mode I stress is exceeded, after initiation, the
Mode I crack grows into a decreasing tensile stress
eld such that less of the stress intensity factor K
I
is
positive, despite the opposing increase due to crack
length thus crack growth progressively slows with
the possibility of arrest.
3. In the region between the Mode I limit and the
Mode II threshold according to Matsunaga
et al.,
22
crack arrest is possible. Thus, butteries
can be NP. However, depending upon initial inclu-
sion size and the extent of crack growth under
Mode I, the buttery may become propagating (P
buttery). For example, at a contact stress of
1700 MPa for a steel of hardness 650 HV, inclu-
sions above 5 mm would form butteries, but an
inclusion crack size of 70 mm is necessary to prop-
agate the buttery to failure.
4. At higher contact stresses, the region between the
Mode I limit and the Mode II threshold narrows,
Figure 12. Estimated boundary for Mode II growth (HV650).
Figure 11. Example of advanced cracking.
Lewis and Tomkins 399
suggesting that the likelihood of NP butteries
decreases.
5. At lower contact stresses (e.g. below 1500 MPa),
typical of many industrial applications, the inclu-
sioncrack size is estimated to be in excess of
150 mm for propagation under Mode II this sug-
gests the possibility of innite life.
6. An NP buttery may become propagating if the
inclusion crack size exceeds the boundary for
Mode II propagation when subjected to a higher
contact stress, even if applied for a short period.
Alternatively, short-term application of a higher
contact stress may grow it suciently under
Mode I to exceed the Mode II boundary. In
either case, continued crack growth under Mode
II is possible at progressively decreasing contact
stresses until normal control stresses continue
growth to failure (e.g. 1300 MPa for an inclu-
sioncrack size of 200 mm or more).
7. There is a cycle order eect, that is, high stress
cycles applied early in life are more damaging.
This is in contrast to use of Miners rule for fatigue
damage accumulation which cannot distinguish the
order of high and low stress cycles.
8. Inclusion or inclusion crack aspect ratio is of sec-
ondary importance for the stress intensity factor of
Mode I cracks (less than 5% according to
Murakami
19
) and for the conversion of inclusion

area
p
to principal dimension.
9. Inclusion crack aspect ratio is of importance for
the stress intensity factors of Mode II cracks but
the stable growth ratio of b/a 0.49 is the govern-
ing factor.
10. A value of contact stress of 1500 MPa is assumed
for the calculation of fatigue load limit P
u
or C
u
in
modern rating life methods this would be consis-
tent with a maximum inclusion size of 10 mm, i.e.
AISI 52100 steels of high cleanliness. However, for
steels of poorer cleanliness (e.g. bainitic steels in
larger bearings) where the maximum inclusion size
is in excess of 100 mm, a lower contact stress
(approaching 1000 MPa) may be more appropriate.
11. The eects of residual compressive stress may be
incorporated by means of a decrease in the von
Mises factor below 0.55.
12. The eects of tensile hoop stress could be
accounted for by increasing the von Mises factor
above 0.55.
Extreme value analysis
Since the allowable maximum contact stress is critically
dependent upon the maximum size of inclusion within
the steel, it is necessary to have a method to assess this. It
is widely accepted that the fatigue resistance of bearing
steels is reduced by the presence of non-metallic inclu-
sions.
25
Conventional methods such as ASTM E45 used
to assess the cleanliness of steels do not appear to give a
good correlation with fatigue behaviour.
An assessment of the maximum likely size of inclu-
sions may be performed according to ASTM E2283-
08.
26
The maximum inclusion in a sample of area A
o
is noted and the sizes tabulated in increasing order for
many such samples. Then, the probability position for
each inclusion is calculated according to
P
i

i
N1
23
where i is the ith inclusion and N the total number of
inclusions.
