Materials Science and Engineering A284 (2000) 268 – 279

www.elsevier.com/locate/msea

The effects of residual macrostresses and microstresses on fatigue

crack initiation
J.D. Almer, J.B. Cohen 1, B. Moran *
Department of Materials Science and Engineering, Robert R. McCormick School of Engineering and Applied Science, Northwestern Uni6ersity,
E6anston, IL 60208, USA

Received 13 July 1999; received in revised form 18 November 1999

Abstract

The combined techniques of X-ray diffraction and Finite Element (FE) analysis have been used to analyze and predict fatigue
crack initiation in the presence of residual stresses. These stresses were introduced into polished double-edge notched 1080 steel
specimens by pre-straining and press-fitting operations, and crack initiation was monitored during high-cycle fatigue tests using
surface replicas. The local stress–strain behavior in the vicinity of the initiation sites was tracked using FE analysis. Microbeam
X-ray diffraction measurements were also made in each phase of the steel to determine both residual macrostresses, which
compared well with the FE computations, and microstresses that were employed along with mechanical tests to determine input
parameters for the FE analysis. Microstresses were observed to fade rapidly during fatigue, while macrostresses relaxed less
rapidly and were observed to strongly affect crack initiation behavior. The fatigue behavior was correlated with the observed
stresses via a stress–life relationship that incorporates mean stress effects. © 2000 Elsevier Science S.A. All rights reserved.

Keywords: Residual stresses; Fatigue; Cracks; Crack initiation

1. Introduction Predictions of the fatigue initiation life are contingent

on obtaining accurate values of the local stress–strain
The most common site for fatigue crack initiation around stress concentrators, which can then be input
(FCI) in structural components is at stress concentra- into fatigue–life equations developed for smooth speci-
tors such as notches or fillets. In order to increase mens. Often, this local information is obtained using
fatigue resistance, compressive residual stresses are of- engineering approximations such as Neuber analysis [7],
ten purposely introduced around these concentrators or simplified expressions based on the notch radius and
through processes such as shot-peening [1], press-fitting stress intensity factor [8], which have been found to
[1 –3], and tensile pre-straining [1,4,5]. However, tensile provide reasonable predictions of initiation behavior,
residual stresses may be inadvertently introduced, for especially around sharp notches [9–13]. An alternative
example by compressive pre-straining, and these tensile method for obtaining these local stress–strain values is
stresses may be deleterious to fatigue performance [6]. Finite Element (FE) analysis, coupled with constitutive
Given the potential importance of residual stresses, laws describing the material’s elastic–plastic behavior.
proper characterization of their values and effects on This technique is attractive because more complex load-
fatigue crack initiation are vital. ing sequences and material geometries can be examined
than with other analyses. In one of the few studies
* Corresponding author. Present address: Department of Civil utilizing FE for this purpose [4], good agreement was
Engineering, Northwestern University, 2145 Sheridan Road, found between calculated strains, determined using a
Evanston, IL 60208, USA. Tel.: +1-847-4918793; fax: +1-847- kinematic work-hardening relationship, and those mea-
4914011. sured by a strain gage at a notch root in both low
E-mail address: [email protected] (B. Moran)
1
Deceased. Former Dean, Robert R. McCormick School of Engi-
carbon and stainless steels. They also found the Neuber
neering and Applied Science, Northwestern University, Evanston, IL, approach to overpredict notch root strains relative to
USA. the FE analysis. One goal of this work is to directly

0921-5093/00/$ - see front matter © 2000 Elsevier Science S.A. All rights reserved.
PII: S 0 9 2 1 - 5 0 9 3 ( 9 9 ) 0 0 7 7 9 - 0
 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279 269

