STP 1546-2012

Daniewicz | Belsick | Gdoutos
Journal of ASTM International

Selected Technical Papers
STP 1546
Fatigue and
Fracture
Mechanics:
JAI • Fatigue and Fracture Mechanics: 38th Volume

38th Volume
JAI Guest Editors:

Steven R. Daniewicz
Charlotte A. Belsick
Emmanuel E. Gdoutos
STP 1546
www.astm.org
ISBN: 978-0-8031-7532-7
Stock #: STP1546
Selected Technical Papers STP1546
Fatigue and Fracture Mechanics:
38th Volume
JAI Guest Editors:

Steven Daniewicz
Emmanuel E. Gdoutos
ASTM International
100 Barr Harbor Drive
PO Box C700
West Conshohocken, PA 19428-2959
Printed in the U.S.A.
ASTM Stock #: STP1546

Library of Congress Cataloging-in-Publication Data
ISBN: 978-0-8031-7532-7
ISSN: 1040-3094
Copyright © 2012 ASTM INTERNATIONAL, West Conshohocken, PA. All rights
reserved. This material may not be reproduced or copied, in whole or in part, in any printed,
mechanical, electronic, film, or other distribution and storage media, without the written
consent of the publisher.
Journal of ASTM International (JAI) Scope
The JAI is a multi-disciplinary forum to serve the international scientific and engineering
community through the timely publication of the results of original research and critical
review articles in the physical and life sciences and engineering technologies. These
peer-reviewed papers cover diverse topics relevant to the science and research that
establish the foundation for standards development within ASTM International.
Photocopy Rights
Authorization to photocopy items for internal, personal, or educational classroom use,
or the internal, personal, or educational classroom use of specific clients, is granted by
ASTM International provided that the appropriate fee is paid to ASTM International,
100 Barr Harbor Drive, P.O. Box C700, West Conshohocken, PA 19428-2959,
Tel: 610-832-9634; online: http://www.astm.org/copyright.
The Society is not responsible, as a body, for the statements and opinions expressed in
this publication. ASTM International does not endorse any products represented in this
publication.
Peer Review Policy
Each paper published in this volume was evaluated by two peer reviewers and at least
one editor. The authors addressed all of the reviewers’ comments to the satisfaction of
both the technical editor(s) and the ASTM International Committee on Publications.
The quality of the papers in this publication reflects not only the obvious efforts of the
authors and the technical editor(s), but also the work of the peer reviewers. In keeping
with long-standing publication practices, ASTM International maintains the anonymity
of the peer reviewers. The ASTM International Committee on Publications acknowledges
with appreciation their dedication and contribution of time and effort on behalf of ASTM
International.
Citation of Papers
When citing papers from this publication, the appropriate citation includes the paper
authors, “paper title”, J. ASTM Intl., volume and number, Paper doi, ASTM International,
West Conshohocken, PA, Paper, year listed in the footnote of the paper. A citation is
provided as a footnote on page one of each paper.
Printed in Bay Shore, NY

July, 2012
Foreword
THIS COMPILATION OF THE JOURNAL OF ASTM INTERNATIONAL
(JAI), STP1546, on Fatigue and Fracture Mechanics: 38th Volume, contains
only the papers published in JAI that were presented at the Eleventh
International ASTM/ESIS Symposium on Fatigue and Fracture Mechan-
ics (38th National Symposium on Fatigue and Fracture Mechanics) held
during May 18–20, 2011 in Anaheim, CA, USA. The Symposium was jointly
sponsored by ASTM International Committee E08 on Fatigue and Fracture
and the European Structural Integrity Society (ESIS).
The Symposium Co-Chairmen and JAI Guest Editors are Steven R.
Daniewicz, Mississippi State University, Mississippi State, MS, USA,
Charlotte A. Belsick, Lockheed Martin, Sunnyvale, CA, USA, and
Emmanuel Gdoutos, University of Thrace, Xanthi, Greece.
Contents
Overview ............................................................ vii
Fatigue Crack Growth

Dimensional Analysis and Fractal Modeling of Fatigue Crack Growth
A. Carpinteri and M. Paggi ............................................. 3
Unraveling the Science of Variable Amplitude Fatigue
R. Sunder ......................................................... 20
Fatigue Crack Closure in Residual Stress Bearing Materials
M. R. Hill and J. Kim.................................................. 65
Fatigue Crack Growth Rate Behavior of A36 Steel using ASTM Load-Reduction
and Compression Precracking Test Methods
J. C. Newman, Jr., B. M. Ziegler, J. W. Shaw, T. S. Cordes, and D. J. Lingenfelser ...... 87
Crack Closure Behavior on a Variety of Materials under High Stress Ratios
and Kmax Test Conditions
Y. Yamada and J. C. Newman, Jr. ......................................... 109
Modelling of Surface Crack Advance in Round Wires Subjected to Cyclic Loading
J. Toribio, J. C. Matos, B. González, and J. Escuadra .......................... 126
Study of an On-Line Crack Compliance Technique for Residual Stress Measurement
Using 2D Finite Element Simulations of Fatigue Crack Growth
S. Ismonov and S. R. Daniewicz ......................................... 136
Analysis of Notch Effect in Fatigue
K. Yanase and M. Endo ............................................... 157
High Temperature, High Frequency, and Environmental Effects

Estimation of Corrosion Fatigue-Crack Growth through Frequency Shedding Method
R. V. Prakash and S. Dhinakaran ......................................... 179
A Numerical Strip-Yield Model for the Creep Crack Incubation in Steels
G. P. Potirniche ..................................................... 197
Influence Analysis of Application-Specific Phenomena on the Creep-Fatigue
Life of Turbine Housings of Turbochargers
F. Laengler, T. Mao, and A. Scholz ........................................ 215
Fatigue Crack Closure at Near-Threshold Growth Rates in Steels,
Effects of Microstructure, Load Sequence and Environment
M. K. Schaper ...................................................... 231
Temperature and Load Interaction Effects on the Fatigue Crack Growth Rate
and Fracture Surface Morphology of IN100 Superalloy
B. S. Adair, W. S. Johnson, S. D. Antolovich, and A. Staroselsky .................. 255
Fatigue Sensitivity to Small Defects of a Gamma–Titanium–Aluminide Alloy
................
M. Filippini, S. Beretta, L. Patriarca, G. Pasquero, and S. Sabbadini 279
Investigation of Load Control Errors for Spectrum Fatigue Testing
at High Frequencies
D. T. Rusk and R. E. Taylor............................................. 296
Fracture Mechanics
FEM Analysis of a DCP Implant on a Human Femoral Bone With a Fracture Gap
T. Fongsamootr and S. Bernard .......................................... 321
Point Load Weight Functions for Semi-Elliptical Cracks in Finite Thickness Plate
Z. Jin and X. Wang ................................................... 338
Evaluation of Fracture Toughness Test Data for Multilayer Dissimilar Joint Welds
Using a Weibull Stress Model
Y. Takashima, M. Ohata, M. Seto, Y. Okazaki, and F. Minami ..................... 357
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379

Overview
This book compiles the work of several authors who made presentations at
the Eleventh International ASTM/ESIS Symposium on Fatigue and Frac-
ture Mechanics (38th ASTM National Symposium on Fatigue and Fracture
Mechanics) sponsored by ASTM Committee E08 on Fatigue and Fracture
and the European Structural Integrity Society (ESIS). The symposium was
held on May 18–20, 2011 in Anaheim, CA in conjunction with the May 16–17,
2011 standards development meetings of ASTM Committee E08.
We were honored to have Alberto Carpinteri from the Politecnico di Torino
in Torino, Italy give the Jerry L. Swedlow Memorial Lecture. The manuscript
associated with his fascinating presentation regarding dimensional analysis
and fractal modeling of fatigue crack growth is included. The symposium
provided a forum to promote discourse and disseminate state-of-the-art ad-
vances in analysis methodologies and testing techniques in the areas of fa-
tigue and fracture mechanics. Applications of emerging analytical tools and
novel experimental techniques to assess the durability of engineering struc-
tures, components, and devices and to identify the associated damage mech-
anisms in materials were discussed, as were multi-scale and mutli-physics
approaches for predicting the fatigue and fracture response of materials and
structures. A particular focus area was fatigue crack growth behavior.
Steven R. Daniewicz
Emmanuel E. Gdoutos
JAI Guest Editors
vii
FATIGUE CRACK GROWTH
J_ID: DOI: Date: 15-June-12 Stage: Page: 3 Total Pages: 17
Reprinted from JAI, Vol. 8, No. 10

doi:10.1520/JAI104105
Available online at www.astm.org/JAI
Alberto Carpinteri1 and Marco Paggi1
Dimensional Analysis and Fractal Modeling

of Fatigue Crack Growth
ABSTRACT: In the present paper, generalized Paris and Wöhler equations

are derived according to dimensional analysis and incomplete similarity con-
cepts. They provide a rational interpretation to a majority of empirical power-
law criteria used in fatigue. In particular, they are able to model the effects of
the grain size, of the initial crack length, as well as of the size-scale of the
tested specimen on the crack growth rate and on the fatigue life. Regarding
the important issue of crack-size dependencies of the Paris’ coefficient C and
of the fatigue threshold, an independent approach, based on the application
of fractal geometry concepts, is proposed to model such an anomalous
behavior. As a straightforward consequence of the fractality of the crack
surfaces, the fractal approach provides scaling laws fully consistent with
those determined from dimensional analysis arguments. The proposed scal-
ing laws are applied to relevant experimental data related to the crack-size
and to the structural-size dependencies of the fatigue parameters in metals
and in quasi-brittle materials. Finally, paying attention to the limit points defin-
ing the range of validity of the classical Wöhler and Paris power-law relation-
ships, correlations between the so-called cyclic or fatigue properties are
proposed, giving a rational explanation to the experimental trends observed
in the material property charts.
KEYWORDS: S-N curves, fatigue crack growth, short cracks, dimensional
analysis, fatigue property charts
Manuscript received June 3, 2011; accepted for publication July 13, 2011; published
online August 2011.
1
Politecnico di Torino, Dept. of Structural Engineering and Geotechnics, Corso Duca
degli Abruzzi 24, 10129 Torino, Italy, e-mail: alberto.carpinteri@polito.it; marco.paggi@
polito.it
Cite as: Carpinteri, A. and Paggi, M., “Dimensional Analysis and Fractal Modeling
of Fatigue Crack Growth,” J. ASTM Intl., Vol. 8, No. 10. doi:10.1520/JAI104105.
Copyright V
C 2011 by ASTM International, 100 Barr Harbor Drive, PO Box C700, West
Conshohocken, PA 19428-2959.
3
ID: kumarva Time: 12:35 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

4 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS
Nomenclature
a¼ crack length (L)
d¼ microstructural dimension (grain size) (L)
da/dN ¼ crack growth rate (L)
D¼ fractal dimension (-)
E¼ elastic modulus (FL2 )
h¼ characteristic structural size (L)
N¼ number of cycles (-)
R¼ loading ratio (-)
DK ¼ stress-intensity factor range (FL3=2 )
DKth ¼ fatigue threshold (FL3=2 )
Dr ¼ stress range (FL2 )
Drfl ¼ fatigue limit (FL2 )
KIC ¼ fracture toughness (FL3=2 )
x¼ frequency of the loading cycle (T1 )
ry ¼ yield strength (FL2 )
Introduction
As admitted by Paris in a recent review [1], “a specific accumulation damage
model for the computation of damage growth under a wide variety of service
loads is still lacking” and “no computational model is entirely satisfactory
today,” although a general understanding of many aspects of fatigue crack
growth was established since the early 1960s. We know that fatigue damage
increases with applied cycles in a cumulative way, which may eventually lead to
failure. To model this physical phenomenon, the existing approaches for the
prediction of fatigue life can be distinguished in two main categories: those
related to the cumulative fatigue damage (CFD) approach, which is the tradi-
tional framework based on the Wöhler or S-N curves [2] for fatigue life assess-
ment, and those based on the fatigue crack propagation (FCP) approach,
developed since the 1960s after the advent of fracture mechanics and the intro-
duction of the Paris’ law [3,4]. In the empirical S-N curve, the fatigue life, N, is
related to the applied stress range, Dr or S, and a reasonable power-law approxi-
mation was discovered since 1910 by Basquin [5]. A schematic representation
of a typical Wöhler’s curve is shown in Fig. 1, where the cyclic stress range,
Dr ¼ rmax rmin , is plotted as a function of the number of cycles to failure, N.
The loading ratio is the ratio between the minimum and the maximum applied
stresses, R ¼ rmin =rmax . In this diagram, we also introduce the range of stress at
static failure, Dry ¼ rmax rmin ¼ ry rmin ¼ ð1 RÞry , where ry is the material
yield strength, and we define the endurance or fatigue limit, Drfl , as the stress
range that a sample will sustain without fracture for N1 ¼ 1 107 cycles, which
is a conventional value that can be thought of as “infinite” life. Fatigue criteria
based on the CFD approach have the advantage that can be used for the fatigue
life assessment of unnotched or welded specimens, but suffer from the signifi-
cant deficiency that there is no consistent definition of failure. It may

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 5
FIG. 1—Scheme of the Wöhler’s curves with the corresponding fatigue parameters.
correspond to the appearance of the first detectable crack, although it may also
be defined as when the actual failure of the structural component takes place.
With the advent of fracture mechanics, a more ambitious task was under-
taken, i.e., to predict, or at least understand, the propagation of cracks. Plotting
the crack growth rate, da/dN, as a function of the stress-intensity factor range,
DK ¼ Kmax Kmin , most of the experimental data can be well-interpreted in
terms of a power-law relationship, i.e., according to the so-called Paris’ law [3,4]
(see Fig. 2). Note that the power-law representation presents some deviations
for very high values of DK approaching DKcr ¼ ð1 RÞKIC [6,7], where KIC is the
material fracture toughness, or for very low values of DK approaching the thresh-
old stress-intensity factor range, DKth . Again, in close analogy with the concept of
fatigue limit, the fatigue threshold is defined in a conventional way as the value
of DK below which the crack grows at a rate of less than 1 109 m/cycle. The
main drawback of this approach relies in the fact that the Paris’ law is far from
providing a universal representation of fatigue, since several deviations have
been noticed in the last decades. Among them, the anomalous behavior of short
cracks is probably the most important aspect, which led to the development of
more complicated fatigue crack growth criteria (see, e.g., [8–14] for a compre-
hensive discussion).
For a long time, the CFD and the FCP approaches have been considered
as totally independent. The CFD criteria have been mainly confined to the
fatigue life assessment of unnotched or welded components, where the

FIG. 2—Scheme of the Paris’ curves with the corresponding fatigue parameters.
elastoplastic nature of damage, crack nucleation and crack initiation are impor-
tant aspects, whereas the FCP models have been mainly applied to the long-
crack regime, when the concept of small scale yielding holds and LEFM applies
reasonably well.
In the last few decades, the researchers have attempted to extend the field of
application of the FCP approach. Among the various efforts, it is worth men-
tioning the contribution by McEvily and co-workers [14], who proposed a modi-
fied Paris’ law dealing with the elastoplastic behavior of small cracks, and that
by Atzori et al. [15], who proposed a method for the fatigue life prediction of
welded joints based on the notch stress-intensity factor. The effect of surface
roughness was also modeled by Spagnoli [16,17] according to a fractal model,
and a unified interpretation of the anomalous scaling laws in fatigue due to
short cracks has recently been provided by Paggi and Carpinteri [18,19] accord-
ing to fractal geometry. These advances in understanding the complex phenom-
enon of fatigue crack growth shed a new light on the possibility to unify the
CFD and the FCP approaches, and to solve the challenging task of interpreting
the Paris and Wöhler power-law regimes within a unified theoretical frame-
work. A recent effort in this direction was given by Pugno et al. [20,21], who
proposed a generalized Paris’ law based on Quantized Fracture Mechanics for a
unified treatment of long cracks, short cracks and fully yielded regimes.

In the present paper, we extend the dimensional analysis approach pioneer-

ingly proposed by Barenblatt and Botvina [22,23] to derive generalized mathe-
matical representations of the phenomenon of fatigue. It will be shown that
such generalized representations cover almost all the main deviations from the
empirical fatigue laws of Wöhler and Paris.
Using an independent approach based on fractal geometry, we also show
that the incomplete similarity in the crack length represents the effect of the
multiscale fractal roughness of crack profiles. Related implications for the
fatigue threshold are discussed.
Finally, analytical correlations between the fatigue properties of engineer-
ing materials are determined on a theoretical basis and compared with the em-
pirical trends proposed by Fleck et al. [24], giving a rational interpretation to
the fundamental fatigue property charts.
Generalized Cumulative Fatigue Damage Formulation

Let us consider the number of cycles, N, as the parameter representative of
fatigue. Following this route, we can consider the following functional
dependence
N ¼ Fðry ; KIC ; Drfl ; DKth ; E; Dr; x; h; a; d; 1 RÞ (1)
where the definitions of the governing variables are summarized in the nomen-
clature list, along with their physical dimensions expressed in the length-force-
time class (LFT). Considering a state with no explicit time dependence, it is
possible to apply the Buckingham’s P Theorem [25] to reduce the number of pa-
rameters involved in the problem (see, e.g., [26–30] for some relevant applica-
tions of this method in Solid Mechanics). As a result, we have
!
2
Drfl DKth E Dr ry Dr2fl r2y
N¼W ; ; ; ; 2 h; 2
a; 2 d; 1 R ¼ WðPi Þ; i ¼ 1; :::; 8 (2)
ry KIC ry ry KIC DKth KIC
where W is a nondimensional function. The dimensionless number r2y h=KIC 2

is
proportional to the ratio between the structural size h and the critical process
zone size rp , since rp ¼ ðKIC =ry Þ2 =p according to Irwin. This number is responsi-
ble for the size-scale effects and it is proportional to the square of the number Z
introduced by Barenblatt and Botvina [22] and to the inverse of the square of
the brittleness number s introduced by Carpinteri [26]. The dimensionless num-
ber Dr2fl a=DKth
2
is responsible for the crack-size effects, and it is proportional to
the ratio between the crack length and the Haddad [8] characteristic size of
mechanically short cracks a0 ¼ ðDKth =Drfl Þ2 =p.
At this point, we want to see if the number of the quantities involved in the
relationship [2] can be reduced further from eight. This can occur either in the
case of complete or incomplete self-similarities in the corresponding dimension-
less numbers. In the former situation, the dependence of the mechanical
response on a given dimensionless number, say Pi , disappears and we can say

that Pi is nonessential for the representation of the physical phenomenon. In

the latter situation, a power-law dependence on Pi can be put forward, which
usually characterizes a physical situation intermediate between two asymptotic
behaviors. To this aim, we assume incomplete self-similarity in P4 , P5 , P6 , P7
and P8 , obtaining
a1 a2 a3 a4
Dr h a d
N¼ ð1 RÞa5 W ðPi Þ (3)
ry rp a0 rp
The exponents ai cannot be determined from considerations of dimensional analy-

sis alone and may depend on the nondimensional parameters Pi . Equation 3 repre-
sents a generalized Wöhler relationship of fatigue and encompasses the empirical
S-N curves as limit cases. For instance, the S-N curve in Fig. 1 can be approximated
by the Basquin power law in the high cycle fatigue (HCF) regime, stating that
1 Drny ¼ N1 Drnfl ¼ NDrn ¼ k (4)
where k is a constant. Equating the first and the third terms in Eq 4, we obtain
he following power-law equation
n
Dry n ð1 RÞ rny
N¼ ¼ (5)
Dr Drn
Comparing the generalized expression of the S-N curve in Eq 3 with the empiri-
cal one in Eq 5, we find that a perfect correspondence exists when a1 ¼ n, a2 ¼
a3 ¼ a4 ¼ 0 and a5 ¼ n. It is important to notice that the generalization of the
S-N curve including a power-law dependency on the crack size [31,32] and on
the grain size [33,34] permitted to better interpret the experimental trends.
Size-scale effects on the S-N curves are also observed in concrete, as shown in
Fig. 3. The increase in the size of the specimen leads to a lower fatigue life, for a
given applied stress-range.
Generalized Fatigue Crack Propagation Formulation

Following the pioneering work by Barenblatt and Botvina [22], we now assume
that the mechanical response of the system can be fully represented by the crack
growth rate, da/dN, which is the parameter to be determined. This output pa-
rameter is a function of a number of variables
da
¼ Fðry ; KIC ; Drfl ; DKth ; E; DK; x; h; a; d; 1 RÞ (6)
dN
where the governing variables are summarized in the nomenclature, along with
their physical dimensions expressed in the length-force-time class (LFT).
Considering a state with no explicit time dependence, it is possible to apply
the Buckingham’s P Theorem [25] to reduce the number of parameters involved
in the problem. As a result, we have

FIG. 3—Size-scale effects on the S-N curve (experimental data from [35,36]).
2 !
2
da KIC Drfl DKth E DK ry Dr2fl r2y
¼ U ; ; ; ; 2 h; 2
a; 2
d; 1 R ¼
dN ry ry KIC ry KIC KIC DKth KIC

KIC 2 Drfl DKth E DK h a d
¼ U ; ; ; ; ; ; ;1 R (7)
ry ry KIC ry KIC rp a0 rp
where the Pi dimensionless numbers have been rewritten using the same nota-
tion as in the Generalized Cumulative Fatigue Damage Formulation section. At
this point, we want to see if the number of quantities involved in the relation-
ship [7] can be reduced further from eight. Considering the nondimensional
number P1 ¼ DK=KIC , it has to be noticed that it rules the transition from the
asymptotic behavior characterized by the condition of nonpropagating cracks,
when DK ! DKth , to the pure Griffith-Irwin instability, when DK ! DKcr . More-
over, incomplete self-similarity in P1 would correspond to a power-law depend-
ence of the crack growth rate on the stress-intensity factor range, which is
experimentally confirmed by the Paris’ law [3,4]. Therefore, complete self-
similarity in P4 cannot be accepted, whereas incomplete self-similarity gives
2 b1
da KIC DK
¼ U ðPi Þ (8)
dN ry KIC
where the exponent b1 and the nondimensional function U cannot be deter-

mined from considerations of dimensional analysis alone. Incomplete self-
similarity can also be assumed for the nondimensional numbers P5 , P6 , P7 ,
and P8 , obtaining the following generalized representation of fatigue crack
growth

! b2 b3 b4

2b1
da KIC h a d
¼ DK b1 ð1 RÞb5 U (9)
dN r2y rp a0 rp
The experimentally observed deviations from the simplest power-law regime

suggested by Paris (da=dN ¼ CDK m ) is therefore the result of incomplete self-
similarity which gives us the following expressions for the Paris’ law parameters
m and C
m ¼ b1
!
b b
2m
KIC h 2 a b3 d 4
C¼ ð1 RÞb5 U (10)
r2y rp a0 rp
This generalized mathematical representation encompasses several improved

versions of the Paris’ law proposed in the past to cover specific anomalous devi-
ations from the simplest power-law regime suggested by Paris. For instance, as
far as the grain-size dependence of C is concerned, it has recently been demon-
strated in [33,34] that the cycles to failure in many alloys is a decreasing func-
tion of the grain size, suggesting a power-law dependence of C on d as in Eq 10,
with an exponent b4 related to the parameters of the Hall-Petch relationship.
Modified Paris’ laws taking into account the effect of the crack length have
been proposed both for metals and quasi-brittle materials. For metals, several
researchers have questioned the validity of the similitude hypothesis, which
states that “two different sized cracks embedded into two different sized bodies
subjected to the same stress-intensity factor range should grow at the same
rate.” In this context, Molent et al. [37] and Jones et al. [38] have recently pro-
posed a generalized Frost and Dugdale [39] crack growth equation of power-law
type on a.
Finally, as far as the loading ratio is concerned, several Authors have pro-
posed to include in the fatigue crack growth criterion both R and DK on an em-
pirical basis [40–42]. They obtained the so-called “two-parameters” formulations
with an exponent b5 less than zero, confirming the experimental evidence that
the crack propagation rate is an increasing function of the loading ratio.
The Anomalous Crack-Size Dependency of The Paris’ Law:

An Interpretation According to Fractal Geometry
As shown in the previous section, a crack-size dependency of the Paris’ law cor-
responds to the incomplete similarity in the dimensionless number a=a0 . In this
section, we demonstrate that this phenomenon can be ascribed to the multiscale
fractal roughness of crack surfaces. An early application to fatigue of the inno-
vative concepts of fractals and multifractal measures, introduced by Mandel-
brot in [43], can be traced back to the work by Williford [44,45]. He modeled
the fracture surfaces near the crack tip as an invasive fractal and proposed a
modified Paris’ law where both the Paris’ parameters are functions of the sur-
face fractal dimension. In the 1990s, experimental evidences in [46] and [47]

pointed out a dependence of the crack growth rate on the specimen size, i.e., a
size effect on fatigue crack growth. Thus, exploiting the renormalized quantities
related to fractal cracks (whose surfaces can be modeled as invasive fractals
according to Carpinteri [48,49]), Spagnoli [17] proposed the following size-
independent fatigue crack growth law
da
¼ CðDK Þm (11)
dN
where:
a ¼ a1þdG
DK ¼ DKaðdG =2Þ
and dG is a parameter related to the fractal dimension of the invasive rough

crack profile, D ¼ 1 þ dG . A scaling law can be obtained by rewriting Eq 11 in
terms of the nominal crack propagation rate, da/dN, and the nominal stress-
intensity factor range, DK
da C
adG ð1þ 2 Þ DK m
m
¼ (12)
dN 1 þ dG
This model was referred to as monofractal approach to size effect on fatigue

crack growth [17,50]. Comparing Eq 12 with Eq 10, we note that the incomplete
similarity exponent b3 can be theoretically related to the fractality of the crack
profiles, i.e., b3 ¼ dG ð1 þ m=2Þ.
The use of a multifractal approach was also suggested in [17,50] to model
the propagation of cracks over a wider size range. Recently, Paggi and Carpin-
teri [18] have proposed a multifractal scaling law for the Paris’ law parameter C,
as an interpolating function between the asymptote for short cracks, where
dG ! 1=2, and that for long cracks, where dG ! 0, see also Fig. 4
a0 12ð1þm2 Þ
CMF ðaÞ ¼ C 1 þ (13)
a
Equation 13 has also related consequences on the crack-size dependency of the

threshold stress-intensity factor range. In fact, inverting Eq 13 in correspondence
of a conventional crack growth rate corresponding to infinite life, vth, we have
1=m
vth 1 a0 12ð12þm1 Þ
DKth ¼ ¼ DKth 1þ (14)
CMF a
1
where DKth is the value of the fatigue threshold for long cracks, see Fig. 5.
Considering the data collected in [51], an experimental assessment of
Eq 14 is proposed in Fig. 6 for different metals. By performing a nonlinear

FIG. 4—Multifractal scaling law for the Paris’ law parameter C.
regression analysis on the experimental data, the value of a0 and the exponent
of the multifractal scaling law are determined. The characteristic length a0
ranges from 1–10 lm for very high strength steels to 100–1000 lm for very low
strength steels. The exponent 1=2ð1=2 þ 1=mÞ of the scaling law [14] ranges
from 0.33 to 0.48.
FIG. 5—Multifractal scaling law for the fatigue threshold.

FIG. 6—Experimental assessment of the proposed multifractal scaling law for the
fatigue threshold (experimental data taken from [51]).
Analytical Correlations Between the Fatigue Properties

of Engineering Materials
Let us consider the limit points of the Paris’ curve defining the range of validity
of the power-law approximation relating the crack growth rate, da/dN, to the
stress-intensity factor range, DK. They correspond, respectively, to the points
with horizontal coordinates equal to the fatigue threshold, DKth , and to the frac-
ture instability limit DKcr . In this range, the Paris’ curve is usually defined in
terms of the parameters C and m.
Now, let us consider the construction added with dashed line to Fig. 2, as
proposed by Fleck et al. [24]. If a tangent is drawn at the midpoint of the central

linear region of the curve and extrapolated, it is found empirically that it inter-
sects the vertical line DK ¼ DKth in correspondence to a crack growth rate of
approximately vth ¼ 1 109 m/cycle, and it intersects the line DK ¼ DKcr ¼
ð1 RÞKIC at about vcr ¼ 1 105 m/cycle. Evaluating the Paris’ law in corre-
spondence to the latter point, the following correlation between the parameters
C and m of the Paris’ curve can be obtained
vcr
C¼ (15)
½ð1 RÞKIC m
Repeating this reasoning for the point defined by the fatigue threshold, we have
vth
C¼ (16)
ðDKth Þm
Equating 15 to 16, we express the Paris’ law parameter m as a function of the fa-
tigue properties
logvth logvcr
m¼ (17)
logDKth log½ð1 RÞKIC
or, in a bilogarithmic form

DKth 1 v
log ¼ logð1 RÞ þ log th (18)
KIC m vcr
Equation 18 establishes a correspondence between DKth , KIC and m in the long-

crack regime and was experimentally confirmed by Fleck et al. [24] for a wide
range of materials. Considering the fatigue property chart reported in Fig. 7(a),
we observe a very good agreement between the experimental trend and the pro-
posed correlation, being R ¼ 0 and logðvth =vcr Þ ffi logð1 109 =1 105 Þ ¼ 4.
A relationship between the fatigue stress-intensity factor threshold and the
fatigue limit can be derived by considering the propagation of a Griffith crack of
length 2a0 in an infinite elastic plate subjected to cyclic loading with Dr ¼ Drfl
acting at infinity and R ¼ 0. The initial crack length is representative of the size
of the existing microdefects, i.e., a0 ¼ ðKIC =ry Þ2 =p. If Dr ¼ Drfl , then DKth ¼
pffiffiffiffiffiffiffiffi
Drfl pa0 and the life of the specimen would tend to infinity. On the other hand,
when the applied load Dr ¼ ry , the stress-intensity factor at the crack tip
reaches the fracture toughness: KIC ¼ ry pa0 . Eliminating a0 from the previous
equations we obtain an important relation
DKth ¼ Drfl pa0 DKth Drfl
pffiffiffiffiffiffiffiffi ) ¼
KIC ¼ ry pa0 KIC ry
that can be rewritten as follows
2KIC Dr
logDKth ¼ log þ log fl (19)
ry 2

FIG. 7—Interpretation of the material property charts adapted from [24].

A direct comparison between this correlation DKth versus Drfl /2 and the experi-
mental trend observed for a wide range of materials and collected in the fatigue
property chart by Fleck et al. [24] is proposed in Fig.7(b). A linear relation is cor-
rectly reproduced and the intercept depends on the ratio KIC =ry , which is pro-
portional to the square root of the critical process zone size. Engineering
ceramics present a lower value of KIC =ry as compared to steel alloys, and there-
fore their position in the diagram is shifted downwards.
Conclusions
The Wöhler and Paris curves were originally thought as “universal laws” in the
sense that they should have been able to provide a universal description of fatigue.
Actually, the experimentally observed deviations led to a proliferation of modified
fatigue criteria, very often represented by power laws. Therefore, if on the one
hand the research efforts were directed towards the extension of the original fields
of application of the Wöhler and Paris representations of fatigue, on the other
hand the fundamental problem of finding the link between the cumulative fatigue
damage and the fatigue crack propagation approaches remained largely unsolved.
In the present contribution, a dimensional analysis approach and the con-
cepts of complete and incomplete self-similarity have been applied to the Wöhler
and Paris’ curves. As a main conclusion, it has been shown that the large number
of power laws used in fatigue are the result of an incomplete self-similarity in
the corresponding dimensionless numbers. This gives a rational interpretation
to such empirically-based fatigue criteria, towards a unified description of fa-
tigue and a possible standardization. Special attention has been paid to the
anomalous crack-size dependencies of C and DKth , proposing a model based on
the fractality of fatigue crack paths. The results confirmed by experiments pro-
vide a way to estimate the incomplete similarity exponent of the crack length on
a theoretical basis and to link it to the fractal dimension of the crack profiles.
Finally, analytical correlations between the cyclic properties of engineering
materials have been established, providing a rational interpretation to the em-
pirical correlations existing in the Literature and to the well-known fatigue
property charts.
Acknowledgments
The financial support of the Italian Ministry of Education, University and
Research (MIUR) to the Project “Advanced applications of Fracture Mechanics
for the study of integrity and durability of materials and structures” within the
“Programmi di ricerca scientifica di rilevante interesse nazionale (PRIN)” pro-
gram for the year 2008 is gratefully acknowledged.
References
[1] Paris, P. C., Tada, H., and Donald, J. K., “Service Load Fatigue Damage – A Histori-
cal Perspective,” Int. J. Fatigue, Vol. 21, 1999, pp. S35–S46.

[2] Wöhler, A., “Über die Versuche zur Ermittlung über die Festigkeit von Achsen,” Z.
Bauwesen, Vol. 13, 1863, pp. 233–258.
[3] Paris, P., Gomez, M., and Anderson, W., “A Rational Analytic Theory of Fatigue,”
Trend Eng., Vol. 13, 1961, pp. 9–14.
[4] Paris, P., and Erdogan, F., “A Critical Analysis of Crack Propagation Laws,” J. Basic
Eng. Trans. ASME, Vol. 58D, 1963, pp. 528–534.
[5] Basquin, O. H., “The Exponential Law of Endurance Tests,” Proc. ASTM, Vol. 10,
1910, pp. 625–630.
[6] Forman, R. G., Kearney, V. E., and Engle R. M., “Numerical Analysis of Crack Propa-
gation in Cyclic-Loaded Structures,” ASME J. Basic Eng., Vol. 89, 1967, pp. 459–464.
[7] Carpinteri, A. and Paggi, M., “Self-Similarity and Crack Growth Instability in the Corre-
lation Between the Paris’ Constants,” Eng. Fract. Mech., Vol. 74, 2007, pp. 1041–1053.
[8] Haddad, M. E., Topper, T., and Smith, K., “Prediction of Nonpropagating Cracks,”
Eng. Fract. Mech. Vol. 11, 1979, pp. 573–584.
[9] Kitagawa, H. and Takahashi, S., “Applicability Of Fracture Mechanics to Very
Small Cracks or the Cracks in the Early Stage,” Proc. of Second International Con-
ference on Mechanical Behaviour of Materials, American Society for Metals, Metal
Park, OH, 1976, pp. 627–631.
[10] Kitagawa, H. and Takahashi, S., “Fracture Mechanical Approach to Very Small Fatigue
Cracks and to the Threshold,” Trans. Jpn. Soc. Mech. Eng., Vol. 45, 1979, pp. 1289–1303.
[11] Taylor, D., Fatigue Thresholds, Butterworths, London, 1981.
[12] Miller, K., “The Short Crack Problem,” Fatigue Fract. Eng. Mater. Struct., Vol. 5,
1982, pp. 223–232.
[13] Suresh, S. and Ritchie, R., “Propagation of Short Fatigue Cracks,” Int. Met. Rev.,
Vol. 29, 1984, pp. 445–476.
[14] Endo, M. and McEvily, A.J., “Prediction of the Behaviour of Small Fatigue Cracks,”
Mater. Sci. Eng. A, Vol. 460, 2007, pp. 51–58.
[15] Atzori, B., Lazzarin, P., and Meneghetti, G., “Fatigue Strength Assessment of
Welded Joints: From the Integration of Paris’ Law to a Synthesis Based on the
Notch Stress Intensity Factors of the Uncracked Geometries,” Eng. Fract. Mech.,
Vol. 75, 2008, pp. 364–378.
[16] Spagnoli, A., “Fractality in the Threshold Condition of Fatigue Crack Growth: An
Interpretation of the Kitagawa Diagram,” Chaos, Solitons Fractals, Vol. 22, 2004,
pp. 589–598.
[17] Spagnoli, A., “Self-Similarity and Fractals in the Paris Range of Fatigue Crack
Growth,” Mech. Mater., Vol. 37, 2005, pp. 519–529.
[18] Paggi, M. and Carpinteri A., “Fractal and Multifractal Approaches for the Analysis
of Crack-Size Dependent Scaling Laws in Fatigue,” Chaos, Solitons Fractals, Vol.
40, 2009, pp. 1136–1145.
[19] Carpinteri, A. and Paggi M., “A Unified Fractal Approach for the Interpretation of
the Anomalous Scaling Laws in Fatigue and Comparison with Existing Models,”
Int. J. Fract., Vol. 161, 2010, pp. 41–52.
[20] Pugno, N., Ciavarella, M., Cornetti, P., and Carpinteri, A., “A Generalized Paris’
Law for Fatigue Crack Growth,” J. Mech. Phys. Solids, Vol. 54, 2006, pp. 1333–1349.
[21] Pugno, N., Cornetti, P., and Carpinteri, A., “New Unified Laws in Fatigue: From the
Wöhler’s to the Paris’ Regime,” Eng. Fract. Mech., Vol. 74, 2007, pp. 595–601.
[22] Barenblatt, G. I. and Botvina, L.R., “Incomplete Self-Similarity of Fatigue in the
Linear Range of Fatigue Crack Growth,” Fatigue Fract. Eng. Mater. Struct., Vol. 3,
1980, pp. 193–202.
[23] Barenblatt, G.I., Scaling, Self-similarity and Intermediate Asymptotics, Cambridge
Univ. Press, Cambridge, UK, 1996.

[24] Fleck, N. A., Kang, K. J., and Ashby, M.F., “The Cyclic Properties of Engineering
Materials,” Acta Metall. Mater., Vol. 42, 1994, pp. 365–381.
[25] Buckingham, E., “Model Experiments and the Form of Empirical Equations,”
ASME Trans., Vol. 37, 1915, pp. 263–296.
[26] Carpinteri, A., “Size Effect in Fracture Toughness Testing: A Dimensional Analysis
Approach,” Analytical and Experimental Fracture Mechanics, Proceedings of an
International Conference, Roma, Italy, 1980, edited by G. C. Sih, M. Mirabile, Sijth-
off & Noordhoff, Alphen an den Rijn, 1981, pp. 785–797.
[27] Carpinteri, A., “Static and Energetic Fracture Parameters for Rocks and Con-
cretes,” RILEM Mater. Struct., Vol. 14, 1981, pp. 151–162.
[28] Carpinteri, A., “Notch Sensitivity in Fracture Testing of Aggregative Materials,”
Eng. Fract. Mech., Vol. 16, 1982, pp. 467–481.
[29] Carpinteri, A., “Plastic Flow Collapse Vs. Separation Collapse in Elastic-Plastic
Strain-Hardening Structures,” RILEM Mater. Struct., Vol. 16, 1983, pp. 85–96.
[30] Carpinteri, A., “Strength And Toughness in Disordered Materials: Complete and
Incomplete Similarity,” Size-Scale Effects in the Failure Mechanisms of Materials
and Structures, Proc. IUTAM Symposium, Torino, Italy, 1994, edited by A. Carpin-
teri, E & FN SPON, London, 1996, pp. 3–26.
[31] Ciavarella, M. and Monno, F., “On the Possible Generalizations of the Kitagawa-
Takahashi Diagram and of the El Haddad Equation to Finite Life,” Int. J. Fatigue,
Vol. 28, 2006, pp. 1826–1837.
[32] Ciavarella, M., Paggi, M., and Carpinteri, A., “One, No One, and One Hundred
Thousand Crack Propagation Laws: A Generalized Barenblatt and Botvina Dimen-
sional Analysis Approach to Fatigue Crack Growth,” J. Mech. Phys. Solids, Vol. 56,
2008, pp. 3416–3432.
[33] Chan, K., “Scaling Laws for Fatigue Crack Growth of Large Cracks in Steels,” Met-
all. Trans. A, Vol. 24, 1993, pp. 2473–2486.
[34] Plekhov, O., Paggi, M., Naimark, O., and Carpinteri, A., “A Dimensional Analysis
Interpretation to Grain Size and Loading Frequency Dependencies of the Paris and
Wöhler Curves,” Int. J. Fatigue, Vol. 33, 2011, pp. 477–483.
[35] Murdock, J. W. and Kesler, C.E., “Effect of Range of Stress on Fatigue Strength of
Plain Concrete Beams,” ACI J., Vol. 55, 1959, pp. 221–232.
[36] Zhang, J. and Stang, H., “Fatigue Performance in Flexure of Fiber Reinforced Con-
crete,” ACI Mater. J., Vol. 95, 1998, pp. 58–67.
[37] Molent, L., Jones, R., Barter, S., and Pitt, S., “Recent Developments in Fatigue
Crack Growth Assessment,” Int. J. Fatigue, Vol. 28, 2006, pp. 1759–1768.
[38] Jones, R., Molent, L., and Pitt, S., “Crack Growth of Physically Small Cracks,” Int.
J. Fatigue, Vol. 29, 2007, pp. 1658–1667.
[39] Frost, N. E. and Dugdale, D.S., “The Propagation of Fatigue Cracks in Sheet Spec-
imens,” J. Mech. Phys. Solids, Vol. 6, 1958, pp. 92–110.
[40] Roberts, R. and Erdogan F., “The Effect of Mean Stress on Fatigue Crack Propaga-
tion in Plates Under Extension and Bending,” ASME J. Basic Eng., Vol. 89, 1967,
pp. 885–892.
[41] Walker, K., “The Effect of Stress Ratio during Crack Propagation and Fatigue for
2024-T3 and 7075-T6 Aluminium,” Effects Of Environments And Complex Load His-
tory On Fatigue Life, ASTM STP, Philadelphia, PA, 1970, Vol. 462, pp. 1–14.
[42] Radhakrishnan, V. M., “Parameter Representation of Fatigue Crack Growth,” Eng.
Fract. Mech., Vol. 11, 1979, pp. 359–372.
[43] Mandelbrot, B., The Fractal Geometry of Nature, W. H. Freeman and Company, NY,
1982.
[44] Williford, R., “Multifractal Fracture,” Scr. Metall. Mater., Vol. 22, 1988, pp. 1749–1754.

[45] Williford, R., “Fractal Fatigue,” Scr. Metall. Mater., Vol. 24, 1990, pp. 455–460.
[46] Bažant, Z. P. and Xu, K., “Size Effect in Fatigue Fracture of Concrete,” ACI Mater.
J., Vol. 88, 1991, pp. 390–399.
[47] Baı̂zant, Z. P., and Shell, W., “Fatigue Fracture of High Strength Concrete and Size
Effect,” ACI Mater. J., Vol. 90, 1993, pp. 472–478.
[48] Carpinteri, A., “Fractal Nature of Material Microstructure and Size Effects on
Apparent Mechanical Properties,” Mech. Mater., Vol. 18, 1992, pp. 89–101, 1994, In-
ternal Report, Laboratory of Fracture Mechanics, Politecnico di Torino, N. 1/92.
[49] Carpinteri, A., “Scaling Laws and Renormalization Groups for Strength and
Toughness of Disordered Materials,” Int. J. Solid Struct., Vol. 31, 1994, pp.
291–302.
[50] Carpinteri, An., and Spagnoli, A., “A Fractal Analysis of Size Effect on Fatigue
Crack Growth,” Int. J. Fatigue, Vol. 26, 2004, pp. 125–133.
[51] Tanaka, K., “Fatigue Crack Propagation,” Comprehensive Structural Integrity,
edited by R. Ritchie, and Y. Murakami, Vol. 4, Cyclic Loading and Fatigue, Elsevier,
Amsterdam, 2003, pp. 95–127.


doi:10.1520/JAI103940
R. Sunder1
Unraveling the Science of Variable Amplitude

Fatigue
ABSTRACT: Conventional methods to estimate variable-amplitude fatigue

life revolve either around cumulative damage analysis using the local stress-
strain approach, or, around one of the crack growth load interaction models.
Despite advances in modeling the mechanics of fatigue, none of these meth-
ods can faithfully reproduce the near-threshold variable amplitude fatigue
response that determines the durability of machines and structures primarily
because they fail to model the science behind the residual stress effect. Re-
sidual stress effects have a strong bearing on metal fatigue and owe their
influence to the moderation of crack-tip surface chemistry and surface
physics. This demands the treatment of threshold stress intensity as a vari-
able, sensitive to load history. The correct estimation of crack closure is also
crucial to determining the variable amplitude fatigue response and demands
assessment of the cyclic plastic zone stress-strain response.
KEYWORDS: fatigue crack growth, variable-amplitude loading, crack
closure, residual stress
Introduction
Many complex phenomena of engineering significance including heat transfer,
stress/strain distribution in materials and built-up structures, their dynamic
response, and even fluid flow have been understood to a point where analytical
Manuscript received May 2, 2011; accepted for publication November 1, 2011; published
online December 2011.
1
BiSS Research, 41A 1A Cross, AECS 2nd Stage, Bangalore 560094, India, e-mail:
rs@biss.in
Presented at the 11th ASTM/ESIS Symposium on Fatigue and Fracture Mechanics, Ana-
heim, CA, USA, May 17-20, 2011. Submitted for publication in ASTM STP.
Cite as: Sunder, R., “Unraveling the Science of Variable Amplitude Fatigue,” J. ASTM
Intl., Vol. 9, No. 1. doi:10.1520/JAI103940.
Copyright V
20
ID: vasanss Time: 02:36 I Path: Q:/3b2/STP#/Vol01546/120220/APPFile/AI-STP#120220

SUNDER, doi:10.1520/JAI103940 21
and numerical modeling, practically from first principles, can simulate the
actual process with amazing consistency. In stark contrast, the science of metal
fatigue has remained largely empirical even after 150 years of intense study. In-
credible improvements have been effected in the safety and useful life of such
heavily stressed transportation vehicles such as aircraft and automobiles.
These were made possible to a large extent by advances in analytical techniques
related to stress-strain distribution in materials and structures under both
static and dynamic conditions, and in the area of materials engineering. The
quality of computer-aided design through solid modeling and finite element
analysis permits even less experienced engineers to ensure a uniform distribu-
tion of stresses and avoid localized stress concentration, so that adequate
safety factors can be provided without substantially increasing weight. Finally,
fracture mechanics combined with improvements in non-destructive evalua-
tion (NDE) allows “on-condition maintenance,” whereby structures and
machines can be periodically inspected and repaired or retired only if neces-
sary—“if NDE does not reveal a defect, the structure must be good till the next
inspection.”
A brief review of progress in understanding metal fatigue is made below in
an attempt to explain its enigmatic nature. This is followed by a description of
two major operative mechanisms that control variable-amplitude fatigue, crack
closure, and residual stress. The implications of the synergy of the two inde-
pendent phenomena are discussed. The paper concludes with a description of
new avenues for research that follow from the discovery of the science behind
the residual stress effect and improved crack closure measurement.
Metal Fatigue—A Chronological Brief
Crucial Early Observation—Railway engineers in the early 19th century

were shocked to discover that wagon axles made from high quality ductile
steel could inexplicably break like glass, even though operating stress levels
were far less than the tested static strength of these superior quality steels.
Thus, the same material would show a “fibrous” (ductile) fracture when it fails
statically and a “crystalline” (brittle) one when it fails under very long term
repeated loading of low magnitude [1]. This gave birth to the speculation
(‘theory’ at the time), that cyclic loading can induce metallurgical transforma-
tions even at ambient temperature, forcing local brittle failure along crystallo-
graphic planes. Steam from the locomotive flowing past axles was cited as one
possibility [2]. The present study proposes, in part, to show that while such
conclusions may seem delusive, the factual significance of the “crystalline”
appearance of high cycle fatigue fractures appears to have been overlooked for
too long.
Significance of Cyclic Loading—Wohler’s experiments in the mid-

nineteenth century opened up metal fatigue to engineering applications [3].
He established the concept of the S-N curve that relates fatigue life to the am-
plitude of cyclic loading. By performing tests at higher stress amplitudes,
Wohler showed that fatigue fractures could retain the “fibrous” appearance

associated with static fracture.2 He also established the idea of a fatigue limit
and its relationship with mean stress. In so doing, Wohler put in place the idea
of fatigue being sensitive to both the amplitude and mean level of cyclic load-
ing and also the machinery of empirical correlation that continues to serve as
the foundation of fatigue analyses. The significance of Wohler’s work must be
judged against the background of prevailing speculative interpretations of the
time along with the backdrop of the Industrial Revolution. Scientific advance
of the discipline came much later through its association with cyclic slip, as
summarised in Fig. 1. This perception served as virtual blinders, clouding for
more than a century, a pertinent but inconvenient question: if fatigue is indeed
driven by cyclic slip, why is fatigue life and particularly, fatigue limit, so sensi-
tive to mean stress?3 The link between cyclic plastic strain, reversed slip, and
dislocation dynamics appeared to hold much more promise given the nebu-
lous nature of the mean stress effect. Additionally, with the subsequent discov-
ery of crack closure (to which we will return), the mean stress effect also
appears to have been treated as effectively ‘closed.’
Cumulative Damage and Service Load Environment—Service loading typi-

cally involves a mix of cycles of varying magnitude and asymmetry, with the
largest load occurring extremely rarely in actual usage, if at all.4 Merely ensur-
ing that stresses due to the largest expected load do not exceed the fatigue limit
is an impractically safe design proposition except, perhaps, in civil structures.
The Miner Rule5 introduced in the early 20th century attempts to resolve this
problem by suggesting that the remaining life in a given variable-amplitude
load history undergoes a continuous cycle-by-cycle fractional decrement
expressed as the inverse of total fatigue life after each load cycle [13]. Thus, for
any given arbitrary load sequence, failure is associated with the sum of
2)
In commenting on Wohler’s collection of laboratory fatigue fractures displayed at the
Paris Exhibition in 1867, Anon. prophetically observed “M. Wohler’s modest exhibition
may have been overlooked by ninety nine out of a hundred professional visitors to the Ex-
hibition, yet we believe ourselves justified in saying that his scientific and patient experi-
ments will be referred to long after the majority of those things which have drawn a
shower of medals and ribbons upon themselves at present will be dismissed and for-
gotten” [4]. Indeed, in terms of value, Wohler’s lifetime effort appears formidable even
given today’s experimental resources. Just consolidating the results of his fatigue experi-
ments under a vast variety of conditions involving axial, shear, and torsional loading
would constitute a meaningful research effort.
3)
Particularly considering that cyclic slip is mean stress insensitive! From the published
literature, only Manson’s expression of hope that “a meaningful rationale for the mean-
stress effect would be a noteworthy achievement over the coming 25 years” [5] appears to
suggest awareness of the enigma surrounding an important but unresolved phenomenon.
4)
Examples are the occasional potholes for automobiles and turbulent weather for air-
craft. Careless driving over deep potholes and a flight straight into a storm may serve as
extreme design considerations.
5)
Though it is known this way, actually, the rule was proposed some 20 years earlier by
Palmgren in Europe.

SUNDER, doi:10.1520/JAI103940 23
FIG. 1—A brief on metal fatigue. (a) Typical fatigue test results obtained in Wohler’s
time [7] shown as tables of max applied stress (fully reversed in tension and compres-
sion by rotation-bending) versus cycles to failure. (b) Test results of Wohler and Baush-
inger for different steels showing that the fatigue limit is mean stress sensitive [8].
Many decades later, these came to be better known as the Goodman diagram [9]. (c) A
new understanding of fatigue emerged with the association of yield with dislocation
movement. Mott’s analog between slip and the ease of moving a fold in a carpet and
[10,11] helps explain the formation of persistent slip bands (PSBs) (d) [12]. This, in
turn, readily explains why fatigue life is controlled by the plastic strain range (e). (f)
Cycles A, B, and C, being identical in magnitude, will cause the same extent of reversed
slip or cyclic plastic strain. They ought to result in the same fatigue life, but do not, as
shown by Wohler and Bauschinger in (b). This has been an enduring enigma surround-
ing metal fatigue.

cumulative fractional damage from successive load cycles attaining unity. The
idea of cumulative damage is purely notional, carries no scientific rationale,
and is not associated with any entity that could be monitored in real time.
Nevertheless, it held out the promise of practical application in designing for
desired finite life, such as the warranty period for non-safety critical engineered
products. Any such optimism was soon dashed by Gassner’s experiments under
multi-step programmed block loading [14]. He established that the actual dam-
age sum at failure can fluctuate wildly, depending on the mix of programmed
loads, i.e., that fatigue damage is not linearly cumulative. In the tumultuous
years preceding WWII, Gassner proceeded to develop empirical procedures
involving testing under a simulated service environment, in order to obtain fa-
tigue life curves valid for a given material, component, joint, or even structural
assembly, subject to the statistical equivalent of a given service load history.
Thus, while Gassner’s effort did finally come up with an engineering solution, it
did so without casting any light on why metal fatigue is so sensitive to load
sequence. Continued emphasis on laboratory testing under a simulated service
environment underscores the significance of load sequence sensitivity. In the
meantime, some four decades after Gassner experiments, the first analytical ba-
sis to account for it emerged in the form of the local stress-strain (LSS)
approach.
Local Stress-Strain Approach

Figure 2 summarises the LSS approach that is based on the principle that notch
fatigue response will be the same as smooth specimen fatigue response to the
simulated notch root stress-strain response. Due to the hysteretic6 nature of the
notch root inelastic stress-strain response, local tensile yield during an overload
will cause a downward shift in the local stress response to subsequent elastic
loading. Assuming that fatigue is a localized phenomenon, it would follow that
accounting for sequence sensitivity of metal fatigue hinges on the capability to
simulate the notch root inelastic response and then translate that response into
local stress-strain cycles, identifiable for the purpose of a cumulative fatigue
damage estimate after correcting for sequence sensitive local mean stress. The
LSS approach is built around several important advances in applied mechanics.
Neuber came up with a simple equation that relates remote elastic loading to
local inelastic stress-strain at a notch root subject to shear [15]. This was
assumed to be extendable to the axial stress-strain response. A simultaneous so-
lution of Neuber’s equation with the Ramberg-Osgood equation [16] yields the
local inelastic stress-strain response to a given applied load. In the late 1960s,
6)
Deviation from linear response due to yield imposes hysteresis upon load reversal. As a
consequence, local stress and strain at any point of time need not be uniquely related to
applied load. They will become sensitive to load history and also to the direction of the
load change. Quite simply, hysteresis induces either reduced local stress at the cost of
increased local strain, or vice versa.

SUNDER, doi:10.1520/JAI103940 25
FIG. 2—Fatigue damage caused by the two sequences shown in (a) would appear simi-
lar, gauging from the smooth specimen elastic response in (b). However, if the two
sequences are applied on a notch root seeing the local inelastic response as in (c), the
local mean stress in cycles B and E will be dissimilar. Thus, if Miner’s Rule appeared to
apply to (b), it needs to be adapted to (c) by accounting for load sequence sensitivity of
the notch root mean stress. (d) and (e) Local Stress Strain (LSS) approach serves as the
foundation of contemporary industrial fatigue design. It incorporates (d) Neuber con-
version based on the Masing model of material stress-strain memory [17,18], (e) Rain-
flow cycle counting to determine closed fatigue cycles, (f) damage estimates using
strain-life data and Miner’s Rule. In practice, case (b) also exhibits load sequence sensi-
tivity, rendering the LSS approach questionable.

Wetzel [17] employed the emerging power of digital computers to combine a

linearized Masing model representing material memory effect in stress-strain
response7 with the Neuber equation into a numerical model, capable of realisti-
cally simulating the notch root cyclic inelastic response to an arbitrary applied
load sequence. This made it possible, for the first time, to visualize the effect of
load history in inducing changes to notch root residual stress and thereby
account for its effect on fatigue damage [18]. Around the same time, Endo [19]
came up with the Rainflow cycle counting technique to identify closed fatigue
cycles from an arbitrary random sequence of peaks and valleys, which is typical
of the service load environment.8
The early 1970s finally saw the emergence of a numerical apparatus built
around the Neuber conversion, the Masing model, Rainflow, and cumulative
damage estimates to calculate notch fatigue life. A timely addition to fatigue
technology in the 1960s were computer controlled servo-hydraulic testing
machines. They permitted the determination of cyclic stress-strain characteris-
tics for use in modeling the material response. They also permitted testing
under both total strain and plastic strain control, so as to obtain strain-life data
under highly controlled conditions.
The LSS apparatus was amenable to variations in terms of equations to cal-
culate damage and correct it for sequence-sensitive mean stress. It was also
open to sophistication in terms of accounting for strain hardening and soften-
ing, stress relaxation, and creep-fatigue interaction.9 Continuous advancement
in computing power combined with its integration with finite element analyses
now permit the digital simulation of the cyclic stress-strain response at hot
spots in a structure for design optimization and durability assurance. Such soft-
ware packages form the backbone of contemporary industrial fatigue design.
Even so, fatigue critical components are released into the market only after first
testing their durability and structural integrity in the laboratory under simu-
lated service conditions.
The continued need for component-level testing may not merely be a mea-
sure of insurance against the unexpected, but an acknowledgment of the
7)
The stress-strain curve of a material can be divided into a number of linear segments.
Metals have this amazing property to remember exactly “how much” they have deformed
along each linear segment and, therefore, how much more they can afford to deform
along the same segment. Thus, having exhausted one, their response will move on along
the next segment and so on. By simulating this response, one can digitally simulate a
tension-compression stress-strain response in a manner that will be remarkably similar
to that of real materials.
8)
The salient feature of Rainflow is its physical consistency. Rainflow counted cycles will
always correspond to fully closed stress-strain hysteresis loops required to estimate cu-
mulative fatigue damage. Previous cycle counting techniques did not carry a physical
basis.
9)
This opened the opportunity for the research community to come up with fairly diverse
ways of computing damage through a variety of corrections employed to suit observed
empirical results, while essentially using the same technique to compute inputs in the
form of local stress and strain.

SUNDER, doi:10.1520/JAI103940 27
unknown with regard to variable amplitude fatigue. This possibility is under-

scored by a serious shortcoming of the LSS approach, as illustrated in Fig. 3.
For all its sophistication, even the most modern machinery of notch fatigue sim-
ulation cannot explain sequence sensitivity under a fully elastic notch root
response. Designers strive to ensure that local stresses never exceed yield. This
effectively implies that if machines and structures respond in real life the way
they do in simulation, there will be no local inelasticity.10 Experience shows
however, that while the notch root stress-strain response in real life may remain
elastic and therefore, sequence insensitive, sequence effects, in fact, become
more significant with reducing overall stress level. This serious anomaly appears
to have remained largely unnoticed in the shadow of the elegance of numerical
simulation.
Limitations of the LSS approach should not come as a surprise. In scientific
terms, advances over what Wohler had originally conceived some 150 years ear-
lier were restricted to the newfound ability to accurately determine the local
stress strain response at fatigue critical locations. Note that local stress and
strain amplitude is load sequence independent.11 Their estimation does not
actually require the elaborate cycle-by-cycle numerical simulation provided by
state-of-the-art software. The only reason for resorting to cycle-by-cycle simula-
tion is to determine sequence sensitive local mean stress. If, indeed, this sensi-
tivity disappears under a fully elastic response, there must be other reasons for
metal fatigue being load sequence sensitive. The LSS approach elegantly han-
dles the mechanics of the notch root response, however. it fails to address the
science behind how such mechanics induce fatigue damage and, particularly,
why such damage may be sensitive to mean stress. Viewing fatigue as largely a
process of crack growth opens the possibility of resolving this problem (Fig. 4).
The impressive analytical machinery upon which the LSS approach is based
may indeed provide an accurate picture of the sequence-sensitive notch root
cyclic inelastic stress strain and cycle-by-cycle variation in residual stress under
service loading. However, fatigue crack growth consumes the bulk of total fa-
tigue life and unlike a notch root, the crack tip will, by definition, always see an
inelastic cyclic response. Thus, once a crack appears, sequence effects will not
only continue to prevail under the elastic notch root response, but may even
become dominant, given the nature of near-threshold crack growth sensitivity
to overloads. Obviously, one cannot hope to harmonize variable amplitude fa-
tigue test results obtained using the LSS and fracture mechanics approaches as
shown in Figs. 4(c) and 4(d).
10)
Note that cyclic inelasticity demands the exceedance of twice the yield stress, render-
ing it even more improbable in durable designs. However, even such designs often ulti-
mately fail in fatigue, suggesting that in real-life cracks can form and grow even in the
event of totally elastic notch root response.
11)
Local stress and strain amplitude are uniquely related to applied stress amplitude by
the Neuber and Ramberg-Osgood equations, stress concentration factor, Young’s modu-
lus, the strain hardening exponent, and cyclic strength coefficient. Applied mean stress
and mean strain do not figure in the relationship.

