International Journal of Fatigue 31 (2009) 943–951

Contents lists available at ScienceDirect

International Journal of Fatigue

journal homepage: www.elsevier.com/locate/ijfatigue

Application of a microstructural model for predicting notch

fatigue limits under mode I loading
V. Chaves, A. Navarro *
Departamento de Ingeniería Mecánica y de los Materiales, Escuela Superior de Ingenieros, Universidad de Sevilla,
Camino de los Descubrimientos s/n, 41092 Sevilla, Spain

a r t i c l e i n f o a b s t r a c t

Article history: Microstructural fracture mechanics provides a description of short crack growth that accounts for the
Received 13 March 2008 interaction between the crack and microstructural barriers. An application to notched components sub-
Received in revised form 22 September jected to mode I loading is presented here. The techniques described are applied to a circular notch. The
2008
fatigue limit has been predicted for different materials and a wide range of hole radii. The predictions are
Accepted 25 September 2008
Available online 7 October 2008
quite close to the experimental results. Discussion and comparison of the results with those calculated
with other methods are presented.
Ó 2008 Elsevier Ltd. All rights reserved.
Keywords:
Short crack
Microstructural barrier
Dislocation
Fatigue limit
Notch

1. Introduction and their heuristic extension to mode I loading. Finally the rigorous
application of the model to mode I loading and the numerical tech-
Fatigue failure in industrial components takes place mainly at nique used to solve the resulting equations will be discussed.
stress concentrations. Methods for predicting fatigue failure must
take their effect into account. There have been many proposals to
2. Model of short fatigue crack growth
tackle this problem, beginning with the pioneering work of Neuber
[1] and Peterson [2] until the recent interpretation of the critical
Building upon the classical work of Bilby, Cottrell and Swinden
distance proposed by Taylor [3]. They all can be included in the
[5] on the representation of a crack and its associated plastic zones
classical fatigue techniques, which correlate stress or strain at
by means of dislocations, Navarro and de los Rios [6–8] developed
the notch area with the number of cycles to failure: the S–N curve
a model for fatigue microcracks growing in plain infinite bodies. It
introduced by Wöhler [4]. They are phenomenological approaches
was assumed that plastic displacement takes place in rectilinear
which do not model the crack, even when its growth is the ulti-
slip bands cutting across the grains of the material. The simplifica-
mate cause of failure. From a different point of view and since
tion of reducing the crack and the plastic zones to a single line
the beginning of the eighties of the last century, some theories of
seems appropriate in the early stages of fatigue failure since it is
short crack growth have been developed, which try to extend frac-
now well-known that the fatigue limit in metallic materials is a
ture mechanics down to the regime where the cracks have the size
threshold condition for the propagation of short cracks growing
of the material microstructure. The present work is included in this
within a few grains. Experimental evidence has shown that in plain
kind of theories. It describes a microstructural fracture mechanics
specimens subjected to uniaxial loads near the fatigue limit cracks
model which is applied to predict fatigue failure at notches under
form on persistent slip bands and grow along them. Thus, it makes
mode I loads. In what follows the microstructural model in which
sense to consider that dislocations are constrained (e.g. by stacking
the work is based will be briefly described. Then the first applica-
fault energy) to remain on their original planes and pile up against
tions of the model to the notch problem, developed prior to this
grain boundaries, giving rise to rectilinear slip bands extending
work, will be discussed. Basically they consist of analytical solu-
through the first few grains.
tions for elliptical notches under antiplane shear mode (mode III)
The microcrack nucleates in the grain which presents the most
favourable size and crystallographic orientation for the formation
* Corresponding author. Tel.: +34 954487311; fax: +34 954487295. of persistent slip bands. Within the first grain the plastic zone
E-mail address: [email protected] (A. Navarro). spreads from the crack tip to the first microstructural barrier

0142-1123/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijfatigue.2008.09.009
944 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951

(e.g. the grain boundary), where it will be stopped until the local
stress at the barrier is high enough for activating plastic slip in
the following grain. The calculation of the stress acting upon the
barrier is thus of fundamental importance in this framework. It will
be shown below that it depends on the applied stress and on the
relative position of the crack tip with respect to the barrier, in such
a way that the nearer the crack tip is to the barrier, the higher the
stress acting upon it is. The fatigue limit is the threshold applied
stress below which plastic slip in the next grain will not be acti-
vated even when the crack itself reaches the boundary. In this case,
Fig. 2. Crack, plastic zone and barrier in a polycrystalline infinite body modelled
the crack becomes non-propagating. If, on the other hand, the ap- with dislocations.
plied stress is above the fatigue limit, plastic slip will be promoted
in the following grain once the crack tip is sufficiently close to the
barrier. In this case, the process of spreading up to the next grain dislocations f ðfÞ. The problem has an analytical solution obtained
and blocking is repeated at every successive grain. Thus the plastic by using the inversion theorem presented by Muskhelishvili [9].
zone is thought to grow in jumps. An oscillating crack propagation In the case where a function f ðfÞ bounded at both ends of the do-
rate with smaller and smaller oscillations is predicted, which main is sought, the solution is [10]:
seems to be in agreement with the experimental evidence.      
The crack, the plastic zones and the barriers are modelled by a 1 1 1  n1 f 1 1 þ n1 f
f ðfÞ ¼ ðr 2  r 1 Þ cosh  n f   cosh  n þf 
continuous distribution of dislocations. The situation is illustrated p2 A 1 1
    