compare measured stresses with computed values from 2. Experimental details

FE analysis.
In multiphase materials, residual stresses may be one 2.1. Material and specimens
of two types: macrostresses, which form when different
macroscopic regions of a material deform non-uni- As-received 1080 steel sheet (6.35 mm thickness) was
formly but are constrained to fit together (such as the heat treated in air at 1073 K for 30 min, then quenched
surface and interior); or microstresses, which form due and tempered at 973 K for 180 min to produce a
to non-uniform deformation between phases or within fine-grained, spheroidized microstructure. The average
grains. Diffraction provides a unique probe to examine ferrite grain size and cementite particle size were found
both types of stress, since the stresses can be measured to be 5(1) and 0.7(1) mm, respectively [18]. These sizes
in each phase and separated into macrostresses and were found to be small enough for proper diffraction
microstresses using equilibrium equations [14]. Here, averaging with the small X-ray beam used herein. The
the stresses were measured in both the ferrite and cementite volume fraction was determined from X-ray
cementite phases of a 1080 steel, in order to gain insight powder measurements [19] as described in a previous
into the relative contributions of macrostresses and paper [17], and found to be 0.142(12). (Numbers in
microstresses on FCI. parentheses represent the standard deviation, e.g.
In this investigation, the residual macrostresses and 0.1429 0.012.)
microstresses formed around a notch due to pre-strain- The geometry for the initiation specimens (Fig. 1)
ing and press-fitting operations were determined using was chosen to simulate fatigue loading in the presence
X-ray diffraction. The measured microstresses were of notch stress concentrators. Fatigue cracks in these
then used, in conjunction with mechanical behavior, as double-edge notch (DEN) specimens have previously
input into the Generalized Plasticity model [15,16] used been observed to initiate in the center surface of the
in the FE analysis to provide macrostress information. notches in 304 stainless steel [20] and 2024 Al alloy [21],
The stress responses at the initiation sites were moni- allowing for optical detection and X-ray stress analysis
tored during fatigue, and their stabilized values were to be performed around the initiation regions.
used as input to a stress – life equation to account for The FCI behavior is expected to be influenced by
residual stress effects on initiation and total life. (In a surface roughness in addition to residual stresses. The
previous paper, we have also examined the effect of specimens were therefore processed to minimize differ-
these stresses on crack growth [17].) ences in roughness between specimens, and provide
controlled residual stresses, as follows. First, holes were
drilled and then reamed (to 6.08 or 6.35 mm diameters)
at 25.4 mm spacing intervals. The insides of these holes
were ground using 320 then 600 grit stones to smooth
the surfaces, and then polished with 6 mm, then 1 mm,
then 0.3 mm diamond suspensions using a Dremel tool
and soft polishing bit. The faces of the specimens were
ground to 600 grit and then polished with 6 mm, then 1
mm, then 0.3 mm diamond suspensions. The resulting
root mean square roughnesses were measured with a
surface profiler (the insides of the holes were measured
after cutting the specimen as described later) and found
to consistently be about 150 A, , with final scratches
running parallel to the (future) loading direction. After
polishing, the sheet was annealed in argon at 923 K for
3 h in order to eliminate residual stresses introduced by
polishing (by flowing argon, the surfaces remained free
of oxidized and/or decarburized layers).
Following the anneal, those specimens with 6.08 mm
diameter holes were radially expanded by hydraulically
pressing a 6.35 mm WC ball (at omisfit = 4.4%) through
the holes. This ‘press-fitting’ operation plastically de-
formed and expanded the holes, leading to residual
Fig. 1. (a) Processing operations to produce double-edge notch
stresses at and around the surface of the hole. The
specimens and (b) expanded view of specimens inside grip region,
with the FE meshes shown (boundary conditions in italics; notch
surface roughness was then 132 A, , and there was no
center element (CE) and edge element (EE), where s –o responses evidence of scratches from the press-fit normal to the
were recorded, are shaded). All dimensions in mm. press direction, nor was there evidence of a change in
270 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279

Table 1 2.2. Fatigue testing

DEN specimen designations

Specimen Initial deformation Dsnom (MPa) Fatigue loading to examine crack initiation was car-
ried out on a 50 kN MTS servo-hydraulic machine
C1 None 200 under load control, with R= smin/smax = 0 and a sinu-
C2 None 225 soidal waveform of 3 Hz. Control samples were tested
C3 None 275
at three different stress amplitudes, 200, 225 and 275
PF Press-fit 225
PT snom = +360 MPa 225 MPa, in order to provide a measure of initiation behav-
PC snom = −360 MPa 225 ior, while specimens containing residual stresses were
fatigued at 225 MPa. Although these nominal stresses
were elastic, some plastic deformation occurred at the
notch in all specimens, since the product of the stress
amplitude and notch stress concentration factor always
exceeded the material flow stress.
In order to determine the crack history versus fatigue
cycles, surface replicas2 were taken at pre-determined
cycle intervals in all DEN specimens. Both the inside of
the notch and flat surfaces surrounding the notch were
imaged using the replicas, which were taken under a
static load of  120 MPa to open and better reproduce
any existing fatigue cracks. The replicas were then
coated with Pt–0.2Pd and examined under a scanning
electron microscope (at 10–15 kV) to determine the
location and length of fatigue cracks.

2.3. X-ray diffraction measurements

X-ray diffraction stress measurements were per-

formed (after removing the specimens from the fatigue
machine at various stages) with a 12 kW rotating anode
Fig. 2. (a) Comparison of monotonic and cyclic stress– strain re- source and V-filtered Cr Ka radiation from a Cr-plated
sponse, and Finite Element-computed response using the Generalized
Al anode. A 500 mm beam, formed using a pinhole
Plasticity model. (b) Ferrite and cementite microstresses versus plastic
strain, and corresponding kinematic hardening component Hkin as collimator, was used in conjunction with a position-sen-
determined from FE analysis. sitive detector, with a 5° angular range, for the Fe 211,
whereas the weaker Fe3C 250 peak was measured with
a solid-state detector. Both the incident and diffracted
beam paths were evacuated, which reduced air scatter
and increased intensity approximately twofold. All indi-
shape of the edges of the hole. The radii were the same cated errors in stress were calculated from standard
to within 12 mm before and after the press-fit, and due techniques [22], which incorporate both statistical and
to the much larger size of the hole, this should have geometric effects. Peaks were fit with a pseudo-Voight
function after background corrections, and the triaxial
little effect on the stress intensity factor. DEN speci-
stress states were determined in V geometry, with multi-
mens were then produced by cutting along the center
ple c and 8 sample fits [23]. Further details of the
(of both the expanded and non-expanded holes) using
diffraction set-up are given in Ref. [24].
electro-discharge machining with a thin (0.25 mm) wire.
Additional residual stresses were introduced into two
2.4. Mechanical properties
DEN specimens (without press-fit) by pre-stressing to
nominal stresses of + 360 and − 360 MPa, which Uniaxial stress–strain behavior under monotonic and
resulted in significant plastic deformation at the notch cyclic loading was determined using ASTM standard
root. Upon unloading, this plastic deformation was monotonic and low-cycle specimens, and is shown in
expected to lead to residual compressive and tensile Fig. 2(a). Cyclic stress–strain behavior was determined
macro-residual stresses, respectively. For convenience, from an incremental strain-controlled test [25], at R=
these processing steps are summarized in Fig. 1, and
specimen designations and fatigue loading conditions, 2
The acetate plastic tape was purchased from Ernest F. Fullam,
described later, are given in Table 1. Inc., 900 Albany Shaker Road, Latham, NY 12110, USA.
 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279 271

