International Journal of Fatigue 82 (2016) 521–529

Contents lists available at ScienceDirect

International Journal of Fatigue

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

Microstructure-sensitive small fatigue crack growth assessment: Effect

of strain ratio, multiaxial strain state, and geometric discontinuities
Gustavo M. Castelluccio a,⇑, David L. McDowell a,b
a
Woodruff School of Mechanical Engineering, Georgia Institute of Technology, 771 Ferst Drive, N.W., Atlanta, GA 30332, USA
b
School of Materials Science and Engineering, Georgia Institute of Technology, 771 Ferst Drive, N.W., Atlanta, GA 30332, USA

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

Article history: Fatigue crack initiation in the high cycle fatigue regime is strongly influenced by microstructural features.
Received 17 March 2015 Research efforts have usually focused on predicting fatigue resistance against crack incubation without
Received in revised form 4 September 2015 considering the early fatigue crack growth after encountering the first grain boundary. However, a signif-
Accepted 10 September 2015
icant fraction of the variability of the total fatigue life can be attributed to growth of small cracks as they
Available online 16 September 2015
encounter the first few grain boundaries, rather than crack formation within the first grain. This paper
builds on the framework previously developed by the authors to assess microstructure-sensitive small
Keywords:
fatigue crack formation and early growth under complex loading conditions. The scheme employs finite
Fatigue crack initiation
Microstructural small cracks
element simulations that explicitly render grains and crystallographic directions along with simulation of
Stress and strain ratios microstructurally small fatigue crack growth from grain to grain. The methodology employs a crystal
Geometric discontinuity plasticity algorithm in ABAQUS that was previously calibrated to study fatigue crack initiation in
Multiaxial loading RR1000 Ni-base superalloy. This work present simulations with non-zero applied mean strains and geo-
metric discontinuities that were not previously considered for calibration. Results exhibit trends similar
to those found in experiments for multiple metallic materials, conveying a consistent physical description
of fatigue damage phenomena.
Published by Elsevier Ltd.

1. Introduction loading conditions. Furthermore, we found [1] that a significant

Fatigue failure in the high cycle regime is dominated by the
formation and early growth of cracks that meander through the
microstructure. Multiple approaches have been proposed to con-
sider a crack initiation length, but such a definition is typically
valid for a specific material and loading conditions, the range of
lengths that can be detected, and the nature of the crack initiation
criterion employed. Nevertheless, it is believed that for engineering
metals, the crack formation process—nucleation and early
growth—concludes after cracking an area corresponding roughly
to several (i.e., two to ten) mean grain diameters; this typically
implies cracking tens to hundreds of individual grains. Once the
crack has reached this size, the crack front effectively samples
enough grains to smear out the microstructural variability, thereby
attaining similitude conditions necessary for application of long
crack linear or elasto-plastic fracture mechanics (LEFM or EPFM,
respectively). The number of grains required along the crack front
to reach this condition depends both on crack geometry and
⇑ Corresponding author now at Sandia National Laboratories, USA.
E-mail address: castellg@gatech.edu (G.M. Castelluccio).
fraction of the variability of total fatigue life can be attributed to
small cracks encountering and crossing the first few grain bound-
aries, rather than fatigue crack formation within the first grain.
Most modeling approaches have either focused on assessing the
nucleation process considering intrinsic microstructure variability
without fatigue crack growth; however, limited modeling efforts
have focused on small crack growth though the microstructure,
e.g., estimating the number of cycles required to grow a crack
through the first few grains. Exceptions can be found in the work
by Krupp and coworkers [2–4], primarily addressing two-
dimensional representations. More recently, we [1,5] developed a
three-dimensional modeling scheme to estimate the early fatigue
life for crystallographic (Stage I) fatigue crack growth across
multiple grains. This approach estimates the fatigue life by assess-
ing nonlocal Fatigue Indicator Parameters (FIPs) [6,7] that serve as
surrogate measures for small crack driving force as expressed by
the range of mixed mode cyclic crack tip displacement. Further-
more, cracks are propagated along crystallographic planes by
degrading the elastic stiffness that respects restoration of elastic
stiffness during roughness- or plasticity-induced crack closure.
The approach is implicitly valid for multiaxial loading conditions,