in Figs. 1 and 2 (where edge dislocations are shown). A polycrystal- 
1 1  n f
2  
1  1 þ n2 f
þ ðr3  r2 Þ cosh  n  f   cosh  n þf  ð2Þ
line infinite body with average grain size D and a crack of length 2a 2 2
is subjected to a uniform applied stress at infinity. A distribution of
screw dislocations subjected to an external shear stress syz ¼ s at And the existence condition ensuring that no singularities ap-
infinity would model a state of antiplane strain and the crack pear in the stress field is:
opened in mode III. The analysis for a distribution of edge disloca- 1 1 p
tions under shear stress sxy ¼ s at infinity leads to a state of plane ðr2  r1 Þ sin n1 þ ðr3  r2 Þ sin n2 þ ðs  r3 Þ ¼ 0 ð3Þ
2
strain and the crack opened in mode II. Edge dislocations with Bur-
The stress r3 required to maintain equilibrium for each position
ger’s vectors perpendicular to the plane of the crack subjected to a
of the crack tip is thus given by
tensile stress ryy ¼ r represents a crack opened in mode I. The
equations determining the dislocation distribution are the same 1 h p i
1 1
in the three cases, except for the value of the constant A given be- r3 ¼ ðr2  r1 Þ sin n1  r2 sin n2 þ s ð4Þ
cos1 n2 2
low. The distribution of dislocations is odd in x, both for edges and
screws, but for the latter there is no traction across the surface As the crack grows, i.e. n1 increases, the stress r3 at the barrier
x ¼ 0. Therefore, in the antiplane mode III case, the solution for must increase to keep the balance. The stress r3 reaches its peak
x > 0 can be used to represent the state of a semi-infinite body value when the crack has grown up to the barrier. Considering a
with a free surface at x ¼ 0. freely slipping crack ðr1 ¼ 0Þ and the crack tip positioned just at
The stresses r1 ; r2 ; r3 represent the resistance to the movement the border of the barrier ðn1 ¼ n2 ¼ ðiD=2Þ=ðiD=2 þ r0 ÞÞ the follow-
of dislocations in the crack, plastic zone and barrier, respectively. ing maximum value is obtained:
The microstructural barrier is modelled as a small region of length 1 p
r0 ðr 0  DÞ, which might be the typical size of the grains interface
r3 ¼ s ð5Þ
arccosðn2 Þ 2
or the distance to the dislocations sources in the next grain. The
equation of the equilibrium of dislocations is a Cauchy-type inte- This equation provides the key for calculating the fatigue limit
gral equation. Its expression, in terms of the non-dimensional vari- for mode III loading. The study of the fatigue limit under mode I
ables x; f, is as follows: loading, however, calls for the consideration of a slightly different
Z problem, an edge short crack growing from the surface, which can
1
1 f ðfÞdf ðs  rf Þ be modelled by means of a distribution of edge dislocations with
þ ¼0 ð1Þ
p 1 xf pA Burger’s vector perpendicular to the crack plane and sustaining a
tensile stress ryy ¼ r (see Figs. 3 and 4). In this case the influence
where A ¼ Gb=2p for screw dislocations and A ¼ Gb=2pð1  mÞ for
of the free boundary gives rise to additional terms in the equilib-
edge dislocations. rf is the friction stress, which takes values
rium equation:
r1 ; r2 ; r3 in each zone. The unknown function is the density of

Fig. 1. Crack, plastic zone and barrier in a polycrystalline infinite body. Fig. 3. Crack, plastic zone and barrier in a semi-infinite body subjected to mode I.
 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951 945

This equation provides a microstructural mechanics representa-

tion of the Kitagawa–Takahashi diagram [12]. Thus, the sequence
ðmi =m1 Þ for increasing i can be obtained from the Kitagawa–Takah-
ashi diagram of the material, as indicated by de los Rios et al. [13].

3. Microstructural model for notches

An important feature of the Kitagawa–Takahashi diagram for

plain specimens is that, generally, rLi is a non-increasing function
of i: there is normally a first horizontal part at the level of the fa-
tigue limit of the material and then, after a few grains, rLi de-
creases steadily until it joints the 1=2 slope line (in log–log
Fig. 4. Crack, plastic zone and barrier in a semi-infinite body subjected to mode I
modelled with dislocations. coordinates) associated to the stress intensity threshold K th . The
existence of a stress gradient at the notch root makes quite a dif-
Z " # ference with respect to the plain case as regards the capability of
1 1
1 1 2f 4f2 ðr  rf Þ the crack to overcome the successive microstructural barriers. Fa-
f ðfÞ   þ df þ ¼0
p 0 x  f x þ f ðx þ fÞ2 ðx þ fÞ3 pA tigue failure can only occur if the crack is able to traverse not just
the first barrier but all the successive ones. But in notched spec-
ð6Þ
imens the stress driving the crack ahead decreases. Thus, it might
It can be shown [11] that for this geometry and in the case of a be possible for the crack to get across the first few barriers but
freely slipping crack the maximum value of r3 , obtained again then to find itself unable to overcome the following ones. If this
when the crack tip reaches the barrier, is: were the case, the crack would become non-propagating. The
k p minimum applied stress necessary to make the crack propagate
r3 ¼ r ð7Þ through all the barriers will be the fatigue limit of the notched
arccosðn2 Þ 2
component, rNFL . If rNLi designates the value of the applied stress
where the constant k has the value k ¼ 1:1215, which ties in with
needed to get across the ith barrier in the notched component,
what is expected from the classical fracture mechanics solution,
then:
where the expression for the stress intensity factor for a small edge 
pffiffiffiffiffiffi
crack in a semi-infinite body is K I ¼ 1:1215r pa, and the factor rNFL ¼ max rNLi i ¼ 1; 2; . . . n ð13Þ
1:1215 may be seen as modifying the ‘‘infinite solution” to yield
The value of i which determines the fatigue limit provides the
the ‘‘semi-infinite” one.
maximum length of non-propagating cracks that may be found in
The crack will overcome the barrier if the stress r3 reaches a va-
the notched component.
lue high enough to activate dislocation sources in the neighbouring
Navarro et al. [14,15] developed a first application of the micro-
grain, and this will happen if the shear stresses projected in the slip
structural model to notch geometries. They obtained an analytical
plane and in the slip directions of the dislocation sources are capa-
solution for an elliptical notch subjected to mode III loading
ble of unlocking them. This critical condition must be written for
(Fig. 5). Using the representation of antiplane stress by means of
each successive barrier, thus calling ri3 the value of the barrier
a complex potential and conformal mapping techniques, the prob-
stress r3 for barrier i, one has:
lem was transformed mathematically to the study of the equilib-
ri3 rium of dislocations in a semi-infinite plane and reduced to a
¼ sic ð8Þ
mi singular integral equation similar to the one already solved for
where mi is the crystallographic orientation factor for the grain con- the infinite plate. Vallellano [16,17] proposed a generalization of
tiguous to barrier number i and sic is the critical stress necessary to this solution for notches with shapes different to the elliptical
activate the dislocation sources there. By substituting in (7), the one and for cracks growing in mode I, by simply substituting, in
minimum applied stress rLi necessary to activate plastic slip in the equation derived for the mode III, the normal stress r for the
the next grain for a crack spanning i half grains is: antiplane shear stress s and by re-writing the equations in terms
of three parameters, namely, the stress concentration factor K t ,
2 arccosðn2 Þ  i the notch root radius q and the notch depth a. It has been claimed
rLi ¼ mi sc ð9Þ
p k that these three parameters can account for the stress distribution
Now by using the approximation arccosðn2 Þ ¼ around the notch [18]. This approximate model was compared
1
1 with experimental results from the literature (El Haddad et al.,
ð2ð1  n2 ÞÞ2  2 ri0 =iD 2 the above equation leads to a very reveal-
Frost and Dugdale, Tanaka and Akinawa, Lukás et al., DuQuesnay
ing expression:
 1 et al., etc.) [15,17] and the predictions effected agreed quite well
1 4 r i0 2  i with the experimental results.
rLi ¼ mi sc ð10Þ
k p iD
The fatigue limit for an unnotched solid is obtained by putting
i¼1
 1
1 4 r 10 2  1
rFL ¼ m1 sc ð11Þ
kp D
If the parameters r i0 and sic are considered to be constant, r i0 ¼ r 0
and sic ¼ sc , the following simple expression relating rLi y rFL can
be obtained:
rLi mi 1
¼ pffi ; i ¼ 1; 3; 5; . . . ð12Þ
rFL m1 i Fig. 5. Crack, plastic zone and barrier in a notched body.
946 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951