− l and a frequency of 0.l Hz, with the cyclic response valued constants. Eq. (1) represents a stress–strain
defined by the tips of the stabilized hysteresis loops. response which asymptotes to the line s=so +b+
The cyclic response is seen from the figure to be charac- Ho p, so that H is the asymptotic plastic modulus, b is
terized by softening at low strain amplitudes, and an the difference in stress between so and the asymptotic
approach to the monotonic curve with increasing stress at o= so/E, and d represents the rate do/ds at
strain. This behavior is consistent with results on a which the asymptote is approached.
modified 1070 steel [26], where a transition from cyclic The plastic modulus H can be decomposed into a
softening to hardening with increasing strain amplitude kinematic part, Hkin, and isotropic part, Hiso, with
was reported. H= Hkin + Hiso, so that the yield function f in Eq. (1)
can be written as:
2.5. Model of mechanical properties — generalized f= s− Hkino P − so − Hisok= s −a − so − Hisok
plasticity (2)

The mechanical response was accounted for in the where k is the isotropic hardening variable defined by
FE analysis using the Generalized Plasticity (GP) dk= do p , and a is the backstress defined by da =
model, recently developed by Lubliner et al. [15]. The Hkindo p, which represents the translation of the yield
GP model was chosen in our study because it provides surface under kinematic hardening. For a precipitate-
a good description of stress – strain behavior under both hardened material, such as 1080 steel, the kinematic
monotonic and cyclic loading, including smooth elas- contribution represents the (directional) microstress ex-
tic–plastic transitions, has a relatively straightforward erted on the matrix by the precipitates, while the
algorithmic implementation, and is more computation- isotropic hardening contribution represents forest hard-
ally efficient than the non-linear kinematic hardening ening due to dislocation–dislocation interactions [28].
models [27] often used to model cyclic plastic behavior. The model, as already outlined, requires five input
A brief description of the GP model, with emphasis parameters which need to be obtained from experiment:
on the plastic flow behavior, is given here (see Refs. so, Hiso, Hkin, b, and d. The total asymptotic hardening
[15,16] for more detail). In our implementation, plastic was determined from a best fit line to the measured
deformation is assumed to obey a von Mises yield monotonic stress–strain curve (Fig. 2(a)) and found to
criterion3 and associative flow rule, with linear isotropic be Htot = 5775 MPa. The kinematic contribution to this
and kinematic hardening behavior. The relationship hardening was then determined from measurements of
between the increment in plastic strain, do p, and stress, the residual ferrite microstress versus plastic strain, as
ds, under uniaxial loading is given by: follows. Five tensile samples were deformed to different
levels of plastic deformation, and then unloaded, and
do p s− so −Ho p  f the triaxial phase-specific residual stresses were mea-
= =
p
ds Hb +d[b − (s −so −Ho )] Hb +d[b −f ] sured. The residual stresses along the loading direction
(1) are shown in Fig. 2(b), and are seen to be exclusively
microstresses, formed due to differential plastic defor-
where Ž’ is the Macauley bracket,4 so is the yield mation between the ferrite and cementite phases. The
stress, f is the yield function and H, b and d are positive kinematic hardening was then determined from the
best-fit line of the ferrite microstress as a function of
Table 2
plastic strain, shown in the figure. This procedure as-
Material parameters used in generalized plasticity model
sumes that the backstress is equivalent to the direc-
Parameter Value (MPa or [no units]) tional (internal) matrix stress [28]. This fit yielded a
value Hkin = da/do P = 3700 MPa. The remaining hard-
E 205 000 ening was taken to be isotropic, so that Hiso =Htot −
n [0.29]
Hkin = 2075 MPa.
so 400
Hiso 2075 The yield stress so was taken to be 400 MPa, which
Hkin 3700 is an average of the measured monotonic and cyclic
b 80 flow stresses. The limit parameter b was then deter-
d 7000 mined by the difference between so and the asymptotic
curve at o= so/E, and found to be b= 80 MPa. The
remaining material parameter, d, was estimated via a
trial-and-error method by using the presented parame-
3
Under uniaxial deformation, the von Mises yield criterion can be
ters and various values of d to obtain a reasonable fit to
expressed by f 3/2(tij tij )-s2y = 0, where f is the yield function, tij is
the deviatoric stress and dy is the uniaxial yield stress. both the measured monotonic and cyclic curves. All
4
The Macauley bracket is defined such that Ž f = f for f]0 and material properties are given in Table 2, along with the
Ž f  = 0 for fB0. elastic constants E and n. The resulting stress–strain
272 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279

effective stress versus strain. These plots were identical

to the uniaxial stress–strain for a single element in the
reference, validating that our 3D implementation was
working properly. Convergence was examined with a
tensile pre-strain and solely elastic behavior. The mesh
employed gave stresses within 2% of the analytical
solution along the notch edge, and reducing the mesh
size gave even closer agreement.
The FE analysis involved first modeling the residual
stresses introduced into specimens PT, PC and PF, and
then simulating fatigue loading in these and control
specimens. For specimens PT and PC, the pre-straining
was simply modeled by axially loading the specimens
(along x2) to nominal stresses of +360 and −360
MPa, and then releasing the loads.
The procedures used to simulate the press-fit opera-
tion in specimen PF are shown in Fig. 3. For these
computations, an additional boundary condition,
dx1 = 0 (x1 = 0) was used since the press-fit was per-
formed before cutting the specimen. Nodes located at
Fig. 3. FE simulations of press-fit operation: (a) nodal displacement
the hole surface (along x3 at Rhole (x 21 +x 22)=3.04
field dnodal applied along hole edge (for nodes located at R =Rhole in mm) were radially displaced by an amount dnodal given