http://dx.doi.org/10.1016/j.ijfatigue.2015.09.007
0142-1123/Published by Elsevier Ltd.
522 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529
thereby avoiding the need for assignment of mesoscopic multiaxial
fatigue criteria at the scale of many grains.
One of the main deficiencies of fatigue models is the introduc-
tion of fitting coefficients that do not have physical interpretation.
These approaches pose a risk when exercising the model for
loading conditions and materials systems different from those used
for calibration. This work expands our previous efforts and applies
those methodologies to multiple multiaxial loading conditions and
geometries without changing any physical parameter. The objec-
tive of this paper is to assess whether the framework presented
and calibrated in previous publications is adequate for loading con-
ditions different from those used for calibration. In particular, we
propose to assess the role on early fatigue life of the applied strain
ratio (that induce different mean stresses) and geometrical discon-
tinuities (e.g., holes, notches).
Given the large number of parameters in crystal plasticity
approaches (which are not uniquely defined), there is value in
assessing capability to predict fatigue trends as opposed to refining
coefficients to match fatigue response of a single material. In other
words, we seek to assess general model capabilities to predict
experimentally observed trends, rather than pursuing detailed
correlation with a fatigue dataset for one material. We argue that
the agreement between simulations for a given set of model
parameters and well-established trends of fatigue behavior is use-
ful to assess the viability of a model to reflect the physics of forma-
tion and growth of microstructurally small cracks (MSCs) in
polycrystalline microstructures.
2. Experimental observations and general trends
Our objective is to evaluate the capability of a microstructure-
sensitive computational fatigue modeling framework to represent
well-established trends in fatigue response observed across multi-
ple material systems and loading conditions. Therefore, this section
summarizes a few general aspects of fatigue crack formation and
early growth (i.e., early stage crack initiation) that we aim to repro-
duce with simulations. Although multiaxial fatigue has long been
an area of active experimental investigation, only limited research
has addressed measuring the number of cycles for fatigue crack
nucleation and early growth within the first few grains; neverthe-
less, some trends are consistent and systematic. For example,
Suhartono and coworkers [8] performed tension–compression
and torsion (i.e., shear-dominated) tests on multiple alloys and
found that the former initiated a larger number of cracks. Akiniwa
et al. [9] assessed the very high cycle fatigue resistance (low
applied strains) of spring steel and found that torsion specimens
endured about two orders of magnitude longer lives than in ten-
sion–compression. Kandil et al. [10] evaluated the effects multiax-
ial loading on low cycle fatigue (high applied strains) of austenitic
stainless steel under equivalent strain ranges and found that fati-
gue lives for tension–compression tests were shorter by a factor
of five to ten than shear tests. In addition, Garud [11] analyzed
the effects of multiaxial loading and compared fatigue experiments
under multiple loading conditions in LCF. His approach employed a
criterion based on plastic work to correlate the finding that torsion
tests resulted in longer lives than the tension–compression tests by
a factor of three to five.
Multiaxial fatigue has been recently revisited in terms of the
assessment of crack nucleation and growth in notched specimens.
Sakane et al. [12] studied tension–compression and torsion fatigue
of 304 stainless steel notched bars to conclude that, under equiva-
lent nominal stress amplitude, torsion is less damaging than axial
loading; Gates and Fatemi [13] reached a similar conclusion for Al
2024-T3. Atzori et al. [14] analyzed notched specimens of C40 nor-
malized steel under nominal load ratios of R =
1 and R = 0, and
found that the lower R ratios had significantly longer fatigue lives
for similar stress amplitudes. Gladskyi and Fatemi [15] studied low
carbon steel and found that the notch effect was more pronounced
in tension–compression loading compared to torsion, in spite of
higher stress concentration factor for the latter. In addition, they
observed that notches have a more detrimental effect (compared
to smooth specimens) for HCF conditions (lower strain ampli-
tudes), which was also concluded by Gao et al. [16] for 16MnR
steel. This is expected, of course, from historical understanding of
notch effects. However, surprisingly few data exist in the literature
regarding coupled effects of multiaxial stress state and notches on
growth of microstructurally small fatigue cracks.
This brief summary of experimental results highlights some
general trends in fatigue crack initiation that are common across
multiple material systems under equivalent nominal applied load-
ing conditions:
(1) Shear loading and low R ratios are less damaging than axial
tension–compression loading for low and high equivalent
applied stresses or strains.
(2) The differences in number of cycles to failure for notches and
smooth specimens is most pronounced with decreasing
applied strain (i.e., higher strains reduce notch sensitivity).
The following sections will assess whether the recently devel-
oped microstructure-sensitive computational fatigue framework
[1,5] for formation and early 3D growth of microstructurally small
cracks can reproduce these trends.
3. Modeling and simulation
3.1. Crystal plasticity constitutive model
The elastic–plastic behavior at the grains scale depends on a
deterministic, physically-based crystal plasticity constitutive
model for RR1000 Ni-base superalloy, as outlined in detail else-
where [1,5,17]. In succinct terms, the crystallographic shearing
rate is given by
2 +p +q 3

* *
4 F0 jsðaÞ
BðaÞ j
SðaÞ l=l0 5sgnðsðaÞ
BðaÞ Þ;
_ ðaÞ
c ¼ c_ 0 exp
1

kb T s 0 l= l0
ð1Þ
where c_ ðaÞ is the shearing rate of slip system a; sðaÞ is the resolved
shear stress, T is the absolute temperature, F0, p, q, c_ 0 , s0, l, and
l0 are material parameters that may differ for octahedral and cube
slip systems, as listed in Table 1 for 650 °C, and kb is Boltzmann’s
constant. The formulation considers 12 octahedral and 6 cube slip
systems along. The model was implemented as a user-material sub-
routine (UMAT) in [18] using an implicit integration scheme based
on the Newton–Raphson with incremental line search and
backward-Euler methods. The slip resistance SðaÞ serves as a thresh-
old stress for plastic flow and the back stress BðaÞ accounts for direc-
tional hardening and depend on the shear rate as follows:
h i 
S_ ðaÞ ¼ hS
dD ðSðaÞ
S0 Þ c_ ðaÞ 
ðaÞ
ð2Þ
 
B_ ðaÞ ¼ hB c_ ðaÞ
r D BðaÞ c_ ðaÞ 
ðaÞ
ð3Þ
n l0 o
1
Here, r D ¼ hSBðlaÞ0
ðaÞ
0
fck

l and S0, hB, hS, dD, l00 , fc, k, are constants
that differ for octahedral and cube slip planes (see Table 1). The con-
stitutive model was calibrated using the cyclic stress–strain exper-
imental data from smooth specimens in Ref. [19]. Further
 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529 523

Table 1
Parameters of the constitutive model at 650 °C for octahedral and cube slip systems. The calibration considered cyclic stress–strain data of smooth specimens [19].