The reduced variate Y is given by
Y lnlnP
i
24
The mean

L and standard deviation Sdev of the inclu-
sion sizes are calculated, from which the following are
calculated
o
mom

Sdev

6
p

z
mom


L 0.5772o
mom
25
The extreme value (Gumbel) distribution is given by
f x
1
o
exp
x z
o
_ _ _ _
exp exp
x z
o
_ _ _ _
26
The maximum likelihood method is used to calculate
o
ML
and z
ML
from which the maximum inclusion size is
estimated according to
L
max
o
ML
ln ln
T 1
T
_ _ _ _
o
ML
27
where T is the return period
T
A
ref
A
o
28
Normally, A
ref
is taken as 150,000 mm
2
, so T could be
about 1000.
A typical plot is given in Figure 13 for the outer race
of a double row spherical bearing (type 24132), in
which a reduced variate of 3 corresponds to a 95%
condence limit. Higher condence may be taken by
a higher reduced variate, for example 7 (99.9%).
400 Proc IMechE Part J: J Engineering Tribology 226(5)
Fatigue and scatter
In structural fatigue, the lives of components or
test specimens under nominally the same loading con-
ditions can be signicantly dierent. Lundberg and
Palmgren
1,2
recognised that a similar situation applied
to rolling bearings and tted two-parameter Weibull
distributions
4
to the lives of rolling bearings. A devia-
tion from the Weibull distribution is observed at low
lives, attributed to an incubation period, often
described as the L
o
life.
However, they did not consider the underlying phys-
ical processes leading to sub-surface initiated fatigue.
Fatigue life covers both the crack initiation and crack
growth periods. In the case of rolling bearings, there is
the additional phase concerning propagation or non-
propagation of butteries. Tables 1 to 3 of Appendix
2 summarise possible sources of scatter that may con-
tribute to the distribution of fatigue lives.
In structural fatigue, the scatter in the initiation
phase is much more sensitive to the various inuences
than in the crack growth phase which usually shows
limited variability. Additionally, scatter tends to
increase at lower stresses close to the endurance limit.
In this approach, it is not necessary to assume a
Weibull distribution in bearing life.
The role of hydrogen
Hydrogen is considered to be a signicant inuencing
factor in the explanation of fatigue or aking failures in
rolling element bearings. The role of hydrogen in such
failures has been reviewed and the reader is referred to
Appendix 3 for details.
In the context of the approach developed here, the
role of hydrogen may cause earlier crack initiation, low-
ering of threshold stress intensity factors and increased
crack growth rate. Therefore, the Mode I and II stress
limits may be revised downwards accordingly to reect
these inuences.
Discussion
Examination of rolling bearings from industrial appli-
cations that operate at maximum contact stresses well
below 2000 MPa has allowed the physical processes of
sub-surface initiated fatigue to be explained using a
fracture mechanics approach. In contrast to higher con-
tact stresses, there is no evidence of strain-induced
DEAs or WEAs whilst microstructural changes princi-
pally consist of micro-cracking at inclusions and the
development of butteries.
A fracture mechanics interpretation has, therefore,
been proposed based upon Murakamis
19
work that
relates maximum inclusion size (as determined by
extreme value analysis EVA) to the contact stress
necessary to initiate micro-cracks. The approach essen-
tially mirrors defect tolerance in structural integrity.
The objective was to establish contact stress limits
for avoidance of micro-crack initiation as a function of
maximum inclusion size. Hence, it should be possible to
avoid sub-surface initiated fatigue and therefore
achieve innite life as the precursors to advanced crack-
ing are eectively suppressed.
Should maximum contact stresses exceed these
limits, then micro-cracking occurs under Mode I load-
ing, but into a progressively decreasing tensile stress
eld. It has been postulated that buttery wing devel-
opment takes place in conjunction with micro-crack
growth due to the formation of nano-crystalline ferrite
under the intense strain eld at the crack tip.
Growth of the crack, mainly along one edge of the
buttery wing, is assisted by micro-cracking close to the
boundary with the steel matrix. However, under some
circumstances, the crack cannot grow out of the wing
into the steel such that arrest can occur and benign NP
butteries formed.