  
(b)) to represent deformation from WC ball; and (b) calculation of by:
maximum nodal displacement Dnodal using eigenstrains and 2D
dx3 2
axisymmetric FE analysis (boundary conditions denoted by italics). dnodal = Dnodal 1− (05dx3 5 lc) (3)
dnodal was then translated along x3 to ‘move’ the ball through the lc
hole. See text for further details.
where Dnodal represents the maximum radial displace-
ment of the hole (determined from an eigenstrain calcu-
response was computed in the FE analysis using these
lation as described later), dx3 is the position of the top
parameters (Fig. 2(a)), and can be seen to provide a
of the ball, and lc represents the length over which the
reasonable fit to both the cyclic and monotonic stress–
ball contacted the hole surface. This quadratic depen-
strain behavior.
dence along x3 was chosen to simulate the shape of the
WC ball (see Fig. 3(a)). The value of lc was calculated
2.6. Finite Element analysis from the geometry to be:
lc = [Dnodal(2Rball − Dnodal)] (4)
FE analysis was used to provide local stress–strain
information around the notch regions, where the fa- and the value of Dnodal was computed from a separate
tigue cracks were expected to initiate. The DEN speci- FE calculation, using eigenstrains to simulate the radial
mens were modeled using three-dimensional (3D) interference between the ball and specimen (Fig. 3). In
elastic–plastic analysis, with material response de- this calculation, the WC ball and 1080 specimen were
scribed by the GP model as already described. The modeled axisymmetrically. The ball was taken to be
meshes used are shown in Fig. 1(b). (The code em- elastic (a good assumption since the maximum calcu-
ployed is a version of the well-known FEAP.) For all lated stress was always well below sy,WC = 6800 MPa),
specimens except PF, only one-eighth of the specimen with an elastic modulus EWC = 640 GPa, while the 1080
was modeled due to the three symmetry planes (corre- specimen was taken to be elastic–plastic (defined by the
sponding to the boundary conditions dx1 =0 (x1 = 12.7 GP model parameters). The interference was modeled
mm), dx2 =0 (x2 =0), and dx3 =0 (x3 =0)) denoted in by subjecting the ball to a radial eigenstrain o *11 =o *22 =
the figure. In specimen PF, the press-fit operation led to omisfit = 0.044, and the resulting radial displacement of
asymmetry in the specimen in the x3 direction, eliminat- the hole was determined, from the relative response of
ing the boundary condition at x3 =0 so that one-quar- the two materials, to be Dhole Dnodal = D−Dball =
ter of the specimen needed to be modeled. 0.135− 0.010 mm= 0.125 mm, where D =Rball −
In Ref. [16], it is shown that the GP model can Rhole = 0.135 mm and Dball = 0.010 is computed from
capture transient cyclic effects, like the Bauschinger the FE analysis (see Fig. 3 for further details).
Effect, ratcheting and cyclic shear. To assure that the Using this value for Dnodal and Eq. (4), lc was calcu-
model was valid in a notch, the cyclic response was lated to be 0.882 mm. These values of Dnodal and lc were
examined for single elements, plotting the Von Mises then input into Eq. (3), and the resulting displacement
 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279 273

field dnodal was translated along x3 in 0.5 mm incre- load cycle. The residual (macro)stresses remaining after
ments, starting at x3 = − 3 mm. After ‘moving’ the ball unloading the specimens are compared with those de-
completely through the hole, the boundary condition termined by diffraction, where stresses were measured
dx1 =0 (x1 = 0) was removed, simulating slitting the in both the ferrite and carbide phases, and separated
hole to produce specimen PF. The simulation was also into macro- and microstresses. The tensile and com-
run with 0.25 mm increments with equivalent results. pressive pre-strains are seen to lead, respectively, to
(The largest finite element was 1 mm in the center and compressive and tensile residual macrostresses, which
this graded to smaller values at the surface (Fig. 1(b)).) result from elastic unloading after plastic deformation.
After modeling these residual stresses, all specimens Good agreement between the measured and FE-com-
were ‘loaded’ to simulate fatigue, and the specimen puted residual macrostresses is observed in both
stress/strain response was recorded at the center ele- specimens.
ment (and edge element in PF) (see Fig. 1(b)). For this Superimposed on these macrostresses are mi-
‘fatigue loading’, 20 steps were used in each load–un- crostresses formed due to differential deformation be-
load cycle to ensure convergence in the plasticity tween phases. In the case of the tensile pre-strain
analysis. (specimen PT), the cementite experiences large tensile
microstresses as this phase takes up a larger fraction of
the load during plastic deformation, while the opposite
3. Results trend is observed in sample PC. The microstresses in
the carbide phase are seen to be comparable in magni-
3.1. Initial residual stresses tude (but of opposite sign) with the macrostresses.
Thus, both residual macrostresses and microstresses are
Measured and FE-computed residual stresses arising significant in these specimens, and these were tracked
from the pre-straining and press-fit operations are pre- during fatigue loading to determine their relative stabil-
sented in Fig. 4. In Fig. 4(a), the FE-computed stress– ity during fatigue.
strain responses of the center notch elements, in Residual stresses in specimen PF are shown along the
specimens PT and PC, are shown for the first load–un- x3 direction (with x1 = x2 = 0 and x3 = − 3 mm repre-
senting the starting position of the WC ball), and are
seen in Fig. 4(b) to be mainly macrostresses, which are
highly compressive. Both the measured and FE-com-
puted stresses increase in compressive magnitude along
x3, in contrast to specimens PC and PT where the
stresses were approximately constant along x3. There-
fore, the region of smallest compressive stress, located
at x3 = − 3 mm, was anticipated to be the location of
crack initiation in this specimen.