FIG. 3—(a) Computed fatigue life versus local elastic design stress using the LSS
approach for an airframe structural Al-alloy under typical fighter aircraft (FALSTAFF
[20]) and transport aircraft (TWIST [21]) load spectra [22]. The shaded area is the esti-
mated potential variation due to load sequence rearrangement. Note that curves for
both spectra merge into a single line below twice the yield stress (800 MPa), when cyclic
slip turns negligible. (b) Schematic notch root response for symmetric load spectrum,
and (c) response for asymmetric spectra such as FALSTAFF and TWIST. Even assum-
ing twice the yield strain at the highest load, only symmetric spectra such as rotating
parts seeing fully reversed loading are likely to experience cyclic inelastic conditions.
Others, as in (c) will not see cyclic inelasticity and, according to the LSS approach,
should not exhibit sequence sensitivity. However,in practice they do, and do so to a sig-
nificant extent, undermining the credibility of the LSS approach. Sequence effects obvi-
ously have to do with the nature of fatigue crack growth. Crack tip response will always
be sequence sensitive because the crack tip will always see a cyclic inelastic response.

SUNDER, doi:10.1520/JAI103940 29
FIG. 4—(a) Fatigue as a crack growth process. Advances in non-destructive inspection

technology are likely to increase demands on the ability to model the growth of smaller
cracks at lower growth rates. (b) Fractograph of natural crack formation and growth
under 3-step programmed loading in an Al-alloy out of an inclusion seen at bottom left.
Each band corresponds to 2000 cycles and is indicative of the reproducibility of the fa-
tigue crack growth process even at small crack size and low growth rates [25]. (a) and
(b) Are suggestive of fatigue as a crack growth process, sensitive to crack tip cyclic
response, rather than of cumulative damage at the notch root. (c) and (d) Range and
damage exceedance (RDE) curves computed for Al-alloy L73/2014-T6 under FALSTAFF
and TWIST load spectra [26]. 1—Rainflow counted cycle range; 2—damage contribu-
tion calculated using the LSS approach at 800 MPa (see Fig. 3(a)), and contribution to
fatigue crack extension for a small crack [3] and long crack [4]. Note that in FAL-
STAFF, just 10% of the cycles (the largest) contribute in excess of 90% of the damage.
This explains why the MiniFALSTAFF and FALSTAFF spectra yield similar results. On
the contrary, in the case of the TWIST spectrum, the LSS and fracture mechanics
approach provide contradictory results, with the former wrongly indicating that just
some 2% of the cycles contribute all the damage, while in actual experience, the smaller
cycles control damage. As shown by curves 3 and 4, when small cycles determine crack
growth, load interaction effects gain in importance. This underscores the significance of
the near-threshold behaviour and its potential load sequence sensitivity.

Modern fatigue critical structures including most airframes are periodically

inspected for cracks. If no cracks are observed, the structure is released for fur-
ther service until the next scheduled inspection. This implies indefinite usage,
provided cracks, if detected, are immediately repaired, or the part is replaced.
The cost of repair will eventually determine “retirement for cause” [23]. The
cost of inspection, along with its periodicity, will determine the overall econom-
ics of operation. In this scheme, the enforced periodicity of inspection is deter-
mined by the quality and reliability of non-destructive inspection (NDI), which
needs to be matched by the ability to correctly estimate the residual life of the
structure with such a crack. Obviously, neither the actual initial defect size
(assuming it is smaller than NDI-detectable size) nor the ability to correctly
model very early growth carry value in a condition monitoring scheme.
From the overall standpoint of durability assessment, understanding fa-
tigue crack growth response below NDI-detectable crack size becomes valuable
in the event there is a demand for an extended period of service before first inspec-
tion. It assumes even more importance when the component is not subject to
inspection. Additionally, it certainly offers the promise of just doing away alto-
gether with the obsolete concept of cumulative fatigue damage. The potential
for doing so is supported by the highly reproducible growth bands in Fig. 4(b)
even at incredibly small crack sizes.
As a rule, the quality of life estimate is inversely proportional to life [24].
Assuming the bulk of that life is exhausted by crack growth, the study of near
threshold variable-amplitude crack growth becomes extremely important.
Indeed, the potential for the advancement and application of fracture mechan-
ics in structural design over the last four decades has largely overshadowed
opportunities presented by the LSS approach.
Fracture Mechanics Approach

With the birth of linear elastic fracture mechanics, the stress intensity factor K
became available, that serves several important purposes. Here, K is, in effect, a
similarity criterion, to which both residual strength and fatigue crack kinetics
can be related (see Fig. 5). Paris showed that the fatigue crack growth rate da/
dN correlates with the cyclic stress intensity range DK [27]. This was a turning
point in the advancement of fatigue research. In contrast to a notional parame-
ter called cumulative damage, a quantifiable parameter in the form of crack size
was now available to characterize damage. Further, K permits the unification of
experimental data for a given material, irrespective of cracked body geometry,
crack size, shape, and applied load level. In effect, K is to a cracked body what
stress is to a smooth uniform section specimen. Using K, experimental crack
growth data obtained on simple laboratory coupons could be readily extrapo-
lated to structural components of engineering interest.
Crack Growth Load Interaction Models

The 1960s saw much progress in unraveling the mystery behind the load
sequence effect researched forty years earlier by Gassner that had debunked the

SUNDER, doi:10.1520/JAI103940 31
FIG. 5—Stress intensity factor K as a similarity criterion for fatigue crack growth. (a)
Stress intensity for crack subject to uniform remote stress [1] increases with crack size
which is the inverse of the case of rivet (point) load [2]. Correspondingly, the growth
rate, da/dN will also vary differently with crack size. Yet, as shown in (b), da/dN for the
two cases will fall into a single scatter band when plotted against the stress intensity
range [28]. Experience shows, however, that the relationship (b) combined with K are
not sufficient similarity criteria for engineering applications. Consider the schematic of
the loads in (c) on a transport aircraft at A—take-off and climb, B—cruise, and C—
descent and landing (load level on a transport liner gradually drops due to mass reduc-
tion from fuel consumption). Crack growth curves will vary as shown in (d), depending
on the mere rearrangement of loads [29]. Cycles covering a few thousand flights and re-
arranged to form a Hi-Lo programmed sequence will yield a crack growth life about
four times greater than if applied as is. This is attributed to load interaction mecha-
nisms including crack closure, residual stress, and crack front incompatibility.

Palmgren-Miner Rule. The advent of precision servo-hydraulics based test sys-

tems allowed systematic experiments on variable amplitude fatigue crack
growth. These permitted the study of crack growth rate transients after over-
loads and underloads superposed on baseline constant amplitude loading.
Experiments came up with the astonishing finding that applying a tensile over-
load, in fact, ends up retarding further crack growth even if the crack would have
substantially incremented during the overload. It was also found that compres-
sive overloads (inappropriately called “underloads”) could, in effect, erase the
retarding effect of a previous overload. These observations revealed that under
variable amplitude loading, the order in which different loads are applied influ-
ences the rate of crack advance in a manner that could not be readily explained
by considerations of solid mechanics (see Fig. 5(c) and 5(d)). Clearly, the mate-
rial at the crack tip appeared to “remember” what previously transpired in a
manner that affected its subsequent fatigue resistance. The search was on for
load interaction mechanisms that may be responsible for sequence effects.
Wheeler [30] and Willenborg [31] came up with empirical models on the
consideration that the tensile monotonic plastic zone ahead of the crack tip will
act as a wedge squeezed by the elastic matrix to create a zone of compressive re-
sidual stresses at the crack tip (see Fig. 6). If an overload is applied, this plastic
zone will increase in size as a square function of the overload ratio, leading to a
substantial increase in the near-tip compressive stress. To account for this
effect, Wheeler introduced a transient retardation factor as a power function of
the ratio of remaining crack extension in the overload plastic zone to the size of
this zone with constants empirically selected to approximate experimental
observations. Willenborg interpreted the same effect in terms of a reduced
“effective” stress ratio due to increased compressive residual stress, also with a
transient function to fit real observations. This model relies on Walker’s equa-
tion correcting the growth rate for the stress ratio [32].
If, in the 1970s, the LSS approach was already incorporated into commer-
cially available industrial software for fatigue design, the Wheeler and Willenborg
models were also brought into the market for the safe-life and fail-safe design of
aircraft structures and later, into the nuclear, piping, energy, railroad, automo-
tive, and other industries. Forty yearslater, software built around these models
continues to dominate industrial fatigue design. Even so, safety critical designs
are invariably tested in the laboratory under simulated service conditions.
Fatigue Crack Closure

Just when it seemed that the Wheeler and Willenborg models appeared to hold
promise in application, if not in scientific conviction, Elber’s [33] discovery of
crack closure (Fig. 6(e)) finally developed a mechanism that actually makes sci-
entific sense and can be analytically modeled using fracture mechanics con-
cepts. Newman [34], de Koning [35], and others came up with numerical
models of how the plastically stretched wake behind the crack tip effectively
closes even under tensile load. This was a milestone in the analytical simulation
of the mean stress effect in metal fatigue. What is more, the new approach was
able to simulate, with reasonable conviction, the consequences of tensile and

SUNDER, doi:10.1520/JAI103940 33
FIG. 6—General scheme of load interaction models in current use. The action of a ten-
sile overload (a) is described in (b)-(d). A is the monotonic plastic zone from baseline
loading and B, the cyclic plastic zone. C is the overload plastic zone and D, the cyclic
plastic zone due to overload, that vanishes upon the next tensile cycle. E is the crack
wake zone squeezed into bearing by the surrounding stretched material from the plastic
zone. (b) Indicates the crack tip picture upon the application of tensile overload. (c)
Shows the picture when the crack is almost through the overload plastic zone, and (d)
indicates crack tip growing through overload stretched wake. (e) Crack tip response to
load sequence 1-5, shown in the inset. Laser interferometry [36] estimates over
0.15 mm gauge length after deducting the elastic response. The loop shape unambigu-
ously underscores the portion of load cycle when the crack was open. Also note that clo-
sure is cycle sequence insensitive (2,4 and 1,5 indicate similar closure level). This is
proof that closure is insensitive to the cyclic plastic zone response (to crack-tip residual
stress). According to both the Wheeler and Willenborg models, compressive stresses in
the overload plastic zone will retard crack growth until the baseline monotonic plastic
zone begins to exit the overload plastic zone, as in (c). Using Elber’s closure model, re-
tarded growth will persist for some distance beyond the overload plastic zone (d). Nei-
ther the Wheeler/Willenborg nor the closure models can explain the possible differences
in crack extension between cycles 2,4 and 1,5. In fact, the first two actually model clo-
sure, even if they may profess to model the residual stress effect!

compressive overloads.12 Clinching evidence appeared by way of the ability of the

closure model to explain accelerated crack growth after a step-wise increase in
load and the finer aspect of delayed retardation after an overload. There was no
way for the Wheeler and Willenborg models to explain such behaviour. Closure
considerations make it obvious and simple. When a tensile overload is applied, it
takes some crack growth for the overload induced wake with extra stretch to take
effect. Therefore, retardation is not immediate. In fact the crack may even mo-
mentarily accelerate because the the overload itself opens up the crack, causing a
reduction in closure stress. However, when a compressive overload is applied the
consequent reduction in closure stress is immediate. Crack closure based models
were thus able to simulate, through mechanics based computations, many seem-
ingly complex load sequence effects that had hitherto appeared inexplicable. In
crack closure, a scientific explanation at long last seemed available for the effect
of both mean stress (stress ratio) and residual stress. All other load interaction
mechanisms appeared either insignificant, were perhaps manifested through clo-
sure, or, an outright figment of imagination. Or so it seemed.
The 1970s and 1980s saw the publication of over a thousand papers related to
crack closure. The bandwagon soon became an overcrowded train, with individual
coaches representing the variety of sources of crack closure. As it were, Elber’s dis-
covery was “merely” of plasticity induced closure. To this were added oxide-induced
closure, roughness-induced closure, and asperity-induced closure. It was then sug-
gested that closure is but one shielding mechanism for a fatigue crack, with the fur-
ther division of shielding into extrinsic and intrinsic. Therefore, closure was now
bracketed with crack tip shielding mechanisms such as uncracked fibres in the crack
wake, or, higher stiffness fibres ahead of it. As a consequence, if everything seemed
simple and straightforward as illustrated by Elber’s early work, a much more com-
plex and confusing picture seemed to emerge from subsequent research.
The cause of closure has not been helped by an unfortunate aspect of its mea-
surement. Unlike parameters that can be directly measured, such as dimensions or
weight, or at least by an easy to define and strictly reproducible process such as mod-
ulus of elasticity, yield stress, or ultimate stress, crack closure measurement carries a
heavy measure of interpretation. An annexure to ASTM E647 with a recommended
practice for closure measurement is a good example of a technique that delivers
measurements of little practical value. Remote measurement of crack opening dis-
placement representing contact response integrated way beyond intervals actually
affecting closure carries only a remote chance of correlation with an actual value.13
12)
The Wheeler and Willenborg models could not account for the effect of compressive
overloads.
13)
Closure induces a certain wedge opening stress intensity to compensate for the applied
stress falling below closure stress. The contribution to the stress intensity of a point force in
the crack wake will be inversely proportional to its distance from the crack-tip. Assuming com-
pressive yield stress upon wake contact, the depth of relevance to closure is of the order of a
monotonic plastic zone size. Displacement measurements made remote from this zone of
influence cannot be expected to sense the crack tip response with the desired sensitivity.
Indeed, there are no published data showing credible closure measurements under variable
amplitude loading.

SUNDER, doi:10.1520/JAI103940 35
The issue is further complicated by difficulties in mechanism isolation to eliminate

ambiguity in the interpretation of the results. For example, would it be fair to attrib-
ute retarded crack growth to roughness induced closure, when the very occurrence
of roughness may have also reduced the intensity of the crack tip stress field by a
ragged crack front and possible multiple plane separation?
The technique in Fig. 6(e) involving near-tip laser indentation interferome-
try14 [36] and fractography using the ‘Closure Block’15 [37] are exceptions that
deliver reproducible and scientifically defendable results. Unfortunately, these
are not amenable to easy implementation in routine engineering laboratory
measurements.
Load Sequence Sensitivity of Individual Crack Extension Mechanisms

We now proceed to analyse different stages of fatigue crack growth associated
mechanisms and how they may be affected by the variable-amplitude environ-
ment. Measurable fatigue crack growth rates range from less than atomic spac-
ing, right up to 1 mm/cycle, a potential variation of at least eight orders of
magnitude (Fig. 4(a)). There are not many phenomena of engineering relevance,
with such a wide swing in kinetics. Crack extension itself occurs in an environ-
ment of several competing mechanisms, with individual mechanisms dominat-
ing selected intervals of growth rate. Add to this the different ways in which
ambient conditions can affect individual mechanisms. It would, therefore,
come as a surprise if any single crack extension mechanism can describe the
process. Even more surprising would be a single load interaction model coming
up with consistent estimates of variable-amplitude crack growth rates. For
clarity, we broadly divide crack kinetics into three distinct ranges of the crack
growth rate and proceed to examine how the dominant crack extension mecha-
nism in each range responds to variable-amplitude loading. Before doing so, we
define a basic assumption that is required to distinguish fracture mechanics
based analysis from cumulative damage concepts.
History Effect on Crack Extension—Consider crack extension in identical

cycles A, B, and C shown in Fig. 7(a) with different loading histories. Case (a)
involves constant amplitude loading. Cases (b) and (c) involve prior cycling at
increased loading amplitude, causing greater near-tip cyclic slip. Based on cu-
mulative fatigue damage considerations, one should expect crack extension C to
exceed B and for both to exceed A due to greater “prior damage,” However,
there appears to be absolutely no empirical evidence to suggest such a possibil-
ity!16 Fracture mechanics based models of variable-amplitude fatigue, in fact,
14)
With a working gage length of the same order as the plastic zone size, this technique is
sensitive to the inelastic stress-strain response within the cyclic plastic zone as seen in
Fig. 6(e).
15)
The technique proceeds on the premise that given constant Kmax, there is no other ex-
planation for equal striation spacing under varying Kmin other than equal DKeff.
16)
Not necessarily because such a possibility does not exist, but rather, because of the lim-
itations in experimental techniques to address the question in quantifiable terms.

FIG. 7—(a) According to the cumulative damage concept, identical load cycles A, B,
and C may extend the crack differently because of the different load history preceding
each of them. In contrast, all crack growth models ignore the possibility of damage to
material ahead of the crack tip. This understanding is central to analytical modeling of
load history effects. (b) The three growth rate regimes and their associated fractures for
an Al-alloy. Crack extension in a cycle under variable amplitude loading may fall into
any of these three regimes, depending on its magnitude. (c) During the rising half cycle
shown in the inset, the crack will first extend by brittle micro-fracture (BMF) over a fi-
nite number of atomic layers embrittled by instantaneous surface diffusion (ii), and
then switch to shear extension (iii), suggesting striation formation by the mode change
(iv) [39]. Any further increase in load beyond 2 may induce a disproportionately higher
quasi-static crack extension. This explains why striations marking individual cycles are
seen only over a very narrow range of growth rate.

SUNDER, doi:10.1520/JAI103940 37
simply ignore it. They assume that the crack extension in the next cycle is driven
only by the magnitude of that cycle. The prevailing understanding of crack
growth load interaction effects is also based exclusively on variables that control
crack kinetics in the next load cycle. It ignores any prior “slip-reversal damage”
to the crack tip. In the absence of compelling arguments to the contrary, we
shall ignore any prior damage and its effect in considering dominant crack
extension mechanisms and how they respond to variable amplitude loading. In
doing so, we make an important assumption that the fatigue crack can extend
under each load cycle.17
Dominant Crack Extension Mechanisms—At the commencement of the rising

half of a new load cycle, the dominant crack extension mechanism is still an
unknown. Crack extension will commence by a yet to be defined mechanism once
the load excursion exceeds a certain threshold value. It will soon transform to stria-
tion mode as the stress intensity falls into the Paris regime and then proceeds to
extend through local quasi-static fracture in the event K approaches critical values
(see Fig. 7(b)). Each of these three stages occupies a finite but overlapping interval
of crack growth rates, with the first transition occurring around 104 mm/cycle
and the second one depending largely on the stress ratio, around 102 mm/cycle.
With the increasing stress ratio, this last transition will progressively move into
lower growth rates because of the onset of quasi-static fracture leading to a short-
ened Paris interval. Note that the different stages in crack growth are associated
with the change in growth rate over several orders of magnitude. Higher order
growth rates will necessarily be associated with a mix of mechanisms18 (see
Fig. 7(c)), though the last mechanism to switch-in would emerge as the dominant
one by virtue of its disproportionately large contribution to crack extension.
The above rationale suggests that in variable amplitude fatigue, a variety of
crack extension mechanisms will continuously leave an imprint on the fracture
surface and their mix will depend on the load spectrum. A corresponding mix of
load interaction mechanisms may also continuously prevail. We now proceed to
consider in greater detail, individual crack extension mechanisms and how
each one may be sequence sensitive. In doing so, less significant load interaction
mechanisms such as crack-tip blunting/resharpening, history-induced phase
transformations, and other such effects whose influence cannot be deemed deci-
sive or quantifiable are ignored.
High-End Growth Rates

The crack tip will see critical conditions associated with catastrophic fracture
when K approaches Kc associated with static fracture. Such local failure is
17)
Crack growth rates less than atomic spacing are readily explained by the possibility of
local crack extension occurring at different points on the crack front at different times
[38].
18)
After all, the crack tip at the commencement of rising load half-cycle, “does not yet
know” the extent to which it will be loaded. It will switch sequentially to the “mechanism
of least resistance to crack extension” corresponding to the instantaneous load increment.

attributed to quasi-static rupture of the material directly ahead of the crack tip.
If the material is inherently brittle, it will simply cleave locally. If it is ductile, as
is the case with most aerospace structural materials, at least two simultaneous
mechanisms are likely. Stable crack growth by shear can be either Mode II or
Mode III. This typically occurs at the specimen edge, where plane stress condi-
tions promote shear ligament formation and gradually spread inward, because
ligament formation demands crack extension.19 A little deeper, and particularly
given a straight crack front, plane strain conditions associated with constraint
can prevail, leading to the buildup of hydrostatic tension20 that can result in
static rupture by microvoid coalescence (essentially, an analog of cavitation in
liquids), seen on the fracture surface as clusters of microscopic cavities, irrefu-
table evidence that local failure was instantaneous. Note that because condi-
tions of constraint develop at some distance from the crack tip, crack jump or
tunneling by microvoid coalescence will invariably be accompanied by a shear
of the interim ligament at the very tip of the crack that remained under plane
stress. A third mechanism is typical of Al-alloys and the proliferation in them of
secondary particulates that are natural barriers to slip. As a consequence, if
sizeable slip is involved that covers a distance exceeding their average spacing, a
strain localization will result, leading to a shear fracture along interconnecting
planes between particulates. This leads to the appearance on the fracture sur-
face of a disproportionately high density of particulate voids, that should not be
confused with microvoid coalescence associated with static fracture as was the
case in. An example of a mix of the two appears in Fig. 7(b) (also, see Fig. 10(b)).
Being a highly localized phenomenon, such ruptures may occur momentarily
and only at one or a few points ahead of the crack front. This, in macroscopic
terms, will show up as increasingly accelerated fatigue cracking as Kmax under
cyclic loading approaches Kc.21 One may expect that as the ratio Kmax/Kc
approaches unity, the crack growth rate will approach infinity (static fracture).
19)
As a rule of thumb, the crack needs to extend over an interval of at least half the speci-
men thickness in order for the front to completely rotate to shear mode. Quite simply,
front rotation also demands extension.
20)
Liquids follow Pascal’s Law. Applying pressure at any point will result in all ends of the
constraining container seeing that pressure. This is what drives fluid power technology.
Solids are different from liquids in their resistance to sliding (shear or slip), which is
infinitely higher than viscosity in liquids. Therefore, when a smooth solid specimen is
pulled, it will readily transversely contract, as seen on a rubber band. However, if for
some reason such a contraction is inhibited by external or internal conditions (con-
straint), a hydrostatic response will result, whereby tension will be experienced in all
directions. An example of hydrostatic tension in the response of secondary particulates is
forthcoming. A stress gradient serves as a natural constraint and can result in a near-
hydrostatic local response.
21)
In the presence of a substantial quasi-static crack extension, one can hear audible pop-
ins. Much lower levels of such an extension can be picked up by acoustic emission, which
often serves as a tool for on-line structural diagnostics. This is used in industry to “hear”
defects growing in a structure and to locate them by triangulation, much like GPS posi-
tioning systems.

SUNDER, doi:10.1520/JAI103940 39
Forman et al. introduced such a correction into the crack growth rate equation
which otherwise only carried two material constants to be determined by statis-
tical analysis of laboratory data. The correction kicks in only at higher values of
DK, or at a very high stress ratio, where Kmax gets closer to Kc even at lower DK.
Critical conditions associated with local quasi-static crack extension require
high stress and strain levels. Since these will be tied to the top end of the local
stress-strain hysteresis loop, they may be immune to hysteretic effects and
therefore insensitive to load history. Also at these levels, crack closure has prac-
tically no role to play because the process is driven by the maximum driving
force, rather than its range. There is, however, some possibility of effects attrib-
utable to strain hardening or softening that may affect local fracture resistance
and will be stress history sensitive. Importantly, the crack-tip stress-strain
response will be extremely sensitive to local constraint. This will vary across the
thickness and will also be determined by instantaneous crack front orientation
as well as shape, that is, in effect, determined by the cumulative preceding crack
extension. Of all the load history related parameters, this one appears worthy of
analytical consideration at a high growth rate. To do so, one may treat Kc as a
crack front related parameter varying between a low of K1c associated with
plane strain and a high of Kc, associated with plane stress and therein introduce
a history sensitive component into the Forman equation to account for
sequence sensitivity of high end growth rates.
In summary, the effect on high end growth rates of the loading history may
be accounted for by correcting K and Kc for crack front shape and orientation.
Parameters such as crack closure and residual stress will have little bearing on
high-end growth rates.
Intermediate (Paris Regime) Range Growth Rates

The Paris Regime is characterized by a log linear relationship between DK and
da/dN over a range of growth rates covering the interval 104–102 mm/cycle
with nonlinearity at the high end coming in due to the quasi-static component
and with the lower end overlapping with near-threshold fatigue response. The
interval is dominated by cyclic slip driven crack-tip extension, according to a va-
riety of schemes proposed in the literature [41]. A reasonably straight crack
front is conducive to transgranular slip along preferred planes and one can
readily accept the possibility within individual grains of highly reproducible
extent of stretch and compression in successive load cycles that leave behind
striation bands with near digital precision.22
There are different ways in which a crack can extend over a load cycle in a pre-
dominantly slip dominated mode. The first is by deformation (as opposed to
22)
Reference [42] describes an experiment that involved “punching” onto the fracture sur-
face of fatigue striations representing binary code of text strings in much the same way as
information is stored on digital media. This would not be possible without precisely re-
producible cycle-by-cycle fatigue crack extension at the microscopic scale and serves as a
compelling argument in favour of fractography as a dependable tool not only in failure
analysis, but also for the quantitative validation of crack growth models.

fracture), whereby the shear stretch produced in the rising half cycle cannot be
fully reversed upon unloading, resulting in a fold, as indicated by the well-known
Laird model [43]. From this, follows the unexpected conclusion that the crack
extends during unloading. The second possibility is that the crack extends by shear
fracture [44], whose extent is determined by rising load excursion exceeding a cer-
tain threshold level over which microscopic stable crack extension occurs, but not
unstable (even if localized) fracture. Reversed deformation during unloading will
essentially prepare a sharp crack for extension in the next cycle.23 The third possi-
bility is a combination of the two, leading to a somewhat greater crack extension
considering that the crack will continue to grow during unloading as well. All three
possibilities are supported by observations of extremely well defined striations that
mark the fatigue fracture surface, though the textbook understanding is of fatigue
crack extension by deformation (slip), not shear fracture.
Assuming that the crack-tip response is controlled exclusively by the cyclic
stress-strain curve and the extent of change in stress intensity, crack extension in
this range should be insensitive to the applied stress ratio and to near-tip mean
stress (i.e., residual stress). Mean stress insensitivity is the very essence of a process
driven by slip alone. It follows that any sensitivity of intermediate range crack
growth rates to the stress ratio and to the load history may be attributed largely, if
not solely, to crack closure. An inevitable conclusion then would be that if the
Wheeler and Willenborg models indeed correctly simulate intermediate range vari-
able amplitude behaviour, they may be merely appearing to do so by the happy
coincidence of fudged closure response. Indeed, if fatigue crack growth is predomi-
nantly slip driven, the only plausible explanation for the stress ratio and load history
effects is fatigue crack closure controlling the effective range of the stress intensity.
All three possible ways of crack extension by slip previously listed carry cer-
tain implications that go beyond insensitivity to residual stress, stress ratio, and
stress history. They imply cycle-by-cycle striation formation. They also imply
relative immunity of the Paris Regime to the environmental effect (assuming
slip is environment independent) and to cycling frequency (assuming rate-
insensitivity of slip over the practical range of frequency). Sensitivity to environ-
ment and frequency increases at lower growth rates associated with thresholds
and at much higher rates associated with sustained load cracking, creep, etc.
There are two curious features of intermediate range crack growth whose
significance appears to have remained largely unnoticed over the five decades of
study by high resolution electron fractography. One is the surprisingly narrow
band of growth rates (usually within one or two orders ofmagnitude of varia-
tion) over which discernible striations are observed.24 The other is the surpris-
ing absence of striations in vacuum.
23)
A blunt crack tip offers multiple parallel slip planes that will contribute to cumulative
stretch by dissipating total strain. A sharp crack restricts the number of shear planes and
thereby encourages shear fracture by focusing shear strain into fewer slip planes.
24)
The resolution of electron fractography is adequate to resolve a crack extension less
than 106 mm/cycle, but one seldom sees striations at growth rate less than 104 mm/
cycle.

SUNDER, doi:10.1520/JAI103940 41
Near Threshold Fatigue Crack Response

Indeed, why are striations not discernible in vacuum? And why are we usually
unable to see striations in atmospheric fractures at growth rates below
104 mm/cycle, even if electron microscopes can resolve features one hundred
times smaller? The controversial brittle micro-fracture (BMF) model of near-
threshold crack growth25 appears to provide the answer [45] (Fig. 8). The fa-
tigue crack tip represents an extreme stress concentrator. Associated with such
stress concentration is an extreme stress gradient that in turn induces condi-
tions of severe near-tip constraint because the surrounding lightly stressed ma-
terial does not permit local necking. This, in turn, induces conditions of
hydrostatic loading: application of tensile load normal to the crack plane causes
increasing tensile stresses in the transverse direction as well. Stress in the third
direction along the major crack axis will be somewhat relaxed, at least at the
tip, because the free crack tip surface is free to move inward into the material.
However, the “diaphragm” stresses stretching the crack tip surface in two direc-
tions will increase the inter-atomic distance along the loading axis while not
allowing transverse spacing to reduce. Such conditions are conducive to the
activation of surface physics (diffusion of active species into surface layers) and
surface chemistry (chemical reaction with active species), leading to accelerated
transgranular26 fatigue crack extension [46].
In a careful study on the near-threshold fatigue fracture mode of an
Al-alloy, Gangloff et al. observe that crack extension occurred along crystallo-
graphic slip planes [47]. This by itself need not imply that crack extension
occurred by slip unless it can also be shown that the fracture plane was oriented
appropriately with respect to the loading direction27 as is the case with ductile
response and striation formation in the Paris Regime. Once a surface layer has
been embrittled, it may not matter whether Mode I (tensile rupture) or II (slip)
is involved. If Mode II was indeed involved, it would lead to the formation of
shear lips and progressive rotation of the fatigue fracture plane by 45 . How-
ever, atmospheric fatigue fracture surfaces in the near-threshold regime do not
tend to develop shear lips. They remain flat and normal to the loading plane.
Compelling evidence in support of the previous rationale comes by way of
fatigue fractures obtained in salt water, air, and vacuum [48,49] under identical
loading conditions. The authors attributed the delayed transition to shear mode
in salt water and air out of early Mode I cracking to the adverse effect of envi-
ronment on resistance to Mode I. A crack tip stress state is determined by shear-
lip formation, which, in turn, is driven by dominant macroscopic mode of crack
25)
Against the general perception of metal fatigue being associated with cyclic slip (defor-
mation), the BMF model suggests that near-threshold fatigue crack extension occurs by
fracture.
26)
This is not to be confused with the mechanism of stress corrosion cracking associated
with the intergranular short circuit diffusion of active species that essentially leads to
crack extension by grain separation.
27)
Just as delamination in composites can occur either by Mode I or Mode II.

FIG. 8—The science behind the residual stress effect in metal fatigue crack growth. (a)
When an argon bubble is inserted under ruthenium monolayers, the stretched top
instantaneously attracts active species, while the compressed region at the root of the
blister repels them [46]. (b) According to the BMF theory, the same holds true at the fa-
tigue crack tip [45]: the active species is moisture at room temperature that is repelled
from the crack tip at minimum load, 1. During the rising half-cycle, moisture molecules
are attracted by the rising stresses at the crack tip. They react with metal to form metal
oxide and hydroxide to release hydrogen that diffuses into the substrate to embrittle and
fracture the affected surface layers under rising stress. (c) The surface physics and
chemistry described in (b) will be affected by the crack tip stress history as shown by the
schematic repeat action of load sequence 1-7. (d) If closure is reduced or absent (Lo-
Sop), cycles 2-3 and 5-6 will see hysteretic crack-tip stress-strain response. Higher stress
causes more BMF at 2-3 than at 5-6. (e) However, if the crack is partially closed during
2-3 and 5-6, both cycles will see similar reduced local stress and therefore, equally re-
tarded crack extension. (b)–(e) Underscores the significance of the cyclic plastic zone
response in controlling atmospheric sub-critical fatigue crack growth. Closure and
Wheeler/Willenborg models are incapable of explaining cycle-by-cycle hysteretic load
interaction effects in fatigue crack growth.

SUNDER, doi:10.1520/JAI103940 43
extension, rather than by applied DK28 or even the growth rate. Thus, in high
vacuum, shear lips will form earlier than in salt water or even air, even if vac-
uum growth rates will be much lower, given similar loading conditions. Pippan
et al. have observed that the fatigue crack stays sharp in air and turns blunt in
vacuum29 [50] but failed to draw conclusions on how this may reflect on the
crack extension mode. Embrittled surface layers will also exhibit reduced elon-
gation. As a consequence, the crack may extend by BMF before the potential
onset of slip on neighboring planes that promote blunting, or, on the same
plane, but deeper into the substrate.
In room temperature atmospheric fatigue, BMF appears to be primarily
promoted by surface diffusion of hydrogen released by the reaction of moisture
with the crack tip surface resulting in oxide and hydroxide formation. Oxidation
appears to be an unlikely factor in BMF, a conclusion prompted by the retarded
near-threshold fatigue crack growth in dry oxygen observed by Bowles [51]. In
tests on an Al-alloy, Bowles also observed that when the environment is
switched from laboratory air to dry oxygen, striations gradually disappear, leav-
ing a surface akin to that obtained in vacuum.30 This observation also points to
the potential role of BMF in striation formation. The BMF controls the near-
threshold fatigue response that extends up to a growth rate of between 105 and
104 mm/cycle, suggesting that surface physics and chemistry do affect tens,
but perhaps not hundreds or thousands of atomic layers at the crack tip. Per-
haps crack extension by the BMF (mode I) over such a distance in the course of
the rising load half cycle, when followed by subsequent crack extension either
by shear in Mode II, or, by folding of shear stretched crack tip surface, or, by a
combination of the two leaves that distinct wavy pattern one associates with
well-defined striations. Striation formation may thus require two distinctly dif-
ferent crack extension mechanisms to operate sequentially (as shown by the
schematic in Fig. 7(c)). If only one of them operates as in the case below the
Paris Regime (only BMF and no slip) or in high vacuum (only slip and no BMF),
no discernible contrasting topographical feature may result to mark the pro-
gress of the crack front.
Just as room temperature near threshold fatigue is closely linked with
cycle-by-cycle crack extension by the BMF of crack-tip surface layers embrittled
by surface physics and chemistry, a similar process may control elevated
28)
The ratio of plastic zone size to thickness is often treated as a reflection of the stress
state. Implicit in this assumption is a flat and straight crack front. In reality, a curved
(tongue shaped) crack front or one that is tilted will both promote plane stress due toliga-
ment response.
29)
Interestingly, having obtained lucid evidence about the cause (sensitivity of crack-tip
deformation to environment), the authors seem to have failed to draw the logical conclu-
sion about its effect (sensitivity of the crack extension mode to the environment)!
30)
Their ‘gradual’ rather than immediate disappearance also raises the intriguing ques-
tion of hydrogen consumption. Does hydrogen get consumed by embrittlement, or does it
escape upon BMF to affect the next layer? Partial consumption can explain the momen-
tary persistence of BMF into vacuum. It may also explain sustained accelerated internal
cracking as in gigacycle fatigue.

temperature transgranular fatigue crack growth. The latter is accelerated by the

enhanced oxidation of crack tip layers that can considerably exceed the depth of
moisture related surface diffusion by hydrogen. In both cases, crack extension
is transgranular and involves cycle-by-cycle crack tip surface activity that accel-
erates crack extension by comparison to vacuum fatigue response. For this rea-
son, in both cases, the threshold stress intensity will be much less than in high
vacuum. It thus emerges, that, if near-threshold behaviour is sensitive to diffu-
sion kinetics, threshold stress intensity ought to be controlled by the cyclic plas-
tic zone response!
Threshold Stress Intensity

While near-threshold fatigue crack growth behaviour has long been connected
with crack-tip surface physics and with surface chemistry [52, 53], there seems
to have been a general failure to appreciate the connection between the kinetics
of surface activity and near tip hydrostatic stress, and the sensitivity of the latter
to the stress history and to stress ratio. This may be attributed to the prevailing
stereotype of the crack tip essentially seeing an elastic, ideally plastic cyclic
response, implying a local stress ratio of R ¼ 1 (zero mean stress), irrespective
of the applied stress ratio and stress history. Such an assumption may have
assisted crack-tip elasto-plastic stress-strain analyses and may also have been
appropriate for the Paris Regime with its sizeable cycle plastic zones. However,
it appears to have clouded the significance of the hysteretic stress-strain
response within the cyclic plastic zone in moderating near-threshold diffusion
activity at room temperature and chemical reactivity at elevated temperatures.
On the contrary, the significance of residual stress is well known and appreci-
ated in stress corrosion cracking, as is practiced in assessing heat affected zones
in welding and crack growth in an aggressive environment. There is, however,
an extremely important distinction between stress corrosion cracking and
atmospheric near-threshold fatigue crack growth. Near threshold fatigue crack
growth is affected by near-tip stress zones that are hydrostatic and microscopic
in comparison to those considered in stress corrosion cracking.
When crack-tip cyclic slip recedes below a certain threshold, crack growth
will practically cease in vacuum. This vacuum threshold stress intensity range is
about three times greater than effective DKth in air. Crack growth in air at DK
less than vacuum DKth cannot obviously be explained on considerations of slip.
The difference between the two when raised to the power of four and above,
covers a growth rate variation in air, exceeding two orders of magnitude. The in-
terim interval thus covers a vital segment of the sub-Paris regime atmospheric
fatigue response that may well account for the bulk of total fatigue life.31 Fa-
tigue kinetics over this interval will obviously be determined by the ability of
31)
In observing fatigue fractures, one may be inclined to associate the bulk of the fatigue
process with the largest observable area of the fatigue fracture. However, the bulk of fa-
tigue life may, in fact, have been consumed in early crack growth. While assessing fatigue
fractures, it may be important not to ignore that, albeit small, region covering the crack
initiation area.

SUNDER, doi:10.1520/JAI103940 45
hydrostatic stresses within the cyclic plastic zone to moderate crack-tip surface
chemistry and surface physics. These stresses are the sum of the crack-tip mean
stress associated with the current stress ratio, history sensitive, residual stress
and their hysteretic variation while the crack remains open.32
The consequences of such a “crack-tip cyclic diffusion pump” may be var-
ied. An arrested crack tip will progressively lose its resistance under the persis-
tent onslaught of diffusing active species. Hydrogen trapped in the rising load
half-cycle will not be released upon unloading. Oxidation at an elevated temper-
ature accelerated during the rising load half-cycle will not be reversed upon
unloading. This implies that the crack front will be inclined to straighten itself
even if the crack does not uniformly extend in successive cycles. Over each cycle
that the tip does not give way, surface layers are likely to see a little more
embrittlement. At the same time, interstitial diffusion is a self-retarding process
because diffused layers represent barriers to newer and deeper diffusion. This is
why the effect in question is unlikely to significantly influence growth rates in
excess of 104 mm/cycle. Another measure of the effect can emerge from a com-
parison of Paris Regime growth rates in air and high vacuum. Their difference
is substantially less than under near-threshold conditions. Thus, while a fatigue
crack in air can grow at 105 mm/cycle, it may just remain arrested under high
vacuum given the same loading conditions.
Significance of Cyclic Plastic Zone Response in Variable Amplitude Fatigue

Near-tip residual stress is highly sensitive to load history and can vary substan-
tially on a cycle-by-cycle basis. Figure 8(c)–8(e) schematically illustrates the
cycle-by-cycle near-tip stress history for a fully open and for a partially closed
fatigue crack. Consider the repeated action of cycles 1-7. Identical embedded
cycles 2-3 and 5-6 will see the consequences of the hysteretic crack tip response
within the cyclic plastic zone. In both cases, near tip stress will be well below
that under constant amplitude loading because of the compressive stress intro-
duced by tensile overload 4. As a consequence, cycles 2-3 and 5-6 will see re-
tarded crack extension. However, the retardation in 5-6 will be greater because
lower local stress reduces the diffusion activity with the crack growth tending
towards vacuum response. Importantly, the difference between 2-3 and 5-6
being hysteretic can be seen on a cycle-by-cycle basis. This is possible only if the
crack is fully open during the embedded cycles as in Fig. 8(d) and with closure
stress well below the minimum stress in the two cycles. The hysteretic variation
will cease in the event of partial closure as shown in Fig. 8(e). Both cycles will
be equally retarded in this case, being rendered cycle-sequence insensitive due
to partial closure. Cycle-sequence sensitivity is a term alien to conventional
modeling based on the Wheeler, Willenborg, or Elber models.33 The significance
32)
An important consequence of this possibility is that a partially closed crack will not see
cycle-sequence sensitivity, a feature to be addressed further in the text.
33)
Curiously, interpretation of notch root fatigue response universally proceeds on this
very understanding, and has remained unquestioned, even in the absence of any scientific
rationale for the notch root mean stress effect!

of the cyclic plastic zone and the possibility of cycle-sequence sensitivity

becomes obvious only when viewed from the perspective of the BMF model.
Figure 9 describes the extension of this new understanding to re-interpret
the simple case of tensile and compressive overloads and how their action may
be modeled. Consider the case of identical cycles AB, DE, and GH in Fig. 9(a)
and 9(b). We assume the stress ratio to be sufficiently high in order to preclude
the possibility of crack closure. Near-tip cyclic stress strain response for these
cycles appears as Fig.9(c). We see that the baseline cycle AB would be associated
with a certain near-tip stress rA. If the stress ratio of this cycle had been higher
such that A and C were equal, the local stress would have risen to rC. In the
event of tensile overload as in Fig. 9(a), the following cycle ED sees a deep drop
in the near-tip stress to rE. However, if a compressive overload followed the ten-
sile overload as in Fig. 9(b), the following cycle GH will see an increased local
mean stress due to the preceding yield in compression at F. The new mean
stress will still be lower than in AB. Assuming a unique relationship between the
near-tip mean stress and threshold stress intensity, the three identical load
cycles in question will follow different near-threshold da/dN curves as, indicated
in Fig. 9(d). Empirical determination of these modified da/dN curves of rele-
vance to variable amplitude fatigue requires specially designed experiments
involving the controlled variation of near-tip mean stress. Note that, given the
impact of compressive overloads leading to increased tensile near-tip stresses,
there is no reason why the left extreme of these curves cannot tend towards
zero. Note also, that the right extreme for the da/dN curve is a high vacuum
response that effectively limits the extent to which compressive residual stress
can retard the fatigue process.
Figure 9(e) and 9(f) assist in understanding transients associated with near-
tip residual stress response after an overload. Here, A and C are the monotonic
baseline and overload plastic zones, respectively; D and B are the associated
cyclic plastic zones, respectively. Figure 9(f) shows the near-tip response at the
boundary of zone D. Since this point will see a fully elastic response, one may
assume that as the crack tip approaches this point, any hysteretic effects seen in
Fig. 9(c) will disappear. This means that beyond this point, from a residual
stress perspective, crack growth will be identical for the two cases in Fig. 9(a)
and 9(b). The memory about the compressive overload stands is erased from
this point. It also appears possible that, after some crack extension and well
before the boundary of C, near-tip stresses will be restored to the levels associ-
ated with the baseline conditions and one should, therefore, not see the extent
of the retardation zone assumed by the Wheeler and Willenborg models.
In summary, quite independent of crack closure, the residual stress effect is
manifested through the response of near-tip elements within the cyclic plastic
zone to stress history. Their response determines instantaneous DKth, suggest-
ing that the near-threshold da/dN versus DK curve is not a material constant
and needs definition on a cycle-by-cycle basis as a function of near-tip mean or
maximum stress. As a consequence, the hysteretic near-tip stress variation indu-
ces cycle-sequence sensitivity in near-threshold crack extension, provided the
crack is fully open during the given cycle. In the event of partial crack closure,
cycle-sequence sensitivity is not possible because the minimum crack-tip stress

SUNDER, doi:10.1520/JAI103940 47
FIG. 9—The new perspective of how tensile and compressive overloads distort the fa-
tigue process. (a) Tensile overload; and (b) compressive overload following a tensile
overload. (c) Crack tip stress-strain response showing the effect of overloads on local
mean stress (crack-tip residual stress). Tensile overload pushes local stress into com-
pression (ED), but if a compressive overload follows, local stress will rise (GH), though
not to the baseline value (AB). (d) Near threshold crack growth rates can swing dramat-
ically depending on crack tip stress. (e) Overload cyclic plastic zone is small by compari-
son to the tensile overload plastic zone. Therefore, any sequence sensitive hysteretic
effect will disappear on its boundary, as seen in (f). This implies that beyond this point,
it will not matter whether a compressive overload followed the tensile one. However,
due to the combined action of closure and residual stress, most of the load-interaction
effect, bordering on crack retardation and possible momentary arrest, would have been
exhausted within the cyclic plastic zone. Conventional modeling techniques cannot
reproduce these effects because they ignore the cyclic plastic zone response and its effect
on threshold.

in the cycle is practically tied to the lowest possible crack-tip stress (see Fig.
8(e)).
Upon application of a tensile overload, the impact of the associated residual
compressive stress is immediate. This combines with the delayed development
of closure awaiting wake build up. As a consequence, retardation will be imme-
diate in the event of the near-threshold response and delayed in the event of the
Paris Regime response or in high vacuum.34 Closure related retardation due to
overload will vanish only after the crack tip has extended well outside the over-
load plastic zone of the crack tip (see Fig. 6(d)). In contrast, the hysteretic na-
ture of residual stress effects will disappear at the boundary of the overload
cyclic plastic zone and the retarding effect of residual stress will altogether dis-
appear well before the crack tip exits the overload plastic zone as the near tip
stresses approach baseline values. This point has no connection with the point
where crack closure reaches its maximum. Thus, the combined action of crack
tip residual stress and closure will be limited in the crack extension interval.
However, over this small interval, retardation is likely to border on crack arrest.
The closure model accounts for only part of what happens except in the partial
case of Paris Regime growth rates. Additionally, the Willenborg and Wheeler
models altogether ignore the cyclic plastic zone response and treat the transient
process as a continuously changing one over the entire monotonic plastic zone.
The ramifications of the deviation from reality of all existing approaches to
crack growth estimates under variable amplitude loading can be judged from
two important practical considerations of computation. First, the baseline
cyclic plastic zone where hysteretic effects dominate will be well under 10% of
the overload monotonic plastic zone size.35 Second, computed residual fatigue
life, being an integral of the growth rate function, will accumulate errors in
computed transient growth rates. This suggests the questionability of obtaining
reasonable crack growth estimates using available models. The suggestion may
appear preposterous when viewed against the operating framework of techni-
ques currently in use to handle variable amplitude fatigue. An examination of
the empirical evidence and definition of the emerging perspective is, therefore,
pertinent.
The Experimental Evidence

A series of experiments were performed to verify each of the conclusions that
follow from interpreting variable-amplitude fatigue using the BMF model com-
bined with closure. To avoid speculation, each experiment was specially
designed to deliver irrefutable fractographic evidence [54–63]. These highlight
34)
Published fractographic data showing delayed retardation are restricted to the Paris
Regime—they show striations.
35)
Plastic zone size ratio is given as the square of the ratio of overload stress intensity to
half the baseline effective stress intensity range because cyclic plastic zone size is deter-
mined by twice the yield stress required for reverse yield.

SUNDER, doi:10.1520/JAI103940 49
quantifiable effects for which there appears to be no alternate interpretation.36

The experiments imply the important criterion of falsifiability.
It was a chance discovery that initiated this research in the late 1990s. Early
experiments on Al-alloy specimens using closure-free high stress ratio cycling
were performed in search of a correlation between the applied stress and short
crack response [54] (the so-called short crack effect). The experiments were per-
formed under programmed loading with three steps of identical amplitude but
varying mean stress. Figure 4(b) shows a typical fatigue fracture from these
experiments. Clearly visible are sets of three bands of crack extension associated
with each set of three steps of loading. Note that each band is caused by a few
thousand load cycles and is not to be confused with striations from individual
cycles seen in the Paris Regime.
A noticeable difference in the crack growth rate is observed between indi-
vidual steps at small crack size in Fig. 4(b). This difference gradually tapers out
to uniform crack growth rate as the crack grows much larger. The embedded
cycles were placed on the rising half of the major cycle.37 The authors correlated
measured crack growth rates with the maximum local notch root stress in indi-
vidual steps and came up with an empirical equation to account for the short
crack effect as a function of local maximum stress [54]. At the time, this
approach was considered consistent with the prevailing notions of the so-called
“short crack effect,” where parameters such as local stress were considered
essential to explain what the stress intensity range could not. We did not con-
sider the possibility that the steps with lower mean stress may experience the
beneficial effect of preceding stressing at a higher level. We believed that having
ensured the crack was fully open by keeping stress ratios high, no load interac-
tion effects were possible. Sometime after the publication of this work, a
chance38 discovery was made of equally spaced concentric circular bands
within voids on the fatigue fracture surface left behind by secondary particu-
lates [55].
Several conclusions crucial to unraveling the nature of metal fatigue
emerged from the detailed study of fatigue voids. While it has long been known
that fatigue cracks form at the notch root, the new evidence confirms the possi-
bility that early fatigue kinetics are the consequence of several competing mech-
anisms operating at different locations. At a high applied stress level promoting
intense reverse slip, the notch root surface is likely to develop several crack
36)
A few early experiments involved the analysis of striation patterns. The rest involved
estimates of spacing between marker bands employed to unambiguously characterize mi-
croscopic crack extension over thousands of near-threshold load cycles that cannot, in
their individual capacity, produce discernible growth marks. This technique permits
quantitative estimates of crack extension without a limitation on the minimum growth
rate. The pictures reproduced in this paper reach down to 108 mm/cycle.
37)
Had they been placed on the falling half, the retardation effect would have been much
more dramatic given the hysteretic response. At the time, the authors were not aware of
the phenomenon involved.
38)
In routine electron microscopy particulate voids are usually ignored as dark, feature-
less cavities.

origins almost simultaneously [61]. Plane stress conditions at the surface com-
bined with assistance from the environment39 appear to dominate. With a
decreasing stress level, the number of such sites will diminish, with a general
tendency towards eventual sub-surface initiation.40 One may speculate that con-
straint in the interior will promote local defect growth by microscopic failure
through modes other than planar slip, which prefers plane stress conditions.
In Al-alloys, innumerable secondary particulates lying beneath the notch
root appear to bear evidence to the consequences of cyclic hydrostatic stresses
operating in the constrained region beneath the notch. These induce the grad-
ual separation by interfacial fatigue cracking of the secondary particulate from
the matrix. Cyclic hydrostatic loading action is apparent from the simultaneous
onset and identical growth rate of typically six (even more in the case of the
irregular shape of the particulate) penny shaped interface cracks covering all six
sides of the particulate (see Fig. 10). The smallest crack size seen is of the order
of 0.125 lm, which may represent the smallest reproducible and traceable fa-
tigue crack observed in research practice. The bands also indicate an incredibly
low growth rate down to 108 mm/cycle. The generally uniform spacing of the
concentric bands is of practical significance, suggesting that the interfacial
crack growth rate appeared to be insensitive to change in the mean stress in
individual steps of the programmed load sequence employed. This was in con-
trast to the major short crack at the same proximity to the notch root! Surely,
the effect that caused growth rates to be different between steps in the major
short crack as seen in Fig. 4(b) ought to have also have influenced the interfacial
crack growth! However, they apparently did not, after all.
There was, however, an important difference between the conditions under
which the two cracks grew. Unlike the major crack originating from the surface
and continuously exposed to the environment, interfacial cracks around sec-
ondary particulates grow in ideal vacuum. This is confirmed by simultaneous
cracking around the particulate that could not have progressed without cyclic
hydrostatic tensile stresses, and these in turn will disappear once the particulate
is exposed and constraint disappears. There was obviously something linked
not with the macro-mechanics of the notch response, but rather, with the
micro-mechanism of crack extension that seemed to determine fatigue resist-
ance. A possibility has now emerged that vacuum inhibits the root cause for the
mean (residual) stress effect in metal fatigue. It was also possible that in air, it
was not the applied mean stress itself, but the sequence of its change (load his-
tory) that was responsible. Perhaps, indeed, vacuum does disable residual
(mean) stress related effects?
Reference [56] describes an experiment dedicated to conclusively isolate
the role of environment in near-threshold fatigue by falsification. The experi-
ment involved testing to failure under the same three-step programmed loading,
but alternating between air and vacuum every given number of blocks. The
39)
In Al-alloys, interfacial environmental attack causes early pitting through the separa-
tion of secondary particulates on the notch surface. Each pit is a potential initial defect.
40)
Gigacycle fatigue is almost always associated with internal crack formation.

SUNDER, doi:10.1520/JAI103940 51
FIG. 10—Fatigue voids and microvoids [55]. (a) Proof that individual voids seen on Al-
alloy fatigue fractures were formed by fatigue-separation of secondary particulates from
the matrix and not due to high Kmax quasi-static failure as claimed in [40]. Evidence of
interfacial cracking under three-step programmed loading (inset). Clear, equally spaced
bands marked by marker loads between steps indicate that the change in the mean
stress level did not have any effect on the crack extension due to the 2000 cycles in each
step. The schematic shows cyclic hydrostatic forces responsible for the cracking. (b)
Rare picture of the secondary particulate that remained bonded to the fatigue fracture.
The area immediately around the particulate is evidently formed by fatigue. The sur-
rounding area is marked by clusters of microvoids that coalesced to cause quasi-static
crack extension. Microvoids are formed by very high hydrostatic stresses leading to
microcavitation, with the walls between cavities failing in ductile fashion due to local-
ized plane stress conditions. Note the vast difference in size between particulate voids
and microvoids, indicating that one cannot be confused with the other (as was the case
in [40]). (c) Multiple interfacial cracks separating an irregularly shaped particulate sit-
ting on the boundary of three grains suggesting the action of tensile cyclic hydrostatic
stress.

vacuum steps were twice the cycle count to account for retarded growth. The
switch from vacuum to air took a mere few minutes. However, the switch from
air to high vacuum (108 Torr) required more than 48 h, with the entire experi-
ment lasting several weeks. The experiments provided conclusive evidence in
support of the BMF model (see Fig. 11(a) and 11(b)). In air (Fig. 11(a), left), the
notch root small crack growth rate in the three steps varied exactly as in the ear-
lier experiment (Fig. 4(b)). However in high vacuum, the three steps caused
identical crack extension as seen at top right of Fig. 11(a) and magnified as in
Fig. 11(b). The instant air was released into the chamber, and the growth rates
in the three steps once again became different. This confirmed the absence of
the crack-tip residual stress effect in high vacuum. It also provides an alternate
explanation for the so-called short crack effect.
If, indeed, residual stress operates by the moderation of cycle-by-cycle envi-
ronmental action, it should reproduce on all metals and in the presence of any
active species that can diffuse and thereby adversely affect fatigue resistance.
To confirm this possibility, experiments were repeated on a Ni-base superalloy
at an elevated temperature, once again on the same machine, in air and in vac-
uum. In this case, the results were even more dramatic, apparently because of
the sensitivity of the crack tip oxidation to the near-tip residual stress [58].
If, indeed, the near-tip stresses within the cyclic plastic zone control diffusion
kinetics and through it, near-threshold crack extension, they should exhibit hyste-
retic sequence sensitivity. This hypothesis was successfully verified by tests per-
formed under two different programmed sequences, one directed at growth rates
closer to the Paris Regime and another, closer to threshold [57]. Figure 11(c)
shows a typical fractograph obtained from the second experiment performed
using the sequence shown in the inset. The three steps are of identical small am-
plitude set way above expected closure levels, in order to induce hysteretic near-
tip stress variation between steps 1 and 3, as shown schematically in Fig. 8(d). As
expected, the growth rate in step 3 is dramatically retarded by comparison to step
1. If the same experiment were to be performed in high vacuum, the crack exten-
sion would be identical in all three steps and close to that in step 3.
Finally, another experiment was designed; this time, to demonstrate the
synergy of crack closure and the residual stress effect [58]. The results are
briefly summarized in Fig. 12. The load sequence was specially designed to
selectively induce full crack closure, partial crack closure, and a fully open
crack. A key-hole notched C(T) specimen was chosen to induce natural crack
formation under conditions of notch root compressive residual stress due to
monotonic yield at maximum stress. Notch root crack closure is known to be
sensitive to local residual stress [59]. Steps of identical small amplitude were
embedded at three different mean stress levels on the rising and falling half of
the major cycle. The fractographs provide a graphic illustration of how the
notch root residual stress affects crack closure and how crack closure combines
with crack-tip residual stress effects to control variable amplitude fatigue crack
growth. Initial notch root yield in tension induced residual compressive stress
that reduced local stress ratio and thereby increased closure levels in the initial
stage of fatigue when the crack was barely 0.05 mm deep. As a consequence,
steps 1and 5 were fully closed and steps 2 and 4 partially closed. Furthermore,

SUNDER, doi:10.1520/JAI103940 53
FIG. 11—(a) and (b) Proof of residual stress operative mechanism by falsification. (a)
Crack growth under three step programmed loading in air and high vacuum (top right)
[56]. Noticeable retardation in crack extension in the second and third step is repro-
duced across multiple blocks. However, upon switching to high vacuum (top right and
magnified picture (b), the crack extension in all three steps is identical. The switch in
growth rate response was instantaneous in both the air-vacuum and vacuum-air tran-
sitions suggesting the virtual absence of any transient effects and also the impossibility
of crack closure playing a role. The Wheeler, Willenborg, and closure models cannot
explain these observations. (c) Proof of the effect of the hysteretic crack-tip stress-strain
response on atmospheric near-threshold crack growth rate. Note the substantial retar-
dation in step 3 because of compressive crack-tip stresses due to load cycles lying on the
falling half of the major cycle, as explained in Figs. 8(c), 9(d). This effect tapers out into
the Paris Regime, a phenomenon that the Wheeler/Willenborg and closure models can-
not simulate.

FIG. 12—Experiment on Al-alloy to demonstrate the synergy of the transient notch root
crack closure and residual stress [58]. (a) Multi-step programmed load sequence
designed to induce hysteretic residual stress variation in steps 2 and 4. Note that the du-
ration of step 3 is half of the others. Selected max load induced notch root tensile yield
leaving compressive residual stress at the notch root. (b) Macro showing the notch root
at left and the locations of fractographs c and d. (c) Identical growth from steps 2 and 4
indicates partial crack closure (at about 50% stress) and also explains why steps 1 and
5 did not extend the crack. (d). Almost identical growth in 2 and 4 and equal growth in
1 and 5 suggests that the closure level was around 40%. (e) A large difference in the
crack extension between 2 and 4 suggests a noticeable hysteretic variation in crack tip
mean stress. The closure level must have dropped to the long crack level of 30% (crack
size 1.5 mm). However, steps 1 and 5 are partially closed, causing equal crack exten-
sion. No model or software in commercial use today is capable of simulating the crack
extension patterns shown.