F0 (kJ/mol) p q c_ 0 (s
1) s0 (GPa) S0 (MPa) fc hB (GPa) hS (GPa) dD (MPa) l00 (GPa)

Oct. 295 0.31 1.8 120 810 350 0.42 400 10 6024 72.3
Cube 295 0.99 1.6 4 630 48 0.18 100 4.5 24 28.6

Other parameters: k = 0.85, l0 = 192 GPa. Elastic constants: C11 = 166.2 GPa, C12 = 66.3 GPa, C44 = 138.2 GPa.

information about the calibration and implementation of the model regimes in which the microstructure exerts a reduced influence,
can be found in Refs. [1,5,17]. such as physically small fatigue cracks and ultimately long fatigue
Following prior work [1,5,17], the early fatigue damage is quan- cracks.
tified using a Fatigue Indicator Parameter (FIP) based on a modified In Eq. (5), N Nuc operationally corresponds to the number of
Fatemi–Socie parameter [20], in which the FIP is computed for cycles required to fully crack the first grain, and it is assumed to
each octahedral slip system a, i.e., follow the semi-empirical empirical law [22–24].

 ag
ðFIPa Þ ;

2
Dcap ra Nnuc ¼ ð6Þ
FIPa ¼ 1þk n ð4Þ dgr
2 ry
where FIPa is the band average of the quantify in Eq. (4), ag is an
where Dcap is the cyclic plastic shear strain range at each slip system, irreversibility coefficient and dgr is the equivalent size considered
ran is the peak stress normal to the slip system, ry is the cyclic yield for extending the crack, given by
strength of the polycrystal, and k = 0.5, as proposed by Fatemi and X
n
Socie [20]. To regularize the FEM discretization and represent the dgr ¼ Dst þ xi Dind ð7Þ
nonlocal influence of the fatigue damage process zone, the FIPsa i

computed on each integration point are averaged along bands par- which is the sum of the length of the band in the ith grain (Dst ) and
allel to slip planes across an entire grain, as depicted in Fig. 1. A
length(s) of bands (Dind ) in neighboring grains having low disorien-
detailed analysis of different averaging volumes has been presented
tation ith grain, the disorientation factor (xk ) is computed as
elsewhere [21], which concluded that band averaging effectively
* +
mitigates FEM mesh dependence while retaining most of the intrin- hkdis
sic variability due to microstructure randomness. x ¼ k
1
: ð8Þ
20

3.2. Life prediction using a multistage fatigue model where hkdis is the disorientation between the crystallographic orien-
tations of the bands, and the Macaulay brackets are defined accord-
Following Castelluccio and McDowell [1], the number of cycles ing to hai ¼ a if a > 0, hai ¼ 0 if a 6 0. For further details please refer
required to grow small cracks can be partitioned into nucleation to Ref. [1]. The irreversibility coefficient ag ¼ 41:6 cycles lm
and MSC growth stages, i.e., depends on the loading conditions, including the environment
and strain rate, and it was calibrated previously [5] using the exper-
NSC
f ¼ N Nuc þ N MSC ð5Þ imental data from replica crack growth measurements on U-notch
specimens in Ref. [25].
This formulation differentiates between crack formation The MSC crack growth rate is assumed to be controlled by the
(within one grain) and growth mechanisms (across multiple mechanical irreversibility of dislocations emitted from the crack
grains). Additional terms may be incorporated to represent tip and to be proportional to the DCTD, i.e.,
a * +
da  dgr a
¼ / A FIP
D CTDth ; ð9Þ
dNmsc d
ref
gr

Here, A = 2 lm is a constant scales the FIP with the DCTD [26,27],

dgr ¼ 8 lm is the mean grain size from the data used for calibration,
ref

DCTDth ¼ 5  10
4 lm is a threshold that has a value close to the

Fig. 1. Schematic representation of the bands in which FIPs are averaged to
estimate transgranular fatigue crack growth. The figure represents a single
spherical grain as modeled by FEM with unstructured, voxellated meshing.
Different bands are color coded and numbered. (For interpretation of the references
to colour in this figure legend, the reader is referred to the web version of this
article.)
magnitude of the Burgers vector. The irreversibility factor / mea-
sures the mechanical irreversibility in the process zone and
depends on environment and temperature in addition to material
slip processes. The magnitude of / usually varies from 0.01 to 0.2,
we estimated [5] a value of 0.077 based on early crack growth
experimental data [25].
The number of cycles to extend the crack front through the ith
grain along slip system a in the MSC regime, N Gi jaMSC , is determined
by integrating Eq. (9) with respect to the crack length. The total life
consumed in the MSC growth regime (N MSC ) to reach a certain
crack length is the sum of the lives for each grain that has been
involved in the growth process for a given 3D crack, i.e.,
X a
NMSC ¼ Ni jMSC ð10Þ
i
where
Ni jaMSC ¼ NGi jaMSC
N History
ð11Þ
524 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529