When the problem of a crack in a notched component subjected Once the set of integral equations is numerically solved it is
to mode I loading is analyzed by the continuous dislocations tech- possible to obtain a relationship between the applied stress r
nique, a singular integral equation is obtained which does not and the stress at the barrier ri3 for each notch geometry, expressed
seem amenable to close-form analytical solution. The kernel of in general through the variables K t , q and a. This relationship de-
the equation has a Cauchy term, coming from the interaction pends on the microstructural variables of the problem, the grain
among dislocations, and there are also inverse quadratic and cubic size D and the ‘‘width” of the barrier r0 , as well as on the crack
terms, similar to those in Eq. (6), coming from the interaction with length i. The relationship can be written as:
the free surface of the notch. In general the equation is of the type:
ri3 ¼ uðD; i; r0 ; K t ; q; aÞr ð16Þ
Z 1
½ry ðxÞ  rf ðxÞ 1
 ¼ f ðfÞKðx; fÞdf If the value of the applied stress r at which the stress at the bar-
pA p 1 rier ri3 reaches the critical stress that makes the crack propagate
Z 1  
1 1 into the next grain is named rNLi , then one has:
¼ f ðfÞ þ K 0 ðx; fÞ df ð14Þ
p 1 xf
ri3 ¼ mi sic ¼ mi sc ¼ uðD; i; r0 ; K t ; q; aÞrNLi ð17Þ
with jxj < 1. Kðx; fÞ ¼ 1=ðx  fÞ þ K 0 ðx; fÞ is the so-called kernel of
the equation and will adopt different expressions depending on The critical values s can be related to the plain fatigue limit
mi ic
the geometry of the solid and the notch. For the particular case of through Eqs. (10) and (12). The fatigue limit rNFL for the notched
a crack in a semi-infinite plate the kernel was shown in Eq. (6). component can, thus, be expressed as:
But the presence of the notch further complicates the matter. For (  12 )
the notch geometry discussed in this work its expression will be N
 N
p D mi 1
r ¼ max r
FL Li ¼ max k rFL  ;
given further down. The term ry ðxÞ is the stress induced at the crack 4 r0 m1 uðD; i; r0 ; K t ; q; aÞ
line by the external applied load r (as in Bueckner’s superposition i ¼ 1; 3; 5; . . . ð18Þ
principle [19]). Its value depends on the applied load and the geom-
etry of the solid and the notch. The term rf ðxÞ is the friction stress, Alternatively, it might just be easier to read rLi directly from the
with different values for the crack, the plastic zone and the barrier, experimental Kitagawa–Takahashi diagram and obtain:
as explained above. (  12 pffi )
A numerical integration technique has been applied in the pres- N
 N
p D 1
r ¼ max r
FL Li ¼ max k rLi i ;
ent case, based on the method developed by Erdogan, Gupta and 4 r0 uðD; i; r0 ; K t ; q; aÞ
Cook [20]. The usual procedure for solving regular integral equa-
i ¼ 1; 3; 5; . . . ð19Þ
tions employs quadrature formulae for calculating integrals, reduc-
ing them to an addition of terms [21]. The method of Erdogan, The algorithm for calculating the notch fatigue limit would then
Gupta and Cook, which invokes some special properties of the consist of the following steps:
orthogonal polynomials used in the quadrature, can be applied to
obtain quadrature formulae for the singular terms of the integral  Given the notch geometry ðK t ; q; aÞ, the material properties
equations, yielding a system of algebraic equations of the type: ðD; r0 ; rFL Þ and the applied load, writing the equilibrium equa-
  tions for the crack and the barrier in the presence of the notch.
½ry ðvk Þ  rf ðvk Þ X N
1 This requires the knowledge of the kernel for the particular
 ¼ Wðuj Þ/ðuj Þ þ K 0 ðvk ; uj Þ ð15Þ
pA j¼1
vk  uj geometry (see below). The equations can be written for a generic
length of the crack iD=2; i ¼ 1; 3; 5; . . ..
where the unknown function /ðuj Þ is evaluated at the integration  Solving the integral equation (e.g., by using the modified numer-
points uj within the definition interval and the equations are only ical method of Erdogan et al. as described above) for successive
valid at the collocation points vk . Both the uj ’s and the vk ’s corre- crack lengths iD=2ði ¼ 1; 3; 5; . . .Þ. This provides the numerical
spond to the zeros of the appropriate Jacobi polynomials [20]. relationship between the applied stress r and the stress at the
Wðuj Þ are the weights of the quadrature. The specific values of barrier ri3 for each i, Eq. (16).
m
vk ; uj ; Wðuj Þ, as well of the number of collocation and integration  Obtaining the values of mi or of rLi ði ¼ 1; 3; 5; . . .Þ by using the
1
points, depend on the kind of solution sought, bounded or un- Kitagawa–Takahashi diagram of the material. If the experimen-
bounded at the interval ends. For the present case, interest lies in tal diagram were not available the value of rLi could be approx-
the bounded solution. The standard method of integration [20] imated by means of the following equation proposed by
was originally devised for calculating stress intensity factors in lin- Vallellano [15,17]:
ear elastic fracture mechanics problems, where no plastic zones or pffiffiffiffiffi
a0
barriers exist and therefore unbounded solutions capable of captur- rLi ¼ rFL h i1 ð20Þ
pffiffiffi
ðiD=2Þf þ ða0 Þf  ðD=2Þf
2f
ing the 1= r singularity in the stress field at the crack tip were
sought. The straightforward application to the present problem
meets with some difficulties, arising from the inherent jump in where a0 ¼ ½K th1 =rFL 2 =p. A value 2:5 for the exponent f has
the value of the friction stress, which induces logarithmic singular- proved appropriate for a range of materials [15].
ities at the crack tip and plastic zone-barrier interfaces [22,23]. To  Obtaining the notched fatigue limit rNFL by calculating the max-
solve this limitation the interval of integration is divided in imum of the sequence rNLi as expressed in Eqs. (18) and (19).
subintervals, placing the singularities at their ends. For the study
of fatigue limits, since the crack tip is just touching the barrier, only 4. Application to a circular notch
two subintervals are required, one for the crack and another for the
barrier. The solution from both subintervals must be coupled by As a first application of the proposed methodology, a circular
imposing continuity of the plastic displacement at their common notch (K t ¼ 3, q ¼ a ¼ R), as shown in Fig. 6, has been chosen. It
point and, in this case, this provides the equation to calculate ri3 . is a circular hole in an wide plate subjected to pull–push loading.
The values of the unknown function required at the end of the sub- The objective is to obtain the fatigue limit as a function of the ra-
intervals may be obtained by using the extrapolation formulae dius R. The analytical solution of the problem of a dislocation in
introduced by Krenk [24]. an infinite plate located near a circular hole is known and this
 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951 947