SUNDER, doi:10.1520/JAI103940 55
in Fig.12(c), we see equal bands from 2,4 and no crack extension during 1 and 5.
Figure 12(d) shows a fractograph at a location about 0.2 mm from the notch
root, where closure level has by now dropped somewhat. As a consequence,
steps 1 and 5 are partially open and steps 2 and 4 are fully open. This is indi-
cated by discernible and equal crack extension in steps 1 and 5 and marginal re-
tardation in step 4 by comparison to 2 due to the hysteretic difference in near-
tip stress. Figure 12(e) shows a fractograph from a location about 1.5 mm from
the notch root where closure has dropped to a long crack level of about 30%.
This causes equal crack extension in steps 1 and 5 and considerable retardation
in step 4 by comparison to step 2. Also note that the crack extension in step 2 is
retarded by comparison to step 3 (which is of half the duration). If this test were
to be conducted in high vacuum, crack extension in steps 2 and 4 would have
been identical and exactly twice that in step 3 (merely because of twice the cycle
count). Crack growth in steps 1 and 5 would have remained less, due to partial
crack closure.
Implications of Empirical Evidence

Some 150 years after the effect was first noticed, the underlying science behind
variable-amplitude fatigue is finally unraveling. The connection between crack-
tip hydrostatic stress and near-threshold growth rates resulting in DKth ceasing
to be a material constant completely changes one’s perspective of laboratory
test data and their relevance to engineering applications. In view of its sensitiv-
ity to hysteretic crack-tip stress-strain response, DKth emerges, in effect, as the
sole reason for fatigue being cycle-sequence sensitive. For this reason, DKeff can
no longer be treated as a similarity criterion for near-threshold fatigue crack
growth. Crack closure can no longer be assumed to account by itself for stress
ratio effects and for load interaction phenomena. Retardation is possible with-
out closure. A fully open crack can accelerate under conditions of increased ten-
sile residual stresses at the crack tip. As it turns out, all these are possible
“merely” because of the moisture in the air.
Interestingly, Marci and Lang came close to intuitively judging the signifi-
cance of crack-tip hysteretic stress-strain [64]. They proposed the idea of a mini-
mum K called KPR (for propagation) that needs to be exceeded by DKth, for the
onset of fatigue crack extension. A simple yet time-consuming experimental
technique was developed to determine KPR, by cycling at DKth (treated as a ma-
terial constant) with gradually stepped up Kmin until the detectable onset of
crack extension. In simple variable-amplitude experiments on Al-alloys that
were later extended to other alloys, it was found that KPR was sensitive to load
history and as a rule, exceeded Kop. It was also found that depending on load
history, KPR changed in ways that could not be explained by closure. On the ba-
sis of these two considerations and on the basis of KPR capability for improved
estimates of spectrum load crack growth, the authors concluded that in KPR, a
variable had been found that is an effective replacement for Kop. They specifi-
cally noted its ability to not only account for closure, but also for a certain
notional residual stress effect at the crack tip. Notional, because its potential for
cycle-by-cycle hysteretic variation and the significance of the cyclic plastic zone

were both ignored. Unfortunately, Marci and Lang failed to realize that KPR
may have been actually accounting for the instantaneous change in DKth, which
they had wrongly assumed to be a material constant. The difference KPR – Kop,
may, in fact, represent instantaneous DKth changing as a function of load his-
tory. One may speculate that the authors would have found this value to remain
virtually constant in high vacuum and equal to the difference in DKth between
vacuum and air! Like many others before them, Lang and Marci appear to have
succumbed to the perception of DKth as a material constant.
This new understanding finally allows for reassessment of the residual
stress effect by separating its mechanics from operating mechanisms. Residual
stress in the cyclic plastic zone ahead of the crack tip is controlled primarily by
load history and the associated cycle-by-cycle stress intensity sequence. By vir-
tue of its immediate proximity, crack tip surface resistance to fracture is directly
affected in atmospheric near-threshold fatigue. On the contrary, the notch root
residual stress, and other such (remote) macroscopic stress distributions con-
trol local stress ratio and through it, crack closure [59, 65]. Previous work on
the subject may have offered powerful tools to address the mechanics of fatigue
and fracture mechanics, yet they did not have the benefit of clarity in scientific
understanding, without which realistic analytical modeling or even targeted ex-
perimental research appears rudderless.
Emerging Avenues for Future Work

Historically, research on the residual stress effect was restricted to technologies
for its introduction, for its removal, and for its measurement. Its operating
mechanism in moderating fatigue damage seems to have elicited little intellec-
tual curiosity. The discovery of the science behind how near-tip residual stress
affects near threshold fatigue crack growth and its distinction from how the
notch root residual stress affects crack closure together open avenues for future
research towards improved modeling in engineering applications. Listed in the
following sections are a few such avenues.
Crack closure—Of the different experimental techniques available, only

fractography [37] and crack tip laser interferometry [36] appear to deliver
authentic measurements of crack closure. Of these, the first requires special
load sequences and cumbersome microscopy and is restricted to materials that
are “fractography friendly.” Furthermore, since crack extension is also sensitive
to crack-tip residual stress, one must ensure in designing the load sequence,
that the crack-tip stress remains unchanged at the applied maximum stress.
Failure to ensure this condition can lead to distortions as seen in [58]. The sec-
ond technique is expensive and demands special equipment and in the end, only
provides surface measurements. All other techniques deliver far field measure-
ments that cannot possibly serve as a reliable measure of a near-field phenom-
enon. Until laboratory techniques become available that can plot local strain
within or very close to the cyclic plastic zone versus the applied load, it is
unlikely that one can make useful measurements of crack closure in routine

SUNDER, doi:10.1520/JAI103940 57
testing practice. The development of such techniques appears to hold much

promise in future applications.
Elaborate finite-element solutions have been developed to compute stress
intensity for a given crack path, given applied loading, and a given “resident” re-
sidual stress field in a material including processing induced fields, fields due to
interference fit fasteners, along with response induced fields such as at notches
subject to local yield. State-of-the-art numerical simulation as demonstrated by
Seshadri et al. [66,67] appears to provide realistic cycle-by-cycle estimates of
the stress-strain response at any point in a cracked structure, under any given
load sequence. This includes the cyclic inelastic response around the crack tip
in the presence of crack closure, which appears to confirm that closure is a
mechanics driven phenomenon involving crack wake development into the
monotonic plastic zone, applied cyclic loading, and possible residual stress
fields in the material. The crack-tip load-displacement data reported in Refs.
[66, 67] appear to accurately reproduce the laser interferometry measurements
shown in Fig. 6(e). They clearly show the hysteretic response and the associated
loop formation in displacement versus load. Closure load can be unambigu-
ously associated with loop closure. Yet, the authors have preferred to interpret
closure in terms of change in compliance response as per ASTM E647, leading
to exaggerated closure estimates.
One may conclude that contradictory measurements and estimates of crack
closure stress may, in part, be attributed to the selection of compliance offset
points in measurements, that is synonymous with wake contact point in analy-
ses. In such a definition, the focus is deflected from the primary objective, which
is to define which fraction of the applied load cycle is responsible for creating
the observed cyclic plastic zone size.
Threshold Stress Intensity—The association of threshold fatigue with the

BMF, crack tip diffusion, and reaction kinetics implies that the highest possible
DKth will be in high vacuum and may be treated as a material constant. This pa-
rameter characterizes the maximum beneficial value that compressive residual
stresses can have on fatigue in air. The ultimate goal would be to come up with
an analytical model relating DKth to crack-tip diffusion kinetics as a function of
history-sensitive instantaneous hydrostatic stress ahead of the crack tip.
An intermediate objective may be to characterize DKth under a variety of
controlled near-tip stress conditions, while at the same time ensuring the ab-
sence of closure. If closure is present, an error in its measurement will carry
over to the DKth estimate. Such errors are unacceptable in threshold studies
because they may be of the same order as DKth. By keeping the stress ratio suffi-
ciently high, closure free DKth measurements are possible.41 An exploratory
study of Hi-R DKth under the action of periodic overloads provided a linear rela-
tionship between the overload plastic zone ratio and closure-free DKth [61]. The
41)
Actual mid-thickness closure levels seldom exceed 25-30% of the max load under con-
stant amplitude loading, when measurements are made using techniques such as fractog-
raphy or laser indentation interferometry.

study on two Al-alloys confirmed that variable-amplitude DKth does indeed

approach vacuum levels. It is proposed to repeat the experiments with compres-
sive overloads in order to determine the lower bound of DKth at a given test fre-
quency. Experiments are also proposed to examine the effect of the hysteretic
near-tip response on thresholds. Finally, hold times at periodic tensile and com-
pressive overloads are also likely to affect the atmospheric threshold fatigue
response because they induce stress relaxation.
Cyclic Stress-Strain Response—The modeling effort requires answers to new

questions. Is there a “characteristic distance” behind the crack tip whose stress-
strain response controls diffusion kinetics? What stress-strain curve would apply
at this point, considering constraint to be a controlling factor? Recent research
[68] reconfirms diffusion and reactivity of active species such as moisture as the
source of accelerated near-threshold fatigue response. An inter-disciplinary
effort to connect the crack-tip cyclic mechanical response to diffusion and chem-
ical reaction kinetics may be the next step towards modeling near-threshold fa-
tigue resistance in variable-amplitude fatigue. Determining the connection
between DKth, near-tip constraint, and cyclic strain hardening coefficient, com-
bined with the subsequent incorporation of sensitivity to hold time serve as
attractive long-term goals to tie in material cyclic stress-strain response with fa-
tigue. It is likely that materials with low strain hardening properties and reduced
constraint will exhibit a reduced sensitivity of DKth,eff to the stress ratio. This is
because increasing the stress ratio may not induce much increase in the near-tip
stress. However, all materials will exhibit a hysteretic stress-strain response and
will therefore exhibit stress history effects. Obviously, experiments restricted to
constant amplitude loading are unlikely to carry much practical value, apart
from underscoring the significance of the phenomenon as in [68].
The emergence of MEMS and biomedical applications of metallic compo-
nents and the application of nano-structured materials holds much scope for
the application of future work because of the potential dominance of surface
phenomena in these cases.42 Finally, the unification of near-threshold variable-
amplitude fatigue at room and elevated temperatures through the near-tip
response holds promise in gas and steam turbine applications. It is likely to
assist in improved modeling of the effect of overloads at high temperature,
hold-time, and creep-fatigue interaction effects. For example, it now appears
obvious that crack extension during hold at a given load at an elevated tempera-
ture will be driven by diffusion kinetics moderated by crack-tip residual stress.
This opens up the possibility of modeling the interaction of overloads with the
hold-time.
The bulk of fatigue damage due to many spectra including transport air-
craft load spectra is from the smallest load cycles that arguably advance fatigue
through near-threshold mechanisms yet are subject to the history effects from
periodic overloads. It is plausible that the B737 Aloha Airlines incident and,
42)
For a given volume, the total exposed surface area increases with the decreasing size
and scale of constituents.

SUNDER, doi:10.1520/JAI103940 59
more recently, with a Southwest Airlines fuselage panel may have been associ-
ated with stress intensity ranges deemed to be “sub-threshold” from laboratory
test data on coupons tested at a higher frequency. In the course of about fifteen
years of service, such aircraft would experience over 80,000 flights or 107 small
load cycles.
Overall, modeling of the residual stress effect holds the promise of advance-
ments in the quality of fatigue life estimates with a greater reliance on simula-
tion and a reduced emphasis on expensive empirical inputs. Lack of it will
continue to force dependence either on corrections of cumulative damage to
match experimental data, or on corrections of the crack driving force to com-
pensate for the inability to account for the change in material resistance. In the
meantime, disciplines other than fatigue and fracture mechanics will continue
to determine the safety and durability of engineered products, while in the long
term, metal fatigue may be simply rendered less relevant by advances in the
application of engineered composites that would be immune to the type of
mechanisms that induce metal fatigue.
Summary
1. The practical relevance of cyclic-slip to metallic component durability
is overrated. Slip-driven fatigue dominates low-cycle fatigue and crack
growth at rates exceeding 104 mm/cycle. In durable fatigue designs
most of the fatigue life is expended at crack growth rates below the
Paris Regime. Atmospheric metal fatigue under these conditions is
controlled by the near-threshold response, where the consequences of
cyclic crack-tip surface activity overshadow the possible consequences
of cyclic slip.
2. Crack-tip surface activity progresses during each rising load half-cycle
with rising near-tip stress acting as a diffusion pump to promote
embrittlement or chemical weakening of surface atomic layers and
associated accelerated crack extension. At ambient temperature, reac-
tion with moisture releases hydrogen for diffusion. At elevated temper-
ature, oxidation is involved. The depth and extent of such an attack is
moderated by local hydrostatic stress, that in turn, is determined by
the stress ratio and cycle-sequence sensitive near-tip residual stress.
The effect is restricted to crack-tip surface atomic layers and therefore
becomes insignificant as the growth rate progresses into the Paris Re-
gime. It is totally absent in high vacuum.
3. For the purpose of understanding its effect on metal fatigue, residual
stress may be divided into remote (or macroscopic, or crack-free)
stress distribution and the local (microscopic) field associated with the
crack tip response.
4. The macroscopic field, including residual stresses left by mechanical
processing, and those induced by local inelastic static or cyclic
response such as at notches, control the local stress ratio and associ-
ated fatigue crack closure in conjunction with applied cyclic load

conditions and crack wake development into the monotonic plastic

zone. These can be computed using state-of-the-art analytical tools
that determine the material and structural stress-strain response along
with the stress intensity function for a given crack size, shape, and
path.
5. Crack closure can be unambiguously determined only from the shape
and closure of the loop formed either by the inelastic near-tip strain,
or, by non-linear hysteretic near-tip wake displacement when plotted
against the applied load. The conventional approach of monitoring
wake contact therefore appears misdirected and may be the root cause
of incorrect closure estimates reported in the literature, be it computed
values, or those obtained experimentally using standard practices such
as ASTM E-647.
6. The microscopic crack-tip stress field is the result of the action of the
next loading cycle, superposed on the residual crack-tip stress-strain
field at the end of the previous unloading half cycle. This field will
reflect the effect of the macroscopic residual stress distribution in the
material (through stress intensity), along with that of the monotonic
plastic zone.
7. In atmospheric fatigue, by shifting the near-tip stress up or down, the
microscopic field moderates diffusion kinetics to determine instanta-
neous threshold stress intensity, strictly speaking, for a given ambient
partial pressure of active species (humidity), temperature, and cycling
frequency. Vacuum threshold stress intensity serves as its upper limit.
There is no known lower limit.
8. The microscopic field will exhibit a significant cyclic inelastic hyste-
retic response while the crack is fully open. The associated change in
the threshold stress intensity is the root cause for cycle-by-cycle (hyste-
retic) load sequence sensitivity in variable-amplitude metal fatigue.
This component of sequence sensitivity will be absent in high vacuum
and diminish in air into the Paris Regime because its effect is restricted
to crack-tip surface atomic layers.
9. Conventional modeling of notch fatigue under variable amplitude load-
ing using the local stress-strain (LSS) approach does not carry any sci-
entific rationale. If it does correctly describe trends in fatigue response,
it is by the coincidental qualitative similarity of the cyclic notch root
and crack-tip response. This similarity vanishes under a fully elastic
notch root response, exposing the invalidity of the LSS approach.
10. Commercially available models of variable amplitude fatigue crack
growth, including the Wheeler and Willenborg models, focus on the
monotonic plastic zone and therefore, essentially address crack clo-
sure. These models treat threshold stress intensity as a material con-
stant even under variable amplitude loading. Therefore, correct
estimations by such models of the atmospheric variable amplitude fa-
tigue growth rate below 104 mm/cycle can only be by accident.
11. For a given applied load cycle, crack front orientation, and tortuosity
moderate crack-tip stress-strain response, crack closure determines its

SUNDER, doi:10.1520/JAI103940 61
effective magnitude and the near tip stress response superposed on re-
sidual stress determines instantaneous resistance (threshold stress in-
tensity). The first is sensitive to the crack extension history. The second
is sensitive to the crack extension and loading history. The third is sensi-
tive to the load cycle-sequence and loading history. Variable-amplitude
fatigue response needs to be modeled as the synergy of all three.
12. Further improvements to analytical modeling of variable-amplitude
fatigue demand consideration of threshold stress intensity as a cycle-
sequence sensitive variable. They would also benefit from reliable labo-
ratory measurements and analytical estimates of crack closure and
from improved characterization of the crack-tip response to variations
in crack front geometry (shielding effects). Such studies should include
the effect of constraint.
Acknowledgments
Some of the experiments and all of the reported fractography were performed
at the Air Force Research Laboratories (AFRL), WPAFB, OH, USA. Other
experiments were performed at BiSS Research, Bangalore. The author deeply
appreciates the support and encouragement provided by colleagues in both lab-
oratories and also the University of Dayton Research Institute (UDRI).
References
[1] Thorneycroft, T., “On the Form of Shafts and Axles,” Proceedings of the Institution
of Mechanical Engineers, London, Oct 1850, pp. 35–41 pp. 4–15.
[2] Braithwaite, F., “On the Fatigue and Subsequent Fracture of Metals,” Proc. Inst.
Civ. Eng, London, May 1854.
[3] Wohler, A., Uber die Festigkeitsversuche mit Eisen und Stahl, Berlin, Ernst und
Korn, 1870.
[4] Anon, “Wöhler’s Experiments on the strength of Metals,” Engineering, Vol. 4, 1867,
pp. 160–161.
[5] Manson, S. S., Future “Directions for Low Cycle Fatigue,” Low Cycle Fatigue, ASTM
Spec. Tech. Publ. 942, H. D. Solomon, G. R. Halford, and B. N. Leis, Eds., American
Society for Testing Materials, Philadelphia, 1988, pp. 15–39.
[6] Miner, M. A., “Cumulative Damage in Fatigue,” Trans. ASME J. Appl. Mech., Vol.
12, 1945, pp. A159–A164.
[7] Anon., “Wöhler’s Experiments on the “Fatigue”of Metals,” Engineering, June 1871,
pp. 199–441.
[8] Bauschinger, J., “On the Change of the Elastic Limit and Strength of Iron and Steel
by Tension and Compression, by Heating and Cooling and by Often Repeated
Loading,” Technical Report, Munich Technical Univ., Munich, Germany, 1886 (in
German).
[9] Goodman, J., Mechanics Applied to Engineering, Longmans-Green, London, 1899.
[10] Raju, K. N., Workshop on Fatigue, Fracture and Failure Analysis, Notes, Vol. 1,
National Aeronautical Laboratory, Bangalore, March 1979.
[11] Hull, D., Bacon, D. J., Introduction to Dislocations, Fourth Edition, Butterworth-
Heinemann, Oxford, 2001.

[12] Grosskreutz, J. C., “A Critical Review of Micromechanisms in Fatigue,” Fatigue –

An Interdisciplinary Approach, Syracuse University Press, Syracuse, 1964, pp. 27–
62.
[13] Schijve, J., Fatigue of Materials and Structures, Springer, New York, 2003, p. 317.
[14] Gassner, E., “Strength Experiments Under Cyclic Loading in Aircraft Structures,”
Luftwissen Vol. 6, 1939, pp. 61–64 (in German).
[15] Neuber, H., 1937, Kerbspanmungslehre [Theory of notch stresses], J.W. Edwards,
Ed., Springer, Berlin, Ann Arbor, MI, 1946.
[16] Ramberg, W. and Osgood, W. R., “Description of Stress-Strain Curves by Three
Parameters,” Technical Report No. 902, National Advisory Committee for Aeronau-
tics, Washington, D.C., 1943.
[17] Wetzel, R. M., 1971, “A Method for Fatigue Damage Analysis,” Ph.D. thesis, Dept.
of Civil Engineering, Univ. of Waterloo, ON, Canada.
[18] SAE Fatigue Design Handbook, 3rd ed., AE-22, Society for Automotive Engineers,
Warrendale, PA, 1997).
[19] Matsuishi, M. and Endo, T., “Fatigue of Metals Subjected to Varying Stress,” Trans.
Jpn. Soc. Mech. Eng., March, 1968.
[20] van Dijk, G. M. and de Jonge, J. B., “Introduction to a Fighter Aircraft Loading
Standard for Fatigue Evaluation ‘FALSTAFF’,” Report No. NLR MP 75017U,
National Aerospace Laboratory, Amsterdam, The Netherlands, 1975.
[21] de Jonge, J. B., Schutz, D., Lowak, H., and Schijve, J., “A Standardized Load
Sequence for Flight Simulation Tests on Transport Aircraft Wing Structures,”
Report No. NLR TR 73029U, National Aerospace Laboratory (NLR), Amsterdam,
The Netherlands, 1973.
[22] Sunder, R., “Rainflow Applications in Accelerated Cumulative Fatigue Analysis,”
The Rainflow Method in Fatigue, Y. Murakami, Ed., Butterworth -Heinemann,
Oxford, 1992, pp. 67–76.
[23] Neggard, G. R., “The History of the Aircraft Structural Integrity Program,” Report
No. 680.1B, Aerospace Structural Information and Analysis Center (ASIAC), Flight
Dynamics Directorate, WPAFB, Ohio 1980.
[24] Schijve, J., “Cumulative Damage Problems in Aircraft Structures and Materials,”
Aeronaut. J., Vol. 74, 1970, pp. 517–532.
[25] Sunder, R., Porter, J., and Ashbaugh, N. E., “The Effect of Stress Ratio on Fatigue
Crack Growth Rate in the Absence of Crack Closure,” Int. J. Fatigue, Vol. 19, 1997,
pp. S211–S221.
[26] Sunder, R., “Contribution of Individual Load Cycles to Fatigue Crack Growth
under Aircraft Spectrum Loading,” M. R. Mitchell and R. W. Landgraf, ASTM Spec.
Tech. Publ., Vol. 1122, 1992, pp.176–190.
[27] Paris, P. C., Gomez, M. P., and Anderson, W. E., “A Rational Analytical Theory of
Fatigue,” Trend Eng., Vol. 13, 1961, pp. 9–14.
[28] Figge, I. E., and Newman, J. C., “Fatigue Crack Propagation in Structures with
Simulated Rivet Forces,” ASTM Spec. Tech. Publ., Vol. 415, 1967, pp. 71–93.
[29] Schijve, J., “Effect of Load Sequences on Crack Propagation Under Random and
Program Loading,” Eng. Fracture Mech., Vol. 5, 1973, pp. 269–280.
[30] Wheeler, O. E., “Spectrum Loading and Crack Growth,” ASME J Basic Eng., Vol.
94, 1972, pp. 181–186.
[31] Willenborg, J., Engle, R. H., and Wood, H. A., “A Crack Growth Retardation Model
Based on Effective Stress Concepts,” Report No. AFFDL-TM-71-1 FBR, WPAFB,
OH, 1971.
[32] Walker, E. K., “Effects of Environment and Complex Load History on Fatigue
Life,” ASTM Spec. Tech. Publ., Vol. 462, 1970, pp. 1–14.

SUNDER, doi:10.1520/JAI103940 63
[33] Elber, W., “The Significance of Fatigue Crack Closure,” ASTM Spec. Tech. Publ.,
Vol. 486, 1971, pp. 230–242.
[34] Newman, J. C., A Crack-Closure Model for Predicting Fatigue Crack Growth Under
Aircraft Spectrum Loading. Methods and Models for Predicting Fatigue Crack Growth
Under Random Loading, ASTM Spec. Tech. Publ., J.B. Chang and C. M. Hudson,
Eds., Vol. 748, 1981, pp. 53–84.
[35] de Koning, A. U., “A Simple Crack Closure Model for Prediction of Fatigue Crack
Growth Rates Under Variable-Amplitude Loading,” Fracture Mechanics, ASTM
Spec. Tech. Publ., R. Roberts, Ed., Vol. 743, 1981, pp. 63–85.
[36] Ashbaugh, N. E., Dattaguru, B., Khobaib, M., Nicholas, T., Prakash, R. V., Rama-
murthy, T. S., and Seshadri, B. R., “Experimental and Analytical Estimates of Fa-
tigue Crack Closure in an Aluminum–Copper Alloy. Part I. Laser Interferometry
and electron Fractography,” Fatigue Fract. Eng. Mater. Struct., Vol. 20(7), 1997, pp.
951–961.
[37] Sunder, R., and Dash, P. K., “Measurement of Fatigue Crack Closure Through Elec-
tron Microscopy,” Int. J. Fatigue, Vol. 4, April 1982, pp. 97–105.
[38] Schijve, J., “Four Lectures on Fatigue Crack Growth,” Eng. Fracture Mech., Vol. 11,
1979, pp. 176–221.
[39] Sunder, R., “A Unified Model of Fatigue Kinetics Based on Crack Driving Force and
Material Resistance,” Int. J. Fatigue, Vol. 29, 2007, pp. 1681–1696.
[40] Riddell, W. T. and Piascik, R. S., “Stress Ratio Effects on Crack Opening Loads and
Crack Growth Rates in Aluminum Alloy 2024,” Fatigue Fracture Mechanics, ASTM
Spec. Tech. Publ., Vol. 1332, T. L. Panontin and S. D. Sheppard, Eds., Vol. 29, Amer-
ican Society for Testing and Materials, West Conshohocken, PA, 1999, pp. 407–25.
[41] Forsyth, P. J. E., “Fatigue Damage and Crack Growth in Aluminium Alloys,” Acta
Metall., Vol. 11, 1963, pp. 703–719.
[42] Sunder, R., “Binary Coded Event Registration on Fatigue Fracture Surfaces,” J.
Soc. Env. Engrs., SEECO, London, 1983, p. 197.
[43] Laird, C., Mechanisms and Theories of Fatigue, Fatigue and Microstructure, ASM,
Metals Park, OH,1978, pp. 149–204.
[44] Zhang,. J. Z., “A Shear Band Decohesion Model for Small Fatigue Crack Growth in
an Ultra-Fine Grain Aluminum Alloy,” EFM, Vol. 65, 2001, pp. 665–681.
[45] Sunder, R., “Fatigue as a Process of Brittle Micro-Fracture,” FFEMS, Vol. 28(3),
2005, pp. 289–300.
[46] Gsell, M., Jakob, P., and Menzel, D., “Effect of Substrate Strain on Adsorption,” Sci-
ence Vol. 280, 1998, pp. 717–720.
[47] Ro, Y., Agnew, S. R., and Gangloff, R. P., “Environmental Fatigue-Crack Surface
Crystallography for Al-Zn-Cu-Mg-Mn/Zr,” Metall. Mater. Trans. A, Vol. 39A, 2008,
pp. 1449–1465.
[48] Schijve, J. and Arkema, W. J., “Crack Closure And the Environmental Effect on
Fracture Mode Transition in Fatigue Crack Growth,” Report No. VTH-217, Delft
Univ., Delft, The Netherlands, 1976.
[49] Vogelesang, L. B. and Schijve, J., “Environmental Effects on Fatigue Failure Mode
Transition Observed in Aluminium Alloys” Report No. LR-289, Delft Univ. of Tech-
nology, Delft, The Netherlands, 1979.
[50] Gach, E. and Pippan, R., “Cyclic Crack Tip Deformation – the Influence of Environ-
ment,” Proceedings of the Tenth International Conference on Fracture, International
Congress of Fracture, Hawaii, Dec 2001 [Paper ICF 100420OR].
[51] Bowles, C. Q., 1978, “The Role of Environment, Frequency and Wave Shape During
Fatigue Crack Growth of Aluminum Alloys,” Ph.D. thesis, Report No. LR-270, Delft
Univ. of Technology, Delft, The Netherlands.

[52] Petit, J., Henaff, G., and Sarrazin-Baudoux, C., “Mechanisms and Modeling of
Near-Threshold Fatigue Crack Propagation, Fatigue Crack Growth Thresholds, En-
durance Limits and Design,” ASTM Spec. Tech. Publ., J. C. Newman, Jr. and R. S.
Piascik, Eds., Vol. 1372, American Society for Testing and Materials, West Consho-
hocken, PA, 2000.
[53] Bradshaw, F. J. and Wheeler, C., “The Effect of Gaseous Environment and Fatigue
Frequency on the Growth of Fatigue Cracks in Some Aluminium Alloys,” Int. J
Fract. Mech., Vol. 6, 1969, pp. 255–268.
[54] Sunder, R., Porter, W. J., and Ashbaugh, N. E., “Stress-Level Dependent Stress Ratio
Effect on Fatigue Crack Growth, Fatigue and Fracture Mechanics: Twenty-Ninth
Volume,” ASTM Spec. Tech. Publ., Vol. 1332, T. L. Panontin and S. D. Sheppard,
Eds., American Society for Testing and Materials, West Conshohocken, PA, 1999.
[55] Sunder, R., Porter, W. J., and Ashbaugh, N. E., “Fatigue Voids and Their Signifi-
cance,” Fatigue Fract. Eng. Mater. Struct., Vol. 25, 2002, pp. 1015–1024.
[56] Sunder, R., Porter, W. J., and Ashbaugh, N. E., “The Role of Air in Fatigue Load
Interaction,” Fatigue Fract. Eng. Mater. Struct., Vol. 26, 2003, pp. 1–16.
[57] Sunder, R., “On the Hysteretic Nature of Variable-Amplitude Fatigue Crack
Growth,” Int. J. Fatigue, Vol. 27, 2005, pp. 1494–1498.
[58] Sunder, R., “Fractographic Reassessment of the Significance of Fatigue Crack Clo-
sure, Fatigue and Fracture Mechanics,” ASTM Spec. Tech. Publ., Vol. 1461, S. R.
Daniewicz, J. C. Newman, and K. H. Schwalbe, Eds., American Society for Testing
Materials, Philadelphia, Vol. 34, 2005, pp. 22–39.
[59] Anandan, K. and Sunder, R., “Closure of Part-Through Cracks at the Notch Root,”
Int. J. Fatigue, Vol. 9, 1987, pp. 217–222.
[60] Ashbaugh, N. E., Porter, W. J., Rosenberger, A. H., and Sunder, R., “Environment-
Related Load History Effects in Elevated Temperature Fatigue of a Nickel-Base
Super-Alloy,” Proceedings Fatigue, Stockholm, Sweden, June 2-7, 2002, EMAS (2002).
[61] Sunder, R., “Effect of Periodic Overloads on Threshold Fatigue Crack Growth in
Al-Alloys,” Fatigue Fracture Mechanics, ASTM Spec. Tech. Publ., S. R. Daniewicz, J.
C. Newman, K. H. Schwalbe, Eds., Vol. 1461, American Society for Testing Materi-
als, Vol. 34, 2005, pp. 557–572.
[62] Sunder, R., Prakash, R. V., and Mitchenko, E. I., “Growth of Artifically and Natu-
rally Initiating Notch Root Cracks under FALSTAFF Spectrum Loading,” Report
No. 797, AGARD, Paper 10, 1994.
[63] Sunder, R., Prakash, R. V., and Mitchenko, E. I., “Fractographic Study of Notch Fa-
tigue Crack Closure and Growth Rates,” ASTM Spec. Tech. Publ., J. E. Masters and
L. N. Gilbertson, Eds., Vol. 1203, 1993, pp. 113–131.
[64] Lang, M., “A Model for Fatigue Crack Growth, Part I: Phenomenology,” Fatigue
Fract. Eng. Mater. Struct., Vol. 23, No. 7, 2000, pp. 587–601.
[65] Lados, D. A., Apelian, D., and Donald, J. K., “Fracture Mechanics Analysis for Re-
sidual Stress and Crack Closure Corrections,” Int. J. Fatigue, Vol. 29, 2007, pp. 687–
694.
[66] Seshadri, B. R. and Newman, Jr., J. C., “Elastic-Plastic Finite Element Contact
Stress Analyses of Tapered Fasteners,” 4th Joint DoD/FAA/NASA Conference on
Aging Aircraft, St. Louis, MO, May 2000.
[67] Seshadri, B. R. and Newman Jr., J. C., “Numerical Investigation of Interference-Fit
Tapered Fasteners,” USAF Aircraft Structural Integrity Program Conference, San
Antonio, TX, Dec 2000.
[68] Ro, Y., Agnew. S. R., Bray, G. H., and Gangloff, R. P., “Environment-Exposure–
Dependent Fatigue Crack Growth Kinetics for Al-Cu-Mg/Li,” Mater. Sci. Eng., A
Vol. 468–470, 2007, pp. 88–97.

J_ID: DOI: Date: 16-July-12 Stage: Page: 65 Total Pages: 22

doi:10.1520/JAI104071
Michael R. Hill1 and Jihwi Kim2
Fatigue Crack Closure in Residual Stress

Bearing Materials
ABSTRACT: During fatigue crack growth, the two opposing faces of a fa-
tigue crack can make physical contact while unloading from a maximum level
of cyclic load, so that the crack tip state at the minimum cyclic load depends
on the host geometry, material properties, and loading history. Although sig-
nificant work has been performed in order to examine the effects of crack
face contact, often called crack closure, under variations of applied loading
history, little work has been done to understand the details of crack closure in
materials that contain bulk residual stress fields. For an elastic material, var-
iations of applied load history create changes in the crack tip behavior that
are directly related to the current levels of cyclic stress, with no effect of prior
loading. For an elastic-plastic material, variations of the applied load history
cause the crack tip behavior to depend on the current and former loading
cycles, because of plastic deformation in the crack wake. In an elastic mate-
rial with bulk residual stress, crack closure occurs because the strain fields
locked into the material, which are the source of the residual stress, alter the
shape of the crack faces, so that the details of closure depend on the residual
stress field and crack geometry. Residual stresses might therefore affect fa-
tigue crack growth in two distinct ways: first, by combining with applied loads
to affect the stress intensity factor (at the current crack size), and second, by
altering crack closure. We emphasize that the effect of bulk residual stresses
on crack closure described here is an elastic effect, which distinguishes it
from the more commonly discussed forms of closure, such as arise from
Manuscript received June 3, 2011; accepted for publication October 18, 2011; published
1
Dept. of Mechanical and Aerospace Engineering, Univ. of California, One Shields Ave.,
Davis, CA 95616 (Corresponding author), e-mail: mrhill@ucdavis.edu
2
Dept. of Mechanical and Aerospace Engineering, Univ. of California, One Shields Ave.,
Davis, CA 95616.
Cite as: Hill, M. R. and Kim, J., “Fatigue Crack Closure in Residual Stress Bearing Materi-
als,” J. ASTM Intl., Vol. 9, No. 1. doi:10.1520/JAI104071.
Copyright V
65
ID: kumarva Time: 11:43 I Path: Q:/3b2/STP#/Vol01546/120221/APPFile/AI-STP#120221

plasticity or roughness. The paper describes a means to forecast crack

closure due to bulk residual stress fields and assesses schemes to account
for its effects on fatigue crack growth.
KEYWORDS: fatigue crack growth, fatigue crack closure, residual stress,
contact pressure
Nomenclature
a¼ crack size
da/dN ¼ fatigue crack growth rate
E0 ¼ effective elastic modulus
f (xi, Nj) ¼ crack face displacement at xi due to piecewise linear pressure at
node j
F¼ crack face displacement due to piecewise pressure matrix
K(a) ¼ stress intensity factor at crack size a
Kapp,max ¼ stress intensity factor due to maximum applied load
Kapp,min ¼ stress intensity factor due to minimum applied load
Kcp ¼ stress intensity factor due to contact pressure
Krs ¼ stress intensity factor due to residual stress
Ktot,max ¼ maximum total stress intensity factor
Ktot,min ¼ minimum total stress intensity factor
m(x, a) ¼ weight function
MS ¼ modified superposition method
Nj(x) ¼ piecewise linear basis function for node j
NSC ¼ new superposition contact method
p¼ contact pressure vector
pj ¼ contact pressure at node j
p(x) ¼ contact pressure distribution along the crack line
Rapp ¼ applied stress ratio
Rtot ¼ total stress ratio
S¼ superposition method
SC ¼ superposition contact method
uapp ¼ crack face displacement vector due to applied load
ucp ¼ crack face displacement vector due to contact pressure
urs ¼ crack face displacement vector due to residual stress
u(x, a) ¼ crack face displacement
W¼ coupon characteristic width
DKtot ¼ total stress intensity factor range
r(x) ¼ crack-line stress in uncracked configuration
Introduction
Residual stresses affect fatigue crack growth behavior, and this paper describes an
analytical and numerical approach for predicting crack closure in bodies contain-
ing residual stresses with long length scale, called bulk residual stresses, and dem-
onstrates the approach with comparisons of predicted and observed fatigue crack
growth rate behavior. Bulk residual stresses often exist in mechanical components

HILL AND KIM, doi:10.1520/JAI104071 67
as an undesired consequence of manufacture, and sometimes of attempts to

beneficially affect performance (e.g., compressive residual stress treatments such
as the tempering of plate glass or peening of metallic parts). Cutting a surface
through a residual stress bearing part results in two pieces that do not fit back to-
gether on account of the misfit strain fields that cause the residual stresses; there-
fore, crack face contact during fatigue crack growth in residual stress bearing
material is distinct from that in stress-free material. Predictions of fatigue crack
growth in residual stress bearing components should therefore consider crack clo-
sure due to elastic behavior at the minimum applied load, just as they consider
closure due to plasticity or roughness [1].
Fatigue crack closure under cyclic loading has been studied extensively
because it is associated with the fatigue crack growth rate (FCGR). Since Elber
[2] described the crack closure phenomenon, many researchers have investi-
gated the effects of crack face contact with experimental and numerical methods
in variations of coupon geometry, crack geometry, material properties, loading
conditions, and closure mechanisms [3–12]. McEvily [13], from many of the ear-
lier works, identified various mechanisms of crack closure: plasticity induced,
roughness induced, crack filling, transitional, transformation induced, and grain
boundary closure. Excluding grain boundary closure, all these mechanisms are
related to the crack wake that contains plastic or other permanent deformation.
Compared to the large amount of earlier work for bodies without residual
stress, little research has been conducted for residual stress bearing bodies
[1,14–17]. The present paper considers crack closure in a fully elastic residual
stress bearing material and its effect on FCGR behavior. When the displacement
of the crack face is negative due to residual stress in a fully elastic material, over-
closure of the crack faces is prevented by crack face contact [18] and resulting
contact pressure, which alters the crack tip fields and FCGR. For FCGR predic-
tion in residual stress bearing material, the stress intensity factor due to applied
loading and the residual stress are commonly added together by the principle of
superposition. When crack face contact occurs, the stress intensity factor due to
contact pressure also might be included in the analysis through superposition,
but finding the pressure distribution requires a non-linear analysis (often per-
formed using finite element methods).
Here we describe a practical approach for determining the contact pressure
distribution for a one-dimensional crack that combines analytical calculations
for crack face displacement with iterative numerical calculations for the pres-
sure distribution. The crack face displacement at a specific crack size is
obtained from a crack-line stress field and the weight function by integration
[1,19–22]. Given crack face displacement fields for the minimum applied load
and for residual stress, a minimum piecewise-linear crack face pressure distri-
bution that provides an admissible crack shape (without over-closure) is deter-
mined via techniques of quadratic programming. We apply the proposed
approach to a set of aluminum C(T) coupons that contain residual stress due to
laser shock peening [23]. The predicted crack face shape and contact pressure
are verified against finite element analysis. Comparison of the predicted and
observed FCGR behavior for three variations of constant amplitude applied
loading illustrates the value of including contact pressure in FCGR predictions.

Methods
Here we consider a straight, one-dimensional crack along a symmetry plane of
a two-dimensional body for which a weight function is available. The applied
loading and residual stress are assumed to be symmetrical about the crack
plane and create only an opening-mode stress intensity factor (mode I). At the
minimum applied load, there can be four manifestations of crack face contact
(Fig. 1): a fully closed crack, a partially open crack with a closed crack tip
region, a partially open crack with an open crack tip region, and a fully open
crack. For simplicity, we assume no crack face contact at the maximum applied
load.
Contact Pressure on Crack Faces

For an edge crack of size a, Parker [24] showed that the vertical displacement of
the crack face u(x, a) can be computed from the stress intensity factor and the
weight function by
ða
1
uðx; aÞ ¼ mðx; aÞKðaÞda (1)
E0 x
where:
E0 ¼ effective elastic modulus (E for plane stress and E/(1 2) for plane
strain),
m(x, a) ¼ weight function,
K(a) ¼ stress intensity factor, and the coordinate system has x along the
cracking-driving direction with the origin at the crack mouth. Further, the
stress intensity factor is a function of the crack line stress in the uncracked
body r(x)
ða
KðaÞ ¼ rðxÞmðx; aÞdx (2)
0
FIG. 1—Four types of crack face shape for a body subjected to applied load and residual
stress: (a) fully closed crack, (b) partially open crack with closed crack tip region,
(c) partially open crack with open crack tip region, and (d) fully open crack.

Therefore, if a weight function for a specific geometry and the stress distribu-
tion at the crack-line are known, the shape of the crack face can be predicted via
integration.
In order to find the contact pressure between crack faces, the pressure is
expressed as a combination of piecewise linear basis functions. Using n evenly
spaced node points (x1, x2,…, xn) along the crack face (Fig. 2), the contact pres-
sure along the crack face is
X
n
pðxÞ ¼ pj Nj ðxÞ (3)
j¼1
where:
pj is the contact pressure at node point j, and
Nj is the usual piecewise linear basis function
8xx
j1
>
> ðxj1 x xj Þ
>
> x x
>
< j j1
Nj ðxÞ ¼ x jþ1 x (4)

> ðxj x xjþ1 Þ
>
> xjþ1 xj
>
>
:
0 elsewhere:
Given values of pj, and taking the crack line stress as the pressure, the crack face
displacement at nodal location xi due to contact pressure is
X
n
ucp ðxi ; aÞ ¼ f ðxi ; Nj Þpj (5)
j¼1
where
ða ð a
1
f ðxi ; Nj Þ ¼ mðxi ; aÞ Nj ðx0 Þmðx0 ; aÞdx0 da (6)
E0 xi 0
FIG. 2—Crack face schematic showing negative displacement region, n control points,
n 1 evenly-spaced intervals, and piecewise linear basis functions.

For a specific crack size, Eq 5 can be rewritten in a matrix-vector form as
ucp ¼ F p (7)
where:
ucp ¼ n 1 vector having the component ucp(xi, a),
F ¼ n n matrix having the component f(xi, Nj), and
p ¼ n 1 vector having the component pj.
When the crack faces are in contact, two constraints should be met: the
total displacement utot and the contact pressure p at all points must be equal to
or greater than zero (no over-closure and no negative pressure). The total dis-
placement of the crack face is the sum of contributions from applied stress, re-
sidual stress, and contact pressure, so the constraint for the total displacement
can be written as
F p ðuapp þ urs Þ (8)
where uapp and urs are vectors of crack face displacement, analogous to ucp,
determined from crack-line applied and residual stress fields, respectively.
From this inequality, a minimized contact pressure vector p can be computed
by an iterative calculation using techniques of quadratic programming (imple-
mented in commercial software [25]; see Appendix).
Fatigue Crack Growth Prediction

With stress intensity factors known for the applied load, residual stress, and
contact pressure, each as a function of crack size, their combination can be
used to determine FCGR as a function of crack size. To compute FCGR, it is typ-
ical to compute a total stress intensity factor range (DKtot) and total stress ratio
(Rtot) from the available stress intensity factor values. Four different methods
for computing DKtot and Rtot at each crack length from available values of stress
intensity factors are given in Table 1, which includes three methods suggested
earlier in the literature [1,17] and a fourth that is newly suggested, and in which
TABLE 1—Fatigue crack growth prediction methods assuming Kcp ¼ 0 at maximum applied
load: superposition (S), modified superposition (MS), superposition contact (SC), and new
superposition contact (NSC).
Method Ktot,max Ktot,min
S Kapp,max þ Krs Kapp,min þ Krs

0; 0
MS Kapp,max þ Krs if (Kapp,min þ Krs)
> 0; Kapp;min þ Krs

0; Kapp;min þ Krs
SC Kapp,max þ Krs if (Kapp,min þ Krs þ Kcp)
> 0; Kapp;min þ Krs þ Kcp
NSC Kapp,max þ Krs Kapp,min þ Krs þ Kcp

the differences among the methods are limited to the definition of the minimum
total stress intensity factor Ktot,min.
The first method is referred to as superposition (S), and it ignores crack
face contact. Because the stress intensity factor due to residual stress (Krs) is
added to both the maximum and the minimum applied stress intensity factors
(Kapp,max and Kapp,min), the residual stress appears in Rtot but not in DKtot.
The second method is the modified superposition (MS) method, which also
ignores crack face contact. This method is equivalent to the S method when
Kapp,min þ Krs 0, but replaces with zero negative values of the minimum total
stress intensity factor and stress ratio.
The third method is the superposition contact (SC) method suggested by
Jones and Dunn [17], which includes crack face contact. They used a finite ele-
ment model to obtain Ktot,min that included contributions from the applied load,
residual stress, and crack face contact. In the present paper, we include the
stress intensity factor due to contact pressure (Kcp) explicitly, which is super-
posed with Kapp,min and Krs to give
Ktot;min ¼ Kapp;min þ Krs þ Kcp (9)
Jones and Dunn applied this definition of Ktot,min when it gave a positive value,
but when it gave a zero value, they ignored crack face contact and used
Ktot,min ¼ Kapp,min þ Krs (which had values less than 0 in their work). They
argued that using Ktot,min ¼ 0 would provide a non-conservative FCGR assess-
ment because residual stress free materials show higher FCGR when tested
under a negative applied stress ratio. Because Jones and Dunn employed finite
element derived values of the stress intensity factor, and because the finite ele-
ment software they used does not report negative stress intensity factor values,
they could not have encountered negative values of Ktot,min.
The fourth method is the new superposition contact (NSC) method, which
uses Eq 9 to give the minimum total stress intensity factor regardless of value.
Depending on the specific distributions of stress fields and details of crack face
displacements, we find it possible to have negative values of Kapp,min þ Krs þ Kcp.
For each of the four methods, we compute DKtot and Rtot as
DKtot ¼ Ktot;max Ktot;min

Ktot;min
Rtot ¼ (10)
Ktot;max
With DKtot and Rtot defined, FCGR is determined through a correlation such as
the NASGRO equation, the multi-linear approach suggested by Newman [26],
or other, similar equations.
Application to C(T) Coupon

To illustrate the above approach, we apply it to the conditions of an experimen-
tal program performed earlier to investigate the correlation of residual stress
effects in fatigue and fracture [23,27]. Fatigue crack growth experiments were

carried out on standard C(T) coupons that were in various conditions of resid-
ual stress. Here we consider results for as-machined (AM) coupons (which had
negligible residual stress) and coupons with three-layer laser shock peening
(LSP) applied in a square region near the front-face of the coupon (Fig. 3). Cou-
pon processing details can be found in our earlier work [23,27].
Coupon Description—The earlier work used a set of C(T) coupons with a

characteristic width W ¼ 50.8 mm and a thickness of 3.8 mm that were cut from
a single sheet of 7075-T6 aluminum alloy sothat cracking was in the L-T orienta-
tion. The coupon dimensions are shown in Fig. 3. LSP was appliedidentically to
a subset of coupon blanks that had holes but did not have crack starter notches.
Some coupons were used for residual stress measurements, and others were
used for fatigue crack growth testing.
Residual Stress—Through-thickness average residual stress on the crack

plane was measured in LSP coupons using the slitting method, and results are
shown in Fig. 4. Replicate measurements on identically prepared coupons
showed very similar results, and the distribution in Fig. 4 is used for further anal-
ysis. The residual stress distribution in the coupons represents a combination of
compressive residual stress in the laser peened region (x ¼ 12.7 to 35.6 mm) with
plate bending and axial stresses that arise in the coupon to achieve residual stress
equilibrium (zero net force and moment across the crack plane).
Fatigue Crack Growth Testing—The earlier fatigue crack growth testing fol-
lowed ASTM E 647 and was performed for a variety of applied loadings. Here,
we consider the four tests listed in Table 2. Each of the four tests has a designa-
tion in Table 2 (AM1, LSP1, LSP2, or LSP3) that will be used for further discus-
sion. Tests were performed under constant amplitude applied load (at stated
values of maximum load Pmax and applied stress ratio Rapp) or constant
FIG. 3—Compact tension coupon geometry and LSP region; dimensions in mm [23].

FIG. 4—Measured residual stress in the C(T) coupon versus position from the front
face (data adapted from Ref. [23]).
amplitude stress intensity factor DKapp. One AM and one LSP coupon were
tested under constant amplitude load, with the AM coupon (AM1) tested at a
lower load level than the LSP coupon (LSP1) (Table 2). Two other LSP coupons
were tested under constant DKapp, one at DKapp ¼ 22.0 MPa m0.5 and Rapp ¼ 0.1
(LSP2) and the other at DKapp ¼ 11.0 MPa m0.5 and Rapp ¼ 0.5 (LSP3). Prior to fa-
tigue testing, a machined notch was cut into each coupon via wire electric dis-
charge machining to a notch length of 10.2 mm (measured from the hole
center). Fatigue precracking was not performed, but post-test data analysis sug-
gested that the fatigue crack growth rate was unaffected by the notch after 1 to
2 mm of crack growth, and only unaffected data were reported.
Weight Function—For the C(T) coupon, we used a recently published weight

function [28] to calculate the stress intensity factors and crack face shape. The
TABLE 2—Coupon designation, loading, and condition [27].
Designation Rapp Pmax, kN DKapp, MPa m0.5 Condition
AM1 0.1 0.98 … AM

LSP1 0.1 2.22 … LSP
LSP2 0.1 … 22.0 LSP
LSP3 0.5 … 11.0 LSP

weight function is valid for all values of x but only a limited range of crack sizes
(0.2 a/W 0.9), which is sufficient for the present tests that have a notch
length of 0.2W and a final crack size of about 0.8W.
Crack Face Displacement and Contact Pressure Calculation—Crack face dis-

placement and contact pressure calculations were performed for 26 crack sizes
from 15 mm to 40 mm in 1 mm increments for all four applied loading cases.
For each crack size, 40 evenly spaced nodes were used along the crack line,
excluding the crack tip, so F was a 40 40 matrix. Crack face displacements
were computed for applied and residual stress using an equation analogous to
Eq 7. The calculations used crack-line stress for applied pin load from Ref 28
and the residual stress shown in Fig. 4. The contact pressure was computed
using the code in the Appendix. The calculations used an elastic modulus
E ¼ 71.7 GPa and assumed plane stress.
Finite Element Verification—Results from two finite element method (FEM)

analyses are used to verify the crack face displacement calculations; one analy-
sis ignores crack face contact, and the other includes it. The FEM analysis used
a two-dimensional plane stress formulation and an elastic material having
E ¼ 71.7 GPa and a Poisson’s ratio of 0.33. The mesh was half-symmetric and
composed of four-node bilinear plane stress quadrilateral elements. The analy-
sis with contact assumed small sliding contact without friction and used an ex-
ponential pressure-overclosure behavior to enhance convergence. The analysis
without contact was performed for residual stress only (i.e., for Kapp,min ¼ 0),
which was applied as a non-uniform traction on the crack face. The analysis
with contact included residual stress and the minimum load for condition
LSP1. Deformed crack face shapes computed via FEM are compared to results
from the new model proposed here in order to provide verification; the residual
stress only case (without contact) verifies Eq 1, and the case with contact veri-
fies the contact pressure determination.
FCGR Calculation—With stress intensity factors for each type of applied

loading (and each type of stress: applied, residual, and contact pressure), we
use the approach suggested by Newman [26] to compute FCGR from DKtot and
Rtot using a piecewise linear log(DK)log(da/dN) relation. Stuart et al. [29]
applied this method in 7075-T6 sheet using
da C1i ðDKeff ÞC2i

¼ 2 (11)
dN DKtot
1
C3 ð1 Rtot Þ
where

1 So =Smax
DKeff ¼ DKtot (12)
1 Rtot
So =Smax ¼ 0:32566 þ 0:0819R þ 0:85923R2 0:26679R3 (13)

TABLE 3—Points to create a piecewise power-law FCGR-DKeff tabular lookup curve for
7075-T6 Al [29].
DKeff, MPa m0.5 FCGR, m/Cycle
1.00 1.50 1013

1.50 9.95 1010
3.36 8.03 109
4.87 8.36 108
13.52 9.31 107
39.63 3.44 105
with the fitting coefficient C3 ¼ 60 MPa m0.5, and where C1i and C2i correspond
to piecewise power-law fits between the points in Table 3.
Results
Crack Face Displacements and Contact Pressure Distribution

Crack face displacements due to only residual stress are shown for crack sizes
a ¼ 20 and 30 mm in Fig. 5. The results show very good agreement between cal-
culations based on Eq 1 (Theory) and FEM. (Note that whereas FEM results are
FIG. 5—Crack face displacement for residual stress alone: “Theory” calculated from the
weight function compared to FEM results.

available for the whole crack face, including along the initial notch, the results
of Eq 1 are available only for positions where the weight function is valid: 0.2 W
beyond the loading holes, which is 22.86 mm from the front face. In these
experiments, this was not a problem because contact did not occur nearer the
crack mouth due to the clearance afforded by the machined notch; however, a
large sample with a small notch height could exhibit closure in the area of the
machined notch, and this would introduce a complication not encountered
here.) Crack face displacements due to residual stress, the minimum applied
load for LSP1, and contact are shown in Fig. 6, and there is good agreement
between the new calculation method (Theory) and FEM.
At the minimum applied load for test LSP2, crack face displacements, with
and without contact, are shown for selected crack sizes in Fig. 7, and contact
pressure is shown for the same loading in Fig. 8. Figure 7 shows that the crack
is fully open at a ¼ 15 mm, the crack tip region is closed from a ¼ 17 to 21 mm,
and the crack tip region is open for cracks 23 mm and longer. This illustrates
the four types of closure behavior mentioned earlier: fully closed, partially open
with crack-tip closed, partially open with crack-tip open, and fully open (Fig. 1).
Crack size ranges according to this categorization are shown in Table 4 for the
three LSP test conditions.
FIG. 6—Crack face displacement for residual stress, applied minimum load, and con-
tact: “Theory” includes contact via quadratic programming, “FEM” includes contact
implementation.

FIG. 7—Crack face displacements for a range of crack size, computed with and without
accounting for contact at minimum applied loading (DK ¼ 22.0 MPa m0.5, Rapp ¼ 0.1).
FCGR for Constant Amplitude Load Tests

FCGR data observed for test AM1 (constant amplitude load) are shown in Fig. 9
along with the FCGR prediction. The piecewise FCGR correlation shows a good
fit to the experimental data and verifies the use of the piecewise power-law
FCGR correlation.
The results for the LSP constant amplitude load test (LSP1) are shown in
Fig. 10. The figure has six sub-figures that show (a) stress intensity factors
except Ktot,min, (b) Ktot,min for all four methods, (c) DKtot for all methods, (d) Rtot
for all methods, (e) DKeff for all methods, and (f) FCGR for all methods along
with experimental FCGR results (legend key “Exp”). Because Kcp is similar to

FIG. 8—Crack face pressure for a range of crack size at minimum applied loading
(DK ¼ 22.0 MPa m0.5, Rapp ¼ 0.1).
TABLE 4—Crack size ranges according to crack face shapes depending on contact condi-
tions based on the crack face displacement calculation with contact.
Fully Partially Open Partially Open

Designation Closed (Crack Tip Closed) (Crack Tip Open) Fully Open
LSP1 N/A 15 < a < 23 23 a < 38 a ¼ 15, a 38

LSP2 N/A 15 < a < 23 a 23 a ¼ 15
LSP3 N/A 19 < a < 23 23 a < 33 a 19, a 33

FIG. 9—Results for AM, Pmax ¼ 0.98 kN, Rapp ¼ 0.1, fatigue crack growth rate as a
function of DK.
Krs from the initial crack size to 22 mm but of opposite sign, Ktot,min for MS and
NSC is close to zero before a ¼ 22 mm. In the same crack length range, Ktot,min is
highly negative for S and SC [Fig. 10(b)]. This gives rise to a significant differ-
ence in Rtot for these methods [Fig. 10(d)]. It is also noteworthy that Ktot,min crosses
from negative to positive near a ¼ 22 mm for the SC and NSC methods but is nega-
tive or zero until much larger crack sizes (a 28 mm) for S and MS. Relatively
small differences in DKeff [Fig. 10(e)] make significant differences in the predicted
FCGR [Fig. 10(f)]. For smaller crack sizes (a < 20 mm), there is reasonable agree-
ment among all methods and the experimental data. For crack sizes between
20 mm and 25 mm, there are significant differences among prediction methods,
with S and SC falling nearest the experimental data. For crack sizes larger than
25 mm, SC and NSC show good agreement with the experiment. Overall, the SC
method appears to provide the best prediction for the constant amplitude load test.
FCGR for Constant DKapp Tests

The results for the two constant DKapp tests are shown in Fig. 11 and Fig. 12,
which include the same six sub-figures as in Fig. 10. Results for LSP2
(DKapp ¼ 22.0 MPa m0.5, Rapp ¼ 0.1) show that Kcp increases until a ¼ 22 mm and
then decreases linearly, whereas Krs decreases until a ¼ 22 mm and then
increases asymptotically [Fig. 11(a)]. Similar to the results for LSP1, Fig. 11(b)
shows that for SC and NSC, Ktot,min becomes positive at a 23 mm, whereas

FIG. 10—Results for LSP1, Pmax ¼ 2.22 kN, Rapp ¼ 0.1: (a) stress intensity factors,
(b) Ktot,min for all methods, (c) DKtot for all methods, (d) Rtot for all methods, (e) DKeff
for all methods, and (f) FCGR for all methods.

FIG. 11—Results for LSP2, DK ¼ 22.0 MPa m0.5, Rapp ¼ 0.1: (a) stress intensity factors,
(b) Ktot,min for allmethods, (c) DKtot for all methods, (d) Rtot for all methods, (e) DKeff
for all methods, and (f) FCGR forall methods.

FIG. 12—Results for LSP3, DK ¼ 11.0 MPa m0.5, Rapp ¼ 0.5: (a) stress intensity factors,
(b) Ktot,min for allmethods, (c) DKtot for all methods, (d) Rtot for all methods, (e) DKeff
for all methods, and (f) FCGR forall methods.

for S and MS it becomes positive much later (a 28 mm). All four methods pro-
vide similar values of DKeff for cracks smaller than about 20 mm, somewhat dif-
ferent values for cracks between 20 and 23 mm, and very different values for
cracks longer than 23 mm [Fig. 11(e)]. The trends in DKeff are reflected directly
in trends for FCGR [Fig. 11(f)], which show the contact-based methods SC and
NSC to be in better agreement with the experimental data than S or MS, espe-
cially for long cracks.
The higher level of minimum applied load for LSP3 (DKapp ¼ 11.0 MPa m0.5,
Rapp ¼ 0.5) results in a smaller range of crack lengths at which Kcp is non-zero
[19 a 33 mm; Fig. 12(a)] than found with LSP1 [Fig. 10(a)] or LSP2 [Fig.
11(a)]. The SC and NSC methods are in better agreement with the data for
cracks longer than 23 mm, but SC provides a somewhat better prediction of
FCGR for cracks from 20 to 22 mm [Fig. 12(f)].
Discussion
The good correlation between the new calculation method and FEM in Fig. 5
and Fig. 6 validates Eq 1, the C(T) weight function [28], and the new method for
computing contact pressure based on a piece-wise linear basis and quadratic
programming. The good correlation between observed and predicted FCGR for
the AM1 test shows that the FCGR prediction scheme suggested by Stuart et al.
[29] is reasonable for the coupon material, though the prediction is somewhat
higher than the data (but well within a factor of two) throughout the test.
There are two distinct ranges of crack length for which FCGR predictions
for the LSP tests exhibit noteworthy trends. The first range has cracks longer
than 23 mm, and the different methods for computing DKtot and Rtot give rise to
significant differences in the stress ratio, DKeff, and FCGR. In this range, the
methods that ignore contact (S and MS) predict FCGR significantly above the
data, whereas the methods that include contact (SC and NSC) match the FCGR
data very well. The crack is partially open with an open crack tip [Table 4 and
Fig. 1(c)] in this crack length range, and the residual stress field causes remote
crack closure that reduces the crack-tip stress cycle and lowers FCGR.
The second interesting range of crack length is where the crack is partially
open with the crack tip closed [Fig. 1(b)]; this range starts at 15 or 19 mm,
depending on loading, and runs to 23 mm (Table 4). Here Kapp,min þ Krs þ Kcp is
negative and the S and SC methods provide the same values of DKtot and Rtot
and identical FCGR values. The FCGR values from S and SC are higher (more
conservative) than the FCGR values provided by MS or NSC. In this second
range of crack length, SC agrees better with the data for LSP1 and LSP3, and
NSC agrees better with the data for LSP2.
Noteworthy discrepancies exist between the data and the predictions for all
LSP conditions near strong gradients in FCGR, where, in general, the predic-
tions transition to lower or higher FCGR differently than do the experimental
data. Near a 25 mm, the data for all LSP conditions exhibit increasing FCGR
that transitions to stabilized FCGR for a few millimeters of crack growth, but
the predictions do not show a corresponding region of stabilized FCGR. Early
in the R ¼ 0.1 tests (LSP1 and LSP2), the FCGR data generally follow the rapid

decrease of the predicted FCGR, but near a ¼ 20 mm, the data exhibit an earlier
increase of FCGR than predicted by any of the models. For the R ¼ 0.5 test, the
opposite is true: the predicted increase in FCGR occurs earlier than shown by
the test data. Some of this discrepancy might be due to variations in the location
of the peened patch, but inspection of the samples indicated only minor varia-
tions in patch position. Perhaps more likely is a shortcoming of the prediction
approach that arises from the use of Eq 13, which is based on crack closure lev-
els for steady-state crack growth [26]. Because the loading of the LSP coupons
has significant gradients of DK and R, the use of the steady-state crack opening
level is an approximation. It would be very useful to combine the present
approach for predicting elastic crack closure with a capability for predicting
elastic-plastic crack closure (e.g., FASTRAN [30]). This is left for future work.
Conclusions
A method was described to predict crack closure in a fully elastic material con-
taining a long length-scale bulk residual stress field. The method relies on the
computing of crack face displacements using the weight function for crack-line
distributions of applied stress, residual stress, and contact pressure. Whereas
the applied and residual stress are defined by the application, the unknown
crack face contact pressure was found by expressing it as a piecewise linear dis-
tribution along the crack face and finding minimized point-wise values using
quadratic programming. With the contact pressure defined, stress intensity fac-
tors at a minimum applied fatigue load could be defined via superposition of
stress intensities due to applied stress, residual stress, and contact pressure.
The method described for predicting crack closure due to bulk residual stress
was demonstrated for a set of C(T) coupons having residual stress from LSP.
Crack face displacements due to applied and residual stress, but ignoring contact,
from the proposed method agreed with the results of finite element stress analy-
sis. Crack face displacements including contact agreed with a finite element stress
analysis that included contact. Superposition was used to predict FCGR for LSP
coupons tested in three conditions of applied cyclic loading (one constant ampli-
tude load and two constant amplitude applied stress intensity factor). The effects
of contact pressure on the stress intensity factor range and stress ratio enabled
improved estimates of FCGR compared to estimates that ignored contact stress.
Of the models that ignore crack face contact, the MS model (which takes
Ktot,min ¼ 0 when Ktot,min < 0) provided non-conservative FCGR predictions when
compared to the S model (which admits Ktot,min < 0). Of the models that account
for crack face contact, the SC model, suggested by Jones and Dunn [17], provided
more conservative results than the NSC model described here, and on that basis
it might be the most useful for further validation and eventual application.
Acknowledgments
This work was supported by the Federal Aviation Administration, Rotorcraft
Damage Tolerance Program, FAA contract DTFACT-06-C-00025 (“Analytical
Tools for Residual Stress Enhancement of Rotorcraft Damage Tolerance”).