P butteries are formed if cracking extends beyond
the buttery wing under Mode II shear growth. Thus,
Mode II thresholds become important in dening a
second contact stress limit. In this respect, the works
Figure 13. Typical EVA plot.
Lewis and Tomkins 401
of Kassir and Sih
21
and Matsunaga et al.
22
have been
used to dene this limit as a function of overall crack
size, taking into account that the buttery will tend to
grow with an aspect ratio b/a 0.49 (at which Mode II
and III stress intensity factors are equal).
A defect-tolerant approach demands that a factor of
safety be incorporated. In this instance, there are three
possible levels of safety. The most rigorous uses a
reduced variate of 7 in the estimation of maximum
inclusion size (99.9%), whilst a reduced variate of 3
may also be considered (95%), but this gives a lower
margin. The corresponding maximum contact stresses
are then determined using a plot similar to Figure 12
that reects actual steel hardness.
Lower margins are obtained by allowing micro-
crack initiation (i.e. contact stresses greater than the
Mode I stress) but preventing crack growth to failure
by non-propagation of butteries. This means that con-
tact stresses must be below the limit for Mode II
thresholds.
It must be understood that the approach here does
not consider micro-crack initiation time. Experience
suggests that prior cracking of the matrix at the bound-
ary with the inclusion should be assumed. Bearing life is
principally determined by the growth rate from micro-
crack to P buttery, since the nal stage of crack
growth to the formation of a spall is also likely to be
short.
Conclusions
A fracture mechanics interpretation of sub-surface ini-
tiated fatigue in rolling bearing steels containing non-
metallic inclusions has led to an allowable contact stress
limit as a function of maximum inclusion size (as esti-
mated by EVA). An alternative, less conservative stress
limit is the prevention of propagation of butteries by
Mode II loading. Observation of the physical processes
suggests that the concept of an endurance limit is valid
for bearing steels.
For AISI 52100 steels of high cleanliness, a conser-
vative limit of 1700 MPa is applicable for avoidance of
micro-crack propagation, increasing to 2500 MPa for
non-propagation of butteries.
For typical bainitic steels used in larger rolling bear-
ings with maximum inclusion sizes in the range
5080 mm, a conservative limit of 1000 MPa may be
used for maximum contact stress. This limit may be
revised downwards to include the eects of hydrogen.
The approach extends the work of Ioannides and
Harris, who incorporated the concept of an endurance
limit to the pioneering work of Lundberg and
Palmgren. It provides a physical interpretation of the
metallurgical changes that occur under repetitive
stressing leading to non-propagation of micro-cracking
on inclusions or non-propagation of butteries.
In practical engineering applications, it provides
guidance on maximum contact stresses for avoidance
of sub-surface initiated fatigue as a function of steel
cleanliness in order that bearings will attain long life
and thus high reliability. In terms of current rating
life methods based upon ISO 281, this approach sug-
gests that dynamic capacity and fatigue load limit
values should be reduced for bearing steels containing
inclusions of size greater than those in the bearing tests
of Lundberg and Palmgren in the late 1940s (with sub-
sequent improvements), and those of Harris and
Barnsby in the late 1990s.
Funding
This research was funded by the National Centre of
Tribology.
References
1. Lundberg G and Palmgren A. Dynamic capacity
of rolling bearings. Acta Polytech Mech Eng Ser 1947;
1(3): 7 (Royal Swedish Academy of Engineering
Sciences).
2. Lundberg G and Palmgren A. Dynamic capacity of roller
bearings. Acta Polytech Mech Eng Ser 1952; 2(4): 96
(Royal Swedish Academy of Engineering Sciences).
3. Zwirlein O and Schlicht H. Rolling contact fatigue mech-
anisms accelerated testing vs field performance.