Z Rþa
1 1
 ½r ðx^ Þ  r1  ¼ f1 ð^f1 ÞKð^x1 ; ^f1 Þd^f1
pA y 1 p R
Z Rþc
1
þ f3 ð^f3 ÞKð^x1 ; ^f3 Þd^f3 ð24Þ
p Rþa

and
Z Rþa
1 1
 ½r ðx^ Þ  r3  ¼ f1 ð^f1 ÞKð^x3 ; ^f1 Þd^f1
pA y 3 p R
Z Rþc
1
þ f3 ð^f3 ÞKð^x3 ; ^f3 Þd^f3 ð25Þ
p Rþa

As usual, these intervals of integration are normalized to ½1; 1,

through the following changes of variables (Note that a is the crack
Fig. 6. Infinite plate with a circular hole subjected to tension.
length and c ¼ a þ r 0 , n ¼ a=c):

greatly simplifies the obtention of the kernels of the equilibrium ^f1 ¼ ðu1 þ 1Þ n  c þ R
2
equations. n
Fig. 7 shows a sketch of the problem modelled by means of dis- ^x1 ¼ ðv1 þ 1Þ  c þ R ð26Þ
 2
 
locations. The case of two small cracks of length a growing from 1n
^f3 ¼ ðu3 þ 1Þ þ n  c þ R
diametrically opposed locations is considered. Symmetry with re- 2
spect to the vertical-axis is assumed to simplify the problem.   
1n
The stress r is the remotely applied stress while ry ð^ xÞ is the ^x3 ¼ ðv3 þ 1Þ þ n  c þ R
2
stress that would appear at the crack line if the crack were not
there [19]. To obtain this stress, the classical solution for the hoop Now the method of Erdogan, Gupta and Cook [20] can be ap-
stress rhh around a circular hole in an infinite body subjected to plied to transform the integral equations in a set of algebraic equa-
tension and particularized to the plane y ^ ¼ 0 is used [25]: tions. The quadratures employed in each subinterval are shown in
"  2  4 # Table 1. They have been chosen to account for singular behaviour
r R R
ry ð^xÞ ¼ 2þ
^x
þ3
^x
^x P R ð21Þ of the dislocation density at the crack-barrier joint and bounded
2 behaviour at the other two ends. It must be pointed out that the
quadrature used for the crack region represents only an approxi-
The kernel K c ð^
x; ^fÞ for a crack emanating from a circular hole in
mate solution since it does not capture the right behaviour of the
an infinite body can be obtained from the classical solution of Dun-
crack faces at the mouth, which are overconstrained to remain par-
durs and Mura [26,27]:
allel, something which is physically not possible. Hills [27,28] has
" #
1 ^f ð^f2  R2 ÞR2 ^f2 ^f2  R2 discussed this problem of the quadrature for the surface-breaking
K c ð^x; ^fÞ ¼    crack. Although more appropriate quadrature schemes are avail-
^x  ^f ^x^f  R 2 ^fð^x^f  R2 Þ2 R2 ^x^f  R2
able for this solution, it has not been deemed necessary to use
^f2  R2 1 R2 them since according to Dewynne, Hills and Nowell [28], the influ-
þ þ þ 3 R 6 ^x; ^f 6 R þ c ð22Þ
^x2^f ^x ^x ence of this incorrect representation at the crack mouth does not
appreciably affect the calculated values for the stress intensity fac-
The kernel for the two opposing cracks is as follows: tor at the crack tip. Translating this to the present framework, one
does not expect any significant influence on the calculations of the
Kð^x; ^fÞ ¼ ½K c ð^x; ^fÞ  K c ð^x; ^fÞ ð23Þ friction stress r3 at the barrier. There is also another approxima-
tion involved in the fact that the real singularity at the point where
The integral equations defining the equilibrium of a generic dis- the friction stress experiences the jump, which is logarithmic, is
location located at ^x for each one of the two subintervals, crack and pffiffiffi
represented by the quadrature as a 1= r singularity. Tests done
barrier, is given by (variables with subindex 1 refer to the crack and with this procedure on infinite problems where analytical solu-
3 refer to the barrier): tions are available for comparison have shown the error to be neg-
ligible, though.
The resulting equations are:

1 XN
2ð1 þ u1j Þ n
 ½ry ðv1k Þ  r1  ¼ Kðv1k ; u1j Þ /1 ðu1j Þ
pAc j¼1
2N þ 1 2
XN
2ð1  u3j Þ 1n
þ Kðv1k ; u3j Þ /3 ðu3j Þ;
j¼1
2N þ 1 2
k ¼ 1; 2; . . . N ð27Þ

Table 1
Quadratures for the crack and the barrier.

Subinterval f ðuÞ xðuÞ WðuÞ

qffiffiffiffiffiffiffiffiffi
Crack f1 ðu1 Þ ¼ x1 ðu1 Þ/1 ðu1 Þ x1 ðu1 Þ ¼ 1þu1
1u1 W 1 ðu1 Þ ¼ 2ð1þu
2Nþ1
1Þ

qffiffiffiffiffiffiffiffiffi
Fig. 7. Infinite plate with a circular hole subjected to tension modelled with Barrier f3 ðu3 Þ ¼ x3 ðu3 Þ/3 ðu3 Þ x3 ðu3 Þ ¼ 1u3
W 3 ðu3 Þ ¼ 2ð1u3Þ
1þu3 2Nþ1
dislocations.
948 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951

and Kt  1
Kf ¼ 1 þ ð33Þ
1 þ a=R
1 X
N
2ð1 þ u1j Þ n
 ½r ðv Þ  r3  ¼ Kðv3k ; u1j Þ /1 ðu1j Þ
pAc y 3k j¼1
2N þ 1 2 where the material constant ‘‘a” has dimensions of length and, for a
given material, depends essentially on the critical depth below the
XN
2ð1  u3j Þ 1n surface at which the stress is measured [2,14]. It has been adjusted
þ Kðv3k ; u3j Þ /3 ðu3j Þ;
j¼1
2N þ 1 2 to best fit a large set of experimental data on notch sensitivities. In
the case of steels, its value can be obtained from the graph proposed
k ¼ 1; 2; . . . N ð28Þ
by Langer [29], which relates it to the ultimate tensile strength of
where the integration and collocation points are, respectively: the material. Alternatively, the SAE and ASM Fatigue Handbooks
    [30,31] provide the following practical empirical relationship:
2j  1 2j  1:8
u1j ¼ cos p ; u3j ¼ cos p j ¼ 1; 2; . . . N 2069
2N þ 1 2N þ 1 a ¼ 0:0254 ð34Þ
    rUTS
2k 2k  1
v1k ¼ cos p ; v3k ¼ cos p k ¼ 1; 2; . . . N ð29Þ
2N þ 1 2N þ 1 for ‘‘a” in mm and rUTS , the ultimate tensile strength, in MPa. No
clear relationship of this kind seems to be in usage for aluminium
If the friction stress at the crack r1 is considered to be zero then alloys. A value of a = 0.51 mm for aluminium alloys sheets and bars
this is a system of 2N algebraic equations and 2N þ 1 unknowns, is suggested in Peterson’s stress concentration factors book [32].
which are /1 ðu1j Þ, /3 ðu3j Þ and the stress r3 required to keep the The second method with which comparisons have been made is
equilibrium. One further equation is thus needed and it is obtained the so-called point method introduced by Taylor [3,33]. The idea is
from the coupling of the solutions in the two intervals, expressed essentially that of Peterson: failure will occur if the stress at a dis-
as: tance L=2 from the notch root equals the fatigue limit. The applica-
tion of this criterion directly to a sharp crack, rather than to a
f1 ð1Þ ¼ f3 ð1Þ ð30Þ notch, has allowed Taylor to derive a very simple and practical rec-
Which leads to ipe to obtain the critical distance L for the material in terms of lin-
ear elastic fracture mechanics parameters. It turns out that L
/1 ð1Þ ¼ /3 ð1Þ ð31Þ coincides with the El Haddad’s distance a0 (see below). L is given
The values of the function / at the end-points 1 must be ob- by:
tained by extrapolation, since the solution is given for a discrete  2
1 DK th
set of points not including the extremes of the interval. Krenk’s L¼ ð35Þ
extrapolation formulae [24] may be used, as suggested by Hills
p DrFL
[27], giving: This procedure has extended enormously the field of applicabil-
0 h i 1 0 h i 1 ity of the old ideas of Neuber and Peterson, not only by giving a
2j1 2j1
XN sin 2Nþ1 Np XN sin 2Nþ1 Np practical formula to calculate the critical distances, but also by
@ h i /1 ðu1j ÞA ¼ @ h i /3 ðu3ðNþ1jÞ ÞA
2j1 p 2j1 p advocating the straightforward application of the idea of looking
j¼1 tan 2Nþ1 j¼1 tan 2Nþ1
2 2 at the stresses below the surface by actually doing so (via FEM
ð32Þ for example) rather than resorting to the formulas for K f , which de-
pend on K t , a parameter that in many practical situations can not
The final result is a system of 2N þ 1 algebraic equations and
be easily defined [33].
2N þ 1 unknowns (/1 ðu1j Þ; /3 ðu3j Þ and r3 ).
The results obtained with the approximate equation of Vallel-
In order to perform the numerical calculations for the model, a
lano et al. [17] described earlier have also been used for
value for the barrier length r0 must be specified. Systematic simu-
comparisons.
lations performed with a wide range of ratios r0 =D reveal that the
Fig. 8 shows the experimental results for push–pull tests on a
calculated values of the notched fatigue limit are quite insensitive
CSA G40.11 steel (Table 2) from El Haddad et al. [34] and the
to the particularly chosen value of r0 =D as long as it is sufficiently
small. The deviations are negligible whenever r 0 =D < 10 and we
would expect this ratio to be far smaller in practice. In the calcula-
tions reported here the value of r 0 has been set arbitrarily to 700
Experimental points
r0 ¼ D=30. With respect to the number of equations ð2N þ 1Þ Peterson (a=0.14 mm)
needed in the numerical solution to obtain both a reasonable rate Point method (L=0.24 mm)
600 Vallellano et al.
of convergence and the necessary accuracy, it has been found out Present model
that values of N around 30 give satisfactory results. For example
the results for the fatigue limit obtained with much higher values 500
ΔσFL (MPa)