APPENDIX: EXAMPLE MATLAB CODE FOR DETERMINATION

OF CONTACT PRESSURE
% Code written for MATLAB release 7.9.0.529 (R2009b)
% Load matrix F, previously computed according to Eq 6
F ¼ load(‘F_matrix.txt’);
% Load vector of crack face displacements for minimum

% applied and residual stress computed earlier from
% equation analogous to Eq 7
u_app ¼ load(‘u_app_min.txt’);
u_rs ¼ load(‘u_rs.txt’);
b ¼ u_app þ u_rs;
% Define lower bound

lb ¼ zeros(size(b));
% Compute contact pressure vector

p ¼ quadprog(F’*F,F’*b,F,b,[],[],lb,[],[],‘MaxIter’);
References
[1] Beghini, M., and Bertini, L., “Fatigue Crack Propagation through Residual Stress
Fields with Closure Phenomena,” Eng. Fract. Mech., Vol. 36, 1990, pp. 379–387.
[2] Elber, W., “Fatigue Crack Closure under Cyclic Tension,” Eng. Fract. Mech., Vol. 2,
1970, pp. 37–45.
[3] Gan, D., and Weertman, J., “Crack Closure and Crack Propagation Rates in 7050
Aluminum,” Eng. Fract. Mech., Vol. 15, 1981, pp. 87–106.
[4] McClung, R. C., and Sehitoglu, H., “On the Finite Element Analysis of Fatigue Crack
Closure—1. Basic Modeling Issues,” Eng. Fract. Mech., Vol. 33, 1989, pp. 237–252.
[5] Chermahini, R. G., Palmberg, B., and Blom, A. F., “Fatigue Crack Growth and Clo-
sure Behaviour of Semicircular and Semi-elliptical Surface Flaws,” Int. J. Fatigue,
Vol. 15, 1993, pp. 259–263.
[6] Liu, J. Z., and Wu, X. R., “Study on Fatigue Crack Closure Behavior for Various
Cracked Geometries,” Eng. Fract. Mech., Vol. 57, 1997, pp. 475–491.
[7] Dougherty, J. D., Srivatsan, T. S., and Padovan, J., “Fatigue Crack Propagation and
Closure Behavior of Modified 1070 Steel: Experimental Results,” Eng. Fract. Mech.,
Vol. 56, 1997, pp. 167–187.
[8] Wei, L. W., and James, M. N., “A Study of Fatigue Crack Closure in Polycarbonate
CT Specimens,” Eng. Fract. Mech., Vol. 66, 2000, pp. 223–242.
[9] Solanki, K., “Finite Element Modeling of Plasticity-Induced Crack Closure with
Emphasis on Geometry and Mesh Refinement Effects,” Eng. Fract. Mech., Vol. 70,
2003, pp. 1475–1489.
[10] Song, P., “Crack Growth and Closure Behaviour of Surface Cracks,” Int. J. Fatigue,
Vol. 26, 2004, pp. 429–436.
[11] Lei, Y., “Finite Element Crack Closure Analysis of a Compact Tension Specimen,”
[12] Doquet, V., Bui, Q. H., and Constantinescu, A., “Plasticity and Asperity-Induced Fa-
tigue Crack Closure under Mixed-Mode Loading,” Int. J. Fatigue, Vol. 32, 2010, pp.
1612–1619.

[13] McEvily, A. J., “On Crack Closure in Fatigue Crack Growth,” Mechanics of Fatigue
Crack Closure, ASTM STP 982, J. C. Newman and W. Elber, Eds., ASTM Interna-
tional, West Conshohocken, PA, 1988, p. 35.
[14] Ruschau, J. J., John, R., Thompson, S. R., and Nicholas, T., “Fatigue Crack Nuclea-
tion and Growth Rate Behavior of Laser Shock Peened Titanium,” Int. J. Fatigue,
Vol. 21, 1999, pp. 199–209.
[15] LaRue, J. E., and Daniewicz, S. R., “Predicting the Effect of Residual Stress on Fa-
tigue Crack Growth,” Int. J. Fatigue, Vol. 29, 2007, pp. 508–515.
[16] De Matos, P. F. P., and Nowell, D., “Analytical and Numerical Modelling of
Plasticity-Induced Crack Closure in Cold-Expanded Holes,” Fatigue Fract. Eng.
Mater. Struct., Vol. 31, 2008, pp. 488–503.
[17] Jones, K. W., and Dunn, M. L., “Fatigue Crack Growth through a Residual Stress
Field Introduced by Plastic Beam Bending,” Fatigue Fract. Eng. Mater. Struct., Vol.
31, 2008, pp. 863–875.
[18] Tada, H., Paris, P. C., and Irwin, G. R., “Effect of Surface Interference of Partly
Closed Cracks,” The Stress Analysis of Cracks Handbook, 3rd ed., ASME, New York,
2000, p. 31.
[19] Liu, J. Z., and Wu, X. R., “Analytical Expressions for Crack Opening Displacements
of Edge Cracked Specimens under a Segment of Uniform Crack Face,” Eng. Fract.
Mech., Vol. 58, 1997, pp. 107–119.
[20] Wang, G. S., “Crack Surface Displacements for Mode I One-Dimensional Cracks in
General Two-Dimensional Geometry,” Eng. Fract. Mech., Vol. 40, 1991, pp.
535–548.
[21] Beghini, M., Bertini, L., and Vitale, E., “Weight Functions Applied to Fatigue Crack
Growth Analysis,” Fatigue Fract. Eng. Mater. Struct., Vol. 20, 1997, pp. 1093–1104.
[22] Kiciak, A., Glinka, G., and Burns, D. J., “Calculation of Stress Intensity Factors and
Crack Opening Displacements for Cracks Subjected to Complex Stress Fields,” J.
Pressure Vessel Technol., Vol. 125, 2003, pp. 260–266.
[23] VanDalen, J. E., and Hill, M. R., “Evaluation of Residual Stress Corrections to Frac-
ture Toughness Values,” J. ASTM Int., Vol. 5, No. 8, 2008, Paper ID JAI101713.
[24] Parker, A. P., “Stress Intensity Factors, Crack Profiles, and Fatigue Crack Growth
Rates in Residual Stress Fields,” Residual Stress Effects in Fatigue, ASTM STP 776,
ASTM International, West Conshohocken, PA, 1982, pp. 13–31.
[25] MATLAB, version 7.9.0.529 (2009), The Mathworks, Inc., Natick, MA.
[26] Newman, Jr., J. C., “Analyses of Fatigue Crack Growth Databases for Use in a Dam-
age Tolerance Approach for Aircraft Propellers and Rotorcraft,” DOT/FAA/AR-07/
49, Federal Aviation Administration, Washington, DC, 2007.
[27] Van Dalen, J. E., “Observation and Prediction of Fatigue Behavior in Residual
Stress Bearing Metallic Coupons Including: Fatigue Crack Growth, Notched Geom-
etry Effects, and Foreign Object Damage,” M.S. dissertation, Mechanical and Aero-
nautical Engineering, University of California, Davis, 2007.
[28] Newman, Jr., J. C., Yamada, Y., and James, M. A., “Stress-Intensity-Factor Equa-
tions for Compact Specimen Subjected to Concentrated Forces,” Eng. Fract. Mech.,
Vol. 77, 2010, pp. 1025–1029.
[29] Stuart, D. H., Hill, M. R., and Newman, Jr., J. C., “Correlation of One-Dimensional
Fatigue Crack Growth at Cold-Expanded Holes using Linear Fracture Mechanics
and Superposition,” Eng. Fract. Mech., Vol. 78, 2011, pp. 1389–1406.
[30] Newman, Jr., J. C., “A Crack-Closure Model for Predicting Fatigue Crack Growth
under Aircraft Spectrum Loading,” Methods and Models for Predicting Fatigue Crack
Growth under Random Loading, ASTM STP 748, J. B. Chang and C. M. Hudson,
Eds., American Society for Testing and Materials, Philadelphia, 1981, pp. 53–84.


doi:10.1520/JAI103966
J. C. Newman, Jr.,1 B. M. Ziegler,2 J. W. Shaw,2 T. S. Cordes,3

and D. J. Lingenfelser3
Fatigue Crack Growth Rate Behavior of

A36 Steel using ASTM Load-Reduction and
Compression Precracking Test Methods
ABSTRACT: Eccentrically-loaded single-edge crack tension, ESE(T), speci-

mens made of A36 structural steel were tested over a wide range in stress
ratios (R ¼ 0.1 and 0.7) in laboratory air. Two test methods were used: (1)
ASTM Standard E647 load-reduction method and (2) compression precrack-
ing. After compression precracking (CP), three different loading sequences
were used: (1) constant amplitude (CPCA), (2) load reduction (CPLR), and
(3) constant stress-intensity factor (CPCK). The crack-compliance method
was used to determine that the specimens had no residual stresses; and that
the effects of tensile residual stresses from compression precracking dissi-
pated in about 2 compressive plastic-zone sizes. Agreement was found
between the A36 and TC-128B steel DK-rate data tested at both low and
high stress ratio (R) conditions. At R ¼ 0.1 loading, the CPCA and CPLR tests
generated lower thresholds and faster rates than using the standard ASTM
load-reduction method. All load-reduction tests exhibited an accumulation of
debris at the crack front near threshold conditions. A crack-closure analysis
was preformed to calculate the effective stress-intensity factor range (DKeff)
Manuscript received May 11, 2011; accepted for publication December 14, 2011;
published online April 2012.
1
Dept. of Aerospace Engineering, Mississippi State Univ., Mississippi State, MS 39762
(Corresponding author), e-mail: j.c.newman.jr@ae.msstate.edu
2
Dept. of Aerospace Engineering, Mississippi State Univ., Mississippi State, MS 39762.
3
HBM n-Code Federal, LLC, Advanced Applications Center, Mississippi State Univ.,
Mississippi State, MS 39762.
Eleventh International ASTM/ESIS Symposium on Fatigue and Fracture Mechanics
(38th ASTM National Symposium on Fatigue and Fracture Mechanics) on 18 May 2011
in Anaheim, CA.
Cite as: Newman, J. C., Jr., Ziegler, B. M., Shaw, J. W., Cordes, T. S. and Lingenfelser, D.
J., “Fatigue Crack Growth Rate Behavior of A36 Steel using ASTM Load-Reduction and
Compression Precracking Test Methods,” J. ASTM Intl., Vol. 9, No. 4. doi:10.1520/
JAI103966.
Copyright V
87
ID: aip3b2server Time: 17:17 I Path: D:/AIP/Support/ePub_Autopdf/3D_IN_Process/AI-STP#120222

against rate using measured 1 % offset (OP1) values for all R ¼ 0.1 tests.
The DKeff-rate data correlated well with the high-R results.
KEYWORDS: cracks, fatigue crack growth, crack closure, stress intensity
factor, plasticity, steel
Introduction
In the spring of 2009, the Society for Automotive Engineers (SAE) Fatigue
Design and Evaluation Committee (FD&E) reviewed the results of a round robin
on predicting the fatigue behavior of notched and welded A36 steel components
under bending. The results of this study were presented to the ASTM E-08 Fa-
tigue and Fracture Committee in the fall of 2009. The round robin was con-
ducted to discern whether the fatigue community could predict the total fatigue
life (nucleation plus crack growth) to a definable crack size of a typical notched
and welded component for a typical structural material (A36 steel) used in load
carrying members (like frames) of ground vehicles. More information on the
project and a summary of results may be found at www.fatigue.org.
Both strain-life (crack-nucleation) and stress-strain properties, with the
metallurgical pedigree of the microstructure of the components tested, were
made available for the fatigue analyses. Crack-growth properties for the particu-
lar material were not available. As a result, the crack-growth properties used in
the analyses varied widely. In addition, their metallurgical pedigrees were either
not adequately documented or were not available for comparison to the particu-
lar A36 component microstructure.
Because the A36 grade of steel can have a wide variety of microstructures
and properties, it was important that crack-growth properties of the actual com-
ponents tested be characterized. A number of eccentrically-loaded single-edge
crack tension, ESE(T), specimens were machined from bar-stock material used
to make the component test samples. These specimens were provided to Missis-
sippi State University.
From the literature, threshold testing on low-strength steels using the load-
reduction test method [1] has produced DK-rate data that exhibits more spread
with stress ratio (R ¼ Pmin/Pmax) in the near-threshold regime than at higher
rates (a behavior referred to as fanning). Fanning behavior has been attributed
to load-history (plasticity), debris-accumulation, and/or crack-surface rough-
ness effects in the near-threshold regime. Thus, it was of interest to test the low-
strength A36 steel using the new compression precracking test methods to see if
significant fanning occurred with the stress ratio in the near-threshold regime.
To generate fatigue-crack-growth-rate data under constant-amplitude (con-
stant R) loading in the threshold and near-threshold regimes, without load-
history effects, compression-compression precracking methods, as developed
by Suresh [2], Pippan [3], and others [4–9] were used. Using this procedure,
pre-notched specimens are cycled under compression-compression loading to
produce an initial fatigue crack, which naturally stops growing. The specimens
were then subjected to constant-amplitude loading to generate fatigue-crack-
growth-rate data in the near threshold regime at the desired stress ratio.

NEWMAN ET AL., doi:10.1520/JAI103966 89
Compression cycles create tensile residual stresses at the tip of the notch and
the initial crack growth is affected by the residual stress field. Constant-
amplitude (load-history free) crack-growth-rate data is obtained after the crack
has grown several compressive plastic-zone sizes [5,6,10,11].
It is the scope of this paper to determine the fatigue-crack-growth-rate
properties from threshold to near fracture on the A36 steel using the compres-
sion precracking test methods (see Refs. 5 or 6 for further details). Tests were
conducted over a wide range in stress ratios (R ¼ 0.1 and 0.7) on ESE(T) speci-
mens (B ¼ 6.35 mm; w ¼ 38 mm). These results were compared with DK-rate
data generated on the same material using the ASTM E647 [1] load-reduction
test procedure. Comparisons were also made between A36 steel and test data
from the literature on TC-128B steel [12] tested over the same range in stress
ratios. A crack-closure analysis was performed on all R ¼ 0.1 test data using
measured crack-opening loads from remote backface strain (BFS) gages to
determine the effective stress-intensity factor range against rate behavior.
Material and Specimen Configuration

A36 steel is a standard steel alloy that is a common structural steel used in the
United States. The yield stress (0.2% offset), rys, was 420 MPa (60 ksi) and the
modulus of elasticity (E) was 200 GPa (29 000 ksi). (The ultimate tensile
strength, ru, was not available.)
Standard eccentrically-loaded single-edge crack tension, ESE(T), specimens
[1] were machined from 40 mm wide by 40 mm thick bars. The specimens had a
nominal width (w) of 38 mm with a thickness, B, of 6.35 mm, as shown in Fig.
1. A starter notch (cn ¼ 13 mm in length) was electrical-discharged machined
into all specimens with a semi-circular notch-tip-root radius of about 0.13 mm.
In addition, the edges of the pin holes in the specimens were beveled to avoid or
minimize undesired out-of-plane bending moments (pins forced to contact near
mid-thickness of specimen), see Ref. 8. The beveled holes helped to produced
straighter crack fronts during compression precracking and at threshold condi-
tions, as shown in Fig. 2. A specimen tested without the beveled pin holes
(Fig. 2(a)) produced a non-straight crack front at threshold conditions, while
specimens with the beveled holes produced straighter crack fronts at threshold
conditions (like those shown in Fig. 2(b)).
FIG. 1—Eccentrically-loaded single-edge-crack tension, ESE(T), specimen with back-

face strain gage.

FIG. 2—Effects of standard and beveled pin-hole on crack-front shape during threshold
testing. (a) Standard drilled pin holes. (b) Beveled pin holes.
Test Procedures
All fatigue-crack growth tests were performed under laboratory air conditions
at room temperature in a single 5 kN (1.12 kip) servo-hydraulic test machine.
Crack lengths were monitored using backface compliance procedures with a
gage backface strains (BFS), as shown in Fig. 1 and outlined in Appendix A.
Test control was provided by a data acquisition/test control system developed
by Keith Donald, Fatigue Technology Associates (FTA), for fatigue-crack-
growth-rate (FCGR) testing [13]. Crack-growth-rate testing was performed at
stress ratios, R, of 0.1 and 0.7 at a nominal cyclic frequency of 18 Hz. Because
the ESE(T) specimen was not a standard option in the FTA crack-monitoring
system, a new stress-intensity factor solution using the same form as the com-
pact, C(T), specimen was developed (see Appendix B).
Threshold testing to determine very low rates was performed using two pro-
cedures. The first procedure was the standard load-reduction (LR) method
described in ASTM E647 for threshold determination [1]. Initial starting load
levels were carefully selected to ensure that growth rates were less than 1e 8 m/
cycle (4e 7 in./cycle), as required in the standard. A load reduction rate of
C ¼ 0.08 mm1 ( 2 in.1) was maintained in all ASTM LR and CPLR tests.
Upon developing rates at or near the target 1e 10 m/cycle, test control was
changed to constant-amplitude (CA) loading, DK increasing, in order to trace
back up the crack-growth-rate curve.
The second method used was compression-compression precracking (CP);
followed by CA loading, and referred to as CPCA loading, or load reduction,
CPLR; after reaching a specified crack-extension criterion. After CP loading,
one test was conducted at a constant stress-intensity factor, CPCK, immediately
from the starter notch with the initial pre-crack. Figure 3 shows the various
load sequences applied to the ESE(T) specimens.
In the CP method, a small fatigue crack is introduced at the tip of the starter
notch via compression-compression cyclic loading. The resulting crack tip is
enveloped by a small tensile residual-stress field instead of the typical compres-
sive plastic zone normally resulting from tension-tension cyclic loading and in
general, the crack surface is free of any crack closure resulting from either crack
surface roughness and/or the compressive plastic zone. Because of the long test

FIG. 3—Various load sequences using compression precracking. (a) CPCA or CPCK
loading sequences. (b) CPLR sequence.
section for the ESE(T) specimen; in contrast to previously tested C(T) speci-
mens, the maximum compressive load requirement was cut into half of the
previous recommended value. The maximum compressive stress-intensity
factor level (Kcp) required to produce fatigue cracks within 10 000 cycles
was 28 MPaHm for the A36 specimens, as given by the following relationship:

p p
jKcp j=E¼ 0:005 mm¼ 0:001 in: (1)
where:
jKcpj is the maximum absolute compressive stress-intensity factor for pre-
cracking, and
E is the elastic modulus.
(Stress-intensity factors for cracks under compressive pin loading on the
ESE(T) specimen were assumed to be the same as that for tensile loading, see
Appendix B.) Typical crack sizes at the notch tip after compressive precracking
at R ¼ 20 were about 0.4 mm (0.016 in.), which was about a factor-of-2 larger
than previous tests on other materials at the same load levels.
Following compression precracking, constant-amplitude (CA) loading was
selected to be above the anticipated threshold stress-intensity-factor range. If
no appreciable crack growth occurred after approximately 1 106 cycles, then
the loads were increased 2%–5% (maintaining constant R) and, again, cycled
to examine for crack growth. If the crack began to grow, the loads were held
constant and the crack was grown to failure (Fig. 3(a)) or grown to the crack
extension criterion and then a load-reduction test was conducted. This proce-
dure, CPLR, is depicted in Fig. 3(b).
Once crack growth was detected, the fatigue crack was extended by approxi-
mately 2 to 3 compressive plastic-zone sizes (based on the compressive precracking
conditions) from the initial crack size prior to taking any valid crack-growth-rate
data to eliminate potential transient behavior resulting from compressive loading
and the resulting tensile residual stresses. On the basis of extensive testing [5–9]
and analyses [10,11], an expression to determine the required crack extension
beyond which the crack-growth-rate data would not be affected by compressive
yielding at the notch and produce “steady-state” constant-amplitude data (stabi-
lized crack-opening loads) in the near threshold regime is
Dc 3ð1 RÞqc (2)
where:
qc is the compressive plastic-zone size calculated from the plane-stress
equation by
qc ¼ ðp=8ÞðjKcp j=rys Þ2 (3)
where:
Kcp is the compressive stress-intensity factor and
rys is the yield stress of the material.
Residual Stress Measurements

During the past decade, acceptance of the compression precracking procedure
has been plagued by the residual-stress issue. If 6 specimens are machined from
a plate or forging and 3 specimens are randomly selected for testing using the
ASTM LR method and the other 3 specimens are tested using the CP method,
and specimen data within one group agree with one another, but the CP method

generates lower threshold and faster rates, logic dictates that the issue is not re-
sidual stress but the test methods. However, the FTA crack-monitoring system
used herein has the capability to measure residual stress-intensity factors due
to residual stresses present in the plate or forging after machining or from resid-
ual stresses induced by compression precracking. (The crack-monitoring sys-
tem used the crack-compliance method [14–16] to evaluate Krs values.) Thus,
the ESE(T) tests were monitored for the presence of residual stress-intensity
factors, Krs, to determine the extent of the tensile residual stress influence from
compression precracking and the absence or presence of residual stresses in the
specimens machined from the steel bars.
In order to help validate the crack-extension criterion beyond which the ten-
sile residual stresses from compression precracking (CP) do not have an influence
on crack-growth rates, compression precracking loads were applied to the 38 mm
wide ESE(T) specimens. The CP stress-intensity factor (Kcp) was 28 MPa m1/2
and the plastic-zone size was about 2 mm in length from the notch tip. CPCA or
CPCK tests were then conducted and the Krs values recorded as a function of
crack length. (Caution must be exercised during Krs measurements because the
determination is very sensitive to non-linearities in the measurement system and
temperature changes.) Some typical results are shown in Fig. 4 and present
FIG. 4—Residual stress-intensity factors in A36 steel ESE(T) specimens.

measured Krs values from residual stresses induced by CP loading. (Each

residual-stress data point is a 10-point average value to help reduce scatter.) The
CPCK test results show values across nearly the complete test section. The loads
on one CPCA test was changed to a high-R test (crack-length-to-width ratio,
c/w, > 0.55); while the other CPCA test was terminated due to issues of crack-
front straightness. After a crack extension of about 2 compressive plastic-zone
sizes (vertical dashed line in Fig. 4), the Krs values are very small, indicating that
residual stresses from forming, machining, or CP loading would not have an
influence of further fatigue-crack-growth rates. Herein, the 2-compressive-plas-
tic-zone-size criterion was used to indicate the region from the initial crack length
that would be affected by residual stresses (i.e., Krs > 0).
Experimental Results
Fatigue-crack-growth rate (DK-rate) properties from threshold to near fracture
have been determined for A36 steel. Ten (10) ESE(T) specimens (B ¼ 6.35 mm;
w ¼ 38 mm) were machined from bar-stock. Tests were conducted over a wide
range in stress ratios (R ¼ 0.1 and 0.7) using ASTM load-reduction [1] and com-
pression precracking test methods [5–9]. Comparisons are made between the
A36 steel data and test data from the literature on TC-128B steel [12] tested over
the same range in stress ratios. Measured crack-opening loads from the remote
backface strain gage on the R ¼ 0.1 tests were used to conduct a crack-closure
analysis to determine the effective stress-intensity factor range [17] against rate
behavior.
Fatigue-Crack-Growth-Rate Data
The first tests were conducted on the A36 steel at a high stress ratio (R ¼ 0.7)
condition. From previous testing, the high R test conditions have been invariant
to the particular test method, except for tests conducted on Inconel-718 [9]. Fig-
ure 5 shows DK against rate data on A36 and TC-128B [12] steels. After CP load-
ing and crack growth to satisfy the crack-extension criterion, which was about
one compressive plastic-zone size (Eq 2), a load-reduction (CPLR) test was con-
ducted (solid squares). Once the threshold condition was reached (1e 10 m/
cycle), a constant-amplitude (CA) test was initiated at the rate indicated by the
arrows. A CPCA test was also conducted (solid circles) that started at about
2e 10 m/cycle and was grown under CA loading to slightly beyond the ASTM
maximum allow rate (dashed line). The 38 mm wide ESE(T) specimen reached
a plastic-hinge condition at this rate and data could not be obtained at higher
DK values.
For comparison, test data on TC-128B steel [12] compact, C(T), specimens
tested at R ¼ 0.6 and a constant Kmax test are also shown. Test data at higher DK
values could be obtained from the larger width specimens, which appeared to
be a linear extension of the A36 data. The Kmax test produced a slightly lower
threshold than the CPLR test on A36. The TC-128B and A36 data agreed very
well, except in the threshold region.

FIG. 5—Stress-intensity-factor range against rate for high stress ratios (0.6 or 0.7) and
Kmax test results on two steels.
The solid curve with symbols in Fig. 5 is the fit to the high-R data (CPLR
test) on the A36 steel at low rates and the TC-128B steel at high rates. This curve
will be compared with the low R test data and crack-closure analyses later,
because the remote BFS compliance method [1] indicated that the cracks were
fully open (i.e., DKeff-rate curve). However, recent test data from Yamada and
Newman [7–9] has shown that high R and Kmax tests on a variety of materials
have some forms of crack closure in the threshold and near-threshold regimes.
An ASTM load-reduction test was conducted on the A36 steel and these
results are shown in Fig. 6. After CP loading (needed to initiate a pre-crack at
the starter notch), the crack was grown to the maximum allowed rate (1e 8 m/
cycle) and then the standard load-reduction scheme was used. The test gener-
ated a threshold at about 6.5 MPaHm, then a slightly higher load was used to
conduct CA tests (trace back up the DK-rate curve) and grow the crack to near
failure. Test data on the TC-128B steel tested at R ¼ 0.1 are also shown in Fig. 6,
which agreed well with the A36 data. The high-R fit curve is shown for compari-
son. As expected, fanning behavior is apparent—a larger spread in the data is
observed at threshold conditions than at higher rates.

FIG. 6—Stress-intensity-factor range against rate for low stress ratio (0.1) results on
two steels using ASTM load-reduction method.
Figure 7 shows a number of CPCA and CPLR tests conducted at R ¼ 0.1 on

the A36 material. All specimens were compression precracked; and all DK-rate
data are shown. The open symbols show invalid data that have been affected by
the tensile residual stresses caused by CP loading. These data start near the
high-R curve because the cracks are fully open and have to build the plastic
wake (and crack closure) as the cracks grow. The CPCA/LR tests indicated that
using lower initial DK values, before load reduction, produced lower thresholds
and faster rates.
Crack-Opening-Load Measurements
During all crack-growth tests on the A36 steel, the data acquisition system
recorded the 1 % and 2 % offset (OP1 and OP2, respectively) compliance values
as a function of crack length. Figure 8 shows the results from a CPCA test at
R ¼ 0.1 loading that started at a DKi value of 5 MPaHm. The solid and open sym-
bols show OP1 and OP2 values as a function of c/w. The test started at a (c/w)i
value of about 0.34, but the crack-opening measurement method was unable to
detect closure until c/w was about 0.41. The vertical dashed line at c/w ¼ 0.43
(two compressive plastic-zone sizes) indicated where tensile residual stresses

FIG. 7—Stress-intensity-factor range against rate for low stress ratio (0.1) results on
A36 steel using compression precracking methods.
would not have an influence on further crack growth. At c/w ratios less than
0.43, the residual stress-intensity factors, Krs, are greater than zero, as shown in
Fig. 4. The OP1 and OP2 values stabilized at about Po/Pmax of 0.3. At c/w ¼ 0.55,
the loads were changed to R ¼ 0.7 in an effort to generate more high-R test data.
The vertical dashed line at c/w ¼ 0.475 in Fig. 8 indicates the 3-plastic-zone crite-
rion where crack-opening loads should have stabilized, which was verified by
the test data. The horizontal line at Po/Pmax ¼ 0.3 was from a FASTRAN crack-
closure analysis [18]. Test data at low and high R correlated on a DKeff basis
with a constraint factor of 2.5 (nearly plane strain), except in the threshold
regime and at very high rates (an issue that will be discussed later). In the
threshold regime, the plasticity-induced crack-closure model does not account
for debris-induced crack closure and the analysis will produce a conservative
DK-rate curve at low R.
The results from the ASTM load-reduction test are shown in Fig. 9. A CPCA
test was conducted until the crack grew to a rate of 1e 8 m/cycle, then a stand-
ard load-reduction (LR) test was initiated and the crack grew until it had
reached threshold conditions. At this point, a CA test was conducted to trace
back up the DK-rate curve. But immediately upon starting the LR test, the
crack-opening load indications (OP1 and OP2) increased and steadily rose until

FIG. 8—Crack-opening loads during CPCA test at R ¼ 0.1 as a function of crack-length-

to-width (c/w) ratio.
the CA test was initiated. At this point, the crack-opening load indications began
to decrease. The ASTM E647 standard states that the DK-increasing data verifies
the DK-decreasing data, if they are in agreement. Many state that these are inde-
pendent tests. However, the crack-opening-load measurements indicate that
load-history effects are generated during load reduction and the crack is grow-
ing out of the load-history influence during the CA portion. Thus, these test con-
ditions are not independent and do not validate the load-reduction data. The
load-reduction test is basically a variable-amplitude test (changing loads and
DK), and a steady-state crack-opening analysis is not appropriate, as shown in
Fig. 8.
Figure 10 shows a CPCA/LR/CA test conducted where the initial DKi value
was 4 MPaHm. After the crack had grown beyond the 2 plastic-zone require-
ment (no residual-stress influence), a LR test was initiated and, again, the
crack-opening-load (Po/Pmax) ratio indications began to immediately increase.
Threshold conditions were approached as the crack-opening loads rapidly rose.
During the CA portion, the opening loads dropped and leveled off at a Po/Pmax
value of about 0.3. And then the opening loads began to steadily drop for c/w
ratios greater than about 0.6. The reason for this steady drop was not clearly

FIG. 9—Crack-opening loads during ASTM load-reduction test at R ¼ 0.1 as a function

of crack-length-to-width (c/w) ratio.
understood, but the un-cracked ligament was about 2 times the thickness and
became smaller as the crack length increased. Thus, an increase in constraint to
pure plane-strain behavior was expected for the deep-crack bend specimen [19].
Solanki et al. [20] have also shown that cracked bend specimens under pure
plane-strain conditions do not develop crack closure at R ¼ 0 conditions.
Another CPCA/LR/CA test was conducted on an ESE(T) specimen with a
17 mm manual saw-cut notch and the initial DKi value was 4.3 MPaHm after CP
loading. However, here the results on the Po/Pmax values are plotted against
cycles in Fig. 11. After the crack had grown beyond the 2 plastic-zone require-
ment (no residual stress influence), a LR test was initiated and, again, the crack-
opening-load indications immediately began to increase. But the maximum
OP1 value was lower than that achieved in the previous test; and this test went
to a lower threshold than the previous test (Fig. 10). In this test, the threshold
was achieved at a longer crack length than the previous test. Again, the sharp
drop in the crack-opening values occurred at the c/w ratios greater than 0.6.
The measured crack-opening-load (OP1 and OP2) values for a CPCK test
are shown in Fig. 12. After CP loading, the crack was grown at a constant DK
value of 7.6 MPaHm. This test was expected to show a constant crack-growth

FIG. 10—Crack-opening loads during CPLR test at R ¼ 0.1 as a function of crack-

length-to-width (c/w) ratio.
rate after the effects of the CP loading had dissipated. Thus, this test was sup-
posed to be another method to validate the crack-extension criterion, as given
by Eq 2. It was also expected that the OP1 and OP2 values would stabilize at
about 0.3, but two regions of elevated crack-opening values occurred. Once the
test was completed, the specimen was fractured, which revealed a strange event,
as shown in Fig. 13. Two regions of dark debris had occurred along the crack
surfaces. These debris regions correspond well with the elevated crack-opening
values, as shown in Fig. 12. It was also noted that the relative humidity readings,
recorded during the test, showed elevated readings during the regions of debris
accumulation. Forth et al. [4] have also shown that fatigue-crack-growth rates
in D6ac steel were greatly affected by debris accumulation along the crack
surfaces during laboratory-air tests.
Effective Stress-Intensity Factors from Measurements

Measurement of the crack-opening loads during the low stress ratio (R ¼ 0.1)
tests gave an opportunity to determine the effective stress-intensity-factor range
(DKeff) as a function of rate. The remote BFS gage did not record valid OP1 or

FIG. 11—Crack-opening loads during CPLR test at R ¼ 0.1 as a function of cycles.
OP2 values for the R ¼ 0.7 tests or the cracks were fully open (i.e., DK ¼ DKeff).
However, the works of Yamada and Newman [7–9] indicate crack closure for
high-R conditions in the threshold regime for a wide variety of materials. Thus,
it is suspected that the R ¼ 0.7 test results in the threshold regime have devel-
oped some crack closure due to plasticity and the accumulation of debris. The
ASTM E647 standard recommends the 2 % offset (OP2) values, but measure-
ments made by Yamada and Newman [7–9] for R ¼ 0.1 loading on a variety of
materials have shown that OP1 (remote gage) values are closer to values deter-
mined by local strain gages. (Yamada and Newman [7–9] have also proposed
using a zero-offset crack-opening value that gives a slightly higher value than
OP1, and would result in a slightly lower DKeff-rate curve that may be in closer
agreement with the Kmax test results shown in Fig. 5.)
Figure 14 shows the DKeff values determined from the OP1 measurements
as a function of rate (solid curves near the high-R fit curve). All of the CPCA and
CPLR tests correlated well with the high-R curve. The ASTM LR/CA results fell
slightly to lower values of DKeff, but were in fair agreement with the high-R
data. Although correlation of low-and high-R data to generate a unique
DKeff-rate curve is well accepted, Yamada and Newman [7–9] have also found
that high-R tests exhibit crack closure in the threshold and near-threshold
regimes due to plasticity, roughness and/or debris. Results from the local strain
gages correlated low-R, high-R, and Kmax tests onto a unique DKeff-rate curve

FIG. 12—Crack-opening loads during CPCK test at R ¼ 0.1 as a function of crack-

length-to-width (c/w) ratio.
FIG. 13—Fatigue-crack surface during CPCK test showing regions of debris associated
with higher laboratory relative-humidity readings.

FIG. 14—Effective stress-intensity-factor range against rate for low stress ratio (0.1)
tests from measured crack-opening loads.
for a variety of materials [7–9]. But further study is needed to help resolve the
issue of high-R closure and local versus remote measurement methods.
Concluding Remarks
On the basis of testing of eccentrically-loaded single edge crack tension,
ESE(T), specimens made of A36 steel, it was shown that the compression pre-
cracking constant amplitude (CPCA) or compression precracking load reduc-
tion (CPLR) test methods produced more conservative results than the current
ASTM load-reduction (LR) test method for determining low fatigue-crack-
growth-rate data in the threshold and near-threshold regions. Starting load-
reduction tests at lower initial stress-intensity factor ranges produced low
thresholds and faster rates. Current load-reduction test procedures (i.e., starting
load reduction at 1e 8 m/cycle) gave a higher threshold and slower rates than
the CP test methods at a low stress ratio (R ¼ 0.1).
Testing and analyses on A36 steel ESE(T) specimens produced the follow-
ing conclusions:
1. Stress-intensity factor (K) and backface strain (BFS) gage equations
have been developed or verified for monitoring crack growth in ESE(T)
specimens.

2. Beveling pin holes in A36 steel ESE(T) specimens produced straighter

crack fronts in the threshold regime than standard drilled pin holes.
3. A36 steel ESE(T) specimens had no residual stresses due to forming or
machining; and tensile residual stresses due to compression pre-
cracking dissipated in about two (2) compressive plastic-zone sizes.
4. Both low-and high-R test data on A36 agreed with TC-128B steel data
using current ASTM test methods.
5. CPCA and CPLR tests at R ¼ 0.1 loading produced lower thresholds and
faster rates than the ASTM load-reduction test method in the low-rate
regime.
6. Measured crack-opening-load (Po/Pmax) ratios began to immediately
increase during all load-reduction tests and changed with laboratory rel-
ative humidity levels.
7. Effective stress-intensity factors for low-R (0.1) tests using remotely
measured crack-opening loads (1% offset, OP1 values) correlated well
with high-R (0.7) data, which did not show any crack-closure behavior
using the remote method.
APPENDIX A: BACKFACE STRAIN EQUATION FOR THE ESE(T) SPECIMEN

Eccentrically-loaded single-edge crack tension, ESE(T), specimens are widely
used to measure fatigue-crack-growth rates in metallic materials. Two methods
have been used to automatically monitor crack length as a function of cycles using
compliance. They are the crack-mouth-opening-displacement (CMOD) gage and
the backface strain (BFS) gage (as shown in Fig. 1). The CMOD and BFS relations
are standardized in ASTM E647 [1]. Unfortunately, the BFS relation used in E647
is a log functional form [21,22]. The log form was not well suited for use in the
crack-monitoring system used herein without additional programming. During
the past few years, a new BFS crack-length relation has been developed for the
ESE(T) specimen (W. Johnston, NASA Langley, private communication, 2009).
Figure 15 shows a comparison among some previous experiments conducted on
2024-T3 aluminum alloy specimens [21,22], a boundary-element analysis
(FADD2D), and the new equation. The normalized strain, jeEBw/Pj, is plotted
against the crack-length-to-width (c/w) ratio. The new equation was developed
with the same functional form as used in ASTM E647 for CMOD. The curve shows
the equation developed by Johnston (2009). The BFS relation used herein is
c=w¼ A0 þA1 UþA2 U2 þA3 U3 þA4 U4 þA5 U5 (A1)
for 0.1 < c/w < 0.95, where:

U ¼ 1/[A1/2 þ 1] and
A ¼ jeEBw/Pj.
The coefficients are:
A0 ¼ 1:007 A1 ¼ 2:171 A2 ¼ 1:537 A3 ¼ 7:615 A4 ¼ 22:181 A5 ¼ 20:745
This equation was independently verified by Newman using the FADD2D

boundary-element code (open circular symbols).

FIG. 15—Normalized backface strains (BFS) and equations for ESE(T) specimen as a
function of crack-length-to-width (c/w) ratio.
APPENDIX B: STRESS-INTENSITY FACTOR EQUATION

FOR THE ESE(T) SPECIMEN
The stress-intensity factor equation for the ESE(T) specimen had been devel-
oped many years ago [21,22]. But the crack-monitoring software/hardware sys-
tem used herein did not have the ESE(T) specimen as an option. Thus, a new
equation was developed that used the same form as the compact, C(T), speci-
men, so that the crack-monitoring system could be used. Figure 16 shows a com-
parison among various stress-intensity-factor (K) relations and analyses for the
compact C(T) and ESE(T) specimens. The normalized stress-intensity factor,
KBw1/2/P, is plotted against the crack-length-to-width (c/w) ratio. The upper
curve shows the standard equation developed for the C(T) specimen [23 and 24].
For the C(T) specimen the K relation is given by
K¼ P=ðBw1=2 Þð2þkÞG=ð1 kÞ3=2 (B1)
G¼ A0 þA1 kþA2 k2 þA3 k3 þA4 k4 (B2)
for 0.2 < k < 1, where:

k ¼ c/w.

FIG. 16—Normalized stress-intensity factors and equations for compact C(T) and
ESE(T) specimens as a function of crack-length-to-width (c/w) ratio.
The coefficients for the C(T) specimen are:
A0 ¼ 0:886 A1 ¼ 4:64 A2 ¼ 13:32 A3 ¼ 14:72 A4 ¼ 5:6
A new equation was also developed for the ESE(T) specimen that used the same
functional form as that used for the C(T) specimen (Eq 1). The symbols show
the normalized numerical values from a boundary-force method (BFM) [21]
and recent FADD2D boundary-element analyses. The lower dashed curve shows
the equation developed by Piascik et al. [21,22] that covered a range of c/w from
0 to 1. The new equation covered a smaller crack-length range (0.1 < c/w < 1),
but had the same functional form as the C(T) specimen.
For the ESE(T) specimen the K relation is given by
K¼ P=ðBw1=2 Þð2þkÞG=ð1 kÞ3=2 (B3)
G¼ A0 þA1 kþA2 k2 þA3 k3 þA4 k4 þA5 k5 (B4)
for 0.1 < k < 1, where:

k ¼ c/w.
Equation 3 is within 6 1 % of the FADD2D analyses. The coefficients for the
ESE(T) specimen are:
A0 ¼ 0:5 A1 ¼ 2:643 A2 ¼ 6:3 A3 ¼ 8:25 A4 ¼ 5:6 A5 ¼ 1:59

References
[1] ASTM E-647. 2006, “Standard Test Method for Measurement of Fatigue Crack
Growth Rates,” Annual Book of ASTM Standards, Vol. 03.01, American Society for
Testing and Materials, West Conshohocken, PA, pp. 615–657.
[2] Suresh, S., “Crack Initiation in Cyclic Compression and Its Application,” Eng. Fract.
Mech., Vol. 21, 1985, pp. 453–463.
[3] Pippan, R., Plöchl, L., Klanner, F., and Stüwe, H. P., “The Use of Fatigue Specimens
Precracked in Compression for Measuring Threshold Values and Crack Growth,”
J. Test. Eval., Vol. 22, 1994, p. 98.
[4] Forth, S. C., Newman, J. C., Jr., and Forman, R. G., “On Generating Fatigue Crack
Growth Thresholds,” Int. J. Fatigue, Vol. 25, 2003, pp. 9–15.
[5] Newman, J. C., Jr., Schneider, J., Daniel, A., and McKnight, D., “Compression Pre-
cracking to Generate Near Threshold Fatigue-Crack-Growth Rates in Two Alumi-
num Alloys,” Int. J. Fatigue, Vol. 27, 2005, pp. 1432–1440.
[6] Ruschau, J. J., and Newman, J. C., Jr., “Compression Precracking to Generate Near
Threshold Fatigue-Crack-Growth Rates in an Aluminum and Titanium Alloy,”
J. ASTM Int., Vol. 5, No. 7, 2008.
[7] Yamada, Y., and Newman, J. C., Jr., “Crack Closure Behavior of 2324-T39 Alumi-
num Alloy Near Threshold Conditions for High Load Ratio and Constant Kmax
Tests”, Int. J. Fatigue, Vol. 31, 2009, pp. 1780–1787.
[8] Newman, J. C., Jr., Yamada, Y., and Newman, J. A., “Crack-Closure Behavior of
7050 Aluminum Alloy near Threshold Conditions for Wide Range in Load Ratios
and Constant Kmax Tests,” J. ASTM Int., Vol. 7, No. 4, 2010.
[9] Yamada, Y and Newman, J. C., Jr., “Crack Closure under High Load-Ratio Condi-
tions for Inconel 718 Near Threshold Behavior”, Eng. Fract. Mech., Vol. 76, 2009,
pp. 209–220.
[10] James, M. A., Forth, S. C., and Newman, J. A., “Load History Effects Resulting
from Compression Precracking,” ASTM Spec. Tech. Publ., Vol. 1461, 2005, pp.
43–59.
[11] Yamada, Y., Newman, J. C., III, and Newman, J. C., Jr., “Elastic-Plastic Finite-
Element Analyses of Compression Precracking and Its Influence on Subsequent
Fatigue-Crack Growth,” J. ASTM Int., Vol. 5, No. 8, 2008.
[12] McKeighan, P. C., Feiger, J. H., and Riddell, W. T., “Fatigue Crack Growth Rate
Behavior of Tank Car Steel TC-128B,” Iron Steelmaker, Vol. 2, No. 5, 2002, pp.
73–78.
[13] Donald, K., “User’s Reference Manual for Automated Fatigue Crack Growth,” Vol.
2.65, Fracture Technology Associates, LLC, Bethlehem, PA, 2007.
[14] Lados, D. A., Apelian, D., and Donald, J. K., “Fracture Mechanics Analysis for Re-
sidual Stress and Crack Closure Corrections,” Int. J. Fatigue, Vol. 29, 2006, pp.
687–694.
[15] Donald, J. K., and Lados, D. A., “An Integrated Methodology for Separating Closure
and Residual Stress Effects from Fatigue Crack Growth Rate Data,” Fatigue Fract.
Eng. Mater. Struct., Vol. 30, 2006, pp. 223–230.
[16] Schindler, H. J., Cheng, W., and Finnie, I., “Experimental Determination of Stress
Intensity Factors Due to Residual Stresses,” Exp. Mech., Vol. 37, No. 3, 1997, pp.
272–279.
Vol. 486, 1971, pp. 230–242.
[18] Newman, J. C., Jr., “A Crack Opening Stress Equation for Fatigue Crack Growth,”
Int. J. Fract., Vol. 24, 1984, R131–Rl35.

[19] Newman, J. C., Jr., Crews, J. H., Jr., Bigelow, C. A., and Dawicke, D. S., “Variations
of a Global Constraint Factor in Cracked Bodies Under Tension and Bending
Loads,” ASTM Spec. Tech. Publ., Vol. 1244, 1995, pp. 21–42.
[20] Solanki, K., Daniewicz, S. R., and Newman, J. C., Jr., ”Finite Element Modeling of
Plasticity-Induced Crack Closure with Emphasis on Geometry and Mesh Refine-
ment Effects“, Eng. Fract. Mech., Vol. 70, 2003, pp. 1475–1489.
[21] Piascik, R. S., and Newman, J. C., Jr., “An Extended Compact Tension Specimen
for Fatigue Crack Growth and Fracture Testing,” Int. J. Fract., Vol. 76, 1996, pp.
R43–R48.
[22] Piascik, R. S., Newman, J. C., Jr., and Underwood, J. H., “The Extended Compact
Tension Specimen,” Fatigue Fract. Eng. Mater. Struct., Vol. 20, No. 4, 1997, pp.
559–563.
[23] Srawley, J. E., ”Wide Range Stress Intensity Factor Expressions for ASTM Method
E 399 Standard Fracture Toughness Specimens,“ Int. J. Fract., Vol. 12, 1976, pp.
475–476.
[24] Newman, J. C., Jr., “Stress Analysis of the Compact Specimen Including the Effects
of Pin Loading,” Fracture Analysis, ASTM Spec. Tech. Publ., 560, 1974, pp. 105–121.


doi:10.1520/JAI103973
Y. Yamada1 and J. C. Newman, Jr.2
Crack Closure Behavior on a Variety of

Materials under High Stress Ratios
and Kmax Test Conditions
ABSTRACT: Fatigue-crack-growth-rate tests on compact specimens have

been made on a variety of materials (2024-T3, 2324-T39, 7050-T7451, 4340
steel, and Inconel-718) over a wide range in stress ratios from 0.1 to 0.9 (and
0.95 in some cases) and several Kmax test conditions. Test data has been
generated from threshold to near fracture using the compression precracking
constant amplitude or compression precracking load reduction test methods
in the threshold regime; and constant-amplitude loading at higher rates.
A remote back-face strain (BFS) gage was used to monitor crack growth and
to measure crack-opening loads. Local strain gages were also placed along
and slightly off (about one-half thickness) the anticipated crack path to mea-
sure crack-opening loads. Elber’s load-against-reduced-strain method was
used to determine crack-opening loads by means of visual inspection (equiv-
alent to a 0 % compliance offset). For a particular material, the BFS and local
strain gages produced essentially the same crack-opening loads at low
stress ratio (R ¼ 0.1) conditions. But at high stress ratios (R 0.7) and Kmax
test conditions, the local gages produced significantly higher crack-opening
loads than the BFS gage in the threshold and near-threshold regimes.
Previous research had proposed that high stress ratios (R 0.7) and Kmax
test conditions produce closure-free conditions based on crack-mouth-
opening-displacement or BFS gages, and plasticity-induced crack-closure
modeling. However, crack closure under high stress ratios (R 0.7) and Kmax
test conditions is attributed to residual-plastic deformations, crack-surface
Manuscript received May 12, 2011; accepted for publication October 4, 2011; published
online October 2011.
1
Senior Researcher, Ohio Aerospace Institute, NASA Glenn Research Center, Cleveland,
OH 44135.
2
Dept. of Aerospace Engineering, Mississippi State Univ., Mississippi State, MS 39762
(Corresponding author), e-mail: j.c.newman.jr@ae.msstate.edu
Cite as: Yamada, Y. and Newman, J. C., Jr., “Crack Closure Behavior on a Variety of Mate-
rials under High Stress Ratios and Kmax Test Conditions,” J. ASTM Intl., Vol. 9, No. 2.
doi:10.1520/JAI103973.
Copyright V
109

roughness, and/or fretting-debris. From local crack-opening load measure-

ments, the effective stress-intensity-factor range (DKeff) appears to be
uniquely related to the crack-growth rate in the threshold and near-threshold
regimes.
KEYWORDS: Cracks, fatigue-crack growth, crack closure, Kmax effect,
threshold, load ratio, compression precracking
Introduction
In the past, fatigue cracks grown under high load ratio (R ¼ Pmin/Pmax) or Kmax
test conditions, which generated data at extremely high load ratios in the near-
threshold regime, had been assumed to be crack-closure free. A Kmax test holds
the maximum stress-intensity factor constant and reduces the range as thresh-
old conditions are approached. Measurements of crack-opening loads using
“remote” displacement or strain methods had indicated no crack closure at high
load ratios. In addition, plasticity-induced crack-closure (strip-yield) model
analyses had also predicted crack-closure-free behavior under high load ratios.
However, the use of “local” strain gages mounted near the crack-tip location has
produced significant indications of crack closure under high load ratio and
Kmax test conditions in the near-threshold regime. This paper is a review of
crack growth and closure behavior under high R conditions on a wide variety of
materials.
Fatigue-crack-growth (FCG) tests on compact specimens were conducted
on several materials (2024-T3, 2324-T39, 7050-T7451, 4340 steel, and Inconel-
718) at load ratios of 0.1 to 0.95, and for Kmax test conditions under laboratory-
air conditions. Test data were generated from threshold to near fracture using
compression pre-cracking constant-amplitude (CPCA) or compression pre-
cracking load-reduction (CPLR) test methods in the threshold regime, and
constant-amplitude (CA) loading at higher rates. Remote back-face strain (BFS)
gages were used to monitor crack growth. The BFS and local strain gages placed
along the crack path were both used to measure crack-opening loads. Elber’s
load-reduced-displacement (or strain) method was used to determine crack-
opening loads by means of visual inspection (equivalent to a 0 % compliance
offset). Comparisons have been made on the crack-opening loads determined
from both remote and local strain gages. Additionally, the results from the local
strain gages were used to determine the effective stress-intensity-factor range
(DKeff) against crack-growth rate on a wide variety of materials in the threshold
and near-threshold regimes.
Specimen Configuration and Materials

Compact C(T) specimens were used, except the pin-holes were beveled to help
minimize the out-of-plane bending influence on crack-front shapes and to help
produce linear load-against-strain records from side-face (local) strain gages
with only a crack-starter notch. Because of slight misalignments in the
compact-clevis pin-loading fixtures or pin-holes, the pins may contact the outer
edges of the pin-holes and cause out-of-plane bending. Thus, the stress-intensity

YAMADA AND NEWMAN, doi:10.1520/JAI103973 111
FIG. 1—Specimen configuration, pin-hole modification, and strain-gage placement.