3.2. Fatigue crack initiation

Cracks in all specimens, with the exception of speci-

men PF as described later, were observed to form in the
center regions of the notches (xi  0), in accord with
results of studies on similar geometries in HY-130 steel
[29], 304 stainless steel [20], and 2024 Al alloy [21]. In
many cases, a number of cracks were observed to
initiate independently inside the notch, always normal
to the fatigue load, and these linked up to form the
main fatigue crack responsible for failure. We looked
for evidence that inclusions were important, but we
could find no sulfur or manganese signatures around
the cracks. (It is still possible that such inclusions
played a significant role, but this could not cause the
differences between specimens, as all came from the
same stock and would have the same inclusion
Fig. 4. Measured versus FE-computed stresses in DEN specimens
from (a) pre-straining operations in PC and PT, and (b) press-fitting content.)
operation. Horizontal error bars represent the X-ray beam size, while The formation of fatigue cracks in specimen PF is
the vertical bars represent the measurement error. shown in Fig. 5, where scanning electron microscopy
274 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279

crocrack initiation is seen to occur at three distinct sites

near the edge of the specimen (Fig. 5(a)), which is the
position where the compressive residual macrostresses
were found to be smallest in magnitude (Fig. 4(b)). This
provides evidence that the residual stresses are con-
trolling initiation. With repeated fatigue, the microc-
racks linked up, and a second crack was initiated
parallel to the first (Fig. 5(b)). Eventually, these two
cracks joined to form the macrocrack responsible for
final failure, and the growth of these cracks just prior to
link-up is shown in Fig. 5(c).
The measured crack history of all specimens fatigued
at 225 MPa is shown in Fig. 6, where a represents the
length of the crack responsible for final failure. This
length was determined from the replicas, by backtrack-
ing from the (known) critical crack location at later
fatigue cycles to determine the site of initiation. Com-
paring the initiation behavior with control specimen
C2, it is clear that the press-fit increased resistance to
initiation, while the compressive pre-strain had the
opposite effect. The tensile pre-strain, however, had no
measurable effect on initiation. The small crack growth
rates are seen to be well represented by power law
behavior in these specimens.
The number of cycles to fatigue initiation for all
specimens examined, where Ni is taken to correspond to
a crack length of 0.1 mm (approximately the shortest
length consistently detected in all specimens), are sum-
marized in Table 3. The number of cycles to failure, Nf,
Fig. 5. SEM images of replicas showing crack initiation in the notch is also included. Comparing the control specimens, it
of specimen PF. (a) Microcrack initiation near notch edge (x3 = −3 can be seen that the fraction of life taken up by the
mm) at N = 90 × 103 cycles, (b) link-up of microcracks and initiation initiation stage, Ni/Nf, increases with increasing Nf,
of a second crack at N= 115× 103 and (c) growth of cracks just prior corresponding to decreasing Ds, as is well known. In all
to link-up to form main crack at N =145×103. Loading direction is
vertical.
specimens, Ni/Nf is at least 0.44, underlying the impor-
tance of accounting for initiation behavior to predict
the total fatigue life.
Consulting Table 3 and the results for specimens C2
and PF, it can be seen that the cycles for both initiation
and failure are higher due to the press-fit, and the
ration Ni/Nf is similar for the two specimens, so that the
press-fit reduced crack growth rates as well as delaying
initiation, due to the plasticity around the notch in-
duced by the fit.

Table 3
Number of cycles to initiation and failure in DEN specimens

Fig. 6. Small crack initiation and growth behavior for DEN speci- Specimen Ni (a =0.1 mm) Nf Ni/Nf
mens vesrus fatigue cycles, N, at a nominal stress range Ds=225
MPa, with power law fits to the data shown. The length of the crack C1 89 550 116 550 0.77
at different cycles that led to failure. C2a 45 410 84 910 0.53
C3 19 500 43 520 0.45
PFb 91 030 197 380 0.46
PT 47 540 107 200 0.44
(SEM) images of notch replicas taken at N = 90× 103,
PC 17 730 33 600 0.53
115× 103 and 145 × 103 cycles are displayed. Only the
region near x3 = −3 mm is shown since this was the a
A repeated fatigue test gave Nf =91 525.
only region where cracks were observed to form. Mi- b
A repeated fatigue test gave Nf =221 320.
 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279 275

The stress–strain response of specimen PT is shown

for the first two loading cycles in Fig. 8(b). Due to the
large compressive residual stress introduced in the first
cycle, upon further cycling, the superposition of the
applied and residual stress is always less than the
absolute value of the flow stress. Thus, the response is
predicted to be entirely elastic for the remaining fatigue
cycles, and s22,res remains at the value attained in the
first cycle. Similar behavior is predicted in specimen PF
(not shown), due to the large compressive residual
stresses introduced by press-fitting.
In contrast, the tensile residual stresses introduced in
PT are predicted to lead to significant plastic deforma-
tion in subsequent cycles, since the sum of the residual
and applied stresses exceeds the flow stress during
tensile loading (Fig. 8(c)). Upon elastic unloading after
the second loading cycle, the magnitude of the tensile
residual stress is reduced, which leads to a decrease in
plastic deformation in the third cycle. The values of
s22,res and o P22 continue to decrease with cycling, until at
the fifth cycle, the response stabilizes and is dominated
by elastic behavior. The stabilized value of s22,res is
predicted to remain tensile, mainly as a result of the
Fig. 7. Changes in the measured residual stresses at notch center with
fatigue cycles, N: (a) macrostresses and (b) cementite microstresses.
The average values of the stresses in control specimen C2 are shown
by a dashed line for reference.