rffiffiffiffiffi
1 c2
NGi jaMSC

1
¼ pffiffiffiffiffiffiffiffiffi tan h Dst
c1 c 2 c1
Pn
Dst þ xi Dind
c1 ¼ / i
ref
2FIPa
/DCTDth
dgr
1 2FIPa0
c2 ¼ / Pn i i
i x Dnd Þdgr
2 ðDst þ ref
Here, N History corresponds to the number of loading cycles applied
since the band considered has intersected the crack perimeter.
These equations represent the number of cycles required for a crack
to meander through the microstructure with an oscillatory crack
growth rate, which is typical or the MSC regime. For a detailed
explanation of the algorithms please refer to Ref. [1], while further
details about the calibration can be found in Ref. [5].
3.3. Cases studied
This paper assesses microstructurally small fatigue crack forma-
tion and early growth for three strain ratios (Re = emin =emax = 0.5, 0,
and
1), under shear or tensile loading with and without geomet-
ric discontinuities. These strain ratios manifest different crack tip
conditions (e.g., degree of roughness- and/or plasticity-induced
crack closure) and sample the effects of non-zero mean strains.
Fig. 2 provides examples of FE meshes employed for modeling
smooth and through-hole specimens, in which each color repre-
sents a grain with a different crystallographic orientation. These
models are not large enough to be regarded as representative vol-
ume elements (RVE), but they should be considered as statistical
volume elements (SVE); accordingly, we evaluate multiple equiva-
lent microstructural realizations.
In order to reduce the computational effort, the region far from
the hole is assigned an artificially high slip resistance ðSðaÞ Þ to inhi-
bit plastic deformation. This simplification does not significantly
influence the results since only heterogeneous microplasticity
develops far from the hole, without any appreciable effect on the
material stiffness, and with much lower magnitude than the
intense values of local plasticity in grains near the notch root.
We note that somewhat different boundary conditions might be
required to represent specific experimental conditions, but we
expect that fatigue crack growth trends would be qualitatively
similar, including the variability of response [21].
To roughly assess the variability due to the microstructure, each
loading case considers ten equivalent models with random assign-
ment of grain size and orientation according to a target distribu-
tion (no texture). The loading sequence consisted of imposition
of quasistatic relative displacement of the upper and lower bound-
ary planes under shear or tensile mode loading to achieve an over-
all nominal strain range between 0.4% and 0.8%, typical of the HCF
regime (below initial yield); lateral faces are free of traction.
Fig. 2. Example of voxellated (structured) meshes representing the polycrystalline microstructure for axial loading of (left diagrams) smooth and (right diagrams) through-
hole specimens. The latter specimen shows two regions, one in which only elastic deformation is allowed, and one in which plastic deformation can occur, near the notch root.
Simulations with different applied strain ratios employed the same
ð12Þ
applied strain range to facilitate comparison regarding the effect of
applied strain ratio under strain/displacement controlled
conditions.
ð13Þ To assign a notion of equivalence with uniaxial loading, the
magnitude of the boundary displacement in shear loading was
computed by assuming an elastic model with cubic symmetry,
from which the value of Poisson’s ratio was deduced to compute
ð14Þ
the equivalent pure shear strain for Re = 0, as depicted by Fig. 3.
The e = 0.8% uniaxial strain is equivalent in shear to
ceq ¼ ð1 þ mÞe ¼ ð1 þ mÞ0:8% ¼ ð1 þ 0:399Þ0:8% ¼ 1:119% ð15Þ
Thus, the upper and lower faces were displaced in shear up to a
nominal shear strain of 1.119%. It is understood that Eq. (15) is
employed for cyclic shear simulations to produce the same uniaxial
equivalent nominal strain range De = Dc=ð1 þ mÞ as the uniaxial
tension–compression case, De = De. The Poisson’s ratio was
deduced using an elastic model with cubic symmetry [28], i.e.,
m ¼ C 12 =C 11 ¼ 66:3=166:2 ¼ 0:399:
4. Results
4.1. Smooth specimens
Cyclic shear and tension–compression straining was applied to
smooth specimens for three strain ratios, (Re =
1, 0 and 0.5) with
an equivalent nominal strain range of De = 0.8%, at 650 °C and
quasistatic strain control at constant strain rate of 5  10
4 s
1.
Following previous work [5], the simulations consisted of three
applied cycles to assess the grain that nucleates the first crack,
while two cycles are applied thereafter for extension of the crack
front through each successive grain to assess stress redistribution.
For each loading condition, a total of ten equivalent microstructure
realizations were employed. Unidirectional periodic boundary con-
ditions were considered, such that the net sum of the displace-
ments of the nodes on top and bottom faces is null, with traction
free lateral faces.
Fig. 4 reproduces the simulated crack length vs. cycles results
on a logarithmic scale. Horizontal broken lines represent crack
arrest events (fatigue lives exceeding 108). The predicted number
of cycles was computed using Eq. (6) for crack formation in the first
grain and Eq. (11) iteratively for growth in subsequent grains.
These equations are employed to predict the number of cycles to
damage each grain, and the crack growth life across consecutive
adjoined bands with the lowest life estimation. By summing these
cycle counts from successively cracked grains and tracking the
total square root of the cross sectional areas of the grains to mea-
sure crack length, we compute the data shown in Fig. 4. Further
details regarding the model can be found in Ref. [1].
 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529 525

range in Fig. 4, fatigue lives increased by a factor of ten. Values

for Re =
1 are not included since the model predicted fatigue lives
beyond 108 (crack arrest). These results suggest that reducing the
applied strain range induces overlap of tension–compression and
shear results, which was also observed for Re =
1 in Fig. 4.