(a) Compact specimen, and (b) 4340 steel with and without beveled pin holes.
factors at the crack tip on one side of the specimen will be higher than on the
other side and cause a non-straight crack front as threshold conditions are
approached. The beveled pin-holes, as shown in Fig. 1(a), causes the pin to auto-
matically contact near the centerline of the specimen and produces a straighter
crack front, as shown in Fig. 1(b) on 4340 steel. One specimen had the standard
pin-hole configuration and produced a non-straight crack front as threshold
conditions were approached; whereas the specimen with the beveled pin-holes
produced a nearly straight crack front during a similar threshold test. The C(T)
specimens tested on a variety of materials were nominally 51, 76, and 152 mm
wide (W). The crack-starter V-notch had either a 45 or 60 included angle. The
notch-length-to-width (cn/W) ratio varied from 0.33 to 0.35 in order to increase
the sensitivity of the BFS gage crack-monitoring system. A summary of the
materials and C(T) specimen configurations tested are listed in Table 1.
Fatigue Crack Growth Test Procedures

The FCG tests were performed using closed-loop servo-hydraulic fatigue test
machines that applied a sinusoidal wave form for constant- and variable-
amplitude loading for threshold and constant Kmax tests. A computer controlled
crack monitoring system [1] was used to continuously monitor crack lengths
TABLE 1—Materials and compact specimen configurations tested.
Material B, mm W, mm Yield Stress, MPa Tensile Strength, MPa
2024-T3 2.3a 152 360 495

2324-T39 6.35 76 450 500
7050-T7451 6.35 51 470 525
4340 steel 6.35 51 … 1145
Inconel-718 9.5 76 1060 1350
a
Pin holes were not beveled.

during tests using the BFS compliance technique [2]. For a given material, the
crack length was determined using an improved compliance equation for the
C(T) specimen [3]. Additionally, the required loads for all of the K-control tests
were computed by the crack-monitoring system. Periodically, crack lengths
were verified by visual measurements using an optical microscope. Compliance
crack lengths were recalibrated when the visual crack lengths deviated by more
than 0.05 mm. After testing, the FCG rates and stress-intensity factors were cor-
rected by considering the deviation between visual and compliance crack length
measurements. The FCG rates were evaluated after crack-growth increments of
Dc/W ¼ 0.001–0.002.
There are two types of threshold tests available in the American Society for
Testing and Materials (ASTM) standard E-647 [4]. The first type of test is one
where the load ratio R is held constant during the test. Near-threshold data for
large fatigue cracks are generated by reducing the applied loads (i.e., Kmax and
Kmin) as the crack grows. Threshold is achieved when the crack grows very
slowly (i.e., dc/dN ¼ 1010 m/cycle) [4]. The concern with this standardized
load-reduction test method is a possible load-history effect due to the reduction
of the plastic-zone size as the crack propagates and the development of remote
closure [5,6]. The ASTM E-647 standard suggests using the load-shed rate
C ¼ 0.08 mm1, for constant R threshold tests to ensure consistent results and,
presumably, to eliminate load-history effects. However, evidence suggests that
the load-shedding procedure in this standard is insufficient [5–7].
In order to avoid undesirable remote closure effects during threshold tests,
a compression-compression pre-cracking (CP) method was proposed [8,9].
Since pre-cracking was performed under compression-compression constant-
amplitude (CA) loading, a crack from the notch will be fully open at the zero-
load condition. However, the first compressive load will create a tensile
residual-stress field that grows the crack faster than steady-state behavior under
tensile CA loading. In order to generate valid FCG rate data, a crack must be
grown under the desired constant-amplitude loading (R ¼ constant) at least two
compressive plastic-zone sizes from the notch [10]. After the crack-extension
criterion is met, a load-reduction test can be performed to generate threshold
conditions (CPLR) or maintain tensile CA loading (CPCA) to generate data from
threshold to fracture (if the initial load level was higher than threshold condi-
tions). The advantage of using the CP method is that the initial loading condi-
tion to start the FCG rate test is at a much lower rate than what the current
standard allows, however, it is also very effective when the material around a
notch is influenced by residual stresses and/or a recast zone due to electrically-
discharged-machining (EDM) the notch.
The second type of threshold test is performed by holding Kmax constant
and reducing the DK value as threshold conditions are approached. Constant
Kmax threshold tests are considered to have two major advantages over constant
R threshold tests. First, remote closure is less likely during constant Kmax
threshold testing because the monotonic plastic zone (which is responsible for
crack-wake plasticity) remains constant during the test. Second, because the
effects of load history have been eliminated, the minimum load may be
increased at a faster rate. For constant Kmax tests performed, a K-gradient of

C ¼ 0.4 mm1 was used. As a constant Kmax test progresses, R increases and
fatigue crack closure may be eliminated, whereas for constant R tests, in gen-
eral, fatigue crack closure is not eliminated near threshold conditions and
remote closure may occur.
Crack Closure Measurement Technique

During FCG testing, crack lengths were monitored using compliance data from
a BFS gage. Compliance data from the closure-free portion of the load cycle is
used to determine crack length, enabling the tests to be automated and com-
puter controlled. Load-against-strain data can also be used to measure fatigue-
crack-closure events. As a fatigue crack closes, the effective load range is
reduced. A typical load against BFS record is presented in Fig. 2(a) for R ¼ 0.1.
For CA loadings, the compliance is constant at high loads (open crack), which
appears as a linear section in the upper right portion of the figure. As the load
decreases, crack surfaces contact and produce a change in compliance. In cases
where a large portion of the crack surface closes during unloading, this compli-
ance change is very dramatic. When only a small portion of the crack closes
very near the crack tip, this change in slope may be difficult or impossible to dis-
tinguish on a load-against-strain plot. The reduced compliance technique was
developed to improve detection of these subtle compliance changes [11,12]. The
reduced strain De is the deviation from closure-free compliance behavior (i.e.,
Figure 2(b)). Closure loads are more easily detected from load-against-reduced-
strain plots. Closure-free behavior on these plots becomes a vertical line, mak-
ing compliance deviations easier to detect. Using the reduced strain technique,
the deviation from the fitted line due to closure is clearly seen; significant
FIG. 2—Elber’s method to determine crack-opening loads. (a) Load against BFS record,
and (b) load against reduced strain record.

deviation at low loads (P/Pmax < 0.5) is observed in Fig. 2(b). Fitting lines
through closure-free data (P/Pmax > 0.5) allows closure to be defined as the load
corresponding to the intersection of the fitted lines. However, this technique
does not provide information about the location of the crack face contact since
this method relies on changes in compliance to determine closure (or crack-
opening) levels.
FCG Test and Crack-Closure Measurement Preparation

Some researchers have proposed that remote gages (e.g., crack-mouth-opening-
displacement and BFS) are not sensitive enough to measure crack-opening
loads [13]. Thus, to improve the sensitivity of load-against-strain records, strain
gages were bonded on the surface of C(T) specimens along the crack path offset
about a notch height (W/32). One of the concerns with the local gage measure-
ments is out-of-plane bending. All holes on the C(T) specimens were beveled
(except for the thin-sheet 2024-T3) to ensure that pin contact occurs near the
center of the pin-hole (minimizing out-of-plane bending).
The original idea of the crack-closure concept was that there was no crack
growth below the crack-opening load. The crack-opening load can be deter-
mined by a deviation point from the upper linear portion of the load-against-
reduced-strain record. Therefore, if there is no crack-surface contact, then the
load-reduced-strain records should be linear. Before pre-cracking, target cyclic
loads were applied to notched (un-cracked) specimens to ensure that there were
no non-linearities in the local strain-gage readings. For example, a demonstra-
tion was done on the 7050-T7451 aluminum alloy, which had an EDM notch
with a 0.2 mm notch-root radius. Load levels were chosen to be at DK ¼ 2.6 and
1.8 MPa m1/2 for R ¼ 0.1 and 0.7, respectively, which are equivalent to a rate
of 109 m/cycle. Figure 3(a) shows the load-reduced-strain record on the un-
cracked (notched) specimens. Since there was no crack closure and no out-of-
plane bending, the local load-strain records showed only linear response. It also
demonstrated that there were no other disturbances in the testing system to
cause a non-linear response in the local strain-gage readings. During testing, op-
timum measurement signals were obtained when the crack tip was located
almost 2 gage widths from the center of the gage. Strain-gage sizes were chosen
to be about 5 % of the specimen width. A comparison between local and remote
(BFS) gage readings on a C(T) specimen made of the 2024-T3 alloy is shown in
Fig. 3(b). This figure shows the load-reduced-strain records measured on a test
at R ¼ 0.1. The results from the BFS shows the tail-swing associated with crack
closure and the compliance-offset values of 1 % (OP1) and 2 % (OP2). The
compliance-offset values gave progressively lower values of the crack-opening-
load ratio for larger offset values. The circular symbol shows the crack-opening-
load ratio determined by visual inspection (deviation from upper linear portion
of the load-reduced-strain record, such as a 0 % compliance offset method)
from the local gage. The near crack-tip gage showed a similar load-reduced-
strain record as the BFS gage, however, it showed a slightly larger tail-swing
and indicated that the crack-opening load would be about 5 % higher than the 1
% offset value. Based on the difference between 1 % and 2 % offset opening

FIG. 3—Load against reduced strain records for notched and cracked C(T) specimens.
(a) Notched specimen, and (b) cracked specimen.
values (0.465 and 0.425) and linear extrapolation to 0 % offset, the opening load
would be about 0.5, which agreed very well with the crack-opening load deter-
mined from the local reading. Also, by comparing remote and local gage read-
ings, it indicated that the measurement location was not a problem as long as
the load-strain records were measured ahead of the crack tip. The curvature
below the crack-opening load showed noticeable differences between local and
remote gage readings. Local gage readings showed an aggressive change below
the opening load, while the BFS gage showed a gradual change. Thus, local
gages enhance the fidelity to determine crack-opening loads.
Fig. 4(a) shows a comparison of load-strain records measured on the
R ¼ 0.7 test from a near crack-tip strain gage (local) and the BFS gage (remote)
at a FCG rate of 1 1010 m/cycle. Some researchers have tried to determine
the opening load from load-strain records [14]. From these records, it would
have been concluded that the crack was fully open. However, the opening load
is a very subtle change in the load-strain record, so it is impossible to determine
one unique point. Figure 4(b) shows load-reduced-strain records [11,15,16] for
the same load-strain records as shown in Fig. 4(a). The levels of noise were
almost the same between the local and remote gages, but the shape of the load-
reduced-strain records was different. Obviously, the signal-to-noise ratio in
these data is poor. But the local gage did measured a clear indication of crack
closure, even at R ¼ 0.7. The local gages almost always showed some amounts
of crack closure in the near threshold regime; whereas the remote gage consis-
tently showed no indication of crack closure at high R-values. This indicated
that the remote gages are not sufficient to determine crack-opening loads from

FIG. 4—Load against strain and reduced strain records for remote and local gages.
(a) Load against strain, and (b) load against reduced strain.
remote measurements, especially at high R; and that local measurements have

a great advantage in capturing the near crack-tip behavior. The standard ASTM
E-647 [4] suggests using a 2 % offset compliance change to determine crack-
opening loads from load-reduced-strain or displacement records. However,
because of the amplitude of the noise and the size of crack-closure tail-swings
on the reduced load-strain records, the use of offset values was not practical
and greatly underestimated the true crack-opening loads. In Fig. 4(b), the ampli-
tude of the noise was approximately 1 le and the size of the crack-closure tail-
swing was about 3.5 le for the R ¼ 0.7 local measurements at 1 1010 m/cycle;
whereas other tests at lower R (0.1) showed orders-of-magnitude larger tail-
swings from the BFS and local strain gages, such as that shown in Fig. 3(b).
FCG Rate Data and Crack-Closure Measurement Results

Generally, FCG tests were conducted over a wide range in load-ratio conditions
(0.1 R 0.9) and two constant Kmax tests. Figure 5(a) shows an expanded
threshold region for the 2024-T3 aluminum alloy. All specimens were compres-
sion pre-cracked (CP) before testing. After CPCA loadings, the CPLR tests were
conducted from 2 to 3 109 m/cycle. Once the threshold regime was reached,
CA loads were applied to generate the mid-region and near-fracture data. Since
the 2024-T3 C(T) specimens have a thickness of 2.3 mm and a width of 152 mm,
back-face buckling may have occurred when the crack lengths were large or

FIG. 5—Crack-growth rates and load against reduced strain records for 2024-T3 at
R ¼ 0.1. (a) DK against rate, and (b) load against reduced strain.
when high loads were used for the high load-ratio tests, such as R ¼ 0.9. Hence,
it was unable to generate near-fracture data for R ¼ 0.7 and 0.9. The CPCA tests
were performed at R ¼ 0.1 and 0.9 only. The data for the CPCA tests at R ¼ 0.9
agreed well with the CPLR data, but the CPCA R ¼ 0.1 results showed a slightly
higher threshold of 3.2 MPa m1/2, while the CPLR test produced 3.0 MPa m1/2.
A constant Kmax test was chosen to have a much lower Kmax value (7.3 MPa m1/2)
than the reference test (22 MPa m1/2) from the literature [17]. Each constant
Kmax test produced different load ratio data; the Kmax test at 22 MPa m1/2 had an
R value from 0.72 to 0.94, while the Kmax test at 7.3 MPa m1/2 had an R value
from 0.1 to 0.8. The R ¼ 0.7 and 0.9 tests produced thresholds of 1.8 and
1.45 MPa m1/2, respectively; while constant Kmax tests (22 and 7.3 MPa m1/2)
showed thresholds of 1.22 and 1.63 MPa m1/2, respectively. As expected, lower
threshold values were obtained from the higher load-ratio tests. In this section,
local strain gages were used again to measure load-strain records during thresh-
old tests on every load conditions.
Because thin and wide C(T) specimens were tested, R ¼ 0.9 and high con-
stant Kmax tests were suspected to experience back-face buckling, which would
disturb the surface stress distribution and corrupt local gage readings. Thus,
local strain gages were used and measured load-strain records on only the
R ¼ 0.7 test and the lower constant Kmax test. First, the load-strain records at
R ¼ 0.1 were measured during a CPCA test and these results are shown in Fig.
5(b). Because of better sensitivity, crack-opening loads from local gages were
quite easy to determine. Crack-opening loads from the local gages were deter-
mined by visual inspection, whereas OP1 values (shown by cross symbols) came
from the BFS gage readings made with the crack-monitoring system [1]. Consis-
tently, the local gages showed higher crack-opening loads than the remote

gages, however, both records showed the same trend; in the sense that, the
crack-opening-load ratios were rising as the crack-growth rate approached the
threshold regime.
For high R tests, all crack-opening loads were determined by visual inspec-
tion for both local and remote gages, since opening readings from the crack-
monitoring system were either not available or not reliable. Figure 6(a) shows
the series of local gage reading during a CPLR threshold test. The results from
remote gages are not shown here because all of the records were similar to the
ones shown in Fig. 4(b) with no indication of crack closure. The local gages
almost always showed some amounts of crack closure in the near-threshold re-
gime, and also a rise in the crack-opening load as the threshold was
approached. This indicated that the remote gages are not sufficient to determine
crack-opening loads from remote measurements, especially at high R; and that
local measurements have a great advantage in capturing the near crack-tip
behavior. Figure 6(b) shows load-reduced-strain records for the constant Kmax
(7.3 MPa m1/2) test. The local gages showed a clear indication of crack-closure
behavior. At the load ratio of 0.74 and 0.77,the local-gage records indicated that
the crack was fully opened at Po/Pmax of 0.8 and 0.84, respectively.
Based on the crack-opening loads determined from local gages, crack-
closure corrections were performed on the R ¼ 0.1 and 0.7 test data, and the low
constant Kmax test data, and these results are shown in Fig. 7(a). All of the
crack-closure corrected (DKeff) data have collapsed together into a fairly tight
band and the results are approaching a (DKeff)th value at a threshold of about
1 to 1.15 MPa m1/2. These DKeff results consistently fell lower than the high con-
stant Kmax test (22 MPa m1/2). At the ASTM defined threshold (1010 m/cycle),
the (DKeff)th ranged from 1.02 to 1.17 MPa m1/2, whereas DKth from the high
constant Kmax test was 1.22 MPa m1/2. These results suggest that the DKeff
FIG. 6—Load against reduced strain for R ¼ 0.7 and Kmax tests. (a) R ¼ 0.7, and
(b) Kmax.

FIG. 7—Effective stress-intensity factor against rate for 2024-T3 and 2324-T39. (a)
2024-T3, and (b) 2324-T39.
against the rate relation may be a unique function over a wide range of R in the
threshold regime of 2024-T3, if more appropriate crack-opening-load values
were measured.
It seems that a high constant Kmax test may be able to generate DKeff base-
line data for near-threshold conditions. The 2024-T3 aluminum alloy had a
fairly flat crack surface. Thus, plasticity-induced-crack-closure should dominate
with some additional fretting-debris-induced-crack-closure and very minor
crack-surface-roughness-induced-crack-closure. In the case of a material with
very rough crack-surface profiles, such as 2324-T39, even a high constant Kmax
test may be experiencing a combination of the three major crack-closure-mech-
anisms (plasticity, roughness, and debris). Figure 7(b) shows near-threshold
FCG rate data and the DKeff region determined from various test conditions
(R ¼ 0.1, 0.7, 0.9 and constant Kmax test) on the 2324-T39 aluminum alloy. All of
the crack-closure corrected data have collapsed together into a narrow band
and the results are approaching a (DKeff)th value at a threshold of about 1 MPa
m1/2. These results suggest that the DKeff against the rate relation may be a
unique function over a wide range of R in the threshold regime, if more appro-
priate crack opening-load values were measured. Even the remote gage at
R ¼ 0.1 produced DKeff values (OP1) quite close to the results from the local
gages at high R; see Fig. 7(b). In an effort to generate crack-closure-free data in
the near-threshold regime, a CPLR test at R ¼ 0.95 was performed, but unfortu-
nately, without local-strain gages. The test was conducted at an initial DKi of
1.65 MPa m1/2 to generate near-threshold data (Fig. 7(b)). However, the data fell
at higher DK values than the DKeff regime at a given rate, but still at slightly
lower DK values than the R ¼ 0.9 and Kmax tests. These results imply that there
may be crack closure at R ¼ 0.95! Since local gages were not used, however,
there was no direct evidence, but must await further test results.

Similar local-gage measurements were made on 7050-T7451, Inconel-718,

and 4340 steel near threshold conditions; the DKeff-rate data were calculated
from threshold to near fracture. Figure 8(a) shows the FCG data at various load
conditions (R ¼ 0.1, 0.7, 0.9 and two constant Kmax tests) on 7050-T7451. The
lowest DK at the threshold rate (1 1010 m/cycle) was 1.25 MPa m1/2, which
was produced by the highest constant Kmax test. Typically, the R ¼ 0.7 test data
is considered to be crack-closure-free data and its DKth value was found to be
1.5 MPa m1/2. The highest constant Kmax test produced a lower DKth value than
the R ¼ 0.7 test. Hence, it has been considered that Kmax is one of the most
important parameters to cause damage around a crack-tip and produces a phe-
nomenon called the “Kmax effect” [13,17]. However, by use of local measure-
ments, which amplifies the fidelity to determine the amount of damage at a
crack tip, and to determine DKeff, the Kmax effects have been explained by crack
closure. Crack-closure effects were found on all load conditions tested in these
FCG rate data sets. Figure 8(b) shows the DKeff-rate data calculated from crack-
opening loads determined from the local measurements (R ¼ 0.7 and two
constant Kmax tests) and from the remote gage (R ¼ 0.1). Near-threshold and
mid-region DKeff-rate data determined from local-gage measurements showed
very good agreement with DKeff-rate data determined from the remote gage at
R ¼ 0.1. The lowest DKth value was produced by the highest constant Kmax test;
however, (DKeff)th was found to be 0.5 to 0.6 MPa m1/2. Thus, the high R and
constant Kmax tests were not crack-closure free, especially on a material with a
rough crack-surface profile, such as 7050-T7451.
Figure 9(a) and 9(b) shows the FCG rate and DKeff-rate data of Inconel-718.
On this material, there were no constant Kmax test conducted, but constant R
FIG. 8—Fatigue crack-growth rate data for 7050-T7451. (a) DK against the rate for a
wide range in R, and (b) high R and DKeff-rate data.

FIG. 9—Fatigue crack-growth-rate data for Inconel-718. (a) DK against rate data, and
(b) high R and DKeff-rate data.
data were generated on R ¼ 0.1, 0.4, 0.7, 0.9, and 0.95. Local measurements to
determine crack-opening loads were performed on only the R ¼ 0.7 test and the
rest of the DKeff-rate data were calculated from remote gage readings. The FCG
rates on the Inconel-718 material showed a small influence of R in the mid-
region, but very large fanning was observed in the near-threshold region. The
lowest DKth was found at about 3 MPa m1/2 from the R ¼ 0.95 test, while DKth at
R ¼ 0.7 was 4.4 MPa m1/2. By using local gage measurements, crack closure was
observed on the R ¼ 0.7 test near the threshold regime and the DKeff-rate data
calculated from local-gage measurements agreed very well with the R ¼ 0.95
results near-threshold conditions. In addition, remote-gage measurements for
R ¼ 0.1 and 0.4 also indicated that the DKeff-rate data agreed well with the local-
gage data at R ¼ 0.7.
Figure 10(a) and 10(b) shows the FCG rate and DKeff-rate data of 4340 steel.
This material exhibited very small crack-closure effects from threshold to near-
fracture, but lower thresholds were found as the load ratio approached unity.
The DKth for R ¼ 0.7 and 0.9 test data was 2.7 and 2.2 MPa m1/2, respectively.
Once again, by using local measurements during the R ¼ 0.7 test, the spread
between the R ¼ 0.95 and 0.7 data was explained by crack closure, as shown in
Fig. 10(b). The DKeff-rate results determined from local gages agreed very well
with standard crack-closure measurement determined from remote (BFS) gages.
Discussion of Results
The local strain-gage measurements revealed that remote strain gages (and pre-
sumably remote crack-mouth displacement gages) were unable to measure

FIG. 10—Fatigue crack-growth-rate data for 4340 steel. (a) DK against rate data, and
(b) high R and DKeff-rate data.
crack-closure behavior at high load ratio ( 0.7) conditions. Using the local
method, high-R-closure effects were found on several types of materials (three
aluminum alloys, a steel, and a nickel-based superalloy). Also, the local-gage
measurements found crack closure during constant Kmax tests, which was
totally unexpected. Moreover, it was shown that crack-opening loads deter-
mined by remote and local gages consistently showed a rise in the crack-
opening (Po/Pmax) ratio as threshold conditions were approached. Figure 11
summarizes the crack-opening-load ratios measured with the local strain gages
for R ¼ 0.1 and 0.7 on the 4340 steel. The differences between the dashed lines
and the measured values indicate the amount of crack closure for each R. These
results are typical of the behavior observed for the other materials. In the past,
the three major crack-closure (or crack shielding) mechanisms were recognized
as contributing to threshold development: (1) plasticity-induced crack closure
(PICC), (2) roughness-induced crack closure (RICC), and (3) debris-induced
crack closure (DICC). Strip-yield model simulations of FCG [6] have also indi-
cated that cracks were fully open at load ratios (R) higher than about 0.7. How-
ever, these simulations were based on only the PICC mechanism. Hence, it can
be concluded that high-R crack-closure in the threshold regime was caused by
RICC and DICC mechanisms, as suspected from past research, but in addition
to plasticity (PICC), which set the crack-opening load at the minimum (Pmin)
load level; see Fig. 11 for the R ¼ 0.7 data. Thus, small amounts of debris accu-
mulation and roughness along the crack surfaces can then contribute to crack-
opening loads above the minimum load level for high-R conditions.

FIG. 11—Crack-opening-load ratios for R ¼ 0.1 and 0.7 test results in the threshold
regime for 4340 steel alloy.
Concluding Remarks
It was shown that there is crack closure at high R ratios, such as R ¼ 0.7 or
higher, which was commonly considered to be crack-closure-free load ratios.
This behavior was considered to be due to either fretting debris and/or crack-
surface roughness, which were identified in the early 1980s [18,19]. It was
known that RICC and DICC mechanisms may have significant influence in the
near-threshold regime. The experimental determination of crack-opening loads
from load-reduced-strain records measured from either local or remote gages is
a combination of, at least, the three major crack-closure mechanisms (PICC,
RICC, and DICC). There may be a way to separate the effects of each mecha-
nism, however, it has not yet been done in the literature. In the development of
fatigue-crack-growth testing standards, the remote-gage method was standar-
dized, since it not only monitored crack lengths, but was also used to simultane-
ously determine crack-opening loads. As previously indicated, however, remote
gages were shown to lack the sensitivity to measure crack-opening loads in the
case of high load ratios. Yet, there are reports [17,20] that indicate that crack-
opening loads determined from remote gages for low R load-reduction tests
tended to be higher than expected based on high R tests (assuming that the high
R test data were crack-closure free). Hence, the lack of data correlation with the
crack-closure concept led to the conclusion that DKeff was an inappropriate
crack-tip parameter. In this study, however, local-strain gages were used to

increase the strain sensitivity during threshold testing and comparisons of

crack-opening load determination between local and remote gage consistently
showed that local gages produced higher crack-opening loads than remote
gages. However, if remote closure was prematurely induced, such as during the
standard load-reduction test, the so called “crack-opening loads” from remote
gages should more correctly be referred to as “crack-surface lift-off loads.” The
lift-off load is higher than the crack-opening load needed to correlate crack-
growth data on a DKeff-rate curve. On the contrary, the test conducted without
inducing remote closure effects showed that the crack-opening loads from local
gages were higher than those measured with remote gages. It was also shown
that the use of the compression pre-cracking methods helped to eliminate
remote closure effects and provided a better crack-opening load determination
method during threshold testing.
Acknowledgments
Most of this paper was first presented at the Fatigue 2010 Conference held in
the Czech Republic in June 2010 and was published as a conference proceed-
ings in Procedia Engineering and is republished here with permission from
Elsevier. The writers thank Dr. Dy Le, formerly of the Federal Aviation Adminis-
tration, and Dr. A. Vasudevan, Office of Naval Research, for supporting develop-
ment of the compression pre-cracking test procedures at Mississippi State
University; and to Dr. Keith Donald, Fracture Technology Associates, for his val-
uable advice on the use of his crack-monitoring software.
References
[1] Donald, K., “User’s Reference Manual for Automated Fatigue Crack Growth (Com-
pliance),” v. 2.65, Fracture Technology Associates, Bethlehem, PA, 2007.
[2] Deans, W. F., Jolly, C. B., Poyton, W. A., and Watson, W., “A Strain Gauging Tech-
nique for Monitoring Fracture Mechanics Specimens During Environmental
Testing,” J. Strain Anal. Eng. Des., Vol. 13, 1977, pp. 152–154.
[3] Newman, J. C., Jr., Vizzini, A. J., and Yamada, Y., 2010, “Fatigue-Crack-Growth
Databases and Analyses for Threshold Behavior in Rotorcraft Materials,” DOT/
FAA/AR-10/3, Dept. of Transportation, Federal Aviation Administration, Washing-
ton, D.C.
[4] ASTM E-647, 2008, – “Standard Test Method for Measurement of Fatigue Crack
Growth Rates,” Annual Book of ASTM Standards, Vol. 3.01, ASTM International,
West Conshohocken, PA, pp. 1–45.
[5] Newman, J. C., Jr., “A Nonlinear Fracture Mechanics Approach to the Growth of
Small Cracks,” Behaviour of Short Cracks in Airframe Components, AGARD Conf.
Proc., Vol. 328, 1983, pp. 6.1–6.27.
[6] Newman, J. C., Jr., “Analyses of Fatigue Crack Growth and Closure Near Threshold
Conditions for Large-Crack Behavior,” ASTM Spec. Tech. Publ., Vol. 1372, 2000,
pp. 227–251.
[7] McClung, R. C., “Analyses Of Fatigue Crack Closure During Simulated Threshold
Testing,” ASTM Spec. Tech. Publ., Vol. 1372, 2000, pp. 35–43.

[8] Suresh, S., “Crack Initiation in Cyclic Compression and Its Application,” Eng.
Fract.Mech., Vol. 21, 1985, pp. 453–463.
[9] Pippan, R., “The Growth of Short Cracks Under Cyclic Compression,” FatigueFract.
Eng. Mater.Struct., Vol. 9, 1987, pp. 319–328.
[10] Yamada, Y., Newman, J. C., III, and Newman, J. C., Jr., “Elastic-Plastic Finite-
Element Analyses of Compression Pre-Cracking and its Influence on Subsequent
Fatigue Crack Growth,” J. ASTM Int., Vol. 5, No. 8, 2008, pp. 1–13.
[11] Elber, W., “Crack Closure and Crack Growth Measurements in Surface-Flawed Ti-
tanium Alloy Ti-6Al-4V,” NASA-TN-D-8010, National Aeronautics and Space
Administration, Washington, D.C., 1975.
[12] Saxena, A., Hudak, S. J., Jr., Donald, J. K., and Schmidt, D. W., “Computer-Con-
trolled Decreasing Stress Intensity Technique for Low Rate Fatigue Crack Growth
Testing,” J. Test. Eval., Vol. 6, No. 3, 1978, pp. 167–174.
[13] Smith, S. W. and Piascik, R. S., “Determining Closure Free Fatigue Crack Growth
Behavior in the Near Threshold Regime,” ASTM Spec. Tech. Publ., Vol. 1372, 2000,
pp. 109–122.
[14] Lang, M., “Explanation of an Apparent Abnormality In Fatigue Crack Growth
Curves in Titanium Alloys,” Acta Mater., Vol. 47, 1999, pp. 3247–3261.
[15] Schmidt, R. A. and Paris, P. C., “Threshold for Fatigue Crack Propagation and the
Effect of Load Ratio and Frequency,” ASTM Spec. Tech. Publ., Vol. 536, pp. 79–94.
Vol. 486, 1971, pp. 230–242.
[17] Paris, P. C., Tada, H., and Donald, J. K., “Service Load Fatigue Damage – a Histori-
cal Perspective,” International Journal of Fatigue, Vol. 21, 1999, pp. S35–S46.
[18] Walker, N. and Beevers, C. J., “A Fatigue Crack Closure Mechanism in Titanium,”
Fatigue Fract. Eng.Mater. Struct., Vol. 1, 1979, pp. 135–148.
[19] Endo, K., Komai, K., and Matasuda, Y., “Mechanical Effects of Corrosion Products
in Corrosion Fatigue Crack Growth of a Steel,” Bull. Jpn. Soc. Mech. Eng., Vol. 24,
1981, pp. 1319–1325.
[20] Bray, H. G. and Donald, J. K., “Separating the Influence of Kmax from Closure-
Related Stress Ratio Effects Using the Adjusted Compliance Ratio Technique,”
ASTM Spec. Tech.Publ., Vol. 1343, 1999, p. 57–78.


doi:10.1520/JAI103996
J. Toribio,1 J. C. Matos,2 B. González,1 and J. Escuadra2
Modeling of Surface Crack Advance in Round

Wires Subjected to Cyclic Loading
ABSTRACT: This paper shows the evolution of the surface crack front in
round bars constituted of different materials (determined by the exponent
m of the Paris law), subjected to fatigue tension loading (with free ends) or
fatigue bending loading. To this end, a numerical modeling was developed on
the basis of a discretization of the crack front (characterized with elliptical
shape) and the crack advance at each point perpendicular to such a front,
according to a Paris-Erdogan law, using a three-parameter stress intensity
factor (SIF). Each analyzed case was characterized by the evolution of the
semielliptical crack front, studying the progress with the relative crack depth
a=D of the following three key variables: (i) crack aspect ratio a=b (relation
between the semiaxes of the ellipse which defines the crack front); (ii) maxi-
mum dimensionless SIF; and (iii) minimum dimensionless SIF.
KEYWORDS: numerical modeling, fatigue crack propagation, cracked cylin-
der, crack front aspect ratio, dimensionless SIF
Introduction
One of the most relevant geometries in the field of fatigue and fracture mechan-
ics applied to structural engineering is a cracked cylinder under tension loading
or bending moment. As a matter of fact, many structural elements, mainly in
civil engineering consist of wires, bolts, shafts, cables, or other components of
cylinder shapes under constant or cyclic loading, so that the risk of surface
cracking by mechanical or environmental actions is not negligible.
Manuscript received May 18, 2011; accepted for publication November 1, 2011; published
1
Dept. of Materials Engineering, Univ. of Salamanca, E.P.S., Campus Viriato, Avda.
Requejo 33, 49022 Zamora, Spain, e-mail: toribio@usal.es
2
Dept. of Computing Engineering, Univ. of Salamanca, E.P.S., Campus Viriato, Avda.
Requejo 33, 49022 Zamora, Spain.
Cite as: Toribio, J., Matos, J. C., González, B. and Escuadra, J., “Modeling of Surface
Crack Advance in Round Wires Subjected to Cyclic Loading,” J. ASTM Intl., Vol. 9, No. 1.
doi:10.1520/JAI103996.
Copyright V
126

TORIBIO ET AL., doi:10.1520/JAI103996 127
Growth of surface cracks in round bars due to fatigue can be modeled using
different criteria. Prediction of the 90 intersecting angle of the crack with the
surface or the iso-K criterion along the crack front exhibit small differences in
their aspect ratio but both lead to a unique fitting [1]. Another criterion is based
on the crack growth according to the Paris-Erdogan law considering the crack
advance perpendicular to the crack front, assuming elliptic geometry of the
crack [2–4], avoiding the shape hypothesis [5,6], or using the modified Forman
model [7].
Characterization of fatigue crack growth, whose crack front has been com-
monly represented as straight, circular, or elliptical with centre on the wire sur-
face, necessarily implies knowing the dimensionless stress intensity factor (SIF)
Y, which makes it essential to discern how it changes along the crack front. The
dimensionless SIF has been obtained by several authors under different loading
conditions (tension, bending, and torsion) and deducted from different proce-
dures: flexibility method, finite element method, contour integral analysis, ex-
perimental techniques, etc [2,3,8–12].
Fatigue crack growth in round bars with different initial geometry leads to
a preferential crack path, with an aspect ratio between 0.6 and 0.7 for a relative
crack depth close to 0.6 for tension [2,5], since the geometry of the crack front
must be defined with, at least, two independent parameters [9]. Growth patterns
are closer for a higher value of the Paris coefficient m. The crack always tries to
propagate towards an iso-K configuration; however, it cannot be maintained
due to the existence of the surface, where the stress has a two-dimensional state
and the singularity of the square root can be lost at the crack tip [5].
Numerical Modeling
In order to study how a crack propagates on the cross section of a round bar
under tension or bending cyclic loading (Fig. 1), a computer program in Java
programming language was developed to determine the geometrical evolution
of the crack front.
The basic hypothesis of the modeling consisted of assuming that the crack
front can be modeled as an ellipse with centre on the bar surface [13] and the fa-
tigue propagation takes place in a direction perpendicular to this crack front,
following a Paris-Erdogan law [14]:
da
¼ CDK m (1)
dN
FIG. 1—Cracked bar under tension loading (left) and bending moment (right).

Every elliptical arc of the crack was divided in z segments with exactly the same
length using the Simpson method to discretize the front. The point on the wire
edge was not taken into account, since it presents some difficulties regarding
the computation of the dimensionless SIF (there is a plane stress state on the
crack edge). After that, every single point was shifted according to Paris-
Erdogan law perpendicular to the front, so as to keep constant the maximum
crack depth increment, Da(max) : max Dai. The advance of every front point
Dai can be obtained from the maximum crack increment and the ratio of the
dimensionless SIF
m
Yi
Dai ¼ DaðmaxÞ (2)
YðmaxÞ
The newly obtained points, fitted by the least squares method [13], generate a
new ellipse with which the process is repeated iteratively until the desired crack
depth is reached. Due to the existing symmetry, only half of the problem was
used for the computations (Fig. 2).
The dimensionless SIF used in the computations is that proposed by Shin
and Cai [4] obtained by the finite element method together with a virtual crack
extension technique, which depends on the crack geometry a=b, the crack depth
a=D, and the position of the point considered on its front x=h (Fig. 3).
The fitting of the results provides three-parametrical expressions which are
defined as a function of the coefficients Mijk for tension with free ends (Table 1)
X
2 X
7 X
2 ai a j x k
Y¼ Mijk (3)
i¼0 j¼0 k¼0
b D h
and of coefficients Nijk for bending (Table 2)
X
2 X
6 X
2 a i a j x k
Y¼ Nijk (4)
i¼0 j¼0 k¼0
b D h
FIG. 2—Process followed to compute the fatigue crack growth.

FIG. 3—Elliptical crack model used by Shin and Cai.
TABLE 1—SIF coefficients for tension with free ends proposed by Shin and Cai (Mijk).
i j k¼0 k¼1 k¼2
0 0 0.220 0.123 0.409

0 1 28.513 0.511 9.764
0 2 354.782 2.034 128.817
0 3 2178.632 19.569 727.078
0 4 7140.202 144.435 2201.067
0 5 12 957.447 359.284 3732.813
0 6 12 227.977 393.518 3343.521
0 7 4721.868 159.206 1240.214
1 0 0.326 0.065 1.011

1 1 3.780 6.878 3.946
1 2 79.489 47.747 41.099
1 3 571.094 119.954 316.682
1 4 1976.255 14.769 1 284.860
1 5 3583.421 423.169 2563.292
1 6 3256.770 661.610 2455.158
1 7 1163.158 306.176 880.302
2 0 0.266 0.118 1.584

2 1 9.118 3.515 45.562
2 2 85.381 75.016 552.891
2 3 465.013 587.594 3322.477
2 4 1475.911 2197.404 10 812.317
2 5 2794.532 4264.810 19 328.127
2 6 2878.868 4138.287 17 829.715
2 7 1261.348 1588.135 6638.698

TABLE 2—SIF coefficients for bending proposed by Shin and Cai (Nijk).
i j k¼0 k¼1 k¼2
0 0 1.346 0.190 0.926

0 1 9.627 1.323 6.767
0 2 82.244 8.317 42.734
0 3 360.650 31.454 162.595
0 4 841.678 66.389 345.453
0 5 973.482 71.557 375.935
0 6 449.146 31.022 165.151
1 0 0.640 0.347 1.399

1 1 6.435 2.839 10.348
1 2 36.062 18.649 71.260
1 3 102.765 70.186 263.786
1 4 151.830 142.227 531.560
1 5 107.831 144.956 544.306
1 6 27.262 58.870 225.705
2 0 0.022 0.175 0.454
2 1 0.207 1.635 2.400
2 2 22.436 9.091 4.388
2 3 148.962 32.253 18.246
2 4 426.773 60.188 110.187
2 5 554.803 55.293 186.619
2 6 276.533 19.041 108.877
Numerical Results and Discussion

The study of the convergence was performed to obtain the number z of seg-
ments in which each ellipse is divided and the value of the maximum crack
increase Da(max) [15]. The geometrical evolution of the crack front, character-
ized as part of the ellipse, was determined for every relative crack depth a=D
through the aspect ratio a=b (Figs. 4–6). These figures plot the evolution of the
aspect ratio a=b with crack growth (represented by the relative crack depth a=D)
for materials with Paris exponent m ¼ 2, 3, and 4.
Under fatigue loading, different initial crack configurations tend to a prefer-
ential path (in a plot a=b-a=D), the convergence (proximity between the curves
representing the crack advance from different initial crack shapes) being faster
for higher values of the m coefficient of the Paris law and greater for the bend-
ing loading than for the tensile loading. It is observed that results depend on the
exponent of the Paris law (Paris coefficients), so that for m ¼ 2 and m ¼ 3 fronts
are more distant between them than for m ¼ 3 and m ¼ 4, where the m ¼ 3 front
is between m ¼ 2 and m ¼ 4.
When subjected to bending, growth curves generally present lower values
for the a=b parameter than under tension, with the exception of the deepest

FIG. 4—Evolution of the aspect ratio a=b with crack growth (represented by the relative
crack depth a=D) for a material with Paris exponent m ¼ 2, starting from different initial
crack geometries (corresponding to the beginning of each curve, i.e., the point of mini-
mum crack depth a=D) under tension loading (left) and bending moment (right).
cracks growing from an initial crack aspect ratio (a=b)0 % 0. If the initial crack
is circular [i.e., (a=b)0 ¼ 1], the aspect ratio a=b diminishes with the crack
growth, whereas when the initial crack is quasi-straight [i.e., (a=b)0 % 0], the as-
pect ratio a=b increases at the beginning and decreases later [with the exception
of initially deep cracks with (a=D)0 % 0.5, where the aspect ratio a=b always
increases], cf. Figs. 4–6. With quasi-circular initial geometries the aspect ratio
acquires a smaller value for higher values of m, whereas for quasi-straight geo-
metries it tends to higher values until crack depths close to half the diameter of
the round bar, after which this tendency reverses [again with the exception of
initially deep cracks with (a=D)0 % 0.5]. In addition, for m ¼ 3 and m ¼ 4 all
cracks in the last stage of growth (with relative crack depth close to a=D ¼ 0.8)
exhibit an increasing aspect ratio a=b.

Generally, the value of the dimensionless SIF increases when so does the
relative crack depth for the considered conditions in the research, converging
for the different geometries of the initial crack (Figs. 7–9). For bending loading,
the dimensionless SIF has a smaller value compared to the bar specimen sub-
jected to tensile loading (even from smaller relative crack depths), where the
dimensionless SIF under bending is roughly one third of that under tension for
a relative crack depth of 0.8. Thus the risk of catastrophic failure is higher in
the case of tensile loading (in relation to the less dangerous bending situation) if
a local fracture criterion (on the basis of the maximum local SIF K along the
crack front) is used, considering that fracture takes place when K reaches
the material fracture toughness KC.
Maximum values of the dimensionless SIF Ymax (Figs. 7–9, left) also show
a greater convergence than minimum values of the dimensionless SIF Ymin
(Figs. 7–9, right). This fact is more noticeable in bending loading (where the
FIG. 7—For m ¼ 2 evolution of max dimensionless SIF (left) and minimum dimension-
less SIF (right).

less SIF (right).
minimum also converges well) than in tension loading. The greater the charac-
teristic m parameter of the material, the better the convergence of the results
for the different initial geometries, both of the maximum and the minimum
SIF, along the crack front.
Conclusions
According to the Paris-Erdogan law, in fatigue propagation the different initial
crack geometries tend to a unique path on the a=b versus a=D plot, this conver-
gence (proximity between the curves representing the crack advance from dif-
ferent initial crack shapes) being faster for higher coefficients m of Paris. With
quasi-circular initial geometries, the aspect ratio acquires a smaller value for
higher values of m, whereas for quasi-straight geometries it tends to higher val-
ues until crack depths close to half the diameter of the round bar, after which
this tendency reverses [with the exception of initially deep crack with
(a=D)0 % 0.5].
less SIF (right).

Maximum and minimum dimensionless stress intensity factor SIF along

the crack front are smaller under bending than under tension, while the conver-
gence of such a SIF is better under bending than under tension. Maximum
dimensionless SIF presents lower dispersion than the minimum for the differ-
ent initial cracks. Therefore, fracture risk due to a local fracture criterion (when
the SIF value reaches fracture toughness) is higher under tension than under
bending.
The greater the m coefficient of the Paris law, the greater the convergence
of the different initial crack conditions in almost all the results: geometry of the
crack front (a=b) and dimensionless SIF (Ymax, Ymin). The difference between
the results for the different values of m is always bigger between m ¼ 2 and
m ¼ 3 than between m ¼ 3 and m ¼ 4, which implies that, as this parameter
increases, there is less dependence of results on it.
Acknowledgments
The writers wish to acknowledge the financial support provided by the following
Spanish Institutions: Ministry for Science and Technology (MCYT; Grant
MAT2002-01831), Ministry for Education and Science (MEC; Grant BIA2005-
08965), Ministry for Science and Innovation (MICINN; Grant BIA2008-06810),
and Junta de Castilla y León (JCyL; Grants SA067A05, SA111A07, and
SA039A08).
References
[1] Levan, A., and Royer, J., “Part-Circular Surface Cracks in Round Bars Under Ten-
sion, Bending and Twisting,” Int. J. Fract., Vol. 61, 1993, pp. 71–99.
[2] Carpinteri, A., “Shape Change of Surface Cracks in Round Bars Under Cyclic Axial
Loading,” Int. J. Fatigue, Vol. 15, 1993, pp. 21–26.
[3] Couroneau, N., and Royer, J., “Simplified Model for the Fatigue Growth Analysis of
Surface Cracks in Round Bars Under Mode I,” Int. J. Fatigue, Vol. 20, 1998, pp.
711–718.
[4] Shin, C. S., and Cai, C. Q., “Evaluating Fatigue Crack Propagation Properties Using
a Cylindrical Rod Specimen,” Int. J. Fatigue, Vol. 29, 2007, pp. 397–405.
[5] Lin, X. B., and Smith, R. A., “Shape Growth Simulation of Surface Cracks in Ten-
sion Fatigued Round Bars,” Int. J. Fatigue, Vol. 19, 1997, pp. 461–469.
[6] Lin, X. B., and Smith, R. A., “Fatigue Growth Simulation for Cracks in Notched
and Unnotched Round Bars,” Int. J. Mech. Sci., Vol. 40, 1998, pp. 405–419.
[7] Shih, Y.-S., and Chen, J.-J., “Analysis of Fatigue Crack Growth on a Cracked Shaft,”
[8] Astiz, M. A., “An Incompatible Singular Elastic Element for Two- and Three-
Dimensional Crack Problems,” Int. J. Fract., Vol. 31, 1986, pp. 105–124.
[9] Couroneau, N., and Royer, J., “Simplifying Hypotheses for the Fatigue Growth
Analysis of Surface Cracks in Round Bars,” Comput. Struct., Vol. 77, 2000, pp.
381–389.
[10] Da Fonte, M., and de Freitas, M., “Stress Intensity Factors for Semi-Elliptical Sur-
face Cracks in Round Bars Under Bending and Torsion,” Int. J. Fatigue, Vol. 21,
1999, pp. 457–463.

[11] Shih, Y.-S., and Chen, J.-J., “The Stress Intensity Factor Study of an Elliptical
Cracked Shaft,” Nucl. Eng. Des., Vol. 214, 2002, pp. 137–145.
[12] Shin, C. S., and Cai, C. Q., “Experimental and Finite Element Analyses on Stress In-
tensity Factors of an Elliptical Surface Crack in a Circular Shaft Under Tension
and Bending,” Int. J. Fract., Vol. 129, 2004, pp. 239–264.
[13] Toribio, J., Matos, J. C., González, B., and Escuadra, J., “An Automated Procedure
for the Geometrical Modelling of a Surface Crack Front,” Struct. Durab. Health
Monit., Vol. 123, 2009, pp. 1–16.
[14] Paris, P. C., and Erdogan, F., “A Critical Analysis of Crack Propagation Laws,” J.
Basic Eng., Vol. 85D, 1963, pp. 528–534.
[15] Toribio, J., Matos, J. C., González, B., and Escuadra, J., “Numerical Modelling of
Crack Shape Evolution for Surface Flaws in Round Bars Under Tensile Loading,”
Eng. Fail. Anal., Vol. 16, 2009, pp.618–630.


doi:10.1520/JAI103952
S. Ismonov1 and S. R. Daniewicz1
Study of an On-Line Crack Compliance

Technique for Residual Stress Measurement
Using 2D Finite Element Simulations of
Fatigue Crack Growth
ABSTRACT: There are several methods available to measure residual

stress fields present within a structural component. Recently a new so called
on-line crack compliance technique has been proposed, which is based on
linear elastic fracture mechanics. This experimental method uses incremen-
tal crack mouth opening displacements measured during fatigue crack
growth testing to generate information on the existing residual stresses along
the crack line. The present study employs two dimensional (2D) plane stress
finite element simulations of fatigue crack growth from a cold worked hole to
investigate the performance of this technique. Using the simulation results,
the stress intensity factors due to the residual stress field normalized by the
maximum applied stress intensity factor KIrs/KImax were obtained from the on-
line crack compliance method. For validation, the J-integral approach was
used to calculate KIrs/KImax values from fatigue crack growth simulations in
an elastic material. The two methods generated nearly identical results. Fa-
tigue crack growth was also simulated in an elastic-plastic material. Even
though the stress intensity factor is not the appropriate crack tip characteriz-
ing technique for elastic-plastic material conditions, it is still investigated here
to approximate the actual testing conditions, where plastic deformation near
the crack tip is unavoidable. The KIrs/KImax solutions are presented for differ-
ent cold work levels and applied loadings. Results indicate that the agree-
ment between the elastic and elastic-plastic crack growth solutions is
dependent on the maximum applied loading level, as might be expected.
Manuscript received May 7, 2011; accepted for publication April 26, 2012; published
online May 2012.
1
Dept. of Mechanical Engineering, Mississippi State Univ., Mississippi St., MS 39762
Cite as: Ismonov, S. and Daniewicz, S. R., “Study of an On-Line Crack Compliance
Technique for Residual Stress Measurement Using 2D Finite Element Simulations of
Fatigue Crack Growth,” J. ASTM Intl., Vol. 9, No. 5. doi:10.1520/JAI103952.
Copyright V
136

ISMONOV AND DANIEWICZ, doi:10.1520/JAI103952 137
KEYWORDS: On-line crack compliance technique, residual stress intensity

factors, cold working process, fatigue crack growth simulation
Introduction
Residual stresses are those which remain in a body without any external load.
They may be introduced to structural components during manufacturing proc-
esses such as forging, casting, welding, machining, or from heat treatments
such as quenching. Several life enhancement processes have also been devel-
oped to induce compressive residual stresses. Compressive residual stresses are
beneficial to fatigue life under low amplitude, high frequency loadings, since
they retard crack initiation and propagation. Residual stresses, regardless of the
manner of their introduction, are generally produced by nonuniform plastic de-
formation caused by mechanical or thermal loads or by diffusion processes
such as carburizing and nitriding.
Apart from macro-stresses discussed above, grain scale (intergranular) and
atomic scale stresses exist. Low level intergranular micro-stresses are nearly
always present in polycrystalline structures because of variations in the elastic
and thermal properties of differently oriented neighboring grains. Higher inter-
granular stresses exist when the microstructure contains multiple phases.
Atomic stresses; on the other hand, originate from dislocations and coherency
at interfaces [1]. Except for understanding microcrack growth behaviors, the
grain scale and atomic micro-stresses are often ignored in crack growth life
assessment analysis in a metallic component [2]. This is because micro-stresses
must balance out over the very small distance. The current paper will use the
term “residual stresses” to refer to macro-stresses.
The negative influence of the residual stresses on fatigue life is usually
accounted for by a factor of safety, whereas the positive effects are generally not
explicitly considered during the design process. Understanding the residual
stresses present in a component is important to better quantify their beneficial
or detrimental impact. Numerous experimental and numerical methods have
been developed to measure residual stress. Experimental measurement meth-
ods are typically subdivided into three groups: (a) nondestructive, (b) semi-
destructive and (b) destructive.
In nondestructive methods, a workpiece remains physically unaltered, and
the stress field is obtained from the relationship between the physical or crystal-
lographic parameters and the residual stress [3]. Diffraction methods that use
X-ray, electron, or neutron beams are considered as nondestructive if the
stresses are to be measured near the external surfaces. Semi-destructive meth-
ods do not substantially destroy the specimen and the damage is very localized.
A hole drilling method is an example for this category. In this method, strain
gages are attached to the surface, and a hole is drilled in a nearby location.
Relieved strains are detected by strain gages, which are then related to residual
stresses [4].
Destructive measurement methods require the material to be destroyed
while the stresses are measured. Examples for destructive methods include a
slitting method, in which a part is incrementally cut along the plane where the

residual stresses are to be measured and changes in strain at a suitable location

are recorded. By treating this cut as a mathematically sharp crack, a linear elas-
tic fracture mechanics (LEFM) approach is employed to find the relation
between the incremental change in strain with respect to the crack length (de/
da) and the stress intensity factor (SIF) KIrs due to the residual stress
E0 de
KIrs ¼ (1)
ZðaÞ da
where:
E0 ¼ generalized Young’s modulus and
Z(a) ¼ influence function [5].
Calculated SIFs KIrs for the crack length a can then be converted to residual
stresses via inverse solution methods such as incremental stress [6], series
expansion [7], or pulse method [8]. For further details on the slitting method,
the reader may refer to Refs [5,9–12].
Recently, a new so called an on-line crack compliance technique has been
introduced, which can be used to determine KIrs from the crack opening dis-
placement measurements “on-line,” that is during an actual fatigue crack
growth test [13–15]. Hence, this method generates additional data regarding the
residual stress field as a by-product as the crack growth test is carried out. This
method is based on the slitting method and is derived from LEFM as discussed
further in the next section.
The finite element method (FEM) has become a valuable tool to determine
the residual stress fields by making it possible to simulate a wide range of life
enhancement and manufacturing processes numerically (see for example Refs
[16–18]). The FEM can also be used to study the existing experimental methods
of stress measurement. Prime [19] introduced residual stresses in a finite ele-
ment (FE) model of a compact tension (CT) specimen by overloading the model
beyond the elastic limit of the material. He then simulated the slitting method
by incrementally removing the elements along the crack plane and letting the
model elastically unload. Obtained stress fields from the strain solutions on the
back face of the model compared well with the residual stress distributions pro-
duced in the FE model after the overloading event. De Swardt [20] also
employed FEM to simulate the slitting technique on autofrettaged thick-walled
high-strength steel cylinders. He progressively extended the cut in his model by
modifying the nodal constraints along the line of the cut and recorded the
strains on the outside wall. De Swardt compared the computed strains with the
experimental strain data from the slitting method, and concluded that using an
elastic-plastic material model incorporating the Bauschinger effect produced
the best results.
In this study, the on-line crack compliance method was simulated using a
FE model of a rectangular sheet with a central hole under plane stress condi-
tions. The sheet material was chosen to be an AA7075-T6 aluminum alloy. The
analysis was completed in two stages. In the first stage, a residual stress field
was introduced around the hole by a cold hole expansion simulation. In the sec-
ond stage, crack growth simulation was performed by applying remote cyclic

loads and incrementally propagating the crack during each cycle. The crack
growth stage was conducted with two material behaviors: (a) purely elastic and
(b) elastic-plastic. This was done to better understand the performance of the
on-line crack compliance technique under more realistic conditions with plastic
deformations present behind and ahead of the crack tip. The mode I SIFs due to
the residual stress field KIrs /KImax normalized by the maximum applied SIF
were obtained using the on-line crack compliance method. As part of the valida-
tion process, the elastic crack growth solutions of KIrs/KImax were compared
with the results obtained from J-integral values. Finally, the influence of plastic-
ity is presented by comparing the results from the elastic and elastic-plastic
crack growth simulations for different cold working levels and applied loadings.
Methodology
On-Line Crack Compliance Method

Figure 1 illustrates a close-up view of a crack with a length a advancing by an
amount da. A newly extended crack face is depicted with a dashed line in the fig-
ure. The crack mouth opening displacement (CMOD) of the crack a under a
remote load P is given by d. As the crack length is grown to a þ da, d will increase
by an increment dd under the same applied load P. For linear elastic materials
under plane stress conditions, the Mode I SIF KI can be expressed as (see Ap-
pendix for derivation)
E dd
KI ¼ (2)
ZðaÞ da
where:
E ¼ Young’s modulus of the material and
FIG. 1—Schematic illustration of crack extension.

Z(a) ¼ influence function that depends on the geometry of the specimen as

well as the location of the displacement measurement.
The influence function Z(a) is simply a doubled Green’s function for the
given geometry as shown in the Appendix. Note that Eq 2 is very similar to Eq 1
of the slitting method, except displacements are used instead of strains. Note
also that the crack opening displacement d can be measured at any fixed point
along the crack face, since the influence function Z(a) changes accordingly to
give the same value for KI. In the present study, the CMOD is used for d.
Figure 2 depicts the load-displacement curves as the remote load is increased
from zero to Pmax prior to and after the crack has been grown by da. Figure 2(a)
is for the specimen without any residual stress field, whereas Fig. 2(b) is for the
specimen with a compressive residual stress field present in the crack growth
region. In the absence of any residual stress field, the load-displacement curves
are linear, and the crack starts to open at the onset of load application. The maxi-
mum applied SIF KImax will be
E ddmax
KImax ¼ (3)
ZðaÞ da
where:
ddmax ¼ incremental CMOD at the maximum load Pmax without residual
stress as shown in Fig. 2(a).
With a compressive residual stress field, the crack mouth does not open
until the applied load reaches a certain level. This corresponds to a vertical seg-
ment in Fig. 2(b). As the applied load is further increased, the crack will start
opening at the same rate as in the case with no residual stress field. Thus, the re-
spective slopes of the inclined segments a and a þ da in Fig. 2(b) are the same as
those of the lines a and a þ da in Fig. 2(a). However, the incremental opening
displacement ddrsmax shown in Fig. 2(b) is smaller because of the presence of
compressive residual stress field. This results in a lower maximum SIF, KIrsmax
FIG. 2—Load-displacement curves under linear elastic conditions: (a) no residual

stress field, (b) with compressive residual stress field.

E ddrsmax
KIrsmax ¼ (4)
ZðaÞ da
Superposition can be employed to determine the SIF KIrs due to the residual
stress field alone from the Eqs 3 and 4

E ddrsmax ddmax
KIrs ¼ KIrsmax KImax ¼ (5)
ZðaÞ da
Normalizing KIrs by KImax will result in a simple expression
KIrs ddrsmax ddmax ddrsmax

¼ ¼ 1 (6)
KImax ddmax ddmax
Hence, KIrs /KImax is readily determined from the ratio of the incremental changes
in the displacements ddrsmax and ddmax given in Fig. 2. The effort reported here
involves a study of this nondimensional parameter to investigate the influence of
plastic deformation (ahead and behind the tip of a growing crack) on the per-
formance of the on-line crack compliance method. It must be noted, however,
that even though two different specimens (with and without the residual stress
field) were used to describe the methodology above, a single specimen with the
compressive residual stress field is in fact sufficient to generate KIrs/KImax data.
That is because the slopes of the corresponding curves in Figs. 2(a) and 2(b) are
same, and the incremental crack opening displacement ddmax can be obtained
directly from Fig. 2(b) by extending the load-displacement curves down to the
minimum load and joining the bottom ends as shown in Fig. 3.
FIG. 3—Illustration of obtaining residual stress free data ddmax from load-displacement
curves with compressive residual stress field.

J-Integral Method
The ratio KIrs/KImax in Eq 6 can also be computed using J-integral values com-
puted near the crack tip. For linear elastic materials under Mode I loading, KI
can be obtained from J using the relation
pffiffiffiffiffiffi
KI ¼ JE (7)
where:
E ¼ Young’s modulus of the material [21].
If Jrsmax and Jmax are the J-integrals computed at the maximum applied load
Pmax with and without the presence of the compressive residual stress field,
respectively, the ratio KIrs /KImax can be expressed as
pffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffi
KIrs ð Jmax E Jrsmax EÞ Jrsmax Jmax Jrsmax
¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffiffi ¼ 1 (8)
KImax Jmax E Jmax Jmax
This approach may be used to validate the numerical solutions from the on-line
compliance technique when a linear elastic material model is used.
Finite Element Model

Figure 4 shows the geometry and dimensions of a rectangular sheet with a cir-
cular hole at the center. A single crack perpendicular to the applied load
FIG. 4—Sheet geometry with dimensions: h ¼ 130 mm, 2w ¼ 44.45 mm, r ¼ 3.535 mm,
a ¼ 0.345 mm.

emanates from the hole edge. The size of the sheet is 130 mm by 44.45 mm
(h 2w) and the radius of the hole is r ¼ 3.535 mm. A plane stress condition is
assumed valid with a unit thickness t ¼ 1 mm. The crack grows from an initial
length a ¼ 0.345 mm by a total amount da ¼ 3.57 mm. The remote stress S ¼
F/(2wt) (F ¼ force) is applied in a cyclic fashion from zero to Smax and back to
zero in each cycle giving a zero load ratio (R ¼ 0). The maximum applied load-
ings considered are Smax ¼ 0.3rys and 0.4rys, where rys is the material yield
strength. These loadings generate average KImax of 19.4 and 25.8 MPaHm,
respectively during crack propagation.
A multilinear stress-strain curve used for the sheet material (AA7075-T6) is
given in Fig. 5 [22]. The Young’s modulus and the Poisson’s ratio are E ¼ 72.5
GPa and ¼ 0.3, respectively. The yield strength of the material is rys ¼ 483
MPa. For the elastic crack growth simulation, only the elastic domain of the
curve is used. Thus, E and are sufficient parameters to describe the material
constitutive model. For cold working and elastic-plastic crack growth simula-
tions, the stress-strain data in both elastic and plastic domains are used with
the Von-Mises yield criterion and isotropic hardening behavior.
ANSYS 12.0 FE program was used to conduct the crack growth simulations.
The entire analysis consists of two major stages: (i) cold hole expansion simula-
tion, (ii) crack growth simulation. In the first stage, a FE mesh of the model was
created using an elastic-plastic material. Figure 6 shows a typical FE mesh,
which consists of about 13500 nodes and 13000 4 node quadrilateral plane
stress elements. Only the top half of the sheet is modeled using symmetry
boundary conditions along the crack line. A compressive residual stress field is
obtained by uniformly expanding the hole beyond the elastic limit of the mate-
rial and allowing it to elastically unload. For further details on cold expansion
simulation, the reader may refer to Ref [23]. The radial interferences considered
FIG. 5—Stress-strain curve for AA7075-T6 Aluminum Alloy.

FIG. 6—Finite element mesh.
here are Dr ¼ 0.8 %, 1.2 %, and 1.6 %, where Dr is the amount of the radial hole
expansion (i.e., initial tool misfit).
Two different cases were considered with regard to crack growth through
the residual stress field:
1. Elastic crack growth: the residual stress field obtained in stage 1 was
transferred to another identical FE mesh but with a linear elastic mate-
rial constitutive model. The crack growth simulation was performed
using the new FE mesh.
2. Elastic-plastic crack growth: the model from stage 1 was used to con-
tinue with the crack growth simulation.
An initial crack was inserted by removing the displacement constraints and
using a rigid, frictionless contact surface along the crack line to prevent crack
face overlapping. An augmented Lagrangian contact algorithm was used in
ANSYS. Highly refined elements of equal length occupy the crack growth region
as shown in Fig. 6. At the minimum point of each load cycle, the crack tip was
extended by one element length. This type of incremental crack growth simula-
tions were widely used previously by other authors for crack closure studies
with both 2D and 3D models (see for example Refs [24–26]). However, while the
crack closure studies investigated the near tip behavior, current study focused
on the remote location from the crack tip (i.e., crack mouth node on the hole
edge) to simulate the on-line crack compliance method. The crack mouth open-
ing displacement solutions d versus applied load data was used for each cycle to
calculate the normalized SIF due to compressive residual stress field KIrs /KImax

using Eq 6. J-integral values were also computed using ANSYS built-in com-
mand CINT [27] for different crack lengths to obtain KIrs /KImax from Eq 8. The
J-integral evaluation in ANSYS is based on the domain integral method
described in Ref [28].
Results and Discussion

Figure 7 presents hoop residual stress fields created around the hole after uni-
form hole-expansion simulation with different radial interferences are carried
out. Normalized stress results (rHH/rys) are plotted versus the normalized dis-
tance from the hole center x/w in the figure. It is observed that the greater radial
interference increases both the magnitude and the depth of the resulting resid-
ual stress field.
An example of normalized SIFs due to the residual stress field from the on-
line crack compliance (Eq 6) and the J-integral (Eq 8) methods are shown in
Fig. 8 for the case of elastic crack growth. The KIrs/KImax values are plotted ver-
sus the normalized crack length (a þ r)/w in the figure. The hole is cold worked
with 1.2 % radial interference, and the applied maximum load is Smax/rys ¼ 0.4.
It is observed that the two methodologies produce nearly identical solutions.
Thus, the on-line crack compliance technique and the more traditional J-inte-
gral methods are equivalent when the crack is grown under purely elastic condi-
tions. Similar observations were also made for other applied loadings and cold
working levels, but their results are omitted here for brevity. Note that no addi-
tional residual stress field is introduced to the numerical model during elastic
crack growth. Therefore, the information obtained from the on-line crack com-
pliance and the J-integral methods pertain to the original residual stresses
FIG. 7—Hoop residual stress field for different radial interferences.

FIG. 8—Comparison of KIrs/KImax solutions from crack compliance technique and

J-Integral method for elastic crack growth.
produced from the cold working simulation. This may no longer be true when
material plasticity is included during crack growth, since the plastic deforma-
tion occurring ahead of the crack tip may alter the existing residual stress field.
The wake of plastically deformed material left behind the growing crack may
also modify the original residual stress.
Next, consider crack growth using a more realistic elastic-plastic material.
Figure 9 presents an example of a convergence study performed for elastic-
plastic crack growth with 1.2 % cold work and Smax/rys ¼ 0.4 applied load. The
KIrs /KImax values computed using the crack compliance method (Eq 6) are plot-
ted versus the normalized crack length (a þ r)/w. Three levels of mesh refine-
ment were made with the element lengths da ¼ 0.12, 0.06, and 0.03 mm along
the crack growth line. Solutions did not change significantly with the level of
mesh refinement, although some noisy behavior was observed when smaller ele-
ments were used. Note that the forward plastic zone size for the normalized
applied load Smax/rys ¼ 0.4 can be estimated to be nearly 0.45 mm. Thus, at least
3 elements contain in the forward plastic zone in the meshes considered. The
convergence behavior for other cold working levels and applied loadings consid-
ered were similar. Solutions of the on-line crack compliance technique pre-
sented next are obtained from the model with an element size da ¼ 0.06 mm in
the crack growth region.
Figure 10 compares the KIrs /KImax results from the elastic and elastic-plastic
crack growth simulations with 0.8 % cold work and Smax/rys ¼ 0.3 applied load. It
is observed that the elastic-plastic crack growth simulation initially exhibits the
lower KIrs /KImax magnitudes. As the crack is further grown, the two solutions
approach one another until they meet at the crack length of (a þ r)/w ¼ 0.22. After

FIG. 9—Convergence study for elastic-plastic crack growth results.
FIG. 10—Comparison of KIrs/KImax solutions from elastic and elastic-plastic crack

growth simulations with 0.8 % cold work and Smax/rys ¼ 0.3.

that the elastic-plastic crack growth model generates slightly higher normalized
residual SIF levels. The maximum percentage difference of DKIrs /KImax ¼ 9 %
occurs near the crack length (a þ r)/w ¼ 0.19.
Figure 11 presents the results from the elastic and elastic-plastic crack
growth simulations with 1.2 % cold work and Smax/rys ¼ 0.4 applied load. Note
that a higher applied load was used for the higher level of cold work, because
the lower applied load did not completely open the crack face during the load
cycle. The on-line crack compliance technique cannot be used for closed or par-
tially open cracks since it requires the incremental crack face opening displace-
ments dd for the fully open cracks under the maximum load. As shown in
Fig. 11, variation occurs between the elastic and elastic-plastic crack growth
model solutions for the crack lengths less than (a þ r)/w ¼ 0.24. The maximum
difference of about DKIrs/KImax ¼ 11 % is observed near the normalized crack
length (a þ r)/w ¼ 0.21. It is of interest to know whether the amount of variation
is dependent on the higher residual stress magnitudes or the higher applied
load level. To shed some light on this, following two sets of simulations were
conducted: (a) higher applied load of Smax/rys ¼ 0.4 with 0.8 % cold working
level to accompany results in Fig. 10; (b) greater cold working amount of 1.6 %
with the applied load Smax/rys ¼ 0.4 to compare with Fig. 11.
Normalized KIrs/KImax results from these two different cases are shown in
Figs. 12 and 13 in the same format as of the previous two plots. First, consider
the results given in Fig. 12 in comparison to those in Fig. 10 to understand the
effect of the applied load level. From Fig. 12, it is observed that KIrs/KImax results
from the elastic-plastic model are drastically reduced because of the higher
applied load with the maximum normalized difference of DKIrs/KImax ¼ 19 %.