In: JJC Hoo (ed.) Rolling contact fatigue testing of bearing
steels. ASTM STP 771. Philadelphia, PA: American
Society for Testing and Materials, 1982, pp.358379.
4. Weibull W. A statistical representation of fatigue failures
in solids. Acta Polytech Mech Eng Ser 1949; 5(9): 49
(Royal Swedish Academy of Engineering Sciences).
5. ISO 281-1.1990. Rolling bearings dynamic load ratings
and rating life.
6. ANSI/AFBMA Std 9:1990. Load ratings and fatigue life
for ball bearings.
7. ANSI/AFBMA Std 11:1990. Load ratings and fatigue life
for roller bearings.
8. Zaretsky E. STLE life factors for rolling bearings. Park
Ridge, IL: STLE, 1992.
9. Ioannides E and Harris TA. A new fatigue life model for
rolling bearings. ASME J Tribol 1985; 107: 367378.
10. Harris TA and Barnsby RM. Life ratings for ball and
roller bearings. Proc IMechE Part J: J Engineering
Tribology 2001; 215: 577595.
11. Barnsby R, et al. Life ratings for modern rolling bearings
a design guide for the application of International
Standard ISO 281/2. Tribology, vol. 14. New York:
ASME, 2003.
12. ISO 281:2007. Rolling bearings dynamic load ratings
and rating life.
13. Zaretsky E. In search of a fatigue limit: a critique of
ISO Standard 281:2007. Tribol Lubr Technol 2010; 66:
3040.
402 Proc IMechE Part J: J Engineering Tribology 226(5)
14. Takemura H, Matsumoto Y and Murakami Y.
Development of new life equation for ball and roller bear-
ings. NSK Motion Control 2001; 11: 110.
15. Sadeghi F, Jalalahmadi B, Slack T, et al. A review
of rolling contact fatigue. AMSE J Tribol 2009; 131:
041403.
16. Jalalahmadi B, Sadeghi F and Bakolas V. Material inclu-
sion factors for Lundberg-Palmgren based RCF life equa-
tions. Tribol Trans 2011; 54: 457469.
17. Donzella G and Petrogalli C. A failure assessment dia-
gram for components subjected to rolling contact load-
ing. Int J Fatigue 2010; 32: 256268.
18. Lamagnere P, Fougeres R, Lormand G, et al. A physi-
cally-based model for endurance limit of bearing steels.
ASME J Tribol 1998; 120: 421426.
19. Murakami Y. Metal fatigue: effects of small defects and
non-metallic inclusions. London: Elsevier, 2002.
20. Irwin GR. The crack extension force for a part-through
crack in a plate. J Appl Mech Trans ASME 1962; 29(4):
651654.
21. Kassir MK, Sih GC and Murakami Y (eds) Stress inten-
sity factors handbook. Oxford: Pergamon Press, 1987.
22. Matsunaga H, Shomura N, Muramoto S, et al. Shear
mode threshold for a small fatigue crack in a bearing
steel. Fatigue Fract Eng Mater Struct 2011; 34(1): 7282.
23. Grabulov A, Petrov R and Zandbergen H. EBSD inves-
tigation of the crack initiation and TEM/FIB analyses of
the micro-structural changes around the cracks formed
under rolling contact fatigue (RCF). Int J Fatigue 2010;
32(3): 576583.
24. Todaka Y, Umemoto M, Li J, et al. Nano-crystallization
of carbon steels by shot peening and drilling. Rev Adv
Mater Sci 2005; 10: 409416.
25. Kerrigan A, Kuijpers JC, Gabelli A, et al. Cleanliness of
bearing steels and fatigue life of rolling contacts. J ASTM
Int 2005; 3(6): paper no. JAI14040. (Presented at 7th
ASTM International Symposium on Bearing Steel
Technologies, Reno, NV, 1719 May 2005).
26. ASTM International. Standard practice for extreme value
analysis of non-metallic inclusions in steel and other micro-
structural features. ASTM E 2283-08, 2008. West
Conshohocken, PA: ASTM International.