4.2. Through hole specimens

Geometric discontinuities induce gradients that superimpose

on those due to microstructural variation. To investigate how these
Fig. 3. Mohr’s circles for stress and strain employed to estimate the equivalent
effects are coupled, we assess the fatigue behavior of a specimen
shear strain under uniaxial loading. with a through hole.

4.2.1. Elastic stress concentration induced by a hole

For a given constant nominal applied uniaxial equivalent strain Fig. 7 presents the concentration of the stress field induced by a
range of 0.8% and several strain ratios, these results reveal that: hole of radius 72 lm in the mesh corresponding to a mean grain
size of 18 lm; these results were computed by assuming only an
 Uniaxial tension–compression loading is generally more detri-
elastic behavior (very high slip resistance, SðaÞ ) that incorporates
mental than applied shear loading. For the case Re =
1, fatigue
the elastic anisotropy of grains. Fig. 7(a) presents an example of
crack nucleation (i.e., time require to crack the first grain) under
the maximum principal stress for the tension–compression case.
tension–compression and shear loading is nearly identical,
The average stress concentration from ten realizations with ran-
although some cracks arrested.
dom distributions of grains was kt = 3.75. For each realization,
 For both loading conditions, the variability seems to increase
the stress concentration was defined as the ratio of the maximum
with decreasing strain ratio. This correlates with the fact that
principal stress on any element to the average tensile stress
lower strain ratios produce longer predicted fatigue lives and
applied on the boundary faces of the specimen; all the results coin-
smaller estimated plastic strain ranges.
cided within 10% and were located in the proximity of the hole.
 Tension–compression fatigue life results seem to exhibit similar
When compared to standard solutions for stress concentration
variability as shear loading. We emphasize that, even when fur-
[29], the present calculations result in stress concentration higher
ther simulations are required for a statistically significant
than that of a hole in an isotropic elastic semi-infinite medium
assessment, the models are able to describe crack growth trends
(kt = 3.47) with similar geometric dimensions and certainly higher
under complex loading conditions.
than the solution for a hole in an infinite isotropic elastic medium
(kt = 3.0). This is expected, of course, since local grain orientation
The location of the grains that nucleated cracks for Re = 0,
and crystallographic elastic anisotropy are included in the present
Re = 0.5 and Re =
1 for simulations presented in Fig. 4 are shown
simulations.
in Fig. 5. We note that the minimum distance between any band
A similar analysis was performed for simulations strained under
and the grain center is typically within one and two elements.
cyclic Shear; Fig. 7(b) presents an example of the field for the max-
Overall, crack nucleation appears to be spatially homogeneous
imum difference between principal stresses (i.e., Tresca stress) for
within the model, which highlights that the boundary conditions
the shear case. The elastic shear stress concentration was com-
are not imposing macroscopic gradients. Note that the same 10
puted as the ratio of the maximum shear stress on any element
microstructure realizations are employed for each strain ratio; in
in the periphery of the hole to the average shear stress (associated
some cases cracks nucleated in the same grain, but not always.
with the applied shear displacement) on the upper and bottom face
of the mesh. The shear stress concentration factor is kts = 4.01.
When compared to standard solutions [29], the calculations are
4.1.1. Cyclic straining at De = 0.4%
rather close to the solution for the shear of a hole in an infinite iso-
To better understand effects of strain amplitude, a set of simu-
tropic elastic medium (kts = 4.0) Fig. 7(a) and (b) confirm that ten-
lations were conducted similar to those presented in the previous
sion–compression and shear loading cases result in significant
section, but with a uniaxial equivalent nominal strain range of
stress concentrations close to the hole.
De = 0.4%; the results are presented in Fig. 6 for tension–compres-
sion and shear straining. Compared to the results at larger strain
4.2.2. Through-hole specimen results for cyclic tension–compression
and shear loading
70 ♦ Tension−Compression
R =0.5
ε
R =0
ε
R =−1
ε Fig. 8 presents the simulated crack length vs. cycles for different
• Shear strain ratios for through-hole specimens subjected to a uniaxial
60
equivalent nominal strain range of De = 0.8% (strain amplitudes
Crack length [μm]

50
of 0.8% in tension–compression and 1.12% in shear). Each data-
40
point corresponds to the extension of the crack by one grain; sim-
30 ulations are carried out to fatigue crack lengths on the order of
20 60 lm and corresponding lives below 108 cycles.
10 When compared to smooth specimens (Fig. 4), the results show
0 a reduction in fatigue life by about an order of magnitude, with the
4 5 6 7
10 10 10 10 same observations that lower strain ratios are less detrimental and
Cycles that tension–compression straining results in lower lives than
shear loading. However, the difference in the number of cycles
Fig. 4. Comparison of the results from smooth specimens for strain ratios Re = 0,
Re = 0.5 and Re =
1 using simulations with periodic displacement boundary
between tension–compression and shear loading for an equivalent
conditions imposed on upper and lower boundaries. An equivalent nominal strain crack length increases somewhat with decreasing strain ratio; this
range of De = 0.8% is employed in all cases. trend is not observed for smooth specimens.
526 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529

Tension−Compression
X X X Shear X

• Rε=−1 • Rε=−1

• Rε=0 • Rε=0

• Rε=0.5 • Rε=0.5

Y Z Y Z

Fig. 5. Location of the grains which nucleated fatigue cracks for Re = 0, Re = 0.5 and Re =
1 for 10 realizations with smooth specimens subjected to a uniaxial equivalent
nominal strain range of De = 0.8%. The same 10 microstructure distributions are employed for each strain ratio.