This indicates that, during actual testing, higher loads can produce significantly
lower stress measurements than the originally existing compressive residual
stress field. Next, consider Figs. 11 with 13 to study the influence of residual
stress field on the residual stress intensity factor results. Higher magnitudes
and greater depth of the compressive residual stress field from the increased
cold working level is clearly reflected in the KIrs/KImax solutions in Fig. 13 when
compared with the previous plot in Fig. 11. The absolute KIrs/KImax values
increased further because of the higher compressive stress magnitudes pro-
duced from 1.6 % cold hole expansion (see Fig. 7). The maximum variation
between the elastic and elastic-plastic crack growth is nearly DKIrs/KImax ¼ 11 %
at (a þ r)/w ¼ 0.24. Thus, difference in KIrs /KImax did not change significantly
with the increased level of cold work. Nevertheless, the region of variation is
shifted slightly to the right (compare Figs. 11 and 13). These observations sug-
gest that it is the higher applied loading, not the residual stress magnitude that
increases the deviation between the solutions of elastic and elastic-plastic crack
growth. Residual stress magnitudes seem to affect the location of the bifurca-
tion zone, where the elastic and elastic-plastic curves are different.
Conclusion
The on-line crack compliance technique was studied using a 2D FE model of a
rectangular sheet with a single crack emanating from a central hole under plane
stress conditions. Residual stress fields were produced in the crack growth
region by uniformly cold working the hole to three different levels: 0.8 %, 1.2 %,
and 1.6 %. The applied maximum loads considered were Smax/rys ¼ 0.3 and 0.4
with R ¼ 0. The crack growth simulations were performed under purely elastic
and elastic-plastic conditions. As part of the validation process, the normalized
SIFs KIrs/KImax calculated using the on-line crack compliance technique are
compared with the solutions obtained via J-integral method for elastic crack
growth. Results were in good agreement indicating that the two methods were
equivalent under elastic conditions. Also, the KIrs /KImax values obtained pertain
to the original stress field induced by cold hole expansion simulation since no
additional residual stresses are produced throughout elastic crack growth.
Convergence studies were performed to validate the results of the on-line
crack compliance technique using the elastic-plastic crack growth model. Gen-
erated KIrs /KImax magnitudes were generally lower than those produced from
purely elastic crack growth. The deviation from the elastic solutions grew larger
with increased applied maximum loading. Therefore, the lowest possible load
levels must be used to obtain more accurate data regarding the original residual
stress field present within the component. However, high compressive residual
stresses may prevent the crack faces from fully opening if the applied maximum
load is too low. This must also be considered when selecting the load level,
because the on-line crack compliance technique requires crack face displace-
ments measured for fully open cracks.
The variation between elastic and elastic-plastic results is explained by
recalling the fact that the on-line crack compliance technique was derived using

LEFM principles. Thus, there is no doubt that the existence of plastic deforma-
tion near the crack tip during the fatigue crack growth testing interferes with
the method results. It must also be mentioned that the current study used 2D
plane stress model, which overpredicts the crack tip yielding since the crack tip
generally sees a 3D stress field. Further study needs to be conducted to better
understand the influence of residual stress evolution during the elastic-plastic
crack growth by performing 3D crack growth simulations.
APPENDIX: THE ON-LINE CRACK COMPLIANCE TECHNIQUE

The on-line crack compliance technique was first proposed in [13–15]. This
methodology allows the computation of applied stress intensity factors (SIF)
using the incremental crack opening displacements. This section presents a
comprehensive theoretical background of the technique for a plane stress condi-
tion. Although parts of this discussion are presented elsewhere [14], they are
repeated here for the reader’s convenience.
Consider a cracked plate subjected to a fixed force load P as shown in Fig.
14. The energy release rate of the plate is defined as the rate of change in its total
potential energy P with respect to a newly formed crack surface A [21]
dP
G¼ (A1)
dA
The total potential energy is given by
P¼UW (A2)
FIG. 14—Cracked plate subjected to a fixed load P.

where:
U ¼ strain energy stored in the cracked body and
W ¼ work done by external forces.
Since the applied point force P is fixed (i.e., the specimen is load controlled),
U and W will be
PD
U¼ (A3)
2
W ¼ PD (A4)
Then the total potential energy P and the energy release rate G will become
PD
P¼ ¼ U (A5)
2
dP 1 dU 1 dðPDÞ
G¼ ¼ ¼ (A6)
dA B da 2B da
where:
B ¼ thickness of the specimen.
For a constant applied load P, Eq A6 can further be simplified to
P dðDÞ
G¼ (A7)
2B da
The mode I SIF KIP is related to the energy release rate via the modulus of elas-
ticity E of the material
2 EP dðDÞ
KIP ¼ EG ¼ (A8)
2B da
We recognize that the SIF KIP is a linear function of P, so it may be expressed as
fP ða; w; :::Þ
KIP ¼ P (A9)
B
where:
fP ¼ function that depends on the geometry and crack size. By combining
Eqs A8 and A9 we get
E dD
KIP ¼ (A10)
2fP ða; w; :::Þ da
Thus far, we were able to relate the rate of change of the load point displace-
ment dD/da to the SIF due to the applied force load P via a geometry function
fP(a,w,…) and the modulus of elasticity E. However, the on-line crack compli-
ance method uses an incremental crack face opening displacements dd instead

of dD. Therefore, dD/da will next need to be related to dd/da. We will use the
approach presented in Appendix B of Tada and Paris [29]. Let F be a virtual pair
force applied at a point aF along the crack face, where the crack face opening
displacement d is being measured (see Fig. 15). If KIF is a SIF due to F, then by
superposition the total SIF is
KI ¼ KIP þ KIF (A11)
The strain energy of the cracked body can be decomposed into two parts
ða
@U
U ¼ Uno crack þ dU ¼ Uno crack þ da (A12)
0 @a
where Uno_crack is the strain energy corresponding to the applied forces with no
crack present, and dU is due to introducing a crack a while holding the forces
constant [29]. Then, using the Eqs A6, A8, A11 and A12
ða ða
B
U ¼ Uno crack þB Gda ¼ Uno crack þ ðKIP þ KIF Þ2 da (A13)
0 E 0
For linear elastic materials, Castigliano’s theorem may be employed to deter-

mine the displacements D and d by differentiating the strain energy U with
respect to the corresponding forces P and F and by setting the virtual force F
equal to zero
ða
@U 2B @KIP
D¼ ¼ Dno crack þ KIP da (A14)
@P E 0 @P
FIG. 15—Cracked plate subjected to a fixed load P and a virtual pair load F.

ða
@U 2B @KIF
d¼ ¼ KIP da (A15)
@F E aF @F
where:
Dno_crack ¼ @Uno_crack/@P is the displacement of the uncracked body due to
the applied force P. Note the Eq (A15) does not have a the similar term dno_crack
since, with the absence of the crack, the opposite forces F are applied at the
same point resulting in dno_crack ¼ 0.
By differentiating the above expressions with respect to the crack length a
and recognizing that Dno_crack has no dependence on a we get
@D 2B @KIP
¼ KIP (A16)
@a E @P
@d 2B @KIF
¼ KIP (A17)
@a E @F
We may express KIF in a similar manner to Eq A9 as
fF ða; w; :::Þ
KIF ¼ F (A18)
B
where fF(a,w,…) is regarded as Green’s function that depends on the geometry,
crack length as well as the location of the point load. By substituting Eqs. A9
and A18 into A16 and A17, the following expressions for the displacement rates
can be obtained
dD 2 fP ða; w; :::Þ2
¼ P (A19)
da E B
dd 2 fP ða; w; :::ÞfF ða; w; :::Þ

¼ P (A20)
da E B
Thus, the load point and crack face displacement rates (dD/da and dd/da) are
related to one another as follows
dD=da fP ða; w; :::Þ

¼ (A21)
dd=da fF ða; w; :::Þ
or
dD fP ða; w; :::Þ dd
¼ (A22)
da fF ða; w; :::Þ da
Finally from Eq A10, the mode I SIF due to applied fixed point load P will be
E fP ða; w; :::Þ dd E dd
KIP ¼ ¼ (A23)
2fP ða; w; :::Þ fF ða; w; :::Þ da 2fF ða; w; :::Þ da

And by defining the influence function as
Zða; w; :::Þ ¼ 2fF ða; w; :::Þ (A24)
Equation A10 can be written as
E dd
KIP ¼ (A25)
Zða; w; :::Þ da
Thus, the applied SIF can be represented in terms of the rate of change of the
crack face displacement, the modulus of elasticity of the material, and the influ-
ence function specific to the crack face measurement location. Since, the influ-
ence function is simply Z(a,w,…) ¼ 2fF(a,w,…), there is a direct relation between
Z(a,w,…) and a Green’s function for a pair of point loads applied at the crack
surface, where the opening displacement d is measured.
References
[1] Withers, P. J. and Bhadeshia, H. K. D. H., “Overview: Residual Stress Part 1 - Mea-
surement techniques,” Mater. Sci. Technol., Vol. 17, 2001, pp. 355–365.
[2] Withers, P. J. and Bhadeshia, H. K. D. H., “Overview: Residual Stress Part 2 - Mea-
surement techniques,” Mater. Sci. Technol., Vol. 17, 2001, pp. 366–375.
[3] James, M. R. and Lu J. “Handbook of Measurement of Residual Stresses,” J. Lu, Ed.,
Society for Experimental Mechanics, Lilburn, GA, 1996, pp. 1–4.
[4] E837–01, 2006, “Standard Test Method For Determining Residual Stresses By The
Hole-Drilling Strain-Gage Method,” Annual Book of ASTM Standards, ASTM Inter-
national, West Conshohocken, PA, pp. 724–733.
[5] Schindler, H. J., 1998, “Experimental Determination of Crack Closure by the Cut
Compliance Technique,” Advances in Fatigue Crack Closure Measurement and Anal-
ysis, ASTM STP 1343, R. C. McClung and J. C. Newman, Jr., Eds., ASTM Interna-
tional, West Conshohocken, PA.
[6] Ritchie, D and Leggatt, R. H., “The Measurement of the Distribution of Residual
Stress through the Thickness of a Welded Joint,” Strain, Vol. 23, No. 2, 1987,
pp. 61–70.
[7] Schajer, G. S. and Prime, M. B., “Use of Inverse Solutions for Residual Stress Meas-
urement,” J. Eng. Mater. Technol. Vol. 128, No. 3, 2006, pp. 375–382.
[8] Prime, M. B., “Cross-Sectional Mapping of Residual Stresses by Measuring the Sur-
face Contour after a Cut,” J. Eng. Mater. Technol., Vol. 123, 2006 pp. 162–168.
[9] Schindler, H. J., “Residual Stress Measurement in Cracked Components: Capabil-
ities and Limitations of the Cut Compliance Method,” Mater. Sci. Forum, Vol.
347–349, 2000, pp. 150–155.
[10] Prime, M. B. “Residual Stress Measurement by Successive Extension of a Slot: The
Crack Complance Method,” Appl. Mech. Rev., Vol. 52, No. 2, 1999, pp. 75–96.
[11] Cheng, W. and Finnie, I., “Measurement of Residual Hoop Stresses in Cylinders
Using the Compliance Method,” ASME J. Eng. Mater. Technol., Vol. 108, 1986,
pp. 87–92.
[12] Cheng, W. and Finnie, I., “An Overview of the Crack Compliance Method For
Residual Stress Measurement,” Proceedings of the 4th International Conference on
Residual Stress, Society Experimental Mechanics, Baltimore, 1994, pp. 449–458.

[13] Lados, D. A., Apelian, D., and Donald J. K. “Fracture Mechanics Analysis for Resid-
ual Stress and Crack Closure Corrections,” Int. J. Fatigue, Vol. 29, 2007, pp.
687–694.
[14] Lados, D. A. and Apelian, D. “The Effect of Residual Stress on the Fatigue Crack
Growth Behavior of Al-Si-Mg Cast Alloys - Mechanisms and Corrective Mathemati-
cal Models,” Metall. Mater. Trans. A, Vol. 37A, 2006, pp. 133–145.
[15] Donald, J. K. and Lados, D. A. “An Integrated Methodology for Separating Closure
and Residual Stress Effects from Fatigue Crack Growth Rate Data,” Fatigue Fract.
Eng. Mater. Struct., Vol. 30, 2006, pp. 223–230.
[16] Frija, M. et al. “Finite Element Modeling of Shot Peening Process: Prediction of the
Compressive Residual Stresses, the Plastic Deformations and the Surface Integ-
rity,” Mater. Sci. Eng. A, Vol. 426, 2006, pp. 173–180.
[17] Ding, K. and Ye, L., “FEM Simulation of Two Sided Laser Shock Peening of Thick
Sections of Ti-6Al-4V Alloy,” Surf. Eng., Vol. 19, No. 2, 2003, pp. 127–133.
[18] Ismonov, S., Daniewicz, S. R., Newman, J. C., Jr., Hill, M. R., Urban, M. R., “Three
Dimensional Finite Element Analysis of a Split-Sleeve Cold Expansion Process,” J.
Eng. Mater Technol., Vol. 131, No. 3, 2009, 031007.
[19] Prime, M. B., “Measuring Residual Stress and the Resulting Stress Intensity Factor
in Compact Tension Specimens,” Fatigue Fract. Eng. Mater. Struct., Vol. 22, 1999,
pp. 195–204.
[20] De Swardt, R. R., “Finite Element Simulation of Crack Compliance Experiments to
Measure Residual Stresses in Thick-Walled Cylinders,” J Pressure Vessel Technol.,
Vol. 125, 2003, pp. 305–308.
[21] Anderson, T.L., Fracture Mechanics: Fundamentals and Applications, 3rd ed., CRC
Press, Boca Raton, FL, 2005, pp. 108–110.
[22] Rice, R. C., Jackson, J. L., Bakuckas, J., and Thompson, S., “Metallic Materials
Properties Development and Standardization,” Report No. MMPDS-01 DOT/FAA,
U.S. Dept. of Transportation, Federal Aviation Administration, and Office of Avia-
tion Research, WA, D.C., 2003, p. 3–402.
[23] De Matos, P. F. P., “Numerical Simulation of Cold Working of Rivet Holes,” Finite
Elem. Anal. Design, Vol. 41, 2005, pp. 989–1007.
[24] Newman, J. C., Jr., “A Finite-Element Analysis of Fatigue Crack Closure,” ASTM
STP, Vol. 590, 1976, pp. 281–301.
[25] Roychowdhury, S. and Dodds, R. H., Jr. “Three-Dimensional Effects on Fatigue
Crack Closure in the Small-Scale Yielding Regime – A Finite Element Study,” Fa-
tigue Fract. Eng. Mater. Struct., Vol. 26, 2003, pp. 663–73.
[26] Ismonov, S. and Daniewicz, S. R. “Simulation and Comparison of Several Crack
Closure Assessment Methodologies Using Three-Dimensional Finite Element Ana-
lysis,” Int. J. Fatigue., Vol. 32, No. 8, 2010, pp. 1322–1329.
[27] ANSYS Release 13.0 Online Documentation, ANSYS Inc., Chap. 13.3.1.
[28] Shih, C. F., Moran, B., and Nakamura, T. “Energy Release Rate Along a Three-
Dimensional Crack Front in a Thermally Stressed Body,” Int. J. Fract., Vol. 30, No.
2, 1986, pp. 79–102.
[29] Tada, H., Paris, P. C., and Irwin, G. R. The Stress Analysis Of Cracks Handbook,
Appendix B, 3rd ed., ASME, NY, NY, 2000.


doi:10.1520/JAI103944
K. Yanase1 and M. Endo2
Analysis of the Notch Effect in Fatigue
ABSTRACT: The fatigue-crack propagation at stress concentrations is a

topic of significant importance in a number of engineering applications. Fur-
ther, it is recognized that the fatigue limit of notched components is dictated
by the critical condition for either initiation or propagation of a small crack at
the root of a notch. Moreover, because most fatigue cracks spend the vast
majority of their lives as short cracks, the behavior of such a flaw is of signifi-
cant importance. In the literature, McEvily and co-workers [McEvily, A. J.,
Eifler, D., and Macherauch, E., “An analysis of the Fatigue Growth of Short
Fatigue Cracks,” Eng. Fract. Mech., Vol. 40, No. 3, 1991, pp. 571–584] devel-
oped a modified linear elastic fracture mechanics (LEFM) approach to tackle
a number of fatigue problems, including the growth and threshold behavior of
small fatigue cracks. In this study, a further extension is presented to deal
with notch effects in fatigue. In this method, the elastic–plastic behavior and
the crack closure are taken into account, as the major factors responsible for
the peculiar behavior of small fatigue cracks emanating from notches. In the
present paper, the notch effect in fatigue is systematically investigated by
making use of a mechanism-based computational framework. A series of
parametric studies demonstrate the predictive capability of the proposed
framework. Based on the thorough investigation for notch-fatigue problem,
the novelty of present study is illustrated.
KEYWORDS: notch effect, small fatigue crack, LEFM, Dugdale model
Manuscript received May 4, 2011; accepted for publication January 10, 2012; published
online April 2012.
1
Assistant Professor, Dept. of Mechanical Engineering, Fukuoka Univ. Institute of
Materials Science and Technology, Fukuoka Univ. 8-19-1 Nanakuma, Jonan-ku, Fukuoka
City, Fukuoka, 814-0180, Japan (Corresponding author), e-mail: kyanase@fukuoka-u.ac.jp
2
Professor, Dept. of Mechanical Engineering, Fukuoka Univ. Institute of Materials
Science and Technology, Fukuoka Univ. 8-19-1 Nanakuma, Jonan-ku, Fukuoka City,
Fukuoka, 814-0180, Japan, e-mail: endo@fukuoka-u.ac.jp
Cite as: Yanase, K. and Endo, M., “Analysis of the Notch Effect in Fatigue,” J. ASTM Intl.,
Vol. 9, No. 4. doi:10.1520/JAI103944.
Copyright V
157

Introduction
Geometrical discontinuities in engineering components are unavoidable in
designing machines and structures. Generally, fatigue cracks initiate at the site
of these stress concentrators, called notches, and they often govern the overall
fatigue strength. Therefore, predicting the fatigue strength of notched compo-
nents is an important engineering problem.
Figure 1 schematically represents typical experimental results of the notch
effect in fatigue, in which the nominal stress amplitude, rw is expressed as a
function of the stress concentration factor, Kt [1–3]. This relation is obtained
in tension–compression or rotating–bending fatigue tests for notched steel
specimens of various notch radii with a constant notch depth. It is noted that
point A corresponds to the fatigue limit of smooth specimen rw0 . The fatigue
limit of the notched specimens rw at complete fracture decreases in conjunc-
tion with an increase of Kt , as illustrated by curve A-B for a blunt notch. Fur-
ther, for a sharp notch, the curve tends to approach a horizontal line as shown
by B-C. Within the region surrounded by the curves B-C (termed as rw2 ) and
B-D (termed as rw1 ), non-propagating cracks exist, their size is from several
grains to a few tens of grains [1–4]. The critical point B is called the branch
point [4].
In essence, the fatigue limit of smooth specimens at the point A is not dic-
tated by the critical condition for fatigue-crack initiation because microscopic
non-propagating cracks exist in most steels [2,3,5–8]. For example, the maxi-
mum lengths of non-propagating cracks at the material surface are reported to
be about 100 lm for annealed 0.13 % carbon steel [7,8] and about 50 lm for
FIG. 1—The schematic relationship between the fatigue limit and Kt .

YANASE AND ENDO, doi:10.1520/JAI103944 159
annealed 0.46 % carbon steel [8]. The fatigue damage condition at the material
surface for curve A-B-D for notched specimens is considered similar to that of
smooth specimens at the fatigue limit [9]. If the maximum stress at the notch
root exclusively dictates the fatigue strength of notched specimens, the curve
A-B-D could be easily estimated based on Kt alone. However, the prediction by
rw ¼ rw0 =Kt tends to underestimate the curve A-B-D, as illustrated in Fig. 1.
This is because the fatigue strength is not only influenced by the maximum
stress at a notch tip but is also affected by the stress distribution near the notch.
Therefore, many studies on notch effect take the characteristics of stress distri-
bution into consideration [e.g., see Refs 4, 9–14].
On the line B-C, the fatigue strength is determined by the non-propagation
condition, where cracks stop after initiation at notch roots and propagation
into the specimen interiors. When a notch is sharp and a crack length is rela-
tively long, the notch-root radius has a negligible effect, such that an “equivalent
crack length” can be used as the sum of the notch depth and crack length. Based
on this concept, the stress level of line B-C is frequently assumed constant in
estimating the fatigue limit [9,14]. However, this concept is not applicable for
every notch [2], and the limitation of its applicability is yet to be examined.
In the literature, linear elastic fracture mechanics (LEFM) has been suc-
cessfully used to handle fatigue-crack propagation data. However, LEFM alone
cannot characterize the initial stage of crack growth at sharp notches [15–18].
As is well recognized, the growth of a small fatigue crack from a notch root
exhibits an anomalous behavior [18–20]. Despite the monotonic increase in
stress intensity factor (SIF) associated with crack propagation from notches, it
is sometimes observed that the fatigue-crack growth rate is initially high, then
decelerates once, and thereafter accelerates, merging into the trend of the
growth rate for a long crack. The inelastic notch stress field and the crack clo-
sure development play major roles for this phenomenon [20–23].
As mentioned above, the principal phenomena determining the fatigue
strength of notched components are intimately related to the behavior of a
small fatigue crack emanating from the notch root. Correspondingly, the objec-
tive of this study is to give a rational interpretation for the notch effect in fatigue
by making use of the modified LEFM approach proposed by McEvily et al. [24],
which has been applied to various fatigue problems with regard to small fatigue
cracks [25]. In this paper, this approach is modified to propose a relatively sim-
ple yet moderately accurate method to deal with various fatigue notch prob-
lems. The proposed method is applicable to a wide range of notch geometries,
which cover very small shallow notches, including small flaws, to very deep
notches, and from very blunt notches, including smooth specimens, to very
sharp notches, including cracked specimens. Specifically, a method to predict
the fatigue strength of double-edge-notched plates [26] in a high-cycle fatigue
regime is systematically presented, based on the minimum sets of data obtain-
able from smooth specimens. Further, to demonstrate the predictive capability
of the proposed method, a series of comparisons between theoretical predic-
tions and Frost and Dugdale’s experimental data [26] is provided for fatigue lim-
its, S-N curves, and crack length as a function of number of cycles, and the
lengths of non-propagating cracks.

Analysis of the Behavior of Small Fatigue Cracks from Notches

In this section, McEvily et al.’s [24] fatigue-crack growth equation is revisited
for the sake of illustration and further modifications.
McEvily’s Approach
McEvily et al. have proposed a fatigue-crack growth equation as:
da
¼ A ðDKeff DKeffth Þ2 (1)
dN
where a is the crack length, N is the number of cycles, A is a constant that

depends on the material and environment, DKeff is the effective range of stress
intensity factor (SIF) given by Kmax Kop , where Kmax is the maximum value of
SIF in a cycle, Kop is the value of SIF at the crack opening level, and DKeffth is
the magnitude of DKeff at the threshold level, a material constant. It is worth
noting that Eq 1 is dimensionally correct.
In essence, McEvily’s approach involves a modified LEFM approach, and
the following three aspects are taken into consideration to predict the behavior
of small fatigue cracks:
(1) In the short-crack regime, crack growth is often elastic-plastic in nature
rather than linear-elastic because of a high ratio of fatigue strength to
yield strength. Consequently, the ratio of the plastic zone to the crack
length is relatively large.
(2) As pointed out by Kitagawa and Takahashi [27], in a range of extremely
short cracks, the endurance limit rather than the threshold for macro-
scopic crack propagation is the controlling factor governing the crack
propagation (the Kitagawa effect).
(3) In the wake of a crack of a few microns in length, the crack closure level
is zero. However, when the crack length is a millimeter or so in length,
the crack closure level becomes comparable to that for a long crack.
Accordingly, Eq 1 is rewritten as [24,25]:
da
¼ AM2 (2)
dN
where M, the net driving force for fatigue-crack propagation, is defined as:
pffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffi
M¼ 2pre F þ Y paF Dr ð1 ekk ÞðKopmax Kmin Þ DKeffth (3)
where re is a material constant that can account for the Kitagawa effect and the
value is usually on the order of 1 lm, Y is a geometrical correction factor, k is a
material constant determining the rate of crack closure development, k is the
length of an advancing fatigue crack measured from the tip of an initial crack,
and Kopmax is the level of crack closure for a long crack. Further, F represents an
elastic-plastic correction factor. In practice, at stresses of the order of the

fatigue strength, the maximum value of applied stress is often comparable to

the yield stress. Thus, the elastic plastic behavior of a small crack should be
properly accounted for. To consider this effect, the crack length is modified as
suggested by Irwin [28], by increasing the crack length by one-half of the plastic
zone size. Correspondingly, McEvily et al. [24] calculated the crack-tip plastic
zone size by using Dugdale’s strip yield model [29], and the modified crack
length is defined as:
amod ¼ aF (4)
where amod is the modified crack length. For instance, for a center crack in a
wide panel in tension, F is given as:

1 prmax
F¼ 1 þ sec (5)
2 2rYS
where rmax is the maximum applied stress and rYS is the yield strength. In Eq 3,
the first term represents the crack driving force, D, and the sum of the second
and third terms is the crack resisting force, R. For a crack to propagate, D must
be greater than R. Alternatively, when D is less than R, the crack does not
propagate.
This approach has a wide range of applicability, and the effect of the above
factors can be quantitatively investigated. In addition, the fatigue limit can be
determined from the threshold condition, that is, M ¼ 0, and the fatigue life can
also be estimated by integrating Eq 2. This approach has been successfully
applied in dealing with a number of small fatigue-crack problems, as reviewed
in Ref 25.
Prediction of Notch Effect in Fatigue
Assumptions—To propose a simple yet reasonably accurate method, the fol-

lowing assumptions are employed in this study:
(1) Crack initiation life is only a small fraction of total fatigue life.
(2) Most of the fatigue life is spent in the small fatigue-crack propagation.
(3) A fatigue crack initiated at the notch root grows preferentially along the
notch-root surface or coalesces with another, and then becomes a two-
dimensional crack in the very early stage of propagation.
(4) Once a crack starts to propagate, its tip is closed under compression.
Further, as mentioned earlier, the surface fatigue damage condition at the
notch root on the curve A-B-D in Fig. 1 is considered similar to that for smooth
specimens at the fatigue limit [9]. Thus, it is assumed that the fatigue crack that
appeared under the condition of curve A-B-D is the same as a non-propagating
micro-crack observed in the smooth specimen at fatigue limit. Its size, a0 , can
be estimated by considering the threshold condition (i.e., M ¼ 0 in Eq 3).
Because the value of re in Eq 3 is much smaller than a0 , it is negligible. Accord-
ingly, by making use of DKeffth and rw0 , we estimate the value of a0 by assuming
that crack closure is absent when a crack just starts to propagate:

TABLE 1—Calculated and observed surface lengths of non-propagating cracks in smooth

specimens.
rYS rw0 DKeffth Calculated Observed

Material ðMPaÞ ðMPaÞ ðMPam1=2 Þ 2a0 ðlmÞ 2a0 ðlmÞ
Annealed S10C [7,8] 206 181 3.0 104 100

Annealed S45C [8] 284 245 3.0 63 50

1 DKeffth 2 1 prw0
a0 ¼ where F0 ¼ 1 þ sec (6)
pF0 Yrw0 2 2rYS
For instance, if the non-propagating crack is assumed to be of a semi-circular

shape, Y ¼ 0:73 is adopted. Table 1 shows a comparison of calculated and
observed lengths of non-propagating cracks for annealed 0.13 % carbon steel
(JIS S10C) [7,8] and annealed 0.46 % carbon steel (JIS S45C) [8]. As demon-
strated, Eq 6 reasonably estimates the surface crack length, 2a0 in comparison
with the observed values in the experiments. In this study, it is assumed that a
crack starts its propagation from an initial crack length of a0 . Further, in the
present analysis, Y ¼ 1:12 is adopted to calculate a0 by assuming an edge crack.
Calculation of SIF—In this study, the primary focus is on fatigue-crack

behavior in the double-edge-notched plates [26] (Fig. 2). When the crack length
is relatively short, the notch stress field has a substantial effect. Namely, by con-
sidering the stress concentration caused by the notch, we write SIF based on
[30]:
rffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1:12 Kt a Wt
K ¼ Y1 pða þ tÞr where Y1 ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7)
1 þ 4:5a=q aþt W
FIG. 2—Double-edge-notched plate in Frost and Dugdale’s experiment [26].

where W is the half width of plate (cf. Fig. 2), t is the notch depth, and q is the
notch-root radius. Further, Kt represents the stress concentration factor for a V
notch in a semi-infinite plate, and is given as [31]:
(
Kt 1:000 0:127ðt=qÞ1=2 þ 0:2908ðt=qÞ 0:1420ðt=qÞ3=2 for t=q 1:0
pffiffiffiffiffiffiffi ¼
1þ 2 t=q 1:148 0:160ðq=tÞ1=2 0:0345 ðq=tÞ þ 0:0693 ðq=tÞ3=2 for q=t 1:0
(8)
By contrast, when the crack length is relatively long, the notch stress field has a
vanishing effect. Thus, SIF can be calculated by the following equation:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Wt
K ¼ Y2 pða þ tÞr where Y2 ¼ 1:12 (9)
W
Accordingly, to calculate SIF both for short and long cracks, we write:
( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Y1 pða þ tÞr for a a
K¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where Y1 ða Þ ¼ Y2 ða Þ (10)

Y2 pða þ tÞr for a > a
Here, a is referred to as the transition crack length. It is noted that in Eq 10,

the nominal stress, r, is used instead of the far-field applied stress, r1 for com-
parisons with the experimental data [26]. Therefore, the following relation is
considered for Eqs 7 and 9:
r ðW tÞ ¼ r1 W (11)
Calculation of Plastic Zone Size—To account for the effects of plastic yield-
ing near the crack tip, we make use of the Dugdale strip-yielding [29] for an
edge crack, as shown by Fig. 3. In principle, the closing stress intensity factor
associated with rYS can be computed based on [32]:
ð aþtþpzs pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2rYS pða þ t þ pzsÞ
KYS ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ½1 þ f ðvÞ dx (12)
aþt p ða þ t þ pzsÞ2 x2
where:
f ðvÞ ¼ ð1 v2 Þð0:2945 0:3912v2 þ 0:7685v4 0:9942v6 þ 0:5094v8 Þ;

x
v¼ (13)
a þ t þ pzs
Here, pzs signifies the plastic zone size. Thus, by integrating Eq 12 analytically,
we obtain:

FIG. 3—Application of Dugdale’s model for a crack emanating from edge notch.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
KYS ¼ G rYS pða þ t þ pzsÞ (14)
where:
pffiffiffiffiffiffiffiffiffiffiffiffiffi
2
G¼ 1:762 1:121sin1 c þ c 1 c2 ð0:173 þ 0:113c2 0:141c4
p

6 8
þ 0:131c 0:0509c Þ (15)
with:
aþt
c¼ (16)
a þ t þ pzs
By using Eqs 10 and 14, pzs can be found by satisfying the following condition:
K þ KYS ¼ 0 where K ¼ Kða þ pzsÞ (17)
It is noted that to solve Eq 17, a recursive calculation needs to be performed.

Once pzs is found, we define the modified crack length [28] as follows:
pzs
amod ¼ a þ (18)
2
Finally, to account for the effect of plastic yielding, we modify Eq 10 as follows:
( pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Y1 ðamod Þ pðamod þ tÞr for amod a
Kmod ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where Y1 ða Þ ¼ Y2 ða Þ (19)
Y2 ðamod Þ pðamod þ tÞr for amod > a

Fatigue-Crack Growth Equation—By adopting the aforementioned assump-

tions in the previous sections, we write the fatigue-crack growth equation for
R ¼ 1 as follows (cf. Eqs 2 and 3):
da
¼ AM2 where M ¼ Kmod ð1 ekk ÞKopmax DKeffth (20)
dN
Here, we set:
Kmod ¼ Kmod ðrmax Þ; k ¼ a a0 (21)
To evaluate the initial crack length, a0 , we consider a smooth specimen associ-

ated with rw0 . Therefore, by setting t ¼ 0 and q ¼ 1 to mimic a smooth speci-
men, Eq 17 leads to the following equation:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1:12 pða0 þ pzs0 Þrw0 G rYS pða0 þ pzs0 Þ ¼ 0 (22)
After solving Eq 22 to find pzs0 in terms of a0 , we write:
pzs0
a0 þ ¼ a 0 F0 (23)
2
Finally, by considering the threshold condition (i.e., M ¼ 0), a0 can be rendered

as:

pffiffiffiffiffiffiffiffiffiffiffiffiffi 1 DKeffth 2
1:12 pa0 F0 rw0 DKeffth ¼ 0 ! a0 ¼ (24)
pF0 1:12rw0

To show the predictive capability of the proposed method, a series of compari-
sons between the theoretical predictions and Frost and Dugdale’s [26] experi-
mental data is presented. In the experiments, tension–compression fatigue tests
on mild steel double-edged plates were conducted. The notches were 5.0-mm
deep with various notch radii: 0.10, 0.25, 0.50, 1.3, and 7.6 mm. It has to be
noted that the present study primarily focuses on the data for R ¼ 1.
Determination of Material Constants

For the theoretical prediction, several material constants need to be employed:
rYS , rw0 , DKeffth , Kopmax , k, and A. In the literature [26], rYs ¼ 334 MPa and
rw0 ¼ 200 MPa are given, but other constants are not explicitly provided. In gen-
eral, DKeffth is known to be about 3.0–3.5 MPam1=2 for various steels [25,33].
Thus, it is reasonable for DKeffth to be set as 3:0 MPam1=2 . Kopmax is reported to
be about 3.0–3.3 MPam1=2 for an R of 1 for several steels [25,33], and herein
Kopmax ¼ 3:0 MPam1=2 is adopted. Finally, k is almost 6000 m1 for medium-low
strength steels [25,33].

FIG. 4—S-N curve for smooth specimens.
To obtain the values of A, elaborating experimental works involving mea-

surement of fatigue-crack growth rate are usually required. In the literature
[26], the experimental S-N data for smooth specimens is provided, and the S-N
curve can be predicted with ease by integrating Eq 20 with t ¼ 0 and q ¼ 1.
Accordingly, by seeking the best fit curve for the experimental data, an appropri-
ate value of A can be estimated, as illustrated in Fig. 4. In the identification pro-
cedure, the finite and the endurance regions are treated in the same manner. As
demonstrated, as the applied stress level is decreased, the S-N curve becomes
horizontal (i.e., da/dN ¼ 0), and the corresponding stress level is independent of
the value of A. Table 2 summarizes the material constants adopted for the sub-
sequent theoretical predictions. It is noted that the value of a0 is estimated to be
40 lm based on Eq 24.
Comparisons of Predicted and Experimental Results
Fatigue Limit—Figure 5 shows the relationship between the threshold stress

and the crack length for a relatively sharp notch with a radius of 0.25 mm. The
threshold stress represented by the solid line in the figure is calculated by set-
ting the net deriving force for crack propagation, M ¼ 0 in Eq 20, and it is inde-
pendent of the value of A. The threshold stress means the minimum stress
required for continuous propagation of a crack. In other words, when a condi-
tion of stress and crack length is located above the solid line in Fig. 5, a crack
will propagate and, inversely, any conditions below the solid line result in crack
TABLE 2—Material constants used in the analysis for mild steel.
rYS ðMPaÞ rw0 ðMPaÞ DKeffth ðMPam1=2 Þ Kopmax ðMPam1=2 Þ k ðm1 Þ A ð1=MPa2 Þ
334 200 3.0 3.0 6000 5.0 1010

FIG. 5—The relationship between threshold stress and crack length.
arrest. When the crack length is relatively short (a < 1:0 mm), the threshold
stress is increased because of the gradual development of crack closure. How-
ever, once the applied stress is sufficient to overcome the crack-closure barrier,
the crack can grow further without the increment of the applied stress level.
Accordingly, the predicted maximum stress value in Fig. 5 represents the fatigue
limit of specimens, rw2 (cf. Fig. 1). In this study, to determine rw2 , we applied a
posterior analysis. In other words, after we obtained the relationship between
the threshold stress and crack length as shown by Fig. 5, we seek the maximum
threshold stress or rw2 . The solid mark in Fig. 5 signifies the experimental
results of a broken specimen under a constant stress amplitude. The open
marks signify the unbroken specimens and indicate the lengths of non-
propagating cracks measured from the notch root. As demonstrated, our theo-
retical predictions reasonably capture the behavior of threshold stress associ-
ated with the propagation of fatigue crack.
For demonstrative purposes, the effect of notch-root radius on the thresh-
old stress is simulated in Fig. 6. For a sharp notch, a non-propagating crack can
exist because the fatigue-crack initiation stress is lower than the maximum
threshold stress. By contrast, for a blunt notch, the fatigue-crack initiation
stress itself is the maximum threshold stress. In other words, the fatigue-crack
initiation triggers complete fatigue failure without crack arrest.
Figure 7 illustrates the fatigue limit, rw of notched specimens associated
with Kt . As Kt is increased from unity (i.e., no stress concentration), significant
degradation of the fatigue limit can be observed. As demonstrated, when Kt is
relatively small, rw is determined by the crack initiation stress, rw1 . Accordingly,
initiation of a microscopic fatigue crack directly leads to complete fatigue fail-
ure. When Kt is further increased beyond the branch point (cf. Fig. 1), rw is
determined by a constant value of rw2 irrespective of value of Kt , and rw0 =Kt
tends to underestimate the fatigue limit significantly. In essence, a region

FIG. 6—Simulation for the effect of notch-root radius on the threshold stress.
surrounded by curves rw2 and rw1 represents the region where macroscopic
non-propagating cracks exist. It is found that in comparison with the experi-
mental data, the present prediction can render reasonably accurate results.
Once the necessary material constants are obtained for calculation, the fa-
tigue limits for a variety of notch geometries can be estimated. Accordingly,
based on the proposed calculation, we examine the validity of classical equa-
tions for the fatigue notch factor, Kf , which is defined as the ratio of fatigue
limit of smooth specimen, rw0 to that of notched specimen, rw . For example,
Peterson [36] provided the following equation for Kf :
FIG. 7—The relationship between fatigue limit and Kt . (a) Comparisons between the
present calculation and the experimental data, and (b) comparisons of fatigue limit.

Kt 1
Kf ¼ 1 þ (25)
1 þ a=q
Similarly, according to Neuber [11,37], the fatigue limit is given as:
Kt 1
Kf ¼ 1 þ pffiffiffiffiffiffiffiffi (26)
1 þ b=q
pffiffiffiffi
where a and b are material constants, and we set a ¼ 0:3 mm and b ¼ 0:5 mm1=2
in this analysis [38]. Because no distinction is made for rw1 and rw2 in the above
equations, the estimation of fatigue limit of notched components, rw is simply
made with the relation of rw0 =Kf . Figure 8 shows the comparisons of those equa-
tions for depth of notches, t ¼ 5.0 mm and 1.0 mm. As shown, Eqs 25 and 26 render
reasonable estimates in part (Fig. 8(a)), but in the case of shallow sharp notch (Fig.
8(b)), they estimate too conservative and too non-conservative values in compari-
son with the proposed calculation. Though those classical equations have been rec-
ognized as a practical equation for relatively mild notch geometries, they should be
used carefully by considering their limitation of applicability.
In Eq 20, Kmod is the driving force, D, and the rest of the terms are the
resisting force, R, for fatigue-crack propagation. The prediction of threshold
values, such as simulated in Figs. 5–7, can be made by comparing the values of
D and R. A similar method has previously been proposed for the evaluation of
the fatigue limit of notched components. For instance, El Haddad et al. [15]
added a fictitious crack length, l0 as a material constant to the actual crack
length to accommodate the Kitagawa effect. They simulated the behavior of
small fatigue cracks at notches and presented a relation similar to Figs. 5 and
6. However, l0 does not correspond to a physical quantity [13]. On the other
hand, McEvily and Minakawa [23] and Tanaka and Akiniwa [34] considered
the effect of crack closure for propagation of small fatigue cracks at notches.
They described the notch effect in fatigue successfully. In contrast, in the pro-
posed method extended from McEvily’s approach, the net driving force, M, is
further correlated to the crack propagation rate by da=dN ¼ AM2 [23,24], as
given by Eq 20. Consequently, the proposed method enables one to calculate
the crack length as a function of the number of cycles or the fatigue life of
notched components. Although McEvily’s approach has previously been
extended to deal with the notch problem [35], in this paper, the more general-
ized method is newly proposed by taking the effect of crack tip yielding into
account. Moreover, the previous studies [25,35] have described the effects of
small defects and cracks on the fatigue strength by modifying McEvily’s equa-
tion. In those studies, based on the experimental evidences and the concept of
equivalent crack length, small defects and notches were simplified as a planar
crack in the analysis. In contrast, the present method can evaluate those
effects without using the simplification. Consequently, it is expected that this
new method of calculation can allow us to evaluate quantitatively the influen-
ces of notch geometry, multiaxial stress, etc., on the components with differ-
ent types of defects.

FIG. 8—Comparisons for the fatigue limit. (a) Notch length t ¼ 5.0 mm, and (b) notch
length t ¼ 1.0 mm.
Crack Length versus Number of Cycles—The fatigue-crack growth can be

simulated by integrating Eq 20. To achieve an efficient yet reasonably accurate
computation, the integration is performed with DN ¼ 300cycles in conjunction
with the trapezoid rule. Figure 9 compares the prediction and the experimental
data for the fatigue-crack growth data. As two extreme cases given in the litera-
ture [26], the simulations and comparisons are made for rmax ¼ 56 MPa
(Fig. 9(a)) and rmax ¼ 39 MPa (Fig. 9(b)). It can be said that the predictions mod-
erately simulate the experimentally observed fatigue-crack growth behavior,

FIG. 9—Comparisons for fatigue-crack growth data.
given the inherent large scatter in S-N curve normally observed in the region
beyond 106 cycles. Regarding the theoretical prediction, it is recognized that the
fatigue limit of smooth specimen, rw0 , is one of the dominant material parame-
ters. However, because various material parameters interact with each other to
predict Fig. 9, it is difficult to clarify the significance of each parameter. In our
future work, the sensitivity analysis will be conducted to tackle the issue.
In relation to Figs. 5 and 9, in the case of q ¼ 0:25 mm, the fatigue-crack
growth rate or da=dN is simulated for different stress levels in Fig. 10. When the
stress level is low, da=dN decelerates, and eventually da=dN ¼ 0 or the crack
arrest is attained. On the other hand, when the stress level is increased, though

FIG. 10—Simulated da=dN curves.
da=dN decelerates initially, it accelerates afterward. As is well known, such a

peculiar behavior is closely related to the crack closure, and the present model
can simulate the characteristic behavior at least qualitatively.
S-N Curve for Notched Specimens—In Fig. 11, the S-N curve for notched
specimens is predicted and compared with the experimental data. In the experi-
ment, various notch-root radii with a constant notch depth were investigated.
Regarding the present prediction, fatigue failure is defined when a ¼ 8:0 mm. As
shown, the present prediction can properly simulate an S-N curve for notched
specimens. Further, as the theoretical prediction reveals, the notch-root radius
has a negligible effect on this S-N curve.
FIG. 11—S-N curve for notched specimens.

Concluding Remarks
In this paper, the notch problem in fatigue is systematically investigated by
extending McEvily’s fatigue-crack growth equation. Specifically, the effects of
crack tip yielding and crack closure are properly taken into account in the
computational framework. In addition, a novel yet relatively simple method
to estimate the intrinsic material parameter A is presented. A series of compar-
isons between the theoretical predictions and the experimental data by Frost
and Dugdale strongly supports the predictive capability of the proposed
method.
Acknowledgments
This research is in part sponsored by Japanese Ministry of Education, Science,
Sports and Culture, Grant-in-Aid for Scientific Research (C) (Fund Number
22560092, 2010-2012).
References
[1] Frost, N. E., Marsh, K. J., and Pook, L. P. Metal Fatigue, Oxford University Press,
London, 1974.
[2] ASTM STP-924, 1988, Nisitani, H. and Endo, M., “Unified Treatment of Deep and
Shallow Notches in Rotating Bending Fatigue,” Basic Questions in Fatigue: Vol. 1,
J. T. Fong and R. J. Fields, Eds., ASTM, Philadelphia, pp. 136–153.
[3] Murakami, Y., Metal Fatigue: Effects of Small Defects and Nonmetallic Inclusions,
Elsevier, Oxford, 2002.
[4] Isibasi, T., Prevention of Fatigue and Fracture of Metals, Yokendo, Tokyo, 1954 (in
Japanese),
[5] Hempel, M., “Metallographic Observations on the Fatigue of Steels,” Proceedings of
the International Conference on Fatigue of Metals, Institution of Mechanical Engi-
neers, London, 1956, pp. 543–547.
[6] Wadsworth, N. J., “The Influence of Atmospheric Corrosion on the Fatigue Limit of
Iron-0.5% Carbon,” Philos. Mag., Vol. 6, 1961, pp. 396–401.
[7] Nisitani, H. and Takao, K., “Successive Observation of Fatigue Process in Carbon
Steel, 7:3 Brass and Al-Alloy by Electron Microscope,” Trans. Jpn. Soc. Mech. Eng.,
Vol. 40, 1974, pp. 3454–3266.
[8] Murakami, Y. and Endo, T., “Effects of Small Defects on Fatigue Strength of Met-
als,” Int. J. Fatigue, Vol. 2, 1980, pp. 23–30.
[9] Nisitani, H., “Effects of Size on the Fatigue Limit and the Branch Point in Rotary
Bending Tests of Carbon Steel Specimens,” Bull. Jpn. Soc. Mech. Eng., Vol. 11,
1968, pp. 947–957.
[10] Siebel, E. and Stieler, M., “Dissimilar Stress Distributions and Cyclic Loading,”
Z. Ver. Deutsh. Ing., Vol. 97, pp. 121–131 (in German).
[11] Neuber, H., Theory of Notch Stresses, 2nd ed., Springer-Verlag, Berlin, 1958 (in
German).
[12] Peterson, R. E., “Notch Sensitivity,” Metal Fatigue, G. Sines and J. L. Waisman,
Eds., McGraw-Hill, New York, 1959, pp. 293–306.
[13] Tayler, D., “Geometrical Effects in Fatigue: A Unifying Theoretical Model,” Int. J.
Fatigue, Vol. 21, 1999, pp. 413–420.

[14] Smith, R. A. and Miller, K. J., “Prediction of Fatigue Regimes in Notched

Components,” Int. J. Mech. Sci., Vol. 20, 1978, pp. 201–206.
[15] E1 Haddad, M. H., Topper, T. H., and Smith, K. N. “Prediction of Nonpropagating
Cracks,” Eng. Fract. Mech., Vol. 11, 1979, pp. 573–584.
[16] Shin, C. S. and Smith, R. A., “Fatigue Crack Growth from Sharp Notches,” Int. J.
Fatigue, Vol. 7, 1985, pp. 87–93.
[17] Ogura, K., Miyoshi, Y., and Nishikawa, I., “Fatigue Crack Growth and Closure of
Small Cracks at the Notch Root,” Current Research on Fatigue Cracks, Material
Research Series 1, T. Tanaka, M. Jono, and K. Komai, Eds., The Society of Materials
Science, Kyoto, Japan, 1985, pp. 57–78.
[18] Tanaka, K. and Nakai, Y. “Propagation and Non-Propagation of Short Fatigue
Cracks at a Sharp Notch,” Fatigue Eng Mater. Struct., Vol. 6, 1983, pp. 315–327.
[19] Pearson, S., “Initiation of Fatigue Cracks in Commercial Aluminium Alloys and the
Subsequent Propagation of Very Sharp Cracks,” Eng. Fract. Mech., Vol. 7, 1975, pp.
235–247.
[20] McClung, C. and Sehitoglu, H., “Closure and Growth of Fatigue Cracks at
Notches,” J. Eng. Mater. Technol., Vol. 114, 1992, pp. 1–7.
[21] Smith, R. A. and Miller, K. J., “Fatigue Crack at Notches,” Int. J. Mech. Sci., Vol. 19,
1977, pp. 11–22.
[22] Leis, B. N., “Displacement Controlled Fatigue Crack Growth in Inelastic
Notch Fields: Implication for Short Cracks,” Eng. Fract. Mech., Vol. 22, 1985, pp.
279–293.
[23] McEvily, A. J. and Minakawa, K., “Crack Closure and the Condition for Fatigue
Crack Propagation,”Fatigue Crack Growth Threshold Concepts, D. Davidson and S.
Suresh, Eds., AIME, Warrendale, PA, 1984, pp. 517–530.
[24] McEvily, A. J., Eifler, D., and Macherauch, E., “An Analysis of the Fatigue Growth
of Short Fatigue Cracks,” Eng. Fract. Mech., Vol. 40, No. 3, 1991, pp. 571–584.
[25] Endo, M. and McEvily, A. J., “Prediction of the Behavior of Small Fatigue Cracks,”
Mater. Sci. Eng., A, Vol. 467–470, 2007, pp. 51–58.
[26] Frost, N. E. and Dugdale, D. S., “Fatigue Tests on Notched Mild Steel Plates
with Measurements of Fatigue Cracks,” J. Mech. Phys. Solids, Vol. 5, 1957, pp.
182–192.
[27] Kitagawa, H. and Takahashi, S., “Applicability of Fracture Mechanics to Very
Small Cracks or the Cracks in the Early Stage,” Proceedings of the Second Interna-
tional Conference on Mechanical Behavior of Materials, American Society for Metals,
Metal Park, OH, 1976, pp. 627–631.
[28] Irwin, G. R., “Plastic Zone Near a Crack and Fracture Toughness,” Proceedings of
the Seventh Sagamore and Metallurgical Behavior of Sheet Materials, 1960, pp.
IV63–78.
[29] Dugdale, D. S., “Yielding of Steel Sheets Containing Slits,” J. Mech. Phys. Solids,
Vol. 8, 1960, pp. 100–108.
[30] Lukas, P. and Klesnil, M., “Fatigue Limit of Notched Bodies,” Mater. Sci. Eng., Vol.
34, 1978, pp. 61–66.
[31] Noda, N. and Takase, Y., Strength of Materials with Notch Effect, Nikkan, Kogyo
Shimbun, Tokyo, 2010 (in Japanese),
[32] Hartranft, R. J. and Sih, G. C., Mechanics of Fracture, Vol. 1, Noorhoff, Amsterdam,
1973.
[33] Ishihara, S. and McEvily, A. J., “Analysis of Small Fatigue Crack Growth under
Two-Step Loading Condition,” Small Fatigue Cracks: Mechanics, Mechanisms and
Applications, K. S. Ravichandran, R. O. Ritchie, and Y. Murakami, Eds., Elsevier,
Oxford, 1999, pp. 389–401.

[34] Tanaka, K. and Akiniwa, Y., “Resistance-Curve Method for Predicting Propagation
Threshold of Short Fatigue Cracks at Notches,” Eng. Fract. Mech., Vol. 30, 1988,
pp. 863–876.
[35] McEvily, A. J., Endo, M., Yamashita, K., Ishihara, S., and Matsunaga, H., “Fatigue
Notch Sensitivity and the Notch Size Effect,” Int. J. Fatigue, Vol. 30, 2008, pp.
2087–2093.
[36] Peterson, R. E., Stress Concentration Factor, John Wiley & Sons, New York, 1974.
[37] Kuhn, P. and Hardrath, H. F., “An Engineering Method for Estimating Notch-Size
Effect in Fatigue Tests of Steel,” Report No. NACA TN 2805, Langley Aeronautical
Laboratory, Langley Field, VA, 1952.
[38] Schive, J., Fatigue of Structures and Materials, Springer, New York, 2008.

HIGH TEMPERATURE,
HIGH FREQUENCY, AND
ENVIRONMENTAL EFFECTS

doi:10.1520/JAI103988
Raghu V. Prakash1 and Sampath Dhinakaran2
Estimation of Corrosion Fatigue-Crack Growth

through Frequency Shedding Method
ABSTRACT: Corrosion fatigue-crack growth characteristics are important for

the design of marine and off-shore structures. Design of critical components
requires data on fatigue-crack growth rate at very low frequencies of the order
of 102 to 103 Hz. Experiments at low frequencies pose practical difficulties of
enormous test duration. To address this, it is proposed to estimate corrosion
crack growth characteristics using a frequency shedding method where the
frequency is shed with crack advance using an exponential law. Fatigue-crack
growth rate tests have been conducted on Ni–Mn–Cr steel at a constant DK
range of 18 MPaHm (lower Paris regime) under lab air conditions as well as 3.5 %
NaCl solution. Crack growth rate data plotted as a function of test frequency
presents a straight-line trend in log–log scale for a frequency range of 1–0.1 Hz;
however, there is a change in trend when the frequencies are dropped further,
which could be due to domination of corrosion mechanism. To understand the
role of crack closure, crack closure estimates were obtained at periodic intervals
of crack length and the effective stress intensity graphs suggest acceleration in
crack growth rate due to corrosion as the frequency is reduced.
KEYWORDS: corrosion fatigue, frequency shedding, Ni–Mn–Cr steel, crack
closure, lab air, 3.5 %, NaCl solution
Introduction
Material performance degradation due to combined effect of mechanical load-
ing and environment has been a subject of importance for engineers involved
Manuscript received May 16, 2011; accepted for publication February 29, 2012; published
online May 2012.
1
Dept. of Mechanical Engineering, Indian Institute of Technology Madras, Chennai 600 036,
India, e-mail: raghuprakash@iitm.ac.in
2
Dept. of Mechanical Engineering, Indian Institute of Technology Madras, Chennai 600 036,
India.
Cite as: Prakash, R. V. and Dhinakaran, S., “Estimation of Corrosion Fatigue-Crack
Growth through Frequency Shedding Method,” J. ASTM Intl., Vol. 9, No. 5. doi:10.1520/
JAI103988.
Copyright V
179

with design of long-life components such as power-plants, chemical plants, off-

shore equipment, and marine equipment. Mechanical loading in the presence
of corrosive environments accelerates the damage in materials, resulting in
embrittlement and loss of performance. Conventionally, corrosion behavior of
materials is studied by conducting experiments under salt spray conditions, to
assess the level of corrosion products deposition and its effect on longevity of
materials. In some cases, slow strain rate tests are used to understand the effect
of corrosion in stress–strain response of materials; this assumes that the mate-
rial does not have any pre-existing defect. However, corrosion environment can
cause pitting of surfaces, which act as stress raisers, leading to early crack ini-
tiation and propagation.
The effect of corrosion in case of cracks has been studied often through the
use of stress corrosion cracking parameter, which represents the static parame-
ter to failure in the presence of a crack. However, it is worth mentioning that
even though studies on degradation of materials during cyclic mechanical load-
ing have been actively pursued, in view of the exorbitant time required for
experiments, the data availability is limited in this domain.
When the mechanical fatigue and corrosion happen simultaneously, the mech-
anism of cracking becomes complex; models that are developed for pure mechani-
cal loading conditions do not provide a reliable data. Nikolin and Karpenko [1]
observed that corrosion reduces fatigue life. The damage of a material is reported
to be high when the mechanical fatigue and aggressive environment is acting
simultaneously compared to the sum of damage caused individually [2].
Unlike plain fatigue, the synergistic actions involved in the mechanics of
corrosion fatigue make the damage accumulation faster, once the crack is initi-
ated [3]. Two mechanisms influence the crack growth kinetics at the crack tip—
hydrogen embrittlement and anodic dissolution. Hydrogen embrittlement
involves transport of deleterious species like Hþ, OH, Cl, O2, etc., by diffusion,
convection, and migration. Hydrogen is absorbed into the crack tip region and
diffuses ahead of the crack tip through dislocation transport or pipe diffusion
along the grain boundaries as well as through bulk diffusion processes [4]. The
interaction between hydrogen and metal atoms results in embrittlement and
weakens the bond strength in that region [5]. The local stress field developed
during fatigue loading breaks the embrittled zone further and leads to acceler-
ated crack growth.
On the other hand, the anodic dissolution process involves only the trans-
port process and the crack tip dissolution. In case of passivating metals, a thin,
adherent, and transparent oxide layer is formed at the crack tip, which will pro-
tect the base metal from further corrosion. However, cyclic loading and previ-
ously deposited corrosion products disturb this oxide layer, resulting in
localized cracking of this layer, which exposes the fresh metal surfaces to the
corrosive environment. The corrosion fatigue-crack growth resulting from the
above processes of anodic dissolution is the slowest process [6]. If the loading
frequency decreases, the time available for the corrosive medium to react is
more, which, in turn, results in a faster crack growth rate compared to fatigue
in laboratory air. Electrochemically, both of these mechanisms are different, as
the hydrogen embrittlement is a cathodic process and the metal dissolution is

PRAKASH AND DHINAKARAN, doi:10.1520/JAI103988 181
an anodic process. To further complicate the corrosion fatigue-crack growth

process, corrosion residues on the crack faces prevents the movement of crack
wakes even before reaching the minimum load during the fatigue loading and
causes corrosion-induced crack closure [7–9].
Factors that affect the corrosion fatigue-crack growth rates are: the maxi-
mum stress intensity factor, stress intensity factor range, load (stress) ratio, fre-
quency, waveform, electrochemical potential, environmental parameters, and
microstructure of the material [10]. Among these, the loading frequency plays
an important role in the degradation of fatigue-crack growth characteristics of
a material. Figure 1 presents the effect of loading frequency on fatigue-crack
growth rates under water vapor conditions. It can be noted that the reduction in
test frequency results in faster crack growth rates at all stress intensity factor
ranges.
Marine and off-shore structures are subjected to cyclic loading because of
ocean wave action and during diving [11]. At low frequency of loading, ample
time is available for the corrosive medium to act at the crack tip. The pumping
action of corrosive medium to crack surfaces results in faster cracking due to
FIG. 1—Crack growth rate behavior of AISI 4340 steel in water vapor. [Reproduced
with permission from Ref. 10.]

embrittlement and availability of fresh material at the crack tip [12]. A threefold
increase in the mean corrosion fatigue-crack-initiation life was observed in
A588 Grade A and A517 Grade F steels as the cyclic load frequency was
increased from 1.2 to 300 cpm [13,14].
The effect of environment and loading variables on the rate of fatigue-crack
growth below KISCC (stress intensity for stress corrosion cracking) was studied
on 12Ni–5Cr–3Mo maraging steel tested in 3% NaCl solution [15–17]. The data
showed that the fatigue-crack growth accelerates at stress intensities below
KISCC; the magnitude of this acceleration is dependent on the frequency of the
cyclic stress intensity fluctuations, and the corrosion fatigue-crack growth rate
was observed to be three times faster than the corresponding value in air. Imhof
and Barsom [18] observed that the magnitude of the effect of cyclic frequency
on the rate of corrosion fatigue-crack growth depends strongly on the
environment–material system. The presence of a corrosive medium also results
in a decrease in the material’s residual strength.
Corrosion-assisted fatigue-crack growth in metals has been classified into
three types by McEvily and Wei [2] as shown in Fig. 2. Type A shows the true
corrosion-fatigue behavior where environment accelerates the crack growth under
the cyclic loading conditions by embrittlement of material. It influences cyclic
FIG. 2—Schematic representation of the effect of inert and corrosive environment on

fatigue-crack growth. [Reproduced with permission from Ref. 2.]

fracture even at Kmax < KISCC. It is predominantly a stress-dependent process.