27. Tamada K and Tanaka H. Occurrence of brittle flaking
on bearings used for automotive electrical instruments
and auxiliary devices. Wear 1966; 199: 245252.
28. Kohara M, Kawamura T and Egami M. Study on mech-
anism of hydrogen generation from lubricants. STLE
Tribol Trans 2006; 40: 5360.
29. Ruo P and Olver AV. Hydrogen in lubricated contact. In:
Proceedings of the first PhD conference, Ostravice, Czech
Republic, 1923 February 2007.
30. Vegter R and Slycke J. The role of hydrogen in rolling
contact fatigue response of ball bearings. In: Proceedings
of the 8th ASTM international symposium on bearing
steel technologies, Vancouver, BC, Canada, 2122 May
2009.
31. Gegner J and Nierlich W. Principles and applications of
XRD material response analysis. In: Proceedings of the
8th ASTM international symposium on bearing steel tech-
nologies, Vancouver, BC, Canada, 2122 May 2009.
32. Murakami Y, Nagata J and Matsunaga H. Factors
affecting ultralong life fatigue and design methods for
components. In: Proceedings of the Fatigue 2006,
Atlanta, GA, 1419 May 2006.
33. Matsubara Y and Hamada H. A novel method it evaluate
the influence of hydrogen on fatigue properties of high
strength steels. J ASTM Int 2005; 3(2): paper no.
JAI14048 (presented at 7th ASTM International
Symposium on Bearing Steel Technologies, Reno, NV,
1719 May 2005).
Appendix 1
Notation
a contact major semi-axis, crack major
semi-axis
area cross-sectional area of inclusion
a
1
life adjustment factor for reliability
a
2
life adjustment factor for material
a
3
life adjustment factor for operating
conditions
a
23
combined life adjustment factor
a
ISO
combined life adjustment factor (ISO 281)
a
SKF
life adjustment factor based on new life
theory
a
SL
stress-life factor
A
o
sample area for EVA
A
ref
reference area for EVA
b contact minor semi-axis, crack minor
semi-axis
c LundbergPalmgren exponent ( 31/3)
C bearing dynamic capacity related to
geometry and materials
C
u
fatigue load limit (alternatively P
u
)
D
pw
pitch diameter of the bearing
e Weibull exponent ( 9/8)
E(k) elliptic integral of first order
E() elliptic integral
f
c
factor depending upon geometry, accuracy
and materials
f(x) Gumbel probability
h LundbergPalmgren depth-weighting
exponent ( 7/3)
HV Vickers hardness
i number of roller sets
k, k
0
elliptic parameters
K stress intensity factor
K
I
, K
II
Mode I and II stress intensity factors
K(k) elliptic integral of second order
l raceway length
L
max
maximum estimated inclusion length
L
10
number of cycles survived by 90% of all
bearings (millions of revolutions)
Lewis and Tomkins 403
N number of cycles survived, total number of
inclusions
p exponent ( 3 for ball bearings and 10/3 for
roller bearings)
P equivalent load
P
i
probability
R R ratio (minimum stress in cycle/maximum
stress in cycle)
R
p
Plastic zone size
Q
u
load on a single roller that results in a
maximum contact stress of 1500 MPa
S percentage of bearing population surviving
Sdev standard deviation
T return period
V stressed volume
Y reduced variate
z stress weighted average depth
z
0
depth of maximum orthogonal shear stress
Z number of rollers
o contact angle, R ratio exponent
o crack tip opening displacement, Gumbel
parameter
o
mom
Gumbel parameter
dimension of tensile stress field around
inclusion (effective gauge length)
K
Ith
Mode I threshold stress intensity factor
K
IIth
Mode II threshold stress intensity factor
S survival probability of a volume V
c load distribution factor, strain
k ratio of operating viscosity to required
viscosity
z ratio of EHL oil film thickness to combined
surface finish, Gumbel parameter
z
mom
Gumbel parameter
v operating viscosity, Poissons ratio
v
1
required viscosity
o stress
o
max
maximum contact stress
o
vM
von Mises stress
o
w1
fatigue endurance limit
o
0
global stress
o
y
yield stress
t shear stress amplitude exceeding the fatigue
limit
t
0
maximum orthogonal shear stress
t
u
shear stress fatigue limit
angle
Appendix 2
Fatigue and scatter
Table 1. Possible sources of scatter for crack initiation.