3.3. Stress e6olution during fatigue

3.3.1. X-ray diffraction

The residual macrostresses at the center of the notch
(xi = 0) are shown in Fig. 7. It can be seen that the
macrostresses (Fig. 7(a)) fade in both specimens PT and
PC, but to non-zero values. Little change in stress is
observed for specimen PF. In addition, compressive
residual macrostresses develop in C2 in the first loading
cycle, and remain approximately constant with cycling.
Changes in the residual microstresses are compared
in Fig. 7(b) for specimens C2, PT and PC. In C2,
microstresses are observed to form in the first loading
cycle, and remain constant with cycling similar to the
macrostresses. The microstresses initially present in
specimens PC and PT rapidly fade to levels approach-
ing those of C2. Comparing Fig. 7(a) and Fig. 7(b), it is
clear that the microstresses are less stable than the
macrostresses under identical loading conditions.

3.3.2. Finite Element analysis

The FE-computed stress – strain histories within the
notch of specimens C2, PT and PC are shown in Fig. 8.
In specimen C2, some plastic deformation occurs dur-
ing the first few loading cycles, but upon further cy-
cling, the response is predicted to be dominated by
elastic behavior, and the stabilized value of residual Fig. 8. FE-Computed stress – strain history at notch center element
stress, s22,res, is predicted to be −98 MPa. for specimens (a) C2, (b) PT and (c) PC.
276 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279

independent of prior cold work. Since the ferrite has a

relatively high stacking fault energy [31], significant
cross-slip is anticipated during fatigue. Thus, the large
dislocation pile-ups which form at the ferrite/carbide
interfaces in the first cycle are partially removed by
dislocation annihilation occurring during cross slip,
which may account for the ‘history-independent’ mi-
crostress values attained during fatigue, as shown in
Fig. 7(b). Since these microstresses are constant be-
tween specimens, they are not expected to cause the
observed differences in initiation lifetimes.
Fig. 9. Comparison of measured and FE-computed residual
In contrast to the microstresses, residual
macrostresses at the notch root (xi = 0), as a function of fatigue macrostresses were found to be more stable during
cycles, N, for all specimens fatigued at Ds= 225 MPa. fatigue. The greater stability of these macrostresses
relative to the microstresses is attributed to their longer-
large amount of accumulated plastic deformation, range nature, which decreases the likelihood of disloca-
which increases the yield stress (isotropic hardening) tion–dislocation interactions. Since the magnitude of
and the maximum achievable tensile stress relative to these macrostresses were found to be appreciable at the
the control specimen C2. surface of the notches, where the cracks were observed
to initiate, these residual stresses had a significant effect
on initiation behavior. The press-fitting was found to
3.3.3. Measured 6ersus computed stresses
lead to large compressive macrostresses that remained
A comparison between the measured residual
relatively stable throughout the fatigue life, and thus
macrostresses (Fig. 7(a)) and those computed by FE
greatly increased fatigue crack initiation resistance.
analysis (Fig. 8) as a function of N is given in Fig. 9.
FE predictions of the initial values of these
Good agreement is observed between the FE analysis
macrostresses were found to be in good agreement with
and measured stresses in specimen C2. In addition, FE
measured values for all specimens (Fig. 4). In addition,
analysis predicts the measured decrease in tensile resid-
FE analysis was found to predict the observed decrease
ual stress in PC, although it overpredicts the magnitude
in residual tensile macrostress in specimen PC. This
of this decrease, and does not account for the observed
resulted from the fact that the applied loading was
fading in PT. Good agreement between experimentally
tensile so that significant plastic deformation occurred
determined and FE analysis residual stresses in speci-
during the second cycle, which limited the maximum
men PF is also observed. Overall, the measurements
tensile stress and subsequently decreased the residual
and FE analysis show reasonable agreement of the
macrostress on elastic unloading (Fig. 8(c)). Both the
stabilized residual macrostress, which varied signifi-
measured and predicted values of the stabilized
cantly between specimens.
macrostress are tensile, which leads to an increased
mean stress and decrease in fatigue life. In contrast, the
observed fading in specimen PT was not predicted by
4. Discussion FE. This relaxation of the compressive macrostresses
indicates that the notch surface is microplastically de-
One of the main objectives of this study was to forming relative to the interior during fatigue, which is
examine the relative contributions of macrostresses and an effect not accounted for in the FE analysis.
microstresses on FCI. Fig. 7 clearly demonstrates that, Regarding the choice of a 3D, rather than more
for the material studied, the macrostresses are generally conventional two-dimensional (2D) models, it can be
more stable than the microstresses under identical load- stated that the former is necessary to capture certain
ing conditions. Within the first 100 fatigue cycles, the geometrical effects such as the asymmetry introduced
microstresses which form during tensile and compres- by the press-fitting operation. In this case, stress gradi-
sive pre-straining are seen to approach the values mea- ents along the notch length, which could only be pre-
sured in the control specimen. These stabilized dicted with a 3D model, correlated with the location of
microstresses are small but non-zero, which suggests FCI (Section 3.2). The 3D model also eliminates the
that the microstresses which form during fatigue domi- need for plane-stress/strain assumptions, both of whose
nate over the initial microstresses. This behavior sup- validity is limited in certain regions, e.g. slightly below
ports the results of Feltner and Laird [30], who found the surfaces of the specimens. It is noted that when the
that, in high stacking fault energy materials with a (geometrically) 3D stress–strain distribution is known a
strong propensity for cross slip, the cyclic flow stress priori, use of 2D models in areas of interest may be
and dislocation cell structure formed during fatigue is sufficient. At the center of the notches, for example,
  