R =0.5
ε
R =0
ε
nucleation are similar to those found in experiments (e.g., see
70 ♦ Tension−Compression
• Shear Ref. [15]); however no cracks nucleated at
45° for shear loading,
60
which suggests that the crack nucleation site is sensitive to the
Crack length [μm]

50 direction of the first loading cycle.

40
30
20
10
0 4
5 6
10 10 10
Cycles
Fig. 6. Results for strain ratios Re = 0 and Re = 0.5 using smooth specimens subjected
to uniaxial equivalent nominal strain range of De = 0.4%. Values for Re =
1 are not
included since the model predicted fatigue lives beyond 108 cycles.
Although we do not aim for a rigorous analysis, we note that the
variability of the number of cycles to nucleate or grow a crack is
larger in through-hole specimens than in smooth specimens. We
attribute this finding to the different volumes of highly strained
material and FIP gradient field in notched specimens. Therefore,
we emphasize the importance of considering the volume of highly
strained material around geometric discontinuities for the statisti-
cal assessment of fatigue crack formation and early growth.
Fig. 9 shows the location of the grains that nucleated fatigue
cracks for through-hole specimens. Except for one simulation
under shear, cracks nucleated consistently next to the through-
hole at 90° and 45° from the straining axis for tension–compres-
sion and shear loading, respectively. These locations for crack
4.3. Mean stress effects on smooth specimens
The mean stress is known to have a detrimental effect on S–N
curves (i.e., complete specimen failure), and multiple empirical
7
10 models have been proposed to extrapolate experimental results
for different mean stresses. These relationships are much less stud-
ied and understood for MSCs. In our study, the applied nominal
mean stress is not employed to estimate fatigue lives, but rather
the response is the outcome of the applied loading acting in con-
cert with the constitutive model. This section investigates the cor-
relation between the predicted mean stress and fatigue behavior,
which is then compared to empirical models for mean stress
effects.
4.3.1. Mean stress evolution
To quantify mean stresses we analyzed stress–strain curves for
multiple applied strain ratios in smooth specimens. Fig. 10
presents the nominal stress–strain evolution for ten realizations
of smooth specimens subjected to tension–compression loading
after 50 cycles: Re = 0, Re = 0.5 and Re =
1, at two nominal strain
ranges, De = 0.8% and 0.4%. The slow evolution of the mean stress
suggests that fatigue life estimations are performed at a relatively
constant nominal mean stress level. Further cycling could poten-
tially further relax the mean stress, but given the slow rate of

(a) (b)