Aspect Description
Material
variability
Maximum inclusion size
Number of inclusions within stressed volume
Hardness of matrix
Shape of inclusion
Tensile stress field around inclusion (R ratio)
Location of inclusion (i.e. depthwise)
Neighbouring inclusions
Pre-cracking of matrix
Pre-cracking of inclusion
Carbide size and distribution
Hydrogen content
Loading In-service load spectra
Constant or variable amplitude
Load cycle order (e.g. low/high or high/low)
Crest factor (peak/mean)
Accuracy of load estimate
In-service
changes
Development of sub-surface
compressive stress
(higher stresses only >2500 MPa)
Table 2. Possible sources of scatter for butterfly propagation.
Aspect Description
Material variability Mode I threshold stress intensity factor
as a function of length
Mode II threshold stress intensity factor
Carbide size and distribution
Tensile stress field around inclusion
(R ratio)
Hydrogen content
Loading In-service load spectra
Constant or variable amplitude
Crest factor (peak/mean)
Accuracy of load estimate
In-service changes Development of sub-surface
compressive stress
(higher stresses only >2500 MPa)
Table 3. Possible sources of scatter for crack growth.
Aspect Description
Material variability Carbide size and distribution
Hydrogen content
Loading In-service load spectra
Constant or variable amplitude
Crest factor (peak/mean)
Accuracy of load estimate
404 Proc IMechE Part J: J Engineering Tribology 226(5)
Appendix 3
The role of hydrogen
Hydrogen at a concentration of a few parts per million
is believed to cause brittle aking failures in bearing
steels,
27,28
principally through absorption caused by
breakdown of hydrocarbon lubricant in contact with
sliding surfaces. The failure characteristics, dierent
from classical sub-surface initiated fatigue, are irregular
cracking below the surface with white etching layers
along the crack faces.
A number of theories have been proposed to explain
hydrogen embrittlement:
29
. internal pressure due to hydrogen accumulation at
second-phase particles (e.g. inclusions) leading to
crack propagation along cleavage planes;
. reduction of the crack surface energy, a decrease in
the atomic bond strength or a decrement of cohesive
strength of the cleavage plane; and
. hydrogen-enhanced local plasticity due to enhanced
dislocation mobility.
There is also evidence that sub-surface compressive
stress of the order of 500 MPa developed under
repeated rolling contact at high stress tends to suppress
the eects of hydrogen, whilst contact stresses below
2400 MPa do not induce such compressive stress.
30,31
Thus, the eects of hydrogen may be experienced at
modest operating stresses with cracks appearing much
faster.
Murakami
19
and Murakami et al.
32
considered the
eects of hydrogen up to 0.8 ppm on fatigue of bearing
steels and concluded that the optically dark area
(ODA) around an inclusion (including the area of the
inclusion) should be used in equation (7), rather than
the inclusion area. In this instance, samples were pre-
pared with varying initial hydrogen contents, rather
than absorption from lubricant breakdown.
The ODA has a quite dierent morphology to
normal fatigue, this being attributed to crack growth
under the inuence of hydrogen. At 10
7
cycles, the

ODAinclusion area
_
is about 2.4 times that of
the

inclusion area
_
, suggesting that the Mode I
stress should be reduced by about 16%. This is in line
with fatigue endurance limit reductions reported by
Matsubara and Hamada
33
in which the measured
reduction at a higher hydrogen content of 3 ppm was
about 30%.
Lewis and Tomkins 405

You might also like