Type C is the mixed corrosion-fatigue behavior, which is a combination of Type A
and Type B. It is a time-stress dependent process. In all the three types, the stress-
corrosion threshold is greater than the fatigue thresholds (KISCC > Kmax,th).
To explain the variation of threshold cyclic stress intensity, DKth with stress
ratio R, and its dependence on microstructure, environment, etc., two threshold
stress-intensity criteria have to be satisfied simultaneously instead of singly
(DKth) [19–21]. They are: (a) the critical cyclic stress-intensity threshold DK*th,
and (b) a critical maximum stress intensity (K*max). Kmax governs the breakage
of crack tip bonds allowing for crack extension, and DK controls the extent of
cycle damage ahead of the crack tip [22]. Experimental determination of DK*th
is one of the most time-consuming processes.
One of the practical difficulties associated with generating data for corro-
sion fatigue is the duration of tests. Initiation and growth of a crack from stand-
ard specimen geometries such as compact tension (CT) specimens, would imply
several weeks, if not, months of testing, especially if the frequency is low (less
than 1 Hz).
Buitrago et al. [23] conducted fatigue tests using a frequency scanning
method on riser welds in a sour environment to (a) estimate the critical cyclic
frequency for conducting endurance and crack growth tests, and (b) to under-
stand the contribution of physio-chemical reactions and diffusion processes at
the crack tip. The objective of this frequency-scanning method was to deter-
mine the saturated frequency below which the corrosion fatigue-crack growth
rate (CFCGR) does not increase with decreasing frequency; the frequency
range considered for this study is 0.01–10 Hz. The authors found that 1 Hz can
be used instead of 0.33 Hz to characterize the effect of frequency on endurance
or crack growth tests on C–Mn steel in sour brine. Hudak et al. [24] conducted
the frequency-scanning experiments on high strength steels with different
yield strengths and observed that the slope of crack growth rate versus (a=W)
increases with an increase in yield strength. The crack growth rate at a specific
frequency is higher when the frequency was stepped up from Lo- (0.01 Hz)
to Hi- (10 Hz) frequency compared to the reverse condition (i.e., when the
frequency was dropped from 10 Hz to 0.01 Hz). This could be because of
the change in occluded crack chemistry, as the crack spends considerable
time at 0.01 Hz before the frequency is increased. However, at no frequencies
did the scan test display a steady crack growth rate, which implies that
the fatigue-crack growth was still dominated by mechanical loading. Further,
transients in crack growth rate behavior disturb the crack tip electro-
chemistry, so the real effects of slow corrosion on fatigue cycling is not
captured. This can be overcome, in case there is a slow decrement rate of fre-
quency as a function of crack length. This can also reveal the underlying mech-
anisms relating to change in crack growth kinetics because of the change in
frequency.
The purpose of this study is to understand the crack growth rate behavior
of a Mn-steel at constant DK through a novel frequency-shedding method at
very low frequencies. This method is similar to the load shedding test method
(ASTM E647 [25]) used for threshold fatigue-crack growth rate determination.

Experimental Methodology
Environment affects both the applied crack tip driving force and the resulting
crack growth rates [22]; we can define its effects on fatigue-crack growth in two
ways as shown in Fig. 3.
Path a: This helps to distinguish the damage contribution of chemical and
mechanical driving forces when the crack growth rate is constant. It is possi-
ble to find the deleterious effect of an environment in comparison to the
inert environment for a given material. It is either a displacement or strain
controlled fatigue experiment, which is very difficult, if not impossible, to con-
duct, as the crack opening displacement is dependent on the instantaneous
crack length.
Path b: Change in crack growth kinetics due to environment and loading
frequency can be understood for an applied mechanical driving force. It may be
noted that normally there is an additional crack length increment because of
the environmental interactions described earlier (such as hydrogen embrittle-
ment or anodic dissolution). However, a constant DK can be maintained
through a load controlled fatigue experiment, where the instantaneous load val-
ues are estimated based on the knowledge of instantaneous crack length.
To understand the influence of environment on the crack growth behavior
at a given stress ratio, the damage contribution from mechanical cycling (i.e.,
the stress intensity factor range) is kept constant. This ensures that the cyclic
plastic zone size is constant throughout the test; hence, one could expect con-
stant contribution from plasticity induced crack closure throughout the test
[26]. To accelerate fatigue-crack growth testing and to obtain corrosion fatigue-
crack growth rate data at low frequencies of the order of 0.01 Hz, the frequency
is shed exponentially as per Eq 1 from an initial test frequency of 5 Hz. It may
FIG. 3—Characterization of the environmental effects on crack growth rate using path
a–constant da=dN, and path b–constant DK.

be noted that many literatures indicate that the effect of corrosion fatigue is in-
significant at frequencies above 5 Hz [6].
The choice of exponential method to shed the frequency is based on the pre-
mise that exponential decay allows quick reduction in test frequency when the
corrosion effects are not significant, but ensures that the crack tip is exposed to
a given low frequency for a relatively long duration to capture the combined
effect of environment and stress field. This may appear to be arbitrary at the
first instance, but this was definitely a better method compared to random
sweeping of frequency ranges, or any linear shedding of frequency with crack
advance. Further, as the mechanical crack-driving force is maintained con-
stantly throughout the test, the effect of plasticity-induced crack closure is mini-
mized. It may also be noted that the maximum stress intensity factor is retained
constant with crack advance, all through the test, which ensures that the corro-
sion potential at the crack tip is also maintained constant.
The instantaneous test frequency was derived by using the expression:
f ¼ f0 ecða0 ai Þ (1)
where f ¼ instantaneous frequency, Hz,

f0 ¼ initial frequency, Hz,
C ¼ gradient, mm1,
a0 ¼ final crack length, mm, and
ai ¼ initial crack length, mm.
Figure 4 shows the plot of crack length versus frequency for a set of three
different exponential constants of 0.08, 0.16, and 0.32 mm1. It may be
FIG. 4—Frequency decrease as function of crack length with different exponential constants
of C ¼ 0.08 mm1, 0.16 mm1, and 0.32 mm1.

FIG. 5—Dimensions of C (T) specimen and clip on gauge extender.
noted that too shallow an exponential constant (i.e., 0.08 mm1) would imply
testing over large crack lengths (>40 mm) to reach the desired low frequency,
and, as a consequence, very high duration for experimentation; whereas, too
steep a frequency drop (exponential constant 0.32 mm1) would miss out on
the effect of corrosion on crack growth. Based on the exploratory experiments
carried out in lab air environment, it was decided to use 0.16 mm1 as an ex-
ponential constant for all corrosion experiments, as the crack growth rate was
stable with this exponential constant. It may be noted that more studies are
required before a final decision on the value of exponent for frequency shedding
is made, but, for the present set of materials and experimental conditions,
use of 0.16 mm1 as exponent seemed to provide the required answer to the
problem.
Experimental Setup and Test Procedure

A 100-kN MTS 810 servo-hydraulic test system integrated with a servo-
controller and data-acquisition system was used to perform the constant cyclic
stress-intensity tests. A compact tension (CT) specimen as per ASTM E647 [25]
made of Ni–Mn–Cr steel was used for crack growth rate tests (Fig. 5); the mate-
rial composition is given in Table 1. The specimen was tested using a continu-
ous supply of 3.5 % NaCl solution using a re-circulating pump and the crack tip
area was always kept in constant flow (due to capillarity action) with 3.5 %
NaCl solution by placing a cotton swab around the crack plane. Some research-
ers have proposed dipping a crack tip under the 3.5 %-NaCl solution while
studying the corrosion crack growth response, but as sealing the specimen and
ensuring that the salt water did not damage the test system was not easy; hence,
the present method of soaking the crack tip by cotton swab was resorted to. It
may be noted that the crack tip is always soaked with 3.5 % NaCl in view of the
capillarity action of cotton swab and to ensure that the pH vales are maintained
at 8.2 6 0.1 (as measured in a re-circulating pump tank); the solution was
TABLE 1—Material composition of Ni–Mn–Cr Steel.
Elements Mn Ni Cr Ti Fe
Composition (wt. %) 2.1 0.78 0.31 0.21 Balance

FIG. 6—Photograph of test setup used for corrosion fatigue experiments.
replaced after every 100 h of continuous usage. Figure 6 presents the overall test
setup used for corrosion fatigue experiments.
A crack opening displacement gage was used for crack length monitoring.
As conventional mounting of clip gage as per ASTM E647 [25] guidelines
is not feasible (due to spilling of NaCl solution in the strain gage region), a
special extender as shown in Fig. 5 for clip-on-gage mounting was designed.
Figure 7 presents the details of screw mounting of COD gage extender on the
CT specimen. As the COD gage mounting is a non-standard location as per
FIG. 7—Details of screw mounting of COD gage extender on the C (T) specimen.

ASTM E647 [25] standard, a crack length–compliance correlation was estab-

lished for this arrangement by conducting a few calibration crack growth rate
tests in air. During calibration experiments, crack length was first monitored
on the front and back faces by using a traveling microscope. Further, the crack
growth experiments were conducted using a two-step (Hi-Lo) block loading,
which retained the peak loads during cycling as a constant, to induce beach
marks on the fracture surfaces. The crack length as estimated by compliance
method was found to correlate with visual measurements and post-failure
beach marks within 63 %.
The crack length-compliance calibration data was obtained for the present
case of clip-on-gage mounting as:
a
¼ 0:5489 1010 U5 2:7176 1010 U4 þ 5:3812 1010 U3
w
5:3277 1010 U2 þ 2:6373 1010 U 5:222 1010 (2)
where
1
U¼ ;
BEV 1=2
P þ1
a ¼ crack length (in m),

W ¼ width of the specimen (in m),
E ¼ Young’s modulus (in MPa),
v ¼ crack opening displacement (in m), and
P ¼ applied load (in kN).
Prior to the start of frequency-shedding tests, constant amplitude fatigue-
crack growth rate tests were conducted on CT specimens at two different frequen-
cies of 5 Hz and 1 Hz under 3.5 % NaCl salt solution conditions [27]. Figure 8
presents the fatigue-crack growth rate data obtained from these experiments. The
combined action of mechanical and corrosion driving forces are observed to be
influencing fatigue-crack growth rates in the DK range of 12 to 30 MPa m1=2.
Based on this, subsequent frequency-shedding experiments were proposed to be
conducted at a constant DK range of 15 MPa m1=2 in lab air and at a DK range of
18 MPa m1=2 under 3.5 % NaCl solution to bring out the effect of environment on
corrosion fatigue-crack growth rates.
In case of experiments under lab air conditions, the starting frequency was
set at 5 Hz and the frequency was shed with crack increment. The crack length
was estimated using the crack length–compliance relation given by Eq 2. Initial
experiments under 3.5 % NaCl environment was started at frequency of 5 Hz.,
but as this increased the test duration to more than 40 days, subsequent experi-
ments were carried out with a starting frequency of 1 Hz., after pre-cracking the
specimen under 3.5 % NaCl solution environment. This reduced the test dura-
tion to about 25 days. In all the cases, crack growth was monitored at periodic
intervals of cycles. Initially, crack length was measured at every 2000 cycles,
which was later reduced to 1000, 500, and 100 cycles.

FIG. 8—Crack growth kinetics of Ni–Mn–Cr steel at cyclic loading frequencies of 5 Hz

and 1 Hz.
The load and crack opening displacement data were acquired during the
course of tests at a sampling rate of 1 kHz. These data were used to evaluate the
crack length using compliance technique as well as to estimate crack closure
load using the ASTM compliance offset method. A MATLAB program was used
to evaluate the crack opening loads at those cycles for 1 % and 2 % compliance
offsets.

Figure 9 presents the crack growth rate versus frequency response for fatigue-
crack growth rate test conducted under 3.5 % NaCl solution at a constant
DK (¼18 MPa m1=2) in the log–log scale. The crack growth rate was observed to
increase from an average value of 4 108 to 3 107 m=cycle, as the frequency
is shed from 5 Hz to 0.1 Hz. As noticed from this log–log graph, there is a linear
increase in crack growth rate when the frequency is shed from 5 Hz to 0.5 Hz.,
and thereafter a plateau in crack growth rate is observed for test frequency up
to 0.1 Hz. Linear increase in crack growth rate up to 0.5 Hz would simplify the
task of estimating crack growth rate at lower frequencies under corrosive envi-
ronments, as data generated at 5 Hz can be used to predict the growth rates at
0.5 Hz; this would save time and effort involved in generating the data at low
frequencies. This linear trend in crack growth rate as a function of frequency
requires careful examination for different materials and for different stress-
intensity factor ranges. As shown in later sections, indeed, we noticed a linear
increase in crack growth rate as a function of frequency for different materials
tested at different frequencies, when we compiled data from literature.
To verify, if the above proposal holds good for different materials, crack
growth rate versus frequency data was extracted from literature for constant

FIG. 9—Fatigue-crack growth rate as a function of loading frequency for a Ni–Mn–Cr

steel tested in 3.5 % NaCl solution by frequency shedding method.
amplitude tests conducted at a fixed frequency in corrosive environments and

the same is compiled and shown in Fig. 10. The frequency effect on fatigue-
crack growth rate in AISI 4340 steel in a water vapor environment (585 Pa) is
observed to be more pronounced at lower frequencies [10]. Further, an upward
shift in crack growth rate was observed as the DK values are increased for the
same material, which means higher DK results in higher crack growth rates.
The crack growth rate data of X65 steel tested in an aqueous 3.5 % NaCl
FIG. 10—Crack growth rate behavior of AISI 4340 steel in water vapor, X65 steel in
aqueous 3.5 % NaCl solution and Ti662 alloy in methanol and HCl solution based on
data compiled from Refs. 10, 28, and 29 respectively.

solution shows an increase in crack growth rate as the frequency is reduced

from 10 Hz to 0.01 Hz [28]. A similar trend of increasing crack growth rate with
decreasing frequency is observed in titanium alloy Ti662 tested in methanol-
HCl solution [29]. Figure 10 suggests that the data for a host of materials
compiled exhibit a log-linear increase of crack growth rate with frequency as
the frequency is reduced, which confirms our observation of a linear increase in
crack growth rate with reduction in test frequency up to 0.5 Hz.
Interestingly, the crack growth rate in the frequency range of 10 to 0.5 Hz,
in most materials, follows a linear trend, and the slopes are more or less similar.
This means that one could predict the crack growth behavior at lower frequen-
cies, if the data at test frequencies of 5 Hz is available. This result also suggests
the possibility of generating data at low frequencies for corrosion crack growth
based on the knowledge of crack growth rates at practicable test frequencies.
This will save time and material spent in conducting each test at a single
frequency.
Crack growth rate experiments were repeated at same applied DK values, but
for frequencies below 0.5. Figure 11 presents the crack growth rate data for
experiments carried out up to a frequency of 0.01 Hz. It is noticed that after a fre-
quency of 0.1 Hz, there is a sudden drop in crack growth rate, that is followed by
an increase in growth rate as the frequency is shed further. In the case of tests
conducted in air, a fall in crack growth rate was observed at much higher fre-
quencies of 1 Hz. The frequency at which there is a change in crack growth rate
could be an indication of transition from a combined mechanical corrosion-
dominated crack growth rate mechanism to a pure corrosion-dominated crack
growth rate mechanism. As the frequencies are so low, one could expect a slow
rate of straining under mechanical loading cycles, and, as a consequence, the
time available for corrosion products to interact with the crack tip is high, which
leads to acceleration in crack growth.
FIG. 11—Effect of frequency on crack growth kinetics for tests conducted in 3.5 %
NaCl solution and air.

During repeat tests at DK of 18 MPaHm, the initial frequency (f0) was set at
1 Hz. The initial crack growth rate at a given frequency for this experiment was
found to be less compared to the test initiated at 5 Hz frequency. This could be
due to a small crack wake and reduced crack opening displacements associated
with the lower crack lengths for the same DK. However, it was observed that as
the crack length increases, the growth rate tends to match with the experiment
conducted with f0 of 5 Hz.
Even though the crack growth experiments were conducted under condi-
tions of constant DK, which implies a steady state plastic zone ahead of the
crack tip, one could expect contribution of crack closure due to events that
occur in the crack wake. To examine, if the crack growth rates were affected by
crack closure, crack closure estimates were obtained as per the ASTM compli-
ance offset method proposed in the E647 [25] standard. Figure 12 presents the
crack growth rate data versus effective stress intensity factor range for experi-
ments conducted in lab air and 3.5 % NaCl solution. Crack closure estimates
obtained for 1 % compliance offset.
The effect of crack closure on crack growth kinetics during frequency-
shedding experiments in 3.5 % NaCl solutions with initial frequency of 5 Hz and
1 Hz and in lab air is shown in Figs. 13, 14, and 15, respectively. The figures also
present information on crack closure levels along the second y axis as indicated
by open symbols in the graph. Even though there is an increase in crack closure,
the crack growth rate is significantly increasing. This suggests that even at very
low frequencies, the crack continues to grow due to the combined action of cor-
rosive media and mechanical loading. This also indicates that at low effective
stress-intensity ranges, the crack growth rate is high. This implies that there is a
possibility of very low threshold stress-intensity factor range, when the material
is tested under corrosive medium at very low frequencies. So, the use of
FIG. 12—Crack growth kinetics for tests conducted in 3.5 % NaCl and air, after
accounting for crack closure estimated using compliance offset method.

FIG. 13—Effect of crack closure on crack growth rates for a Ni–Mn–Cr steel obtained by
frequency shedding method; environment: 3.5 % NaCl and starting frequency: 5 Hz.
threshold stress-intensity factor for corrosion crack growth needs to be carefully

considered. It may be noted that the crack closure estimates had some scatter
as the COD values are measured at a distance away from the crack tip. This as-
pect needs fine tuning to arrive at precise estimates of crack closure values. One
possibility is to use other techniques such as laser interferometer or back-face
strain to obtain crack closure information.
FIG. 14—Effect of crack closure on crack growth rate obtained by frequency shedding
method–lab air.

FIG. 15—Effect of crack closure on fatigue-crack growth rates obtained during fre-
quency shedding experiments in 3.5 % NaCl with initial frequency of 1 Hz.
Further experiments are in progress at different values of constant DK and

it is hoped that the results would further strengthen the knowledge of corrosion
crack growth rates at low frequencies. Even under the present conditions of fre-
quency shedding, it was observed that each crack growth rate almost required
about 4 to 5 weeks of testing on the test system.
Summary
The effect of frequency on the corrosion fatigue-crack growth rate of Ni–Mn–Cr
high strength steel in 3.5 % NaCl solution is studied using a frequency-shedding
method at a constant cyclic stress-intensity factor range. The crack growth rate
increases with reducing frequency and a linear trend is observed when the data
is plotted in a log–log scale. Compilation of available literature data on corro-
sion crack growth experiments suggests that a similar trend is observed in other
metals tested in corrosive environments. The frequency-shedding method
appears to be a promising method to generate corrosion fatigue-crack growth
rate data at low frequencies and results in time savings.
References
[1] Nikolin, E. S. and Karpenko, G. V., “Effect of Stress Reversal Frequency on the
Corrosion-Fatigue Strength of Notched Carbon Steel Specimens,” Mater. Sci., Vol.
2, 1967, pp. 128–129.
[2] McEvily, A. J. and Wei, R. P., Corrosion Fatigue: Chemistry, Mechanics and Micro-
structure, NACE-2, National Association of Corrosion Engineers, Houston, 1972,
pp. 381–395.

[3] Menan, F. and Hénaff, G., “Synergistic Action of Fatigue and Corrosion during Crack
Growth in the 2024 Aluminum Alloy,” Proc. Eng., Vol. 2, 2010, pp. 1441–1450.
[4] Sudarshan, T. S., Srivatsan, T. S., and Harvey, D. P., II, “Fatigue Processes in
Metals—Role of Aqueous Environments,” Eng. Fract. Mech., Vol. 36, No. 6, 1990,
pp. 827–852.
[5] Tkachev, V. I., “Some Aspects of Hydrogen Embrittlement of Steels,” Mater. Sci.,
Vol. 15, 1979, pp. 31–35.
[6] Wei, R. P., Fracture Mechanics—Integration of Mechanics, Material Science and
Chemistry, Cambridge University Press, Cambridge, 2010.
[7] Ritchie, R. O., “Mechanisms of Fatigue Crack Propagation in Metals, Ceramics and
Composites: Role of Crack Tip Shielding,” Mater. Sci. Eng., A, Vol. 103, No. 1, 1988,
pp. 15–28.
[8] Bartlett, M. L. and Hudak, S. J., Jr., “The Influence of Frequency-Dependent Crack
Closure on Corrosion Fatigue Crack Growth,” Fatigue’90, Honolulu, HI, 1990,
pp. 1783–1788.
[9] Hudak, S. J., Jr. and Page, R. A., “Analysis of Oxide Welding during Environment
Assisted Crack Growth,” Corrosion, Vol. 39, No. 7, 1983, pp. 285–290.
[10] Pao, P. S., Wei, W., and Wei, R. P., “Effect of Frequency on Fatigue Crack Growth
Response of AISI 4340 Steel in Water Vapor,” Environment-Sensitive Fracture of
Engineering Materials, Z. A. Foroulis, Ed., The Metallurgical Society of AIME, New
York, 1979, pp. 565–580.
[11] Moses, G. G. and Narasimha Rao, S., “Behavior of Marine Clay Subjected to Cyclic
Loading with Sustained Shear Stress,” Marine Geores. Geotechnol., Vol. 25, No. 2,
2007, pp. 81–96.
[12] En-Hou, H. and Ke, W., “Chemical and Electrochemical Conditions within Corro-
sion Fatigue Cracks,” Corros. Sci., Vol. 35, 1993, pp. 599–610.
[13] Taylor, M. E. and Barsom, J. M., “Effect of Cyclic Frequency on the Corrosion-
Fatigue Crack-Initiation Behavior of ASTM A517 Grade F Steel,” ASTM STP 743,
ASTM International, West Conshohocken, PA, 1981, pp. 599–622.
[14] Novak, S. R., “Influence of Cyclic-Stress Frequency and Stress Ratio on the
Corrosion-Fatigue Crack-Initiation Behavior of A588-A and A517-F Steels in Salt
Water,” Fifteenth National Symposium on Fracture Mechanics, University of Mary-
land, College Park, MD, 1982.
[15] Barsom, J. M., Sovak, J. F., and Imhof, E. J., “Corrosion-Fatigue Crack Propagation
below KISCC in Four High-Yield-Strength Steels,” Applied Research Laboratory
Report 89.021-024(3), U.S. Steel Corporation, Arlington, VA, 1970.
[16] Barsom, J. M., “Corrosion-Fatigue Crack Propagation below KISCC,” Eng. Fract.
Mech., Vol. 3, 1971, pp. 15–21.
[17] Barsom, J. M., “Effect of Cyclic-Stress Form on Corrosion Fatigue Crack Propaga-
tion below KISCC in a High-Yield-Strength Steel,” Corrosion Fatigue: Chemistry,
Mechanics and Microstructure, NACE-2, National Association of Corrosion Engi-
neers, Houston, 1972.
[18] Imhof, E. J. and Barsom, J. M., “Fatigue and Corrosion-Fatigue Crack Growth of
4340 Steel at Various Yield Strengths,” ASTM STP 536, ASTM International, West
Conshohocken, PA, 1973, pp. 182–205.
[19] Vasudevan, A. K., Sadananda, K., and Louat, N., “Two Critical Stress Intensities
for Threshold Fatigue Crack Propagation,” Scripta Metall. Mater., Vol. 28, 1993,
pp. 65–70.
[20] Sadananda, K. and Vasudevan, A. K., “Analysis of Fatigue Crack Closure and
Thresholds,” ASTM STP 1220, ASTM International, West Conshohocken, PA, 1995,
pp. 484–501.

[21] Lee, E. U. and Vasudevan, A. K., “Environmentally Influenced Fatigue in High

Strength Steels,” J. ASTM Int., Vol. 2, 2005, pp. 151–163.
[22] Vasudevan, A. K. and Sadananda, K., “Classification of Environmentally Assisted
Fatigue Crack Growth Behavior,” Int. J. Fatigue, Vol. 31, 2009, pp. 1696–1708.
[23] Buitrago, J., Weir, M. S., Kan, W. C., Hudak, S. J., Jr., and McMaster, F., “Effect of
Loading Frequency on Fatigue Performance of Risers in Sour Environment,” Inter-
national Conference on Offshore Mechanics and Arctic Engineering, OMAE 2004-
51641, Vancouver, 2004.
[24] Hudak, S. J., Jr., Feiger, J. H., and Patton, J. A., “The Effect of Cyclic Loading Fre-
quency on Corrosion-Fatigue Crack Growth in High-Strength Riser Materials,”
International Conference on Ocean, Offshore and Arctic Engineering, OMAE 2010-
20705, Shanghai, 2010.
[25] ASTM E647, 2010, “Standard Test Method for Measurement of Fatigue Crack
[26] Park, H.-B., Kim, K.-M., and Lee, B.-W., “Plastic Zone Size in Fatigue Cracking,”
Int. J. Pressure Vessels Piping, Vol. 68, 1996, pp. 279–285.
[27] Chinnaiah, M. and Prakash, R. V., “Corrosion Fatigue Crack Growth Studies in
Ni–Cr–Mn Steel,” Int. J. Mech. Mater. Eng., Vol. 1, No. 1, 2010, pp. 20–25.
[28] Vosikovsky, O., “Fatigue-Crack Growth in an X65 Line-Pipe Steel at Low Cyclic
Frequencies in Aqueous Environments,” ASME J. Eng. Mater. Technol., Vol. 97,
1975, pp. 298–304.
[29] Dawson, D. B., “Fatigue Crack Growth Behavior of Ti–6AI–6V–2Sn in Methanol
and Methanol–Water Solutions,” Metall. Trans. A, Vol. 12A, 1981, pp. 791–800.


doi:10.1520/JAI104187
G. P. Potirniche1
A Numerical Strip-Yield Model for the Creep

Crack Incubation in Steels
ABSTRACT: A numerical strip-yield model was developed to simulate creep

crack incubation in heat-resistant steels. The model is based on a formulation
proposed by Newman (Newman, J. C., Jr., “A Crack-Closure Model for Predicting
Fatigue Crack Growth under Aircraft Spectrum Loading,” Methods and Models
for Predicting Fatigue Crack Growth under Random Loading, ASTM STP 748,
J. B. Chang and C. M. Hudson, Eds., ASTM International, West Conshohocken,
PA, 1981, pp. 53–84) for fatigue crack growth under variable amplitude loading.
The time evolution of the plastic deformation ahead of a crack loaded in tension
is modeled using the Norton law for secondary creep stage, and the primary and
tertiary creep stages are neglected. The model assumes a pre-existing crack in a
specimen and models the behavior of the material prior to the beginning of crack
propagation due to creep loading. The evolution with time of the crack-tip plastic
zone, crack-tip opening displacement, and yield strength in the plastic zone are
computed at constant temperature for center crack panels. Comparison with
two previous strip-yield models and experimental data is performed, and good
correlation is obtained for several Cr-Mo-V steels. This approach to modeling
creep crack incubation has the potential to be applied to other types of cracked
specimens under constant or variable amplitude loading.
KEYWORDS: strip-yield model, creep, crack incubation, Cr-Mo-V steel
Introduction
Heat-resistant alloys are used extensively in coal-fired power plants, as gas tur-
bine materials for gas-fired power plants, or as reactor internals for nuclear
power plants. With ever increasing requirements imposed on structural materi-
als operating at high temperatures, there is a growing need to predict the service
Manuscript received July 5, 2011; accepted for publication December 1, 2011; published
online March 2012.
1
Mechanical Engineering Dept., Univ. of Idaho, P.O. Box 440902, Moscow, ID 83844, e-mail:
gabrielp@uidaho.edu
Cite as: Potirniche, G. P., “A Numerical Strip-Yield Model for the Creep Crack Incubation
in Steels,” J. ASTM Intl., Vol. 9, No. 3. doi:10.1520/JAI104187.
Copyright V
197

life and reliability of components experiencing creep fracture and creep-fatigue

damage. Components operating at high temperatures can develop cracks that
can incubate and grow under large local creep strains, even though the nominal
applied stresses or strains are low. The strip-yield model (SYM), as a numerical
method to simulate crack growth under constant or variable amplitude loading,
was proposed primarily for the modeling of fatigue crack growth in aerospace
alloys and is based on the Dugdale model [1]. Even though numerous SYMs
have been developed for fatigue crack growth, there are very few that have
focused on creep-fracture or creep-fatigue problems. Vitek [2] originally pro-
posed an analytical SYM based on the Bilby-Cotterell and Swinden model [3]
for the dislocation distribution at the crack tip in a center-crack panel of infinite
width. The incubation period of the crack embedded in the creeping material
was modeled by formulating the time evolution of the plastic zone development
and the density of edge dislocations in front of the crack tip. He also modeled
the evolution of the yield stress near the crack tip with time. Ewing [4] devel-
oped another analytical model for creep crack incubation and growth based on
the SYM approach. The material was assumed to undergo isotropic hardening,
and the Norton equation was used for the creep behavior in the secondary stage,
whereas the first and tertiary stages were neglected in his model. In order to pre-
dict the onset of crack growth, Ewing used a critical crack-tip opening displace-
ment (CTOD) criterion. In his work, Ewing demonstrated that the crack
incubation and failure time of a specimen can be correlated with the applied
stress intensity factor K. Ewing compared his model predictions with the experi-
mental data obtained by Batte [5] and Haigh [6,7]. Other researchers have simu-
lated creep crack incubation and growth, and they used either the finite element
method [8–11] or analytical formulations [12–15].
Experimental studies of crack nucleation and growth in heat resistant steels
are numerous. Haigh [6,7] performed creep crack growth tests under both sta-
tionary and variable loading on three different Cr-Mo-V alloys using wedge-
opening-load specimens. He measured crack opening displacements versus
time and correlated the rate of CTOD increase with the increment in crack
length. A thorough review of the creep behavior in Cr-Mo-V steels is presented
by Haigh in Ref 7. Haigh et al. [16] also studied the influence of oxidation on
creep crack growth during high cycle fatigue. They found that oxidation plays a
significant role in the creep crack growth, and the loading frequency is more
marked in air than in vacuum. Several other experimental studies were per-
formed on Cr-Mo-V alloys [17–20].
The failure process by creep crack growth in steels consists of crack incuba-
tion and growth. The goal of this paper is to apply a numerical SYM in order to
study the time evolution of the crack tip parameters during creep crack incuba-
tion in Cr-Mo-V steels. The model is based on the formulation of the crack open-
ing displacement and plastic zone at the crack tip in an isotropic elastic–perfectly
plastic material. For simplicity, the creep behavior is assumed to be described by
a Norton power law. The criterion used for the transition from a non-propagating
to a propagating crack is the critical CTOD, as used by several other authors
[2,4,19,21]. Model predictions are compared with results of two other SYMs
[2,4], and good agreement is obtained. Comparison with experimental data from

POTIRNICHE, doi:10.1520/JAI104187 199
the Haigh study [6] is performed, and good correlation of the evolution of the
crack opening displacement with time is also obtained.
Analytical Strip-Yield Model and Numerical Implementation

For a center crack of length 2a embedded in a panel of width 2b and loaded
under a constant tensile stress S, the resulting crack-tip plastic zone size is q, as
shown in Fig. 1. According to the Dugdale model [1], the elastic-plastic problem
can be solved by considering the superposition of two elastic solutions for the
embedded crack in a finite width panel, as shown in Fig. 2. Figure 2(a) shows
the specimen under the remotely applied load S. According to Tada et al. [22],
the mode I stress intensity factor KI for this configuration is
FIG. 1—Schematic of the strip-yield model for a finite width center-crack panel.

FIG. 2—Superposition principle for two elastic solutions using (a) the remotely applied
load and (b) the local stresses in the crack-tip plastic zone.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffi pd
KIS ¼ S pd sec (1)
2b
where d ¼ a þ q is the fictitious crack length and the secant term becomes one
for an infinite width plate. Figure 2(b) shows the same specimen loaded with
the flow stress r caused by the yielding in the crack-tip plastic zone for a ficti-
tious crack of length 2d. KI in this case is given by
2r pffiffiffiffiffiffip a
KIr ¼ pd sin1 Fða; dÞ (2)
p 2 d
where the geometrical factor F due to the finite width of the specimen is
2 !3
pa
p 1 sin 2b
62 sin 7rffiffiffiffiffiffiffiffiffiffiffiffiffiffi
6 sin pd
2b 7
Fða; dÞ ¼ 6 7 sec pd (3)
6 p 1 a 7 2b
4 sin 5
2 d
For an infinite width specimen, F(a,d) ¼ 1. As the superposition of the two stress
intensity factors is performed, the stress at x ¼ d must have a finite value; thus
the total stress intensity at this point should be zero.

KIS þ KIr ¼ 0 (4)
After solving for the plastic zone size from the above equation, the result is

2b pa pS
q¼ sin1 sin sin a (5)
p 2b 2r
where the stress r ¼ ar0 is the flow stress in the plastic zone adjusted by the con-
straint factor a to account for the stress state, with a ¼ 1 for plane stress and a ¼ 3
for plane strain. Because this is a fracture study, the load applied to the specimen
is a monotonic tensile stress, and the crack does not experience compressive
loads. However, future developments that will involve fatigue loads with alternat-
ing tensile-compressive loading cycles should take into account the fact that for
compressive loads the constraint factor is customarily chosen as a ¼ 1.
In order to formulate the numerical model for the center crack panel, the
crack opening displacements (CODs) must be computed at the maximum load-
ing. To calculate the CODs, a discretization into a certain number of elements
was performed for the entire crack plane, as illustrated in Fig. 3. The crack-tip
plastic zone was divided into 10 elements, numbered from j ¼ 1 to j ¼ 10. The
smallest element is placed at the crack tip (x ¼ a), and the width of the plastic
zone elements increases for elements located farther from the crack tip. The
widths of the plastic zone elements normalized with the plastic zone size
varied from 2w/q ¼ 0.01 (for x ¼ a) to 2w/q ¼ 0.3 (for x ¼ d). The element widths
in the plastic zone used in this study are the ones listed by Newman [23].
The physical crack plane is discretized with elements numbered from j ¼ 11
to j ¼ n.
FIG. 3—Meshing of the crack plane.

FIG. 4—Definition of crack-surface displacements.
For a generic element j located in the plastic zone as shown in Fig. 4,

the crack-surface displacement is Vj, and the element length is Lj. The element
has a width 2wj and is subjected to a compressive stress rj ¼ ar0. Then, for an
arbitrary element i located at x ¼ xi, the crack-surface displacement Vi caused by
the remotely applied stress S and the local stress rj acting on each element j can
be written according to [22,23]
X
n
Vi ¼ Sf ðxi Þ þ rj gðxi ; xj Þ (6)
j¼1
where the functions f and g for a center crack in a finite width panel are defined
as follows:
1=2
2 2 2 pd
f ðxi Þ ¼ ðd xi Þsec (7)
E0 2b
gðxi ; xj Þ ¼ Gðxi ; xj Þ þ Gðxi ; xj Þ (8)
and

2rj d2 b2 xi d2 b1 x i
Gðxi ; xj Þ ¼ 0
ðb2 xi Þcosh1 ðb1 xi Þcosh1
pE djxi b2 j djxi b1 j
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b 2 b 1
þ sin1 sin1 d2 x2i Fðb1 ; b2 ; dÞ (9)
d d
The geometrical factor F(b1, b2, d) for a finite width panel is

2 ! !3
1 sin pb
2b
2
1 sin pb
2b
1
6sin sin 7rffiffiffiffiffiffiffiffiffiffiffiffiffiffi

6 sin pd sin pd
2b 7
Fðb1 ; b2 ; dÞ ¼ 6 2b 7 sec pd (10)
6 b2 b1 7 2b
4 sin1 sin1 5
d d
where:
b2 ¼ xj þ wj , and
b1 ¼ xj wj .
For an infinite width plate, F(b1, b2, d) ¼ 1, and
8
<E for plane stress
E0 ¼ E (12)
: for plane strain
1 t2
where:
E ¼ modulus of elasticity, and
t ¼ Poisson ratio.
Ewing [4] introduced the idea that creep deformation at the crack tip is
experienced by an element of a certain length (k) that can be considered a mate-
rial constant. For instance, the length of the crack tip element k can be related
to the grain size of the material; thus it is independent of the crack length, or it
can be assumed to be related to the plastic zone size, in which case it is depend-
ent on the crack length. Here it is assumed that the length of the crack tip ele-
ment is a material constant independent of the crack length. Assuming, for
brevity, that the CTOD is denoted by /, from the discretization of the crack
plane explained above it follows that
V10 ¼ / (13)
assuming for the creep behavior of the material a power law of the form
r m
0
e_ s ¼ A (14)
G
where:
es ¼ creep strain for the secondary creep stage,
A and m ¼ material constants,
G ¼ shear modulus, and
r0 ¼ flow stress loading the crack-tip element.
Then, the time rate of change of the CTOD becomes
/_ ¼ e_ s k (15)
In this study, the external load S is constant; thus the crack is always open, as
opposed to in a fatigue loading process, in which the crack might experience some
closure for a portion of the loading cycle. It follows that the contribution to the

crack surface displacements from the local stresses rj is given only by the elements
in the plastic zone j ¼ 1,2,…,10, which results in the following equation:
r0 X10
/ ¼ Sf ðaÞ 0
gða; xj Þ (16)
pE j¼1
Two other equations can be derived as

1 pS
d ¼ sec sin (17)
2r0
r m
0
/_ ¼ kA (18)
G
where Eq 17 results directly from Eq 5, and Eq 18 is obtained from Eqs 13–15.

The system of Eqs 16–18, with /, r0, and d as unknowns, was integrated
using an explicit procedure. Assuming that the variables of interest are known
at time k, for a time step Dt ¼ tk þ 1 tk, the updated variables at time k þ 1 are
r m
/_ ¼ kA 0k (19)
G
/kþ1 ¼ /k þ /_ Dt (20)
Sf k ðaÞ /kþ1
r0kþ1 ¼ (21)
R10
j¼1 gk ða; xj Þ

1 pS
dkþ1 ¼ sec sin (22)
2r0kþ1
fkþ1 ¼ f ða; dkþ1 Þ (23)
gkþ1 ¼ gða; dkþ1 Þ (24)
Because the numerical implementation used an explicit scheme, the influence of

the time step on the model results was studied, and the results are shown in Fig. 5.
Several time steps were considered in the analysis, ranging from Dt ¼ 103 h to
Dt ¼ 10 h. The crack opening displacement versus time was plotted, and the results
were almost identical for all time steps considered. Similar results were also
obtained for the plastic zone size and relaxation of the yield strength with time.
Results
This section presents a comparison between the present model’s predictions
and the results from other SYMs, as well as comparisons with experimental

FIG. 5—Influence of time step size on the evolution of the crack-tip opening displacement.
data. Typical evolutions of the CTOD, q, and r0 with time are illustrated in
Fig. 6. For this simulation, the specimen is assumed to be in plane stress, the
crack length is a ¼ 1 mm, and the flow stress is r0 ¼ 600 MPa. The constants in
the creep law are A ¼ 21 h 1 and m ¼ 3, with k ¼ 100 lm. Also, the assumption is
made that the flow stress is the same for the entire crack-tip plastic zone. It can
be observed that whereas the flow stress in the plastic zone r0 decreases from
600 MPa to 265 MPa, the CTOD increases from 1.8 lm to approximately 6 lm,
and the plastic zone size q increases from 1.2 mm to 3.9 mm.
Table 1 shows a comparison between the model results and the experimen-
tal data of creep crack incubation obtained by Batte [5]. The material used in
the experiments was a bainitic Cr-Mo-V rotor steel, and the creep tests were per-
formed at 550 C. The detailed alloy concentration, heat treatment, and proper-
ties are given in the paper by Ewing. Batte used a double-edge notched
specimen. However, Ewing considered in his model the case of a center crack in
an infinite plane in plane stress; thus he used the stress intensity factor
K ¼ rHpa, instead of K ¼ 1.12rHpa. He also did not correct for the finite edges of
the specimens. Ewing performed simulations using the Vitek model and used
the same configuration of a crack in an infinite plane. In this study, the same
crack case as the one used by Ewing was considered, because the goal was to
compare the predictions of this model with those of the other two SYMs. Only
the secondary creep strain effects are taken into consideration, and the creep
law is presented in Eq 14. The creep law constants for this material, as specified
by Ewing, are A ¼ 21 h1 and m ¼ 3. The shear modulus is G ¼ 56.2 GPa, and
k ¼ 100 lm was kept constant for all simulations. Batte tested several specimens
with different crack lengths and applied stress values, as listed in Table 1. The
incubation period ti is defined from experiments, and the three models predict
the critical CTOD at the end of ti. The average critical CTOD predicted with the
Vitek model is 1.96 lm, whereas the prediction of the Ewing model is 1.36 lm,

FIG. 6—Variation of the main crack-tip parameters with time during a typical creep
process: (a) crack-tip opening displacement, (b) plastic zone size, and (c) yield strength
at the crack tip.

TABLE 1—Creep crack incubation results for a bainitic rotor steel. Comparison between
experimental results (Batte [5]) and the predictions of the present model and the strip-yield
models of Vitek [2] and Ewing [4].
Experiment (Batte) Vitek model Ewing model Present model
Applied Crack Yield Yield Yield

Stress S, length a, Incubation stress CTOD, stress CTOD, stress CTOD,
MPa mm time ti, h roi /G lm roi /G lm roi/G lm
221.9 1.0 50 2.50 1.67 3.38 1.18 3.08 1.54

193.8 1.3 40 2.79 1.56 3.81 1.11 3.61 1.40
177.0 1.6 80 3.10 1.89 4.25 1.34 3.63 1.65
148.9 2.2 70 3.55 1.78 4.89 1.27 4.37 1.58
122.5 3.2 150 4.21 2.11 5.82 1.52 4.66 1.75
177.0 1.0 200 2.07 1.66 2.75 1.20 4.03 0.95
132.0 6.0 40 8.35 2.7 11.7 1.33 5.61 2.90
132.0 4.0 75 5.61 2.36 7.80 1.94 5.11 2.10
Average 4.02 1.96 5.55 1.36 4.26 1.73
and the present model predicts a value of 1.73 lm. Even though there is a con-
sistent difference between the predictions of the three models for different
specimens, the present model can be considered satisfactory regarding the criti-
cal CTOD data. A comparison of the yield strength normalized by G at the end
of ti is also performed. The models predict values that do not deviate much from
each other, i.e., 4.02 (Vitek), 5.55 (Ewing), and 4.26 (present model).
Next, a comparison with the experimental data obtained by Haigh [6] was
performed for three Cr-Mo-V steels used in turbine casings. Two of the tested
materials (material 1 and material 3) were 1 % Cr-Mo-V with different heat
treatments, and the third was (1/2)Cr-Mo-V (material 2). Details regarding their
exact composition, heat treatment, and properties are given in the Haigh paper
[6]. All creep tests were performed at 550 C. The tensile strength of these mate-
rials is given in Table 2, along with the constants used for the creep law in Eq
14. The creep constants were obtained by curve-fitting the creep strain versus
time data presented by Haigh. The curve-fits are shown in Fig. 7. Ewing pre-
sented a comparison of his model, the Vitek model, and the experimental results
of Haigh only for material 1. In this paper, the comparison is extended to mate-
rials 2 and 3. In the present model, the ultimate tensile strength (UTS) was
TABLE 2—Creep law constants and ultimate tensile strengths for the three materials in the
Haigh [6,7] experiments.
Material A, h1 m UTS at 550 C, MPa
1 % Cr-Mo-V(1) at 550 C 1.4 1022 10.4 375

1/2 % Cr-Mo-V(2) at 550 C 1.4 1022 10.3 330
1 % Cr-Mo-V(1) at 550 C 1.4 1022 11.3 453

FIG. 7—Curve-fit to obtain the creep power law constants for (a) material 1, 1 % Cr-Mo-V
[1]; (b) material 2, 1/2 % Cr-Mo-V; and (c) material 3, 1 % Cr-Mo-V [2].

TABLE 3—Comparison between experimental results and model predictions for material
1 (1 % Cr-Mo-V) for a constant k ¼ 35 lm.
Experiment Present Vitek Ewing

k ¼ 35 lm (Haigh [6,7]) model model model
Yield
Applied Crack Incubation Measured stress Yield Yield
stress S, length a, time ti, CTOD, CTOD, roi, CTOD, stress roi, CTOD, stress roi,
MPa mm h lm lm MPa lm MPa lm MPa
172.5 32.0 750 230 144.0 209 103 232.0 90.8 246.3
181.5 35.0 180 270 152.7 229 102 263.9 91.3 282.6
221.9 34.4 14 180 284.0 241 110 362.9 100.4 392.2
considered equal to the flow stress, as no information on hardening was pre-

sented by Haigh. Moreover, the Ewing model also considered the flow stress as
the UTS. Thus, for consistency, the same assumption was made in this study.
The same values for the constant k were chosen as in the Ewing study, i.e.,
k ¼ 35 lm and k ¼ 350 lm. For consistency, the specimens were center-crack
panels of infinite width in plane stress, similar to the ones analyzed by Ewing.
Tables 3 indicate the applied stress and crack length for each of the materials
analyzed. Table 3 shows the results for material 1, and the experimental results
are compared with the predictions of the three SYMs. The incubation times, as
measured from experiments, are given for each stress and crack length case.
Given a value for ti, the SYMs predict the critical CTOD. The CTOD was also
measured by Haigh and is presented in each table. Tables 3 and4 show that for
both k ¼ 35 lm and k ¼ 350 lm, the present model consistently calculates larger
critical CTODs than the models of Ewing and Vitek. The predictions of the pres-
ent model are also closer to the experimental values measured by Haigh. The
constant k ¼ 350 lm gives CTOD values closer to the experimental data than the
values predicted using k ¼ 35 lm. The predicted flow stress at the end of the
incubation period r0i is also compared for the three models. From Tables 3 and 4,
1 (1 % Cr-Mo-V) for a constant k ¼ 350 lm.
Experiment Present Vitek Ewing

k ¼ 350 lm (Haigh [6,7]) model model model
Applied Crack Incubation Measured Yield Yield Yield

stress S, length a, time ti, CTOD, CTOD, stress roi, CTOD stress roi, CTOD, stress
MPa mm h lm lm MPa (lm) MPa lm roi, MPa
172.5 32.0 750 230 223 185 137 203.4 90.8 211.0
181.5 35.0 180 270 206 205 140 226.1 91.3 236.5
221.9 34.4 14 180 298 238 149 302.1 100.4 299.2

2 (1/2 % Cr-Mo-V).
Present model
Experiment
(Haigh [6,7]) k ¼ 35 lm k ¼ 350 lm
Applied Crack Measured Yield Yield

stress S, length a, Incubation CTOD, CTOD, stress r0i, stress r0i,
MPa mm Time ti, h lm lm MPa CTOD, lm MPa
131.7 35 550 105 90 185 124 158

147.9 35 250 125 113 194 142 176
181.5 35 130 120 245 196 267 192
133.7 41 30 90 98 196 104 190
124.4 42 18 200 84 196 87 192
it can be observed that the present model predicts r0i values similar to those
predicted by the Ewing and Vitek models for lower applied stresses of S ¼ 172
and S ¼ 181 MPa, whereas for the large applied stress of S ¼ 222 MPa the pres-
ent model predicts a significantly lower value of r0i than the other two SYMs.
Table 5 shows the comparison between the present model predictions and the
experimental data of Haigh for material 2. The experiments were performed on
specimens with several crack lengths ranging between a ¼ 35 mm and a ¼ 42 mm,
and stress levels between S ¼ 131 MPa and S ¼ 181 MPa. In general, the present
model predicts the critical CTOD reasonably well. The larger discrepancies
between experiment and model are recorded for the case of a large applied stress
of S ¼ 181 MPa or for the case of a large crack length of a ¼ 42 mm. Table 6 shows
the comparison between the data from Haigh’s experiments and those of the pres-
ent model for material 3, considering k ¼ 35 lm and k ¼ 350 lm. In general, the
model reasonably predicts the value of the critical CTOD given the incubation
3 (1 % Cr-Mo-V).
Present model
Experiment
(Haigh [6,7]) k ¼ 35 lm k ¼ 350 lm
Applied Crack Measured Yield Yield

stress S, length a, Incubation CTOD, CTOD, stress r0i, CTOD, stress r0i,
MPa mm time ti, h lm lm MPa lm MPa
181.52 35 40 30 69 379 86 320

134.46 35 200 20 to 30 44 318 55 266
155.97 42 4 20 to 50 53 423 60 373

times ti. Some of the discrepancies may probably be attributed to experimental

uncertainties in accurately measuring the critical CTOD. As can be observed from
Tables 3, there is a significant decrease in the flow stress in the plastic zone; thus
it can be assumed that large scale yielding conditions prevail during the creep
crack incubation and at the onset of the crack growth phase. The accumulation of
creep damage around the crack tip and the large increase in the plastic zone, com-
bined with a significant decrease in the flow stress, indicates that the critical
CTOD is a suitable measure that can be used to predict the initiation time ti.
A comparison between experimental and modeling results for the evolution
of the CTOD during the incubation time for material 1 is shown in Fig. 8. In this
case, the crack length was a ¼ 32 mm, and the applied stress S ¼ 172 MPa. From
this comparison, it can be observed that for the large majority of the incubation
time (about 800 h), the model predicts well the evolution of the CTOD; however,
there is a discrepancy between the experimental and modeling results for the first
100 h. The slight mismatch toward the end of the CTOD curve is due to the fact
that after t ¼ 750 h, in the experiment the crack starts growing, marking the end
of the incubation time. This phenomenon is not captured in the present model.
Vitek also presented the variation of the CTOD with time for the same material,
crack length, and applied stress, but he did not compare his model results with
the experiment of Haigh. Vitek used a power exponent of m ¼ 10 in his model.
Figure 9 shows the comparison of the CTOD evolution with time for the present
model and the Vitek model. The present model shows a slight discrepancy with
regard to the Vitek data when a creep power exponent of m ¼ 10 is used; however,
the two models show excellent agreement when a value of m ¼ 10.75 is used in
the present model. Given the scatter in the experimental data and the fact that
the value of m ¼ 10.75 does not produce a significant difference in the power law
curve-fit compared with m ¼ 10 or m ¼ 10.4, as illustrated in Fig. 7(a), the com-
parison between the two models can be considered adequate.
FIG. 8—Comparison between the experimental results and modeling predictions for the
evolution of the crack-tip opening displacement for 1 % Cr-Mo-V steel tested at 550 C.

FIG. 9—Comparison between the present model and the Vitek model [2] prediction for
the evolution of the crack tip opening displacement for 1 % Cr-Mo-V steel tested at 550 C.
Conclusions
A strip-yield model was adapted for the computation of creep crack incubation
periods in cracked components. The method uses a critical crack-tip opening
displacement (CTOD) approach, but other methods can be easily implemented
in the model, such as K-based or C*-based laws for crack incubation. Given
the large scale yielding condition prevailing at the crack tip during creep de-
formation, the critical CTOD was used as a crack incubation criterion. The
creep behavior accounts only for secondary stage effects, and it neglects the
primary and tertiary stages. Future developments should consider these
effects as well for a more realistic simulation of the creep crack incubation
and growth processes. This methodology is versatile and can easily be adapted
to other crack geometries. Moreover, this approach can be modified to include
more detailed models for the creep deformation and damage in the plastic
zone, thereby resulting in more realistic models for controlling the crack incu-
bation or crack growth processes. The model predicts the time-dependent vari-
ation of the CTOD, crack-tip plastic zone sizes, and flow stress in the plastic
zone. In general, the model presented in this paper gives crack incubation
times in agreement with the predictions provided by two other strip-yield
models. Also, the model results compare well to experimental data for critical
CTOD.
Acknowledgments
The writer would like to gratefully acknowledge the Nuclear Engineering Uni-
versity Partnership (NEUP) Program for providing the financial support for this
research through DOE Grant No. 42246, release 59.

References
[1] Dugdale, D. S., “Yielding of Steel Sheets Containing Slits,” J. Mech. Phys. Solids,
Vol. 8, 1960, pp. 100–104.
[2] Vitek, V., “A Theory of the Initiation of Creep Crack Growth,” Int. J. Fract., Vol 13,
1977, pp. 39–50.
[3] Bilby, B. A., Cottrell, A. H., Smith, E., and Swinden, K. H., “Plastic Yielding from
Sharp Notches,” Proc. R. Soc. London, Ser. A, Vol. 279, 1964, pp. 1–9.
[4] Ewing, D. J. F., “Strip Yield Models of Creep Crack Incubation,” Int. J. Fract., Vol.
14, 1978, pp. 101–117.
[5] Batte, A. D., “The Initiation and Growth of Creep Cracks in 1%CrMoV Rotor Forg-
ing Material,” private communication to D. J. F. Ewing, 1975.
[6] Haigh, J. R., “The Mechanisms of Macroscopic High Temperature Crack Growth.
Part I: Experiments on Tempered Cr-Mo-V Steels,” Mater. Sci. Eng., Vol. 20, 1975,
pp. 213–223.
[7] Haigh, J. R., “The Mechanisms of Macroscopic High Temperature Crack Growth.
Part II: Reviews and Re-analysis of Previous Work,” Mater. Sci. Eng., Vol. 20, 1975,
pp. 225–235.
[8] Tvergaard, V., “Analysis of Creep Crack Growth by Grain Boundary Cavitation,”
Int. J. Fract., Vol. 31, 1986, pp. 183–209.
[9] Tvergaard, V., “Effect of Microstructure Degradation on Creep Crack Growth,” Int.
J. Fract., Vol. 42, 1990, pp. 145–155.
[10] Davies, C. M., O’Dowd, N. P., Nikbin, K. M., and Webster, G. A., “An Analytical and
Computational Study of Crack Initiation under Transient Creep Conditions,” Int. J.
Solids Struct., Vol. 44, 2007, pp. 1823–1843.
[11] Yatomi, M., O’Dowd, N. P., Nikbin, K. M., and Webster, G. A., “Theoretical and Nu-
merical Modelling of Creep Crack Growth in a Carbon-Manganese Steel,” Eng.
Fract. Mech., Vol. 73, 2006, pp. 1158–1175.
[12] Nikbin, K. M., Smith, D. J., and Webster, G. A., “Prediction of Creep Crack Growth
from Uniaxial Creep Data,” Proc. R. Soc. London, Ser. A, Vol. 396, 1984, pp.
183–197.
[13] Nishida, K., Nikbin, K. M., and Webster, G. A., “Influence of Net Section Damage
on Creep Crack Growth,” J. Strain Anal. Eng. Des., Vol. 24, 1989, pp. 75–82.
[14] Smith, D. J., and Webster, G. A., “Estimates of the C* Parameter for Crack Growth
in Creeping Materials,” Elastic-Plastic Fracture: Second Symposium, Volume I:
Inelastic Crack Analysis, ASTM STP 803, C. F. Shih and J. P. Gudas, Eds., ASTM
International, West Conshohocken, PA, 1983, pp. 654–674.
[15] Wasmer, K., Nikbin, K. M., and Webster, G. A., “Influence of Reference Stress For-
mulae on Creep and Creep-Fatigue Crack Initiation and Growth Prediction in Plate
Components,” Int. J.Pressure Vessels Piping, Vol. 87, 2010, pp. 447–456.
[16] Haigh, J. R., Skelton, R. P., and Richards, C. E., “Oxidation-Assisted Crack Growth
during High Cycle Fatigue of a 1%Cr-Mo-V Steel at 550,” Mater. Sci. Eng., Vol. 26,
1976, pp. 167–174.
[17] Siverns, M. J., and Price, A. T., “Crack Propagation under Creep Conditions in a
Quenched 2 1/4 Chromium 1 Molybdenum Steel,” Int. J. Fract., Vol. 9, 1973, pp.
199–207.
[18] Barker, E., Lloyd, G. J., and Pilkington, R., “Creep Fracture of a 9 Cr-1Mo Steel,”
Mater. Sci. Eng., Vol. 84, 1986, pp. 49–64.
[19] Piques, R., Molinie, E., and Pineau, A., “Comparison between Two Assessment
Methods for Defects in the Creep Range,” Fatigue Fract. Eng. Mater. Struct., Vol. 14,
1991, pp. 871–885.