Fig. 7. Example of the stress fields resulting from elastic models simulations: (a) maximum principal stress for tension–compression loading, and (b) maximum shear stress
(Tresca).
 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529
Through hole Specimen − Δε = 0.8%
70
• Shear Rε=0.5 Rε=0
60 ♦ Tension/Compression
Crack length [μm]
50
40
30
20
10
0
2 3 4 5
10 10 10 10
Cycles
Fig. 8. Comparison of crack length vs. cycles for strain ratios Re = 0, Re = 0.5 and
Re =
1 using through-hole specimens subjected to a uniaxial equivalent nominal
strain range of De = 0.8%.
relaxation it is useful to compare predicted effects of mean stress
on fatigue life with existing phenomenological mean stress
correlations.
4.3.2. Mean stress correlation with fatigue life nucleation
The detrimental effects of high mean stresses have been recog-
nized by multiple correlative models (e.g., Goodman and Morrow
models for fatigue crack initiation). However, these models are
empirical in nature and are not typically aimed at the regime of
microstructurally small cracks. An exception is the model by Wang
and Miller [30], who studied the effects of the mean stress on fati-
gue crack nucleation and early growth on 1.99% NiCrMo steels and
proposed a simple dependence of the early fatigue life on positive
mean stresses (rm > 0). Their model can be stated as
N0 ¼ A expð
Brm ÞrCa
in which N0 corresponds to the period associated with the develop-
ment of cracks up to surface lengths between two to ten microns
(corresponding to initial crack nucleation).
Fig. 11 presents the number of cycles to nucleate a crack within
the first grain for tension–compression loading (compiled from
Figs. 4 and 6) versus the stabilized (50 cycles) nominal applied
mean stress (Fig. 10) on a semi-log scale from the present simula-
tions; the data correspond to two nominal strain ranges and three
strain ratios. The broken line indicates an exponential dependence
of the minimum nucleation lives with the mean stress values, as
evident in Eq. (16). If we consider the minimum lives among all
the realizations, the mean stress results are in agreement with
the exponential dependence proposed by Wang and Miller. This
is significant as such dependence differs from those of more
Fig. 9. Location of the grains which nucleated fatigue cracks for Re = 0, Re = 0.5 and Re =
1, out of 10 realizations of through-hole specimens subjected to a uniaxial equivalent
nominal strain range of De = 0.8%. The same 10 microstructure distributions are employed for each strain ratio and in some cases cracks nucleated in the same grains.
527
traditional, widely used mean stress models for crack initiation
Rε=−1 such as Morrow [31] or Smith–Watson–Topper (SWT) [31]. We
emphasize that neither the constitutive or fatigue models were
prescribed to follow the Wang and Miller model, so reproducing
these results supports the predictive capability of the model for
other conditions than those used for calibration.
5. Discussion
6
10
Consideration of experimental data in the literature for multiple
materials systems has shown that shear loading is less detrimental
than tension–compression loading under approximately equiva-
lent amplitudes and R ratios. Our simulations of smooth and
through-hole specimens showed that cyclic tension–compression
is indeed more damaging than cyclic shear loading for the same
amount of equivalent applied strain amplitude. The differences in
the number of cycles predicted for equivalent crack lengths are
approximately a factor of 2–3 for smooth specimens and up to
an order of magnitude for through-hole specimens, depending on
the loading conditions. In addition, we showed that lower Re leads
to lower damage and longer fatigue lives, which has also been
found experimentally, as described earlier.
Regarding the effect of geometric discontinuities, Fig. 8 shows
that fatigue lives are longer for cyclic shear than for equivalent ten-
sion–compressions loading, even though shear has a higher stress
concentration factor. Furthermore, by comparing Figs. 4 and 8 we
observe that the reduction in fatigue lives is more significant for
tension–compression loading, suggesting that the notch effect is
less pronounced under shear loading. Both findings are consistent
with experiments described by Gladskyi and Fatemi [15]. Similarly,
ð16Þ our simulations provide mean stress dependence that agrees with
the dependence proposed by Wang and Miller [30] based on direct
experimental observations for crack nucleation on 1.99% NiCrMo
steels.
The consistency of these simulation results with experimental
findings for a rather broad range of polycrystalline alloys is inter-
esting since none of the features of the modeling framework were
in any way prescribed to recover these relations, nor were model
parameters tuned to reflect them. Accurate prediction of fatigue
lives, crack locations, and crack growth paths for specific
microstructures and components would most likely require more
detailed consideration of local material properties, environment,
residual stresses, etc. However, our interest here is in the physi-
cally consistent prediction of trends and functional dependencies
on multiaxial stress state and mean stress as a function of applied
strain amplitude to support comparison of fatigue performance of
528 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529
Fig. 10. Nominal stress–strain responses of 10 smooth specimen microstructure realizations over 50 tension–compression straining cycles. The figures on top row correspond
to a nominal strain range of 0.8%, while the figures on the bottom row correspond to a nominal strain range of 0.4%.
Fig. 11. Number of cycles to nucleate a crack for tension–compression loading
(compiled from Figs. 4 and 6) as a function of the stabilized (50 cycles) mean stress
on a semi-log scale. The line indicates an exponential dependence of the minimum
nucleation lives with the mean values and agrees with the Wang and Miller model
[30].
different microstructure variants that may exist in a given material
system due to thermo-mechanical processing.
We recognize that the small number of cycles employed in sim-
ulations prior to crack nucleation does not seem to be enough to
achieve a saturated condition and may affect the location of crack
embryos (note that saturation can be achieved much more rapidly
than in experiments with an appropriate calibration strategy).
Nevertheless, rough estimations of fatigue resistance and variabil-
ity for a wide range of loading conditions can be obtained with
limited detailed information of a material system. These estimates
are more likely to be pertinent when considering an ensemble of
microstructures rather than one particular microstructure realiza-