of N ðN ¼ 500Þ differ from those obtained with N ¼ 30 by less than

2%.
400

5. Results
300
The predictions effected with the present proposal have been
compared with experimental results and with those obtained by
200
other methods. The first one is based on the idea of notch sensitiv-
ity introduced by Peterson [2]. He assumed that fatigue strength
depended on the stress at a certain depth below the surface. He 100
0.05 0.1 1 10
gave a well-known formula to calculate the fatigue notch factor
R, Radius of the circular hole (mm)
K f (the unnotched fatigue strength divided by the notch fatigue
strength) Fig. 8. Results of El Haddad et al. for the CSA G40.11 steel.
 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951 949

Table 2 700
Material properties, CSA G40.11 steel (R = 1). Tests done by El Haddad et al. Experimental points
pffiffiffiffiffi Peterson (a=0.15 mm)
Material rUTS a(MPa) DK th (MPa m) DrFL (MPa) a0 (mm) Point method (L=0.17 mm)
600 Vallellano et al.
CSA G40.11 steel 800 15.9 580 0.24 Present model
a
Estimated from steels with similar properties.
500

ΔσFL (MPa)
predictions with the different methods. The specimens are plane 400
ones, 70 mm in width, with circular holes of radii 0.2, 0.48 and
4.8 mm. The material constant for the Peterson’s model is
a = 0.14 mm and for the Point method is L = 0.24 mm. 300
The use of the microstructural model requires also the grain size
and the Kitagawa–Takahashi diagram of the material. The grain
200
size was not reported in [34] and thus it has to be estimated. This
can be achieved by using the microstructural equation for the
Kitagawa–Takahashi diagram (see [14])1: 100
0.05 0.1 1 10
 2  2 R, Radius of the circular hole (mm)
2 DK th m1
D¼ ð36Þ
p DrFL m1 Fig. 9. Results of DuQuesnay et al. for the SAE-1045 steel.
m1 1
Taking a value of ¼ m1 3:1
,
representative of fully polycrystalline
behaviour for a long crack, the following estimation is obtained:
300
D = 0.050 mm. Experimental points
El Haddad et al. in their work provide a set of experimental val- Peterson (a=0.51 mm)
Point method (L=0.26 mm)
ues to build the Kitagawa–Takahashi diagram of the material. Be- Vallellano et al.
sides, they propose an equation that accurately interpolates Present model
between these experimental results. The equation is:
200
DK th
ΔσFL (MPa)

Dr ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð37Þ
pða þ a0 Þ
where a0 ¼ ½DK th1 =DrFL 2 =p. Please note that this is really the same
constant used in Eq. (20). Using this equation to calculate the limit-
ing stress ranges DrLi when the crack is stopped at the ith barrier 100

(a ¼ iD=2) for this material one has:

sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a0
DrLi ¼ DrFL iD
ð38Þ
2
þ a0
0
All values of DrLi have been obtained from this equation except 0.05 0.1 1 10
for DrL1 , which has been set equal to the fatigue limit, DrL1 ¼ DrFL , R, Radius of the circular hole (mm)
to maintain consistency with the fact that, in the present frame- Fig. 10. Results of DuQuesnay et al. for the aluminum alloy 2024-T351.
work, the threshold condition to overcome the first grain boundary
is precisely the fatigue limit.
Figs. 9 and 10 show the experimental results for push–pull tests
Table 3
reported by DuQuesnay et al. [35] and the predictions provided by Material properties (R = 1). Tests done by DuQuesnay et al.
the different models. The materials are a SAE 1045 steel and a
2024-T351 aluminum alloy. The material properties are reported Material rUTS DK th
pffiffiffiffiffi
DrFL a0 (mm)
(MPa) (MPa m) (MPa)
by the authors themselves and are given in Table 3. Plate speci-
SAE 1045 steel 745 13.86 606 0.17
mens of 44.45
116.8 mm and 2.54 mm thickness, with circular
2024-T351 aluminum 466 7.04 248 0.26
central notches of radii 0.12, 0.25, 0.5 and 1.5 mm were used. alloy
The fatigue limit was defined at 107 cycles. For the SAE 1045 steel
the material constant for Peterson’s model is a = 0.15 mm and for
the Point method of Taylor is L = 0.17 mm. For the 2024-T351 the
value of the Peterson’s constant has been taken as a = 0.51 mm. 6. Discussion
For the Point method the calculated distance is L = 0.26 mm. The
grain sizes have been estimated again with Eq. (36), giving values For large radii the predictions effected with the four models are
of D = 0.035 mm for the steel and D = 0.053 mm for the aluminum quite similar, converging to the plain fatigue limit divided by the
alloy. For the Kitagawa–Takahashi diagram, Eq. (37) has been used stress concentration factor (surprisingly, the experimental values
following comments in [35] where the authors refer to this equa- for the SAE steel apparently do not follow this trend). The situation
tion postulated by El Haddad et al. is, however, very different when the radii become smaller. It is
interesting to note the big differences in the predictions by the
two methods based on the critical distance concept for the two
1
steels. The main reason for this seems to be the huge discrepancy
Please note that to make the comparison with the data reported in [34] and [35]
between the value of the critical distance calculated through Peter-
more straightforward the equations are written now in terms of ranges of stresses and
stress intensity factors, i.e. DrFL , DrLi and DK th are used from now on in place of rFL , son’s expressions and the value provided by the formula of Taylor.
rLi and K th (recall that R ¼ 1) for the results discussed. Recall that Peterson’s characteristic constant ‘‘a” is not really the
950 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951