[20] Piques, R., Molinie, E., and Pineau, A., “Creep and Creep-Fatigue Cracking Behav-
iour of Two Structural Steels,” Nucl. Eng. Des., Vol. 153, 1995, pp. 223–233.
[21] Ainsworth, R. A., “The Initiation of Creep Crack Growth,” J. Solids Struct., Vol. 18,
1982, pp. 873–881.
[22] Tada, H., Paris, P. C., and Irwin, G. R., The Stress Analysis of Cracks Handbook,
ASME, New York, 2000.
[23] Newman, J. C., Jr., “A Crack-Closure Model for Predicting Fatigue Crack Growth
under Aircraft Spectrum Loading,” Methods and Models for Predicting Fatigue Crack
Growth under Random Loading, ASTM STP 748, J. B. Chang and C. M. Hudson,
Eds., ASTM International, West Conshohocken, PA, 1981, pp. 53–84.


doi:10.1520/JAI103945
F. Laengler,1 T. Mao,2 and A. Scholz2
Influence Analysis of Application-Specific

Phenomena on the Creep-Fatigue Life
of Turbine Housings of Turbochargers
ABSTRACT: The turbine housing of a turbocharger is exposed to extensive

cyclic thermo-mechanical loading. This leads to multiaxial stress states with
local plastifications, so that the design of the turbine housing becomes a major
challenge in ensuring the guaranteed lifetime in relation to the high-temperature
behavior of the materials. In a first step, a phenomenological lifetime approach
in conjunction with a constitutive material model applied in a preceding finite-
element analysis was developed and validated for application on the casting
materials Ni-resist D5S and vermicular cast iron GJV. The present study deals
with the adaption for turbine housing design together with the more detailed
analysis of application-specific phenomena to improve the description of
both the deformation behavior and the creep-fatigue damage behavior. The
influence of different strain rates, mean strain conditions, and aging has been
evaluated. Moreover, a critical plane approach has been investigated to handle
multiaxial stress and strain states. A more accurate damage ratio is derived
by use of specimens subjected to characteristic thermo-mechanical load
conditions, which leads to an improved estimation of the cycle number until
crack initiation on critical component positions.
Manuscript received May 5, 2011; accepted for publication December 13, 2011; published
online March 2012.
1
BorgWarner Turbo Systems Engineering GmbH, D-67292 Kirchheimbolanden, Germany
(Corresponding author), e-mail: flaengler@borgwarner.com
2
Institute for Materials Science, Technical Univ. of Darmstadt, D-64283 Darmstadt,
Germany.
in Anaheim, CA.
Cite as: Laengler, F., Mao, T. and Scholz, A., “Influence Analysis of Application-Specific
Phenomena on the Creep-Fatigue Life of Turbine Housings of Turbochargers,” J. ASTM
Intl., Vol. 9, No. 3. doi:10.1520/JAI103945.
Copyright V
215

KEYWORDS: turbine housing, finite-element analysis, elastoplasticity,

thermo-mechanical fatigue, creep-fatigue life assessment, Ni-resist D5S
Nomenclature
a¼ elasticity tensor
C¼ Larson-Miller constant
Ci, ci ¼ material parameter (kinematic hardening)
c1, c2 ¼ fitting parameter
DtA ¼ creep-fatigue damage/cycle
de ¼ strain rate tensor
E¼ Young’s modulus
f¼ Mises yield function
K, n, m ¼ material parameter (creep)
k¼ yield stress at p ¼ 0
L¼ creep-fatigue damage/life
LA ¼ fatigue damage/life fraction
Lt ¼ creep damage/life fraction
N0f ¼ mean value of fatigue life
Nf ¼ cycle number until crack initiation
n¼ normal vector
PLM ¼ Larson-Miller parameter
p¼ equivalent plastic strain
T¼ temperature
tu ¼ rupture time
X¼ backstress tensor
ath ¼ linear thermal expansion coefficient
De ¼ strain width range
Dec ¼ creep strain increment
Dt ¼ time increment
e¼ strain t time
e¼ strain tensor
ec ¼ creep strain
r¼ stress tensor
r¼ stress
()0 ¼ deviatoric tensor part
()e ¼ elastic part
()j ¼ 1-dimensional value at time step j
()(n) ¼ normal direction
()p ¼ plastic part
Introduction
The most cost-intensive key component of a turbocharger is the turbine housing
(T/H) that provides the kinetic energy required for charging. Inhomogeneous
temperature distributions and the interaction with neighboring components
constrain the thermal expansion and contraction of the T/H, thus causing local

LAENGLER ET AL., doi:10.1520/JAI103945 217
multiaxial stresses and inelastic strains during operation. Numerous test stand
runs are generally needed in order to find the appropriate combination of the
complex design and material. Hence, there is a demand for reliable calculation
methods allowing lifetime assessment with respect to thermo-mechanical fa-
tigue (TMF) early in the design process. Such TMF methods are essential, for
instance, to employ the full potential of materials and to reduce the effort and
expense for component testing by better understanding the cyclic mechanical
long-term behavior at elevated temperatures. Improvements in this respect
could be shown in terms of coupling the numerical component structural analy-
sis with a validated phenomenological lifetime estimation approach based on
creep-fatigue damage calculation. The validation of the TMF lifetime approach
until crack initiation as a post-processing step is performed on casting material
of type Ni-resist D5S and vermicular cast iron GJV [1]. The chemical composi-
tions of both materials are given in Table 1. The lifetime calculation is depend-
ing on results of a preceding finite-element analysis (FEA).
A Chaboche-type constitutive material model applied in the anisothermal
FEA describes the rate-independent elastoplastic material behavior using a ki-
nematic hardening law. Both TMF tests on specimens subject to characteristic
load conditions and thermal shock tests on turbine housings have been con-
ducted for identification and verification purposes to adapt the lifetime
approach for T/H design [2]. The scatter band of the estimated number of cycles
until crack initiation compared with experimental results on various critical
T/H positions has clarified the requirement for more experimental and analyti-
cal work. Therefore, the influence of strain rate, mean strain, and aging has
been investigated as application-specific phenomena, which could occur due to
arbitrary operation conditions. Results are used to more accurately describe
both the deformation and the damage behavior with respect to improving the
estimation quality within the application process.
Component Structural Analysis
Modus Operandi
The modus operandi of the T/H design evaluation process is shown in Fig. 1. As
a result of the variety of different T/H materials used in passenger car and com-
mercial diesel applications, the cost-benefit ratio is an important requirement
and has to be allowed for. Consequently, an efficient level of effort taken in both
determining required material-dependent parameters and numerical calcula-
tion of entire turbocharger models, including computational fluid dynamics
TABLE 1—Chemical composition of D5S and GJV.
Chemical element, % by mass
Sample C Si Mn S P Ni Cr Cu Mo Ti
D5S 2.4 1.5–3.0 0.5–1.0 0.02 0.08 34.0–36.0 2.0–3.0 – – –

GJV 2.8–3.4 4.6–5.0 0.3 0.02 0.07 0.5–0.7 – – 0.6–0.8 0.2

FIG. 1—Modus operandi applied for the evaluation of the T/H design with respect to
different types of materials.
(CFD) or rather conjugate heat transfer (CHT) and heat transfer calculation [3],
is essential to keep the evaluation time as flexible as necessary. Experimental
results of thermal shock tested T/H serve as the reference to assess the calcu-
lated thermo-mechanical T/H behavior under tightened load conditions for
accelerated testing.
The constitutive material model applied in the FEA is limited to rate-
independent plasticity due to the balance between effort and design or rather
the lifing quality, as mentioned earlier. Time-dependent phenomena such as
creep strain and stress relaxation, which are essential to calculate creep dam-
age, are subsequently recalculated within the lifetime estimation approach.
The results of the FEA in terms of time series are in general fully 3 dimen-
sional. But the lifetime approach afterwards requires 1-dimensional input quan-
tities. Therefore, it is inevitable to evaluate an adequate reduction method to
handle the multiaxiality by considering time-variable principal directions. The
equivalent 1-dimensional input variables of 3-dimensional tensors have to be
defined in such a manner that the lifetime estimations are of acceptable accu-
racy. Among other evaluated criteria, the critical plane approach defined by
maximum normal stress (see the section on Lifetime Assessment of T/H) fulfills
this requirement for the casting materials investigated in a first step. The nor-
mal vector of the plane, on which the maximum normal stress component
occurs, is used to reduce the tensor order by applying the double dot product
with the normal vector (see Eq 13).
Material Model
The adapted, constitutive material model applied in the FEA is based on the
work of Chaboche [4], [5] and describes rate-independent elastoplasticity while
elastic strains remain always infinitesimal. The temperature-dependent formu-
lation of the model is implemented within the commercial FEA software code
ABAQUS [6]. Phenomena typical for materials subject to cyclic loading, in which
plastic strain does continuously reverse direction sharply, can be modeled, as is
typically the case in the turbine housing due to the alternating heating-up and
cooling-down phases.
The incremental mechanical equations of the material model are given in
Table 2. Mechanical strain rate tensor de in Eq 1 is a linear combination of
the reversible elastic part dee and the irreversible inelastic part. The latter is

TABLE 2—Constitutive mechanical part of the elastoplastic material model.
Equation Number
de ¼ dee þ dep (1)

r ¼ a : ee qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (2)
@f 2
dep ¼ dp @r ; dp ¼ 3 dep : dep (3)
f ¼ Jðr XÞqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k¼0 ffi (4)
3 0
Jðr XÞ ¼ 2 ðr X0 Þ : ðr0 X0 Þ (5)
P
2 (6)
dX i ¼ Ci 1k ðr XÞdp ci X i dp þ C1i @C
@T X i dT;
i
dX ¼ dX i
i¼1
reflected solely by the plastic strain rate dep. The inelastic response is assumed
to be virtually incompressible, and no volumetric plastic strain occurs. The total
strain is a superposition of the mechanical strain and the thermal strain. The
latter is calculated by use of the linear thermal expansion coefficient multiplied
by temperature change. Stress tensor r is determined on the basis of the gener-
alized Hookean law of Eq 2 for a linear, isotropic, elastic continuum. The
fourth-order elasticity tensor a depends on temperature T and is not affected by
(inelastic) deformation; a consists of the elastic constants. The plastic strain is
defined by the assumed associated flow rule, Eq 3, whereby dp represents the
plastic multiplier known as the accumulated plastic strain rate. The pressure-
independent von Mises yield function f allowing for kinematic hardening is
defined by Eq 4, whereby J(r X) denotes the distance in the deviatoric stress
space. Deviatoric parts of backstress X0 and of stress tensor r0 are included in
the conditional Eq 5. The initial size of the yield surface at zero plastic strain k
remains constant for a certain temperature through the neglect of isotropic
hardening. The non-isothermal and non-linear incremental evolution law of the
strain valued backstress component Xk, formulated in Eq 6, is defined as addi-
tive decomposition, in which a temperature rate term ensures that the material
response is independent of temperature history and; consequently, can be char-
acterized by isothermal uniaxial work hardening data. Ci and ci are
temperature-dependent material parameters, whereby Ci strongly depends on
temperature. The overall backstress X is composed of two backstress compo-
nents: The first one describes the fast increasing primary hardening inside the
hysteresis, the second one describes hardening in the range of higher absolute
inelastic strains.
The model in its present form, which neglects isotropic hardening and
time-dependent effects, contains 8 material parameters for each temperature.
These are Young’s modulus E; Poisson’s ratio ; kinematic hardening parame-
ters C1, C2, c1, c2; initial size of the yield surface k; and linear thermal expansion
coefficient ath. These parameters need to be determined from a suitable set of
experimental data.
In terms of parameter identification, especially concerning kinematic hard-
ening, only the cyclic stabilized elastoplastic material condition at mid-life is of
interest. For this purpose, isothermal strain-controlled low-cycle fatigue (LCF)

tests have to be conducted at constant strain rate, see, e.g., Fig. 3. The adjust-
ment of the material model to cyclic stabilized stress-strain hysteresis loops of
LCF tests performed at several temperatures is shown, e.g., in Ref. [1] with rea-
sonable accuracy. This results in several datasets of best-fitted parameters, each
assigned to a given temperature. In the case of anisothermal calculations, the
parameter set required for a certain temperature has to be interpolated between
the underlying datasets.
Lifetime Approach
The lifetime assessment is based on an empirical phenomenological approach,
see, e.g., Ref. [7], to estimate the number of cycles until crack initiation
Nf. The approach is strain-based and assumes a defect-free material with a
behavior like specimens in LCF tests and time-to-rupture tests. By applying
the life fraction rule [8] for creep/relaxation and Miner’s rule [9] for fatigue,
failure is determined by the cycle at mid-life in terms of the summation of
creep damage Lt and fatigue damage LA, wherein both failure fractions are
assumed to be approximately independent of each other. The damage evalua-
tion follows Eq 7 resulting in the material-specific creep-fatigue damage sum
L. As already mentioned, the lifetime approach requires 1-dimensional input
variables
!
X Dtj ðrj Þ 1
Lt þ L A ¼ N f þ 0 ¼L (7)
j
tu ðrj Þ Nf
Creep damage is calculated with respect to the variation of stress during a cycle.
When fulfilling the creep condition at an individual time step/increment j during
a cycle, the quotient of the corresponding time interval length Dtj ¼ tj tj1 and
the rupture time tu of the active stress rj*, see Fig. 2(b), is calculated.
FIG. 2—(a) Schematic visualization of the incremental creep strain calculation and
(b) assigning the rupture time tu to active stress rj* by using the time-temperature
parameter PLM.

Temperature Tj and stress rj* are considered to be approximately constant

during this incremental time period. Currently, there is no distinction between
tensile and compressive stress. It is assumed in a conservative way that the
damage to both is equal. The creep condition is defined by the use of a limiting
temperature in the range of one third of the material’s melting temperature.
Time-to-rupture tests at several sampling temperatures with respect to varying
load levels are used to describe the creep behavior as well as the creep damage.
To avoid interpolation between different time-to-rupture curves and to compen-
sate the scattering of experimental results, the Larson-Miller time-temperature
parameter PLM [10] is applied
PLM ¼ ð273 þ TÞðC þ logðtu ÞÞ (8)
The rupture time is derived from the master curve by solving the parameter defi-
nition in Eq 8. Experimental data are used to fit the master curve by using an
exponential law with the fitting parameters c1 and c2: r ¼ c1 expðc2 PLM Þ.
The constitutive material model as introduced earlier in this paper is not ca-
pable of capturing time-dependent deformation. Therefore, to obtain the effec-
tive acting (relaxed) stress rj*, the stress relaxation of FEA resultant stress rj is
recalculated within the lifetime approach by the use of both the Norton-Bailey
power law [11]
ec ¼ Krn tm (9)
to describe secondary creep and the time hardening rule defining the path or
rather the continuous creep strain for time-dependent stress changes between
different creep curves, each assigned for constant stress load and constant tem-
perature, Fig. 2(a). The parameter set K, n, m is optimized in each case for a
given sampling temperature to represent varying stress. The stress relaxation
term in Eq 10 is calculated by the subtraction of the continuous accumulated
creep strain ec;j ¼ ec;j1 þ Dec;j multiplied by the Young’s modulus Ej at incre-
ment j. It should be noted that this method of stress correction is a superposi-
tion of relaxation and also the changes to Young’s modulus. Both effects are
strictly dependent on temperature. In addition, to consider the effective stress
acting inside the stress-strain hysteresis loop, the backstress Xj is subtracted
from the FEA resultant stress at each time step, too
rj ¼ ðrj Xj Þ Ej ec;j

rt¼0
Relaxation : e ¼ ee þ ep þ ec ¼ þ ep ¼ const
Et¼0
rt¼0 r
! ee ¼ ec ; ee ¼
Et¼0 E (10)

rt¼0 dr dE dT rt¼0 dec
!r¼E ec ! ¼ ec E
Et¼0 dt dT dt Et¼0 dt
dr dec
! isothermal : ¼ E
dt dt

FIG. 3—(a) LCF life curves and (b) cyclic stabilized stress-strain curves of D5S at strain
rate of 103 s1 (black shaded line) and 104 s1 (gray shaded line), respectively.
In the case of fatigue damage, it is proposed that the LCF life curves (see, e.g.,
Fig. 3) at maximum cycle temperature Tmax and minimum cycle temperature
Tmin together with the maximum strain width range Demax be considered. In
order to calculate a geometric temperature-dependent mean value of fatigue life
Nf [0] at mid-life, the empirical relationship in Eq 11 is introduced

1 1 1 1
¼ þ (11)
N0f 2 Nf ðDemax ; Tmax Þ Nf ðDemax ; Tmin Þ
The estimated number of cycles until crack initiation Nfe is finally determined
by Eq 12 on the basis of n load cycles and the specific creep-fatigue damage DtA
for each cycle. The material-specific damage sum L as the critical value indicat-
ing failure is given, e.g., by the damage evaluation of TMF tests
P
n1
L DtA;i þ nDtA;n
i¼1
Nef ¼ (12)
DtA;n
Application-Specific Phenomena
In general, isothermal strain controlled LCF tests at varying temperatures are
conducted by applying symmetric triangular shaped cycles fully reversed (strain
ratio Re ¼ 1)without a dwell period at a constant strain rate de ¼ 10 3 s 1. The
strain rate of this magnitude is chosen with respect to averaged operation con-
ditions as well as to avoid superimposed creep effects on the strain-life curves.
Smooth specimens with a diameter of 7.9 mm and gage length of 15.395 mm
are used to conduct the tests presented. The total strain width range De is
extended from 0.2 % to 2.0 %. The strain load is driven by the constrained ther-
mal expansion of the T/H, so that several strain rates of lower magnitude as well

as mean strains (Re = 1) appear as a function of different T/H positions and

time. Also the changing of the material behavior or rather the material proper-
ties over operating time is an important issue, which has an impact on lifetime.
Therefore, the influences of the application-specific phenomena
strain rate de,
mean strain em, and
aging
on the LCF life curves and the deformation behavior at mid-life have been stud-
ied. The deformation behavior affects the calibration of the material model
applied in the FEA, and the LCF damage behavior is important for the fatigue
damage calculation. At this point, results are primarily discussed for D5S.
In Fig. 3, the effect of the strain rate de ¼ 10 3s 1 compared to a power of
10 lower strain rate of 10 4 s 1 is shown. At a moderate temperature of around
200 C, the influence on the LCF life curves described by the Manson-Coffin rela-
tionship [12,13], Fig. 3(a), as well as on the stress-strain behavior described by
the Ramberg-Osgood relationship [14] at mid-life, Fig. 3(b), is insignificant.
Results are within the range of the scatter band. The values of the strain width
range and the stress amplitude in the axial direction are standardized separately
on the maximum appearing value.
At higher temperatures, beginning at approximately 500 C, the lower strain
rate has more relevance. The LCF life curves in Fig. 3(a) illustrate two effects. At
a higher strain width range, the cycle number Nf decreases together with a lower
strain rate. As a result of the higher strain load and the lower strain rate, the
superimposed creep damage should be the main damage mechanism. In contrast
to lower strain width ranges, the cycle number increases. It seems that in the lat-
ter case, the ductility of the material dominates the failure mechanism and creep,
especially at lower loads, may become of subordinate relevance. Figure 3(b) dem-
onstrates a significant drop in strength at 700 C caused by lower strain rate.
To investigate the influence of microstructure changes over the material’s life-
time, additional LCF tests on aged unnotched specimens have been conducted.
The aging conditions of 500 h at 700 C have been selected according to the aver-
age lifetime of T/Hs on gas test stands. The results in Fig. 4(a) show in the tempera-
ture range from 200 C to 500 C a significant decrease of the number of cycles until
crack initiation, especially at higher strain width ranges. But at 700 C, this trend
cannot be observed. Moreover, the cycle number remains approximately stable
with a slight increase at higher strain ranges. The full interpretation of this effect is
still an open issue. For this purpose, the microstructure has been analyzed.
After aging, the microstructure in Fig. 5(b) changes significantly compared
with the basic microstructure in Fig. 5(a). The amount of almost spherical
graphite inclusions is much higher at simultaneously reduced diameters. This is
partly caused by the growth of secondary graphite. The pearlite fraction is also
increased after aging, especially as there are a lot of pearlite fringes along the
grain boundaries. The increased pearlite fraction together with an increase in
hardness and a reduced fraction of the more ductile austenitic basic matrix may
indicate embrittlement. In general, the loss in ductility as well as a more complex
network of smaller graphite inclusions, which act as a metallurgical notch in the
same way as the tetrahedron carbide found in the microstructure, downgrade the

FIG. 4—a) LCF life curves of D5S based on basic microstructure (black shaded line)
compared with 500 h at 700 C aged microstructure (gray shaded line). (b) LCF life
curves of D5S at Re ¼ 1 (black shaded line) and Re ¼ 0 (gray shaded line).
LCF behavior, as can be seen, up to approximately 500 C. Further investigations

to evaluate the change in behavior around 700 C are necessary. However, it
seems that due to the renewed increase of ductility at higher temperatures, crack
initiation is decelerated. Apart from scattering, an effect of the aged material con-
dition on cyclic stress-strain behavior at mid-life cannot be observed.
The variation in strain ratio, Fig. 4(b), indicates a loss of cycle numbers to-
gether with higher mean strains, at least at higher strain width ranges. At lower
strain width ranges, this effect seems to be inverted. But due to scattering there
is no explicit conclusion that can be drawn from this.
TMF Lifetime Validation

In terms of damage evaluation and validation of the phenomenological lifetime
approach, a test series of in-phase TMF tests under characteristic load
FIG. 5—(a) Microstructure of the D5S specimen basic material condition compared
with (b) the microstructure of D5S specimen after aging 500 h at 700 C.

conditions was performed on unnotched specimens in a special TMF test rig

[1]. The significant TMF loading history was derived on the surface of T/H criti-
cal positions as results of the FEA within the component structural analysis pro-
cedure described in Fig. 1. Following Eq 7 by applying the underlying
numerical calculation method, creep damage fraction and fatigue damage frac-
tion until crack initiation have been determined by use of a representative cycle
at mid-life on the basis of the TMF tests conducted; the results are shown in
Fig. 6(a). This assumption is justified due to minor hardening and softening
effects during TMF loading of the investigated materials D5S and GJV, respec-
tively. In the case of GJV, the first load cycle also has to account for damage
evaluation due to the significant peak stress drop after first loading, see Ref. [1].
This effect may partly be caused by the thermally damaged graphite inclusions
that are typical for lamellar and vermicular cast iron under high-temperature
loading. As a result of the varying strength/deformation behavior of both materi-
als and their varying creep resistance, TMF loading leads to unequal TMF lifetime
behavior. Whereas for D5S both damage fractions are approximately in the same
range, the damage to GJV is dominated by creep. One reason for this is the lower
plastification during the TMF loading of GJV caused by a lower mechanical strain
width range. This leads to a lower fatigue damage fraction toward Eq 11. To con-
firm these results, additional metallurgical investigations on the microstructure
are necessary to separate the damage mechanism and to determine the damage
fractions quantitatively by detecting persistent slip bands, creep pores on grain
boundaries, and V-notch disconnections/cracks between grain boundaries. The
latter is significant for creep damage at higher load levels.
By applying the evaluated mean creep-fatigue damage sum for each material,
the recalculation of the cycle numbers until crack initiation Nfe provides the TMF
lifetime estimations shown in Fig. 6(b). The results are compared to measured
values Nf. The LCF testing results carried out earlier in this paper are used to
enhance both the description of the deformation behavior of the material model
applied in the FEA and the damage behavior inside the lifetime approach.
FIG. 6—(a) Fatigue and creep damage fractions of D5S and GJV specimens under char-
acteristic TMF load conditions. (b) Recalculated number of cycles until crack initiation
Nfe of TMF tested specimens for D5S and GJV.

Lifetime Assessment of T/H
FEA
The turbine housing of a regulated single stage turbocharger designed in two
variants for D5S and GJV, subject to thermal shock loading, serves as the refer-
ence to assess the numerical component structural analysis. During the thermal
shock test on a gas test stand, the T/Hs were removed time and time again after
a certain number of load cycles for detecting crack initiations as well as moni-
toring crack propagations by dye penetration. In order to simulate the corre-
sponding thermo-mechanical T/H behavior under thermal shock test
conditions, an anisothermal transient FEA was done on the basis of the model
shown in Fig. 7(a). The kinematic hardening parameters of the material model
introduced in the Component Structural Analysis section are calibrated by
using cyclic stabilized stress-strain curves which are the result of LCF tests
under varying conditions described in the Application-Specific Phenomena sec-
tion. The value of each parameter in terms of a quasi mean value for a certain
temperature is optimized on the basis of experience in the adjustment of simu-
lation results to experimental results.
For example, the interior transition wall surface of the wastegate flap valve
chamber downstream from the turbine volute outlet was identified as a critical
position by numerical analysis, in excellent agreement with experimental
results, as can be seen in Figs. 7(b) and 7(c) with the area highlighted in white.
The position of increasing accumulated plastic strain p coincides with the meas-
ured crack initiation and crack propagation, respectively.
Multiaxiality
During loading, the principal stress directions are time variable, and multiple
slip planes, which may cause fatigue damage, are activated. The trace of each
principal stress direction (I, II, III) is plotted in Fig. 8(a) for the critical crack-
FIG. 7—(a) Finite-element model for thermo-mechanical analysis of the T/H marked in
gray. (b) Contour plot of the calculated accumulated plastic strain p compared with (c) the
measured crack position on the interior wall surface of the wastegate flap valve chamber.

FIG. 8—(a) Trace of each principal stress direction and (b) critical plane defined by maxi-
mum normal stress during transient thermal shock load cycle conditions at a critical T/H
position.
initiation position shown in Fig. 7(b) on the T/H under transient thermal shock
load conditions. At each time step, the trace position is visualized on the unit
sphere with respect to the Cartesian engine coordinate system.
Thus, for each time step, the transient stress tensor r of multiaxial stress
states is rotated into the critical plane that is defined by the norm of maximum
normal stress, highlighted in Fig. 8(b) with the gray shaded circular area
maxfjrðnÞ ðx; y; zÞjg: rðnÞ ¼ nT r n (13)

ðx;y;zÞ
The normal stress r(n) results from the double dot product with the normal vec-
tor n of the plane considered, shown in Eq 13. It should be characteristic for
low-/semi-ductile or rather brittle materials such as cast iron that fatigue dam-
age is caused primarily at the plane of maximum normal stress, see, e.g., Ref.
[15]. Other tensors of required variables for lifetime calculation, such as me-
chanical strain and backstress, are rotated into the previous mentioned critical
plane as a 1-dimensional quantity, too.
Compared with other equivalent stress calculations, e.g., both von Mises
stress and first principal stress signed with the first invariant of the stress ten-
sor, the critical plane approach of maximum normal stress provides the best
results for the casting materials considered. Another critical plane approach,
based on the definition of maximum shear stress amplitude related to Ref. [16],
was not leading to the desired results, nor for the more ductile casting material.
Lifetime Estimation
The lifetime calculation method validated with the TMF testing results provides
the recalculated cycle numbers until crack initiation of thermal-shocked T/Hs
shown in Fig. 9(b). The positions, highlighted in white, of increasing equivalent
plastic strain p calculated by the FEA in the manner described, bring out the
positions of crack initiation during the thermal shock test in an excellent way,

FIG. 9—(a) Contour plots of calculated equivalent plastic strain p at positions of crack
initiation for GJV housing 1–4 and D5S housing 5–8, respectively, and (b) correspond-
ing recalculated number of cycles until crack initiation Nfe compared with the averaged
measured results Nf during thermal shock test.
Fig. 9(a). Each measuring point lies within the interval resulting from dye pene-
tration measurements at different T/Hs. Except in the significantly conservative
case 4 for GJV (Fig. 9(b)), the outer connection radius of the volute, all esti-
mated lifetime results are within the scatter band of the scale of two. Unfortu-
nately, some results are slightly non-conservative for both materials, the more
ductile D5S and the more brittle GJV. These results may denote that the mate-
rial failure is adequately characterized by the normal stress approach. Neverthe-
less, further investigations on the microstructure as well as on the material
behavior under different load phenomena are necessary to improve the failure
description, also to avoid non-conservative estimations as an important factor
for component design. Another important aspect to consider is the fact that the
scattering of casting material is also strongly influenced by the graphite mor-
phology on the microstructure. The graphite morphology can be rather differ-
ent on various T/H positions.

Conclusions
The lifetime estimation approach that has been developed demonstrates a satis-
factory estimated number of cycles until crack initiation on unnotched speci-
mens as well as for the adaption on T/H design in a first basic approach.
Moreover, based on critical plane approaches and energy criteria, the handling
of multiaxial stress and strain states on the T/H, for instance to calculate
1-dimensional stress and strain paths required as input quantities of the lifetime
approach, are still being investigated. Alternative fatigue damage methods, e.g.,
based on micro crack growth, are also being focused on. In addition, the analy-
sis of microstructural phenomena should help to enhance the creep fatigue
interaction together with a more detailed failure description. It is assumed that
tensile and compressive stress both have an equal effect on creep, and no dis-
tinction is made when calculating the creep damage. Current testing results
illustrate that for GJV this assumption is well justified, at least at two sampling
temperatures. But there is a need for a refinement in order to describe creep
characteristics. The damage behavior of aged material conditions at elevated
temperatures requires also more investigations in depth.
A proposal has also been made to describe crack propagation following the
estimation of crack initiation as part of the approach. This enables expansion of
the lifetime assessment up until leakage and design failure. First results on both
specimens and the T/H show a clear influence.
References
[1] Laengler, F., Mao, T., and Scholz, A.,“Validation of a Phenomenlogical Lifetime
Estimation Approach for Application on Turbine Housings of Turbochargers,” 9th
International Conference on Turbochargers and Turbocharging, May 19–20, 2010,
London, pp. 193–205.
[2] Laengler, F., Mao, T., and Scholz, A., “Phenomenological Lifetime Assessment for
Turbine Housings of Turbochargers,” 9th International Conference on Multiaxial
Fatigue & Fracture, June 7–9, 2010, Parma, Italy.
[3] Heuer, T., Engels, B., and Wollscheid, P., “Thermomechanical Analysis of a Turbo-
charger Based on Conjugate Heat Transfer,” Proceedings of ASME Turbo Expo
2005, June 6–9, 2005, Reno-Tahoe, NV.
[4] Chaboche, J. L., “Time-Independent Constitutive Theories for Cyclic Plasticity,”
Int. J. Plast., Vol. 2, 1986, pp. 149–188.
[5] Chaboche, J. L., “Constitutive Equations for Cyclic Plasticity and Cyclic
Viscoplasticity,” Int. J. Plast., Vol. 5, 1989, pp. 247–302.
[6] ABAQUS Theory Manual, Version 6.8, Dassault Systèmes, 2008.
[7] Scholz, A., and Berger, C., “Deformation and Life Assessment of High Temperature
Materials Under Creep Fatigue Loading,” Materialwiss. Werkstofftech., Vol. 36,
2005, pp. 722–730.
[8] Taira, S., “Lifetime of Structures Subjected to Varying Load and Temperature,”
Creep in Structures, N. J. Hoff, Ed., Academic Press, New York, 1962, pp. 96–119.
[9] Miner, M. A., “Cumulative Damage in Fatigue,” J. Appl. Mech., Vol. 12, 1945, A1S9.
[10] Larson, F. R., and Miller, J., “A Time-Temperature Relationship for Rupture and
Creep Stress,” Trans. ASME, Vol. 74, 1952, pp. 76.

[11] Norton, F. H., The Creep of Steel at High Temperature, McGraw Hill, New York,
1929.
[12] Manson, S. S., “Behavior of Materials Under Conditions of Thermal Stress,” Tech-
nical Report No. 2933, NACA, 1954.
[13] Coffin, L. F., “A Study of Effects of Cyclic Thermal Stresses on a Ductile Metal,”
Trans. ASME, Vol. 76, 1954, pp. 931–950.
[14] Ramberg, W., and Osgood, W. R., “Description of Stress-Strain Curves by Three
Parameters,” Technical Report No. 902, NACA, 1943.
[15] Sonsino, C. M., “Influence of Ductility on the Multiaxial Fatigue Behaviour by the
Example of Welded Joints of Steel and Aluminium,” Materialwiss. Werkstofftech.,
Vol. 34, 2003, pp. 189–197.
[16] Papadopoulus, I. V., “Critical Plane Approaches in High-Cycle Fatigue: On the Defi-
nition of the Amplitude and Mean Value of the Shear Stress Acting on the Critical
Plane,” Fatigue Fract. Eng. Mater. Struct., Vol. 21, 1998, pp. 269–285.


doi:10.1520/JAI103968
Michael K. Schaper1
Fatigue Crack Closure at Near-Threshold Growth

Rates in Steels, Effects of Microstructure,
Load Sequence and Environment
ABSTRACT: This paper summarizes long-term research on fatigue crack

closure in several labs of the author. Various experimental techniques were
used in different combinations. Primarily, a sophisticated dynamic compli-
ance technique based on highly sensitive resonance vibration measurements
and crack tip diffracted ultrasonic wave analysis is applied. Whereas the first
method reflects the global through thickness behavior, the second method
enables spatial resolution, which is achieved with a reduction of sensitivity
yet has the advantage that the signal generation is independent on the stress
state ahead of the crack tip. The results allow a coherent view into the
effects of the stress state, material plasticity, fracture surface roughness, and
oxidation tendency on the closure behavior and, therefore, on both the
threshold stress intensity and near-threshold crack propagation. Specifically,
the different load ratio, load sequence, and the environmental effects on the
threshold values of ferritic, martensitic, and austenitic steels are addressed.
KEYWORDS: fatigue crack growth, crack closure, steels
Introduction
The crack closure effect describes the phenomenon of mechanical contact in
the wake of a fatigue crack that occurs at above-zero applied loads within a
loading cycle. The effect is caused by bridging between the fracture surfaces
due to roughness, oxide deposits, and residual deformations which lowers the
crack tip loading range as compared to the applied one. Thus, the crack closure
Manuscript received May 11, 2011; accepted for publication October 18, 2011; published
1
Institute of Materials Science, Univ. of Technology Dresden, D-1062 Dresden, Germany,
e-mail: michael.schaper@tu-dresden.de
Cite as: Schaper, M. K., “Fatigue Crack Closure at Near-Threshold Growth Rates in Steels,
Effects of Microstructure, Load Sequence and Environment,” J. ASTM Intl., Vol. 9, No. 1.
doi:10.1520/JAI103968.
Copyright V
231

phenomenon results from load transfer across the crack faces in the crack
wake. It is, therefore, essentially different from the crack tip internal stress field
effects, which are mean stress effects in nature.
Based upon widespread experimental evidence, crack closure is generally
invoked to explain salient features of the fatigue behavior of cracked bodies [1–4].
The anomalous dependence of the crack propagation threshold on crack size, load
ratio, and environment and the large variety of load sequence effects are usually
ascribed to the crack closure phenomenon. Sometimes even the existence of a
threshold is attributed to crack closure. Nevertheless, conflicting results of over 30
years of extensive studies question the significance of crack closure in the predic-
tion of fatigue crack growth and its incorporation into life estimation schemes
[5,6]. It is argued that different measurement techniques and evaluation proce-
dures are responsible for many of the inconsistencies of the published data [7].
Most frequently, crack closure is derived from static compliance measure-
ments using various strain=clip gauge techniques giving a through-thickness aver-
age of the closure effect. On the contrary, methods such as moiré pattern analysis
or digital surface image correlation measure the opening and closing of the crack
on the side surface, which must not be representative for the three-dimensional
stress state in the interior of the specimen. Therefore, interferometric and ultra-
sonic methods were developed to obtain some spatial resolution along the crack
front in order to discriminate between plane stress near-surface and plane strain
regions. BHowever, despite these efforts there is no unanimous agreement in the
fatigue community on the interpretation of the measured data. This includes both
the derivation of a closure load and the possible contribution to the crack driving
force at loads below complete crack opening, i.e., the definition of an effective
loading range [8]. It is suggested that the crack tip driving force should be propor-
tional to the crack tip strain field magnitude and not simply dependent on the
crack opening load as usually determined from load displacement plots.
In the course of more than 30 years of research activity in the field of fatigue
crack growth in ferrous alloys, the crack closure effect has been thoroughly ana-
lyzed by the author’s research groups in the fatigue laboratories of the former
Central Institute of Materials Research of the Academy of Sciences of East Ger-
many, the Leibniz-Institute of Materials Research, and in the Institute of Materi-
als Science of the Technical University of Dresden [9–17]. This paper summarizes
long-term research on fatigue crack closure in the author’s labs, whichhas been
largely unpublished until now. It presents a description of the experimental tech-
niques used and exemplifies the results achieved by typical examples.
Experimental Procedures
Investigated Materials
Over the years, the experiments in the author’s labs were embedded in research
programs directed to develop crack resistant steels through specific alloying
concepts and the optimization of thermomechanical treatments. The investi-
gated model alloys and steel variants covered a strength range from 300 to
2500 MPa with microstructures ranging from austenite to ferrite (including

SCHAPER, doi:10.1520/JAI103968 233
specific iron based model alloys), ferrite-pearlite, bainite, martensite, and also
metastable austenite. Strength data of three selected steels, which form some
part of this publication, are included in Table 1 together with some results
which will be discussed later.
Mechanical Testing
The crack propagation experiments were performed in cantilever bending
using a dynamic compliance resonance technique with frequencies in the
range from 55 to 95 Hz. Mostly, specimens with a cross section of
B W ¼ (10 20) mm2 were used. For the crack propagation tests crack starter
notches having a depth of about 4.0 mm and a root radius of about 0.12 mm
were prepared by spark-machining. The crack growth measurements were
started after fatigue precracking of about 1 mm at DK < 16 MPa m1=2 followed
by K-controlled loading procedures. In general, four types of tests were run:
(a) baseline tests applying different yet, throughout each test, constant DK lev-
els, (b) threshold measurements under continuous load shedding under DK
control at constant load ratio R or constant Kmax (R increasing test for high-R
threshold determination) with subsequent crack growth rate determination
under step-wise load increase, (c) crack initiation experiments after pre-
overloading notched samples, and (d) overload tests under constant base-line
DK and R. In general, the threshold value has been defined for growth rates
da=dN ¼ 5 1011 m=cycle. Experiments were performed under normal envi-
ronmental conditions in gaseous, liquid, and moisture saturated air environ-
ments and at temperatures up to 800 C.
For the threshold determination, the load reduction rate was kept at
dK=da ¼ 5 MPa m1=2=mm or C ¼ (1=K)dK=da ¼ 0.32 mm1, which is some-
what larger than prescribed by ASTM E-647-11[18]. Comprehensive preliminary
tests had shown that the measured thresholds were not influenced by these load
shedding rates, especially when continuous load reduction procedures were
applied. It is well known that load reduction rates that arwe too high result in
erroneously high thresholds, which is due to load interaction effects induced by
remote crack closure. On the contrary, a systematic test series has shown that
TABLE 1—Nominal and effective threshold values for steels with significantly different
microstructures. DKth,eff data are derived from lower and upper bound Kop definitions.
R ¼ 0.05 R ¼ 0.50 R ¼ 0.8
Material DKth DKth,eff DKth DKth,eff DKth ¼ DKth,eff
StE355 NbTi Ferrite-Pearlite, 7.92 5.0–4.0 3.5 3.5–3.3 3.1

Rp=Rm ¼ 426/514 Mpa 7.88 5.3–4.2 3.2
X6CrNiTi18.10 Austenite, 4.42=4.82 2.5–2.4 4.79 3.7–3.5 3.0
Rp=Rm ¼ 336/623 MPa 4.90 3.6–3.0
X210Cr12 Martensite, 3.35=3.5 2.4–1.8 3.46 3.5–3.4 3.2
Rm ¼ 2500 MPa 3.89 2.6–2.1 3.46 3.5–3.4

even very low load shedding rates may result in increased thresholds of ferritic-
pearlitic steels as depicted in Fig. 1. Obviously, this effect is due to excess fretting
oxidation at near-threshold growth rates and extended testing time.
Crack closure measurements were performed at pre-selected K levels using
one of the following techniques:
(a) Differential dynamic compliance analysis using resonance vibrations.
(b) Travel-time evaluation of crack tip diffracted ultrasonic waves.
(c) Quasistatic compliance measurements using near-crack tip strain
gauges.
(d) Surface imaging combined with cross correlation technique.
Method (a) represents the standard technique, which is extensively used in
the laboratories of the author; technique (b) has been developed and used
to obtain specific information on crack closure in near-surface regions as
compared to the interior of the specimen. Techniques (c) and (d) were some-
times used for comparison purposes, however, they are not discussed in this pa-
per. The first two techniques are described in some detail. Typical examples are
shown to illustrate the capabilities of these methods and the achieved results.
Despite the superior sensitivity of the dynamic compliance technique, one of
the crucial questions of the crack closure community arises, i.e., that concerning
an unambiguous and mechanically meaningful measure of the crack closure
effect. Therefore, we determined two limiting values of Kop as illustrated in Fig.
11. Consistent with the often applied practice, the upper limit is defined at the first
detectable deviation from the open crack compliance (which usually corresponds
FIG. 1—Influence of load shedding rate on the fatigue crack growth threshold.

SCHAPER, doi:10.1520/JAI103968 235
to a 0.6 ms reduction of the vibration period from the open crack value), but the
lower limit through the intersection of the tangents on the curves in the transition
to partial closure. The latter definition follows our earlier suggestion (e.g., [4]) that
a reliable measure of Kop should include some part of the gradual opening into
DKeff. When comparing the Kop definition used here with the conventional proce-
dure it is to be accepted that the differential compliance curves as described in the
following text correspond to derivations of conventional load-displacement dia-
grams, thus leading to enhanced sensitivity in deriving Kop data.
Dynamic Compliance Technique

The crack growth and crack closure experiments using the dynamic compliance
technique were performed by means of computer controlled testing machines
(DYNACOMP) developed by Schlät together with the author [9–13]. In these
machines, the specimen represents the frequency determining element of a me-
chanical resonator, which is excited to well-defined resonance vibrations of con-
trolled amplitude. With these machines two physical effects are simultaneously
used; (a) the resonance effect for low-power excitation of cyclic loading, and (b)
the dependence of the frequency of resonance vibrations on the masses and com-
pliances of the resonator. A proper design of the setup not only allows well-
defined loading conditions to be realized (through closed loop cyclic and mean
load control) but also enables high resolution compliance analysis. Therefore
crack length, crack growth, and crack closure can be evaluated without addi-
tional instrumentation, i.e., simply by means of vibration period measurements.
The external and almost continuous crack length measurement technique ena-
bles K controlled testing procedures to be performed, even under aggressive
environmental conditions. Furthermore, specific energy dissipation per crack
growth increment can be derived from measurements of the resonator damping.
The mechanical setup of the DYNACOMP machine is illustrated in Fig. 2.
The single edge notched specimen of thickness B and width W (typically
B ¼ 10 mm, W ¼ 20 mm) is loaded in bending via a cantilever. Cyclic loads are
applied by means of an electrodynamic shaker and calculated from the meas-
ured acceleration amplitude times the vibrating masses. A servo-motor com-
bined with a load cell is used for mean load control. A computer serves for the
on-line evaluation of the instantaneous crack length a (through solving the
crack length function f(a=W)), crack growth rate da=dN, cyclic intensity DK, and
the load ratio R according to Eqs 1 and 2
a4 3 U
f ða=WÞ ¼ a2 þ ¼ K (1)
2 2 ðTf =TÞ2 1
DM pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
DK ¼ 3=2
4:12 a3 a3 (2)
BW
where U and K depend on resonator properties, specifically on vibrating masses

and compliances, T is the measured period of the resonance vibrations, Tf is the
vibration period for the through-cracked specimen, DM the cyclic bending
moment and a ¼ 1 a=W.

FIG. 2—Resonance vibration tester DYNACOMP mechanical setup.
The basic DYNACOMP machines are designed for resonance frequencies

between 55 and 95 Hz. However, frequencies up to 300 Hz are easily achieved by
changing the masses and compliances of the mechanical resonator. The reso-
nance vibration period is usually measured as the mean value over 100 cycles,
i.e., every 1 to 2 s. The technique proved to be capable of resolving increments
in the crack length smaller than 0.5 lm. Interfering crack closure can be, at least
roughly, evaluated through measuring the anharmonicity of the resonance
vibrations and by an oscilloscopic representation of the electronically integrated
acceleration signal versus the instantaneous load (or K value). For more exact
crack closure examination the resonance vibrations are specifically used to
characterize compliance changes within preselected cycles (Fig. 3). To accom-
plish this, the amplitude of the resonance vibrations are reduced to well below
the crack growth threshold. Subsequently, the period of these minor vibrations
is recorded upon quasistatically changing the preload between the maximum
and minimum load of the preceding fatigue cycle. From the corresponding
vibration period, the incremental compliance and a closure affected “effective
crack length,” aeff is calculated. Crack closure curves are then derived in terms
of aeff versus applied stress intensity K as shown in Fig. 4.
Despite being a “far field” method the dynamic compliance technique proved
to be one of superior sensitivity. Due to cantilever loading, the compliance meas-
urements are not affected by friction at loading pins or rollers, which interferes
in testing the compact tension and three point bending specimens. Both of these
advantages make the method especially suited for crack closure measurements.
Additionally, the method is believed to be superior to near-tip surface measure-
ments, which largely reveal the plane stress contribution, since it provides a more
realistic estimate of the through-thickness phenomena. On the contrary, it has to
be admitted that a significantly higher closure level is reported in the literature
(e.g., McEvily and Young [19]) for the plane stress surface region as compared to
the interior, especially after overloading but even in the near-threshold regime.
Therefore, we developed a method which enables some spatial resolution. This

SCHAPER, doi:10.1520/JAI103968 237
FIG. 3—Resonance vibration dynamic compliance technique, principle of crack propa-

gation, and crack closure measurements.
ultrasonic technique is described in the section titled “Crack Tip Diffracted Ultra-
sonic Wave Measurements.”
Crack Tip Diffracted Ultrasonic Wave Measurements

For crack length measurements by the ultrasonic technique a servohydraulic
testing machine, INSTRON 8500, is usually used with sine wave load cycling at
a frequency of 20 Hz. Crack length and crack closure measurements are
FIG. 4—Ultrasonic technique for crack closure analysis with some spatial resolution.

performed using two transducer pairs (emitter and receiver) attached to the
specimen, as shown in Fig. 4. To enable separate measurements of crack closure
in the plane strain and plane stress regions one transducer pair is applied in the
midsection of the specimen and the other one onto the side surface region.
When reaching the receiver the emitted ultrasonic waves traveled through
the specimen along various paths. These different possible travel paths result in a
complicated amplitude-time pattern (A-scan), as detected by the receiver. Of spe-
cific interest is the wave which is diffracted at the crack tip. This signal may be
superimposed by an additional signal of shorter travel time, which results from
the crack closure behind the crack tip. Whereas crack growth is easily seen in the
A-scan from the time shift of the signal at maximum load, the analysis of crack
closure needs complicated data evaluation. To accomplish this and in order to
measure crack closure during 20 Hz fatigue testing the following procedure is
adopted. At selected stages of the fatigue process, 50 ultrasonic signals per cycle
are emitted, which results in 1000 A-scans per second at the receiver. These
A-scans are then converted into a grey scaled map consisting of 50 grey scaled
A-scans per cycle which are plotted in parallel to give a clear indication of crack
closure induced travel time shifts within a loading cycle (Fig. 5). Process control
of the ultrasonic technique and data acquisition of the A-scans is done using a
personal computer with GPI-Bus including a digital storage oscilloscope.
The interpretation of the amplitude maps is finally done by personal deci-
sion. This especially applies to the time of flight measurements of the crack tip
diffracted transverse wave, from which the actual crack length has to be calcu-
lated. Because the test arrangement, specific transducer characteristics, and
sound velocity of the test material must be taken into account, some prelimi-
nary tests are unavoidable.
Fatigue Crack Growth Threshold

In Fig. 6, the near-threshold fatigue crack growth kinetics in a higher strength
ferritic-pearlitic steel is shown. The tests are conducted under continuous DK
shedding and the subsequent step-wise DK increase at R ¼ constant. The strong
dependence of the threshold value on the load ratio R is evident, whereas this
load ratio effect is largely reduced in the Paris region, which is obviously due to
the reduced relative importance of roughness and oxide induced crack closure.
With an increasing load ratio the threshold diminishes and approaches about
3 MPa m1=2 above R ¼ 0.6. This R-effect proved to be influenced by the strength
level and microstructure of the material, environment, and temperature. As
shown in Fig. 7, the low-R threshold is weakly proportional to the square root of
the ferrite grain size of ferritic-pearlitic steels, a proportionality which is largely
reduced at high load ratio. As already reported by Liaw et al. [20], a dry gaseous
environment makes the low-R threshold smaller, whereas it is significantly
enhanced in water vapor saturated air. Correspondingly, the oxide induced
crack closure is largely suppressed in dry gases however, it is largely intensified
in a wet atmosphere. An even lower threshold is measured when electrolytic

SCHAPER, doi:10.1520/JAI103968 239
FIG. 5—Gray scaled map of amplitude scans with 50 A-scans per load cycle and travel
time shift of the crack tip diffracted T-wave within individual load cycles.
hydrogen charging is applied (Fig. 8). This is expected to be a consequence of

an enhanced tendency to intergranular near-threshold crack propagation in fer-
ritic microstructures. As discussed in the section titled “Principle Features of
Crack Closure” these effects are paralleled by the crack closure behavior.
The influence of the load ratio on the threshold of the three steels of Table 1,
which are characterized by largely different strength levels and microstructures,
is shown in Fig. 9. (Threshold values at R> 0.6 were measured using a
Kmax ¼ constant procedure). At R > 0.6 all three microstructures exhibit a
unique threshold of about 3 MPa m1=2. Obviously, the threshold of the investi-
gated martensitic steel is almost independent of R as compared to the ferrite-
pearlite. The load ratio effect of stable austenite is in between both extremes. In
contrast to this fact, it was proven by Che [14] that the high metastability of the
austenite results in a generally increased, however less R dependent threshold
level, i.e., the threshold is increased especially at high R in such a material. The
strength dependence of the threshold for an even wider range of microstruc-
tures as observed over 30 years is summarized in Fig. 10. It is seen that the low-
R threshold decreases with the increasing strength level with a superimposed
minor basic microstructural influence, which might be due to the microstruc-
tural unit size, as already exemplified in Fig. 7. At large R the strength effect
almost disappears. Often the strength effect is explained solely on the basis of
crack closure. The results previously mentioned point to an additional basic

FIG. 6—Fatigue crack propagation curves of a high strength ferritic-pearlitic steel at

different load ratios in the near-threshold regime.
FIG. 7—Effect of ferrite grain size on the fatigue crack growth threshold of a high
strength ferritic-pearlitic steel at different load ratios.

SCHAPER, doi:10.1520/JAI103968 241
FIG. 8—Effect of electrolytic H-charging on the fatigue crack growth threshold of a high
strength ferritic-pearlitic steel at R ¼ 0.05.
microstructural boundary effect as a minor but superimposing influence on

near-threshold crack growth. Furthermore, a remaining mean load effect might
be due to its influence on the fatigue mechanism ahead of the crack tip.
Principle Features of Crack Closure

Using the dynamic compliance technique, crack closure curves are quasistati-
cally measured after interrupting a crack propagation experiment. Each curve
represents the compliance change within a preceding load cycle. In summary,
the following general findings were observed (Fig. 11).
(a) At the low load ratio the crack closes upon unloading down to an effec-
tive crack length which proved to be dependent on the loading history.
When a continuous load shedding procedure is followed the measured
effective crack length at full unloading may approach the depth of the
crack starter notch. Upon reloading, a gradual opening occurs until
above a somewhat arbitrarily definable stress intensity Kop the calculated
aeff,max equals the physical crack length as measured optically on the
fracture surfaces. Apart from minor amplitude effects aeff,max agrees well

FIG. 9—Influence of load ratio on the fatigue crack growth threshold of a ferritic-
pearlitic, an austenitic, and a martensitic steel.
FIG. 10—Strength and microstructure dependence of the fatigue crack growth threshold
of steels and iron based alloys at R ¼ 0.05 and at R ¼ 0.6 and a summary of the results
achieved in the author’s laboratory.

SCHAPER, doi:10.1520/JAI103968 243
FIG. 11—Principle features of the crack closure effect in metallic materials.
with that of the crack length, which is calculated from the vibration pe-
riod of the foregoing high amplitude loading. The basic similarity of the
compliance derived closure curve to a conventional load-displacement
curve (after differentiation) is obvious. Load displacement measure-
ments using a clip gauge mounted at the side surface just behind the
crack tip revealed this correspondence.
(b) When the mean load approaches Kmax of the prior fatigue cycle a slight
decrease in the vibration period is often observed, especially in lower
strength coarse grained materials. This indicates a loss of compliance
which might be due to an exhaustion of the mobility of the dislocations
within the plastic zone near maximum load.
(c) When, after low R cycling, the mean load of the small amplitude vibra-
tions is in the range of partial closure, an enhanced dependence of the

vibration period on their amplitude is observed. This indicates some

form of adhesive fretting between the fracture surfaces. Correspond-
ingly, enhanced damping and specific energy dissipation have been
measured during crack propagation at low R when approaching the
threshold by load shedding [21].
(d) After fatiguing with high R, the incremental compliance (and therefore
aeff) remains constant within the former loading range, but immedi-
ately drops at the first unloading below Kmin of the preceding cycle.
Upon reloading the crack opens at much lower stress intensities. The
partial elimination of asperities in the crack wake is expected to be re-
sponsible for this behavior. The almost complete elimination of this
effect in subsequent measurements implies only a weak influence of
underloads under variable amplitude cycling.
(e) There is some amplitude dependence of the vibration period due to
crack tip plasticity and external damping, which needs careful consid-
eration. By using too small vibration amplitudes, erroneously
enhanced crack opening loads may be derived. Additionally, interfering
crack closure implies some anharmonicity of the loading cycle which,
in principle, can be accounted for by half-period measurements over
the open part of the cycle. Such corrections were applied if necessary.
(f) During crack propagation after overloading a transitional hump
appears in the crack closure curves due to the development of two clo-
sure levels (see the section titled “Crack Closure under Steady-State
Loading Conditions”). The upper closure level is usually expected to
represent the effective minimum load of the fatigue cycle. The lower
closure level obviously indicates closure according to earlier growth
stages. It becomes less pronounced or even disappears after high over-
loading and in later growth stages.
Crack Closure under Steady-State Loading Conditions

As an example, crack closure curves for a microalloyed ferritic-pearlitic steel at
different load ratios are shown in Fig. 12. The shift of succeeding crack closure
curves to larger open crack lengths aeff,max reflects related crack growth incre-
ments. At high R, the crack is propped open during the whole fatigue cycle,
however, the crack closure curves were measured down to almost full unloading
in order to have a clear indication of the closure transition. This unloading
proved to be without influence on the following crack growth behavior. At low R
and near-threshold loading the Kop values of all three steels were found to be
significantly higher than Kmin.
For the ferritic-pearlitic steel, a slight increase in Kop is observed during
load shedding at R ¼ 0.05 when approaching the threshold. Therefore, the rela-
tive importance of crack closure in determining the threshold becomes large. At
higher DK, i.e., in the Paris regime, Kop is found to remain roughly proportional
to DK (Fig. 13). At a high load ratio Kop approaches Kmin early in the Paris re-
gime for R ¼ 0.3, however,it is completely absent even at near-threshold loading
at R ¼ 0.8. It can be concluded that the influence of crack closure in the Paris

SCHAPER, doi:10.1520/JAI103968 245
FIG. 12—Near-threshold crack closure in a ferritic-pearlitic steel under load shedding at

three different load ratios. Crack closure curves are measured down to almost full unload-
ing. Crack growth is indicated by the shift of the upper horizontal part of the curves.
regime is limited to small R. A series of DK ¼ const. crack growth tests has addi-
tionally shown a roughly constant Kop at low and high load ratios for the three
steel microstructures of Table 1. This finding is in accordance with the reduced
R-effect on the crack growth kinetics in the Paris regime and also with the mas-
ter curve of McClung [1]. On the contrary, the influence of R on near-threshold
crack growth is not in all cases fully accounted for by crack closure, at least for
a lower strength material. As shown in Fig. 14, the low load ratio DKth,eff
remains somewhat larger than DKth as measured for R ¼ 0.8, which might imply
a direct mean load effect on fatigue damage ahead of the crack tip.
The Kop and DKth,eff values at threshold are summarized in Table 1 for the
three steels previously mentioned. In this table, both maximum and minimum
values of these data as derived from the described crack closure curves are
given. Important differences in the closure behavior are evident: The Kopvalues
are largest for the ferritic-pearlitic steel yet significantly smaller in the marten-
sitic steel. The austenitic steel shows some intermediate closure intensity. This
finding strongly correlates with the load ratio dependence of the threshold value
previously mentioned, which is strongest for the ferritic-pearlitic steel, but only
minimal for the martensitic one. Although more decisive, these results are in
general agreement with the majority of published data along with the earlier
findings on a wide range of steels and other iron-based alloys.
The differences in the closure behavior of the three steel microstructures are
due to differences in roughness and fretting oxide thicknesses. By detailed topo-
graphic analysis Böhm [21] has shown that the roughness of the fracture surfaces

FIG. 13—Crack closure evolution during near-threshold crack growth in a ferritic-

pearlitic steel.
is significantly smaller in the high strength martensitic steel due to its very fine
microstructure as compared to more pronounced roughness (including second-
ary cracking) in the ferrite-pearlite. A similar finding is reported for bainitic steel
[17]. Additionally, roughness induced closure is exaggerated in ferritic-pearlitic
steels through excessive fretting oxidation, which is pronounced in the near-
threshold growth regime and forms a dark crack front appearance on the fracture
surface. Despite its low strength level and larger grain structure, the austenitic
steel is characterized by a less pronounced closure effect and, therefore, a compa-
ratively weak R effect as compared to the ferritic-pearlitic steel. This is observed
because much less fretting oxidation occurs on the fracture surface of the austen-
itic steel even at low R and near-threshold loading.
Specific investigations were undertaken to clearly prove the influence of
roughness and fretting oxidation on both the closure intensity and the load ratio
dependence of the threshold. In such experiments, superior high threshold values
were measured for coarse grained precipitation hardened FeCu alloys after
underaging due to near-threshold growth along transgranular strain localization
paths, which resulted in extended transgranular fracture facets and large fracture
tortuosity. The extremely intensified crack closure gave rise to a threshold, which
occurred independent of R at Kmax,th ¼ 13.0 MPa m1=2 and Kop ¼ 10.0 MPa m1=2,
i.e., at an effective threshold of DK,th ¼ 3.0 MPa m1=2. The pronounced influence
of fretting oxidation on the low-R threshold has been shown in experiments on
the influence of the environmental humidity by Schlät, together with the author.
A significantly enhanced threshold at R ¼ 0.05 combined with intensified crack

SCHAPER, doi:10.1520/JAI103968 247
FIG. 14—Near-threshold crack growth in terms of nominal and effective cyclic stress
intensity.
closure was found for ferritic steels in water vapor saturated air. On the contrary,
both the crack closure and threshold proved to be reduced in dry hydrogen, dry
nitrogen, or under electrolytic H charging, as discussed earlier.
Crack Closure after Overloading

The role of crack closure during the initial crack growth in a compressive resid-
ual stress field has been analyzed on sharply notched specimens (root radius
q ¼ 0.12 mm) of a high strength low alloy steel after preloading, up to a stress in-
tensity KV ¼ 58 MPa m1=2 [10]. Due to the induced compressive residual stresses
at the notch root, crack initiation becomes more difficult. Therefore, an
enhanced fatigue limit is measured, which becomes higher with the increasing
preload. Simultaneously, an enlarged region of an initially reduced growth rate
is observed together with an enhanced closure level (up to three times the
steady-state value), which continuously diminishes as the crack grows through
the preload affected zone. A typical example for the development of growth and
closure behavior of the crack emanating from the notch root is depicted in
Fig. 15. It is seen that the crack accelerates more slowly after preloading and
reaches its steady-state value at an overall length, which is significantly larger
than that, which would be expected. The affected crack length is larger than the
calculated preload plastic zone size. Thus, a simple residual stress argument
based on plastic zone sizes does not account for the measured behavior. This is
clear proof for the effect of residual plasticity in the wake of the crack.
Overload experiments revealed a similar picture of crack growth retardation
combined with a changing crack closure behavior. In Fig. 16 an example for the

FIG. 15—Crack closure evolution during the initial crack growth from a notch at
DK ¼ 14 MPa m1=2 after overloading with KV ¼ 48 MPa m1=2, HSLA steel. The closure
curves are measured at successive growth stages.
retardation effect of a single tensile overload applied to a metastable austenitic

steel specimen is shown. With regard to the sensitivity of the measuring tech-
nique, it is noted that a slight increase in the incremental specimen compliance
during overload application due to plastic zone growth has been observed in
lower strength materials. Immediately after overload application, crack closure is
clearly suppressed, and sometimes this occurs completely. The initial elimination
of closure combined with some relaxation of the compressive residual stresses is
thought to be responsible for the initial delay of the retardation effect. After some
crack advance, a new crack closure effect develops at a Kop level, which is signifi-
cantly enhanced. During unloading the crack closes first only to the crack length
corresponding to that at the overload application. Further closure requires addi-
tional unloading and occurs below a second Kop, which is usually lower than the
steady state value before the overload application. Once the crack growth recov-
ers, the steady-state rate for both opening levels returns to that one which would
have prevailed in the absence of any sequence effect. The observed behavior is
completely in accordance with the picture of an initial short acceleration due to
crack tip blunting followed by a transitional strong retardation, which is primar-
ily due to the residual displacements left in the wake of the advancing crack.
Accordingly, experiments on steels with largely different strength levels
revealed that the overload affected crack length is not generally related to the
overload plastic zone size but could be much larger, especially in high strength
steels as observed by Böhm [21] and in several unpublished experiments of Schlät
and the author. Despite the markedly different strength level of the material, a
quantitatively similar retardation has been observed. At least in its later stage, the

SCHAPER, doi:10.1520/JAI103968 249
FIG. 16—Crack closure during fatigue crack growth following a 100 % tensile overload for
metastable austenitic steel. The closure curves are measured at successive growth stages.
intensity and extent of the retardation phenomenon cannot be accounted for by

internal stress state induced and plastic zone size related crack closure argu-
ments alone. Therefore, not only closure in the very vicinity of the crack tip but
also closure in the remote wake is to be considered as responsible for the crack
closure behavior and crack propagation kinetics following overloads.
The transitional intensification of crack closure after overloading is similar to
the observed resistance curve behavior of the crack propagation threshold found
at low R. After the elimination of any crack closure by intermittent compression
cycling, an increase in the threshold level during more than 1 mm crack growth at
R ¼ 0.05 has been observed for both the ferritic-pearlitic and the austenitic steels,
which obviously was due to a gradual build-up of crack closure. Accordingly, at a
low load ratio only limited crack advance is possible on cycling with a DK in
between DKth,eff and DKth. The existence of non-propagating cracks on free surfa-
ces and at a sharp notch root is a consequence of this finding, which substantiates
the experiments reported by Ward-