tion/arrangement. Therefore, such methodologies seem well-
suited for the design of microstructures tailored towards fatigue
resistant applications. In fact, one might envision their utility in
constructing microstructure-sensitive multiaxial fatigue criteria
that correspond to different definitions of ‘‘crack initiation,” each
having a different crack length that can be detected, for example.
To do this based on laboratory experiments in the HCF and transi-
tion fatigue regimes is indeed prohibitive, which likely explains
how slow progress has been on material dependent multiaxial fati-
gue criteria and design of microstructure for multiaxial fatigue
resistance.
6. Conclusions
This paper has exercised a computational microstructure-
sensitive framework for a Ni-base superalloy subjected to cyclic
tension–compression and shear conditions for smooth and through
hole specimens. The three-dimensional small fatigue crack growth
algorithms showed their capability to quantify the most damaging
loading conditions in terms of strain ratios, multiaxial strain state,
and mean stress. Although experimental data are lacking for full
validation, we have shown that our results agree with the general
trends observed from experiments reported in the literature.
To assess the effect of stress concentration, this research intro-
duced models with holes under multiaxial loading. Compared to
smooth specimens, the fatigue lives of these models are affected
in multiple ways and statistical differences arise due to the
differences in highly strained volumes and gradients. This kind of
framework therefore has promise to be employed to explore
geometry-induced variability as well as microstructure effects.
 G.M. Castelluccio, D.L. McDowell / International Journal of Fatigue 82 (2016) 521–529
Acknowledgements
G.M. Castelluccio and D.L. McDowell are grateful for the support
provided by Integrated Systems Solutions, Inc. (Technical Monitor:
Dr. Nam Phan, NAVAIR).
References
[1] Castelluccio GM, McDowell DL. A mesoscale approach for growth of 3D
microstructurally small fatigue cracks in polycrystals. Int J Damage Mech
2014;23(6):791–818.
[2] Duber O, Kunkler B, Krupp U, Christ HJ, Fritzen CP. Experimental
characterization and two-dimensional simulation of short-crack propagation
in an austenitic-ferritic duplex steel. Int J Fatigue 2006;28:983–92.
[3] Krupp U, Düber O, Christ HJ, Künkler B, Köster P, Fritzen CP. Propagation
mechanisms of microstructurally short cracks – factors governing the
transition from short- to long-crack behavior. Mater Sci Eng A
2007;462:174–7.
[4] Kunkler B, Duber O, Koster P, Krupp U, Fritzen CP, Christ HJ. Modelling of short
crack propagation – transition from stage I to stage II. Eng Fract Mech
2008;75:715–25.
[5] Castelluccio GM, McDowell DL. Mesoscale modeling of microstructurally small
fatigue cracks in metallic polycrystals. Mater Sci Eng A 2014;598:34–55.
[6] Castelluccio GM, Musinski WD, McDowell DL. Recent developments in
assessing microstructure-sensitive early stage fatigue of polycrystals. Curr
Opin Solid State Mater Sci 2014;18:180–7.
[7] Socie DF, Marquis GB. Multiaxial fatigue. SAE International; 1999.
[8] Suhartono HA, Pötter K, Schram A, Zenner H. Modeling of short crack growth
under biaxial fatigue: comparison between simulation and experiment. ASTM
2000:323–39.
[9] Akiniwa Y, Stanzl-Tschegg S, Mayer H, Wakita M, Tanaka K. Fatigue strength of
spring steel under axial and torsional loading in the very high cycle regime. Int
J Fatigue 2008;30:2057–63.
[10] Kandil FA, Brown MW, Miller KJ. Biaxial low cycle fatigue failure of 316
stainless steel at elevated temperatures. In: Met. Soc. Book 280. Varese,
Italy: Metal Society of London; 1982. p. 203–10.
[11] Garud YS. A new approach to the evaluation of fatigue under multiaxial
loadings. J Eng Mater Technol 1981;103:118–25.
[12] Sakane M, Zhang S, Kim T. Notch effect on multiaxial low cycle fatigue. Int J
Fatigue 2011;33:959–68.
529
[13] Gates N, Fatemi A. Notched fatigue behavior and stress analysis under
multiaxial states of stress. Int J Fatigue 2014;67:2–14.
[14] Atzori B, Berto F, Lazzarin P, Quaresimin M. Multi-axial fatigue behaviour of a
severely notched carbon steel. Int J Fatigue 2006;28:485–93.
[15] Gladskyi M, Fatemi A. Notched fatigue behavior including load sequence
effects under axial and torsional loadings. Int J Fatigue 2013;55:43–53.
[16] Gao Z, Qiu B, Wang X, Jiang Y. An investigation of fatigue of a notched member.
Int J Fatigue 2010;32:1960–9.
[17] Castelluccio GM, McDowell D. Effect of annealing twins on crack initiation
under high cycle fatigue conditions. J Mater Sci 2013;48:2376–87.
[18] ABAQUS. FEM software V6.9, Simulia Corp., Providence, RI, USA. Providence, RI,
USA: Simulia, Inc.; 2009.
[19] Zhan ZL, Tong J. A study of cyclic plasticity and viscoplasticity in a new nickel-
based superalloy using unified constitutive equations. Part I: Evaluation and
determination of material parameters. Mech Mater 2007;39:64–72.
[20] Fatemi A, Socie DF. A critical plane approach to multiaxial fatigue damage
including out-of-phase loading. Fatigue Fract Eng Mater Struct 1988;11:
149–65.
[21] Castelluccio GM, McDowell DL. Microstructure and mesh sensitivities of
mesoscale surrogate driving force measures for transgranular fatigue cracks in
polycrystals. Mater Sci Eng A 2015;639:626–39.
[22] Hunsche A, Neumann P. Quantitative measurement of persistent slip band
profiles and crack initiation. Acta Metall 1986;34:207–17.
[23] Shenoy MM, Zhang J, McDowell DL. Estimating fatigue sensitivity to
polycrystalline Ni-base superalloy microstructures using a computational
approach. Fatigue Fract Eng Mater Struct 2007;30:889–904.
[24] Tanaka K, Mura T. A dislocation model for fatigue crack initiation. J Appl Mech
1981;48:97–103.
[25] Pang HT. Effect of microstructure variation on turbine disc fatigue
live. University of Southampton; 2003.
[26] Castelluccio GM, McDowell D. Assessment of small fatigue crack growth
driving forces in single crystals with and without slip bands. Int J Fract
2012;176:49–64.
[27] Przybyla C, Musinski W, Castelluccio GM, McDowell DL. Microstructure-
sensitive HCF and VHCF simulations. Int J Fatigue 2013;57:9–27.
[28] Dowling NE. Mechanical behavior of materials: engineering methods for
deformation, fracture, and fatigue. Prentice Hall; 1998.
[29] Pilkey WD. Peterson’s stress concentration factors. 2nd ed. Wiley-Interscience;
1997.
[30] Wang CH, Miller KJ. Short fatigue crack growth under mean stress, uniaxial
loading. Fatigue Fract Eng Mater Struct 1993;16:181–98.
[31] Bannantine JA, Comer JJ, Handrock JL. Fundamentals of metal fatigue
analysis. Englewood Cliffs, N.J.: Prentice Hall; 1989.