J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279 277

Ds 1/b
where generally the greatest amount of deformation Nf = (6)
s%f − (sa + Msres)
and FCI occurred, use of a 2D plane strain model
could be expected to provide a reasonable approxima- using the definitions of stress amplitude sa =Eoa,elastic
tion of the stresses and strains experienced by elements and mean stress5 sm = (sa + Msres), where Msres is the
in this region. stabilized value of residual macrostress (at 10 000 cy-
Since the residual macrostresses were observed to cles). This equation assumes that the residual
have a significant effect on fatigue behavior, it is impor- macrostress has the same effect as a mean (applied)
tant to be able to predict these effects for proper fatigue stress, which has been shown experimentally by Wilks
design. This was attempted here by combining the et al. [32] to be a valid assumption. It should be noted
results of measured and FE-computed stress – strains at that some plastic deformation must occur locally in the
the notch root, where the cracks were observed to samples to cause fatigue damage and failure, which is
initiate, with the strain – life relationship [31]: not explicitly accounted for in Eq. (6). However, the
mean stress effects predicted by Eq. (6) account for the
(s%f −sm)
Do = Doel + Dopl = (Nf)b +o%f(Nf)c (5) fact that, under a tensile (compressive) macrostress, a
E larger (smaller) amount of damaging tensile alternating
stress occurs per fatigue cycle, which will accelerate
where s%f and o%f are, respectively, fatigue strength and (limit) these microstructural plastic deformation
fatigue ductility, and b and c are, respectively, the processes.
fatigue strength and fatigue ductility exponents. Results To test the ability of Eq. (6) to account for observed
from our FE analysis show that, in the high-cycle fatigue lifetimes, the material constants s%f and b were
fatigue regime experienced by our specimens, the strain determined from a non-linear least squares fit to the
range is dominated by the elastic strain contribution. variables Dsnotch, Msres and Nf in our six specimens,
Under these conditions, Eq. (5) can be simplified by where the values of Ds and Msres were taken, respec-
considering only the elastic term, and, with Ds, the tively, from the FE and X-ray analyses (Fig. 9). These
stress at the notch at crack initiation: constants were found to be s%f = 1756(21) MPa and
b= −0.157(3), which are in reasonable agreement with
typical values obtained from tests on smooth speci-
mens, where the stress and strain ranges are constant
along the gage length and measured directly [33]. Fur-
thermore, when these parameters were plotted against
Nf using Eq. (6) (Fig. 10(a)), a good correlation was
observed for all samples tested, with r= 0.89. For
comparison, the constants were also determined with-
out taking the residual stresses into account (taking
M
sres = 0 in Eq. (6)), and found to be s%f =3554(41)
MPa and b= −0.238(5). These parameters are unreal-
istically large in magnitude, and when plotted against
Nf, a much poorer correlation coefficient was observed,
with r=0.44 (Fig. 10(a)).
To test the ability of this analysis to account for
initiation life in the presence of residual stresses as well
as to total life, Eq. (6) was modified to:

Ni =
 Ds  1/bi
(7)
s%i − (sa + Msres)

where the material constants s%i and bi were determined

using the procedure already outlined, by replacing Nf
with Ni, and found to be s%i = 2315(26) MPa and b=
−0.197(2). When Eq. (7) was then plotted for all
specimens tested using these parameters (Fig. 10(b)), an
excellent correlation between specimens was again ob-
served, with r= 0.95.
Fig. 10. Relationship between notch stress and (a) Nf (Eq. (6)) and (b)
Ni (Eq. (7)), showing good correlation between specimens when
5
residual macrostresses are considered. Error bars include uncertainty This definition of the mean stress is applicable for R = 0 loading,
in fitting parameters s(i,f) and X-ray values Msres. as performed here, where Msres =smin,applied.
278 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279