Table 4 mount importance in low cycle fatigue problems, we believe its

Critical distances for the Peterson and Taylor models (mm). influence on the notch fatigue limit is much more reduced, since
a d L=2 the applied stresses would not be high enough to produce signifi-
CSA G40.11 steel 0.14 0.0374 0.12 cant plastic flow at the notch root, specially for large hole radii. The
SAE 1045 steel 0.15 0.04 0.085 effect might be more pronounced in the small radii region but the
2024-T351 aluminum alloy 0.51 0.136 0.13 complexities of modeling the co-existence of two plastic zones, the
one due to the notch itself and the one due to the crack, have re-
frained us from undertaking it at the present stage of our
investigation.
critical depth where the stresses are checked. Rather, the critical
distance d is related to ‘‘a” through [2,14]:
7. Conclusions
Kt
a ¼ Cd ð39Þ
Kt  1  We have presented a rigorous formulation of a microstructural
The constant C characterizes the stress gradient at the notch short crack growth model for notched components subjected
root. A value of C = 2.5 was used by Peterson as a reasonable aver- to mode I fatigue loading.
age value for cases of bending and axial loading. Table 4 compares  A numerical procedure for the solution of the (integral) equa-
d (Peterson) and L=2 (Taylor) for the three materials studied here. tions of the model has been presented and discussed.
For the G40.11 steel, Taylor’s distance L=2 is more than three times  The model has been applied to predict the fatigue limit of spec-
Peterson’s d. For the SAE steel L=2 is only twice as big as d. Accord- imens with circular notches of varying radius and compared
ingly the differences in the predictions are much wider for the with experimental results on two steels and an aluminum alloy
G40.11 steel. In the case of the aluminum alloy the values of the available in the literature [34,35].
critical distances are almost identical and thus the disparity be-  Comparisons have also been made with predictions obtained
tween the two predictions with both is minimized: they are not with two other methods based on the critical distance concept.
coincident though, but this is due to the fact that Peterson’s for-  For holes with radius of more than 2 or 3 mm all methods give
mula is only really an approximate relation which involves a num- similar predictions.
ber of simplifying compromises introduced in order to achieve  When the radius become smaller the situation changes, how-
generality, i.e. in order to provide a formula applicable to a wide ever. In particular the predictions made with the two methods
range of stress concentrations as intended by Peterson [2]. based on the critical distance concept (Peterson’s and Taylor’s)
The predictions based on Peterson seems to be better than those are quite different. This has been traced back to the fact that
based on Taylor. This is not surprising after all since, as was said the values of the critical distance calculated with both methods
before, the constant ‘‘a” is really a best fit for numerous experimen- differ quite a lot.
tal results.  The microstructural model gives predictions that are quite close
The microstructural model gives predictions that, in the present to the experimental results and that are markedly better than
case, seem more accurate than those based on the critical distance those based on the critical distance concept. This is particularly
concept. Of course this conclusion is circumscribed to the results so in the case of the results of the G40.11 steel presented by El
described here and can not be generalized on the scant evidence Haddad et al. [34] and it is believed that this is due to the fact
available. The predictions in the case of the G40.11 steel are that in this case the experimentally obtained Kitagawa–Takah-
remarkably good and this is undoubtedly connected to the fact that ashi diagram for this material, rather than a mere approximation
for this steel the experimental Kitagawa–Takahashi diagram is of it, has been used in the calculations.
available. Between the two implementations, the formulation pre-
sented in this paper seems to be slightly better than the earlier one Acknowledgement
developed by Vallellano [16], which is just as well, since Vallellan-
o’s formula was an heuristic generalization of the analytical solu- The authors would like to thank the Spanish Ministry of Educa-
tion for the mode III loading case [14], whereas the development tion for its financial support through Grant DPI2007-04077 and the
given here represents a more rigorous modelling of the mode I Junta de Andalucía through TEP1752. Helpful discussions with Dr.
case. However it is readily admitted that Vallellano’s formula is ex- Carpóforo Vallellano are also gratefully acknowledged.
tremely easy to use, while the calculations in the rigorous formu-
lation are much more involved. References
We would like to conclude this section by explicitly pointing
out two additional influencing factors which may contribute to [1] Neuber H. Kerbspannungslehre. Theory of notches [Edwards JW, Eng Trans].
Ann Arbor MI: Springer-Verlag; 1946.
the problem of fatigue at notches and to the observed discrepan- [2] Peterson RE. Notch sensitivity. In: Sines G, Waisman JL, editors. Metal
cies in the predictions, but which are not taken into account in fatigue. New York: McGraw-Hill; 1959.
any of the models discussed in the present paper. We are referring [3] Taylor D. Geometrical effects in fatigue: a unifying theoretical approach. Int J
Fatigue 1999;21:413–20.
to statistical size effects and residual stresses. Consideration of sta- [4] Wöhler A. Test to determine the forces acting on railway carriage axles and the
tistical fluctuations in the microstructural models can be under- capacity of resistance of the axles. Engineering 1871;11:199.
taken by performing Monte Carlo simulations allowing for [5] Bilby BA, Cottrell AH, Swinden KH. The spread of plastic yield from a notch.
Proc R Soc A 1963;272:304–14.
varying grain sizes and baseline fatigue properties of the material. [6] Navarro A, De los Rios ER. Short and long fatigue crack growth: a unified
However, since the critical distance models do not take that effect model. Phil Mag A 1988;57:12–36.
into account either, it was felt there was little point in introducing [7] Navarro A, De los Rios ER. An alternative model of the blocking of dislocations
at grain boundaries. Phil Mag A 1988;57:37–42.
yet another source of variability and, therefore, it has not been at-
[8] Navarro A, De los Rios ER. Fatigue crack growth by successive blocking of
tempted here. dislocations. Proc R Soc A 1992;437:375–90.
Residual stresses due to the notch plastic zone may also play an [9] Muskhelishvili NI. Singular integral equations. Noordhoff, Groningen, The
Netherlands; 1953.
important role in noth fatigue, and could be responsible for cracks
[10] Vallellano C, Navarro A, Dominguez J. Compact formulation for modelling
being halted at distances greater than and perhaps not directly cor- cracks in infinite solids using distributed dislocations. Phil Mag A
related to the grain size. However, while this should be of para- 2002;2:81–92.
 V. Chaves, A. Navarro / International Journal of Fatigue 31 (2009) 943–951 951

[11] Chaves V. PhD Thesis: Modelo microestructural para la predicción de fallo por [23] Golberg MA, Lea M, Miel G. Superconvergence result for the generalized
fatiga en entallas. University of Seville; 2006. airfoil equation with application to the flap problem. J Integ Eqs 1983;5:
[12] Kitagawa H, Takahashi S. Application of fracture mechanics to very small 175–86.
cracks. Int Conf Mech Behav Mater (ICM2), American Society of Metals; 1976. [24] Krenk S. On the use of the interpolation polynomials for solution of singular
p. 627–31. integral equations. Quart Appl Math 1975;32:479–84.
[13] De los Rios ER, Navarro A. Consideration of grain orientation and work [25] Timoshenko S, Goodier JN. Theory of elasticity. McGraw-Hill; 1934.
hardening on short-fatigue-crack modelling. Phil Mag A 1990;61:435–49. [26] Dundurs J, Mura T. Interaction between an edge dislocation and a circular
[14] Navarro A, Vallellano C, de los Rios ER, Xin XJ. Notch sensitivity and size effects inclusion. J Mech Phys Solids 1964;12:177–89.
by a short crack propagation model. In: Engineering Against Fatigue, Proc Int [27] Hills DA, Kelly PA, Dai DN, Korsunsky AM. Solution of cracks problems, the
Conf, Sheffield UK. Rotterdam: AA Balkema Publishers; 1997. distributed dislocation technique. Kluwer Academic Publishers; 1996.
[15] Vallellano C, Navarro A, Domínguez J. Fatigue crack growth threshold [28] Dewynne JN, Hills DA, Nowell D. Calculation of the opening displacement
conditions at notches. Part I: theory. Fatigue Fract Eng Mater Struct of surface-breaking plane cracks. Comp Meth Appl Mech Eng 1992;97:
2000;23:113–21. 321–31.
[16] Vallellano C. PhD Thesis: Crecimiento de grietas por fatiga en componentes [29] Langer BF. Application of stress concentration factors. Bettis Tech Rev WAPD-
con concentradores de tensión, University of Seville; 1998. BT-18 1960:1.
[17] Vallellano C, Navarro A, Domínguez J. Fatigue crack growth threshold [30] SAE Fatigue Design Handbook, AE-22. 3rd ed. In: Rice RC, editor. Warrendale,
conditions at notches. Part II: generalization and application to experimental PA: Society of Automotive Engineers; 1997. p. 296.
results. Fatigue Fract Eng Mater Struct 2000;23:123–8. [31] Mitchell MR. Fundamentals of modern fatigue analysis for design. In section 3:
[18] Schijve J. Stress gradient around notches. Fatigue Fract Eng Mater Struct fatigue strength prediction and analysis. ASM Handbook, vol. 19. Fatigue and
1980;3:325–32. Fracture, Material Park, OH: ASM International; 1996. p. 242.
[19] Bueckner HF. The propagation of cracks and the energy of elastic deformations. [32] Pilkey WD. Peterson’s stress concentration factors. 2nd ed. New York: John
J Appl Mech 1958;80:1225–30. Wiley and Sons; 1997.
[20] Erdogan F, Gupta GD, Cook TS. Numerical solution of singular integral [33] Taylor D. The theory of critical distances: a new perspective in fracture
equations. In: Sih GC, editor. Methods of analysis and solutions of crack mechanics. Elsevier; 2007.
problems. Leyden: Noordhoff; 1973. [34] El Haddad MH, Topper TH, Smith KN. Prediction of non propagating cracks. Eng
[21] Hildebrand FB. Introduction to numerical analysis. New York: Dover Fract Mech 1979;11:573–84.
Publications, Inc.; 1974. [35] DuQuesnay DL, Yu MT, Topper TH. The effect of notch radius on the fatigue
[22] Golberg MA. Projection methods for Cauchy singular integral equations with notch factor and the propagation of short cracks. In: Miller KJ, de los Rios ER,
constant coefficients on [1, 1]. In: Baker CTH, Miller GF, editors. Treatment of editors. The behaviour of short fatigue cracks. London: Mechanical
integral equations by numerical methods. Academic Press; 1982. Engineering Publications; 1986. p. 323–35.