The presented analysis suggests that the effects of Initially, present residual microstresses were found
residual macrostresses on both fatigue crack initiation to fade rapidly with fatigue (but not to zero), inde-
and total life in notched specimens may be predicted pendent of which phase was in compression or ten-
with good accuracy, provided the stabilized values of sion.
residual stress and local stress – strain behavior at the Combining mechanical data and measurements of
notch are known. Under the high-cycle fatigue tests the residual stress tensor yield all the necessary input
performed here, the combined FE and diffraction for FE analysis.
techniques have been shown to provide accurate The combined FE and diffraction results were
means to determine these quantities. Further compari- found to be capable of accounting for the observed
sons between these techniques, both on different ma- effects of residual macrostress on both fatigue crack
terials and loading conditions, would be useful to initiation and total life, when used in combination
generalize these findings. Specifically, the limited abil- with a stress-based approach to fatigue.
ity of the FE method to account for the observed
changes in residual stress with fatigue are seen to
limit the accuracy of this analysis, when used exclu-
sively, to account for residual stress effects. It is an- Acknowledgements
ticipated that further comparisons with experimental
data would drive refinements in the FE analysis, such This research was partially supported by ONR un-
as including stress fading explicitly in the material der contract no. N000144-90-1374. We thank Laurrie
response, allowing for greater accuracy in the predic- Wittig Cordes for advice and Mark Seniw, Jerry
tions to be realized. Carsello, Ken Lehmann, and Jim Hahn for assistance.
This research is from a thesis submitted (by J.D.A.)
in partial fulfillment of the requirements for the
5. Conclusions Ph.D. degree in Materials Science and Engineering at
Northwestern University.
The principal findings from this study are as fol-
lows.
The type of residual stress introduced varied de- References
pending on the processing operation. Press-fitting was
found to form primarily compressive macrostresses, [1] J.H. Underwood, Exp. Mech. 3 (1995) 61.
while uniaxial pre-straining introduced residual [2] L. Edwards, A.T. Ozdemir, NATO ASI Series E, Kluwer Aca-
macrostresses and microstresses of comparable magni- demic, Dordrecht, 1992, p. 545.
[3] M.O. Lai, J.T. Oh, A.Y.C. Nee, Eng. Fract. Mech. 45 (1993)
tudes.
551.
The measured initial values of these residual [4] D.F. Mowbray, J.E. McConnelee, ASTM STP 519 (1973) 151.
stresses agreed well with FE predictions. [5] M. Toparli, A. Ozel, T. Aksoy, Mater. Sci. Eng. A225 (1997)
The fraction of fatigue life taken up by initiation, 196.
Ni/Nf, was at least 0.44 in the specimens tested, and [6] P.B. Aswath, S. Suresh, D.K. Holm, A.F. Blom, J. Eng. Mater.
this ratio increased with decreasing applied stress am- Technol. 110 (1988) 278.
[7] H. Neuber, J. Appl. Mech. 12 (1961) 544.
plitude. Thus, properly accounting for initiation be- [8] J.M. Barsom, S.T. Rolfe, Fracture and Fatigue Control in
havior can be crucial for predicting fatigue life. Structures, Prentice Hall, Englewood Cliffs, NJ, 1987.
Cycles to crack initiation, Ni, were significantly af- [9] S.S. Manson and M.H. Hirshberg, in: Proceedings of the 1st
fected by processing, and this appears to be due to International Conference on Fracture, 1965, p. 479.
[10] Y.H. Kim, M.E. Fine, T. Mura, Eng. Fract. Mech. 11 (1979)
changes in the residual stresses.
653.
These effects were tied to the stability of the resid- [11] P.S. Maiya, Mater. Sci. Eng. 38 (1979) 289.
ual stresses during fatigue. In particular, the large [12] H.L. Oh, in: P.R. Abelkis, C.M. Hudson (Eds.), ASTM STP
compressive macrostresses resulting from the press-fit 761, ASTM, Philadelphia, 1982, p. 296.
were found to be stable throughout the specimen fa- [13] R.L. Tobler, Q.S. Shu, Cryogenics 26 (1986) 396.
[14] I.C. Noyan, Metall. Trans. 14A (1983) 1907.
tigue life, while macrostresses from uniaxial pre-
[15] J. Lubliner, R.L. Taylor, R. Aurucchio, Int. J. Solids Struct. 30
straining were found be partially stable, decreasing to (1993) 3171.
about 50% of their initial magnitudes within N/Ni  [16] F. Auricchio, R.L. Taylor, Int. J. Plast. 11 (1995) 65.
10%. [17] J.D. Almer, J.B. Cohen, R.A. Winholtz, Metall. Trans. 29A
The FE analysis was found to partially account for (1998) 2127.
[18] J.C. Russ, Practical Stereology, Plenum, New York, 1986.
these observed changes in the specimen pre-strained
[19] A.F. Giamei, E.J. Freise, Trans. AIME 239 (1967) 1676.
in compression, although it did not predict any of the [20] S.M. McGuire, M.E. Fine, O. Buck, J.D. Achenbach, J. Mater.
observed fading in the specimen pre-strained in ten- Res. 8 (1993) 2216.
sion. [21] C.Y. Kung, M.E. Fine, Metall. Trans. 10A (1979) 603.
 J.D. Almer et al. / Materials Science and Engineering A284 (2000) 268–279 279

[22] I.C. Noyan, J.B. Cohen, Residual Stress: Measurement by Dif- [27] Y.F. Dafalias, E.P. Popov, Acta Mech. 21 (1975) 173.
fraction and Interpretation, Springer Verlag, New York, 1986. [28] D.V. Wilson, P.S. Bate, Acta Metall. 34 (1986) 1107.
[23] R.A. Winholtz, J.B. Cohen, Aust. J. Phys. 41 (1988) 189. [29] J.M. Barsom, R.C. McNicol, ASTM STP 559 (1974) 183.
[24] J.D. Almer, The effects of residual macrostresses and mi- [30] C.E. Feltner, C. Laird, Acta Metall. 15 (1967) 1633.
crostresses on fatigue crack initiation and growth, Ph.D. Thesis, [31] S. Suresh, Fatigue of Materials, Cambridge University Press,
Northwestern University, 1998. Cambridge, 1991.
[25] ASTM, Report No. E606-80, 1980, ASTM, Philadelphia. [32] M.B. Wilks, D. Nowell, D.A. Hills, in: M.R. James (Ed.), ICRS
[26] J.D. Dougherty, T.S. Srivatsan, J. Padovan, Eng. Fract. Mech. 4, Society of Experimental Mechanics, 1994, p. 1238.
53 (1996) 829. [33] ASM, Metals Handbook, Metals Park, OH, 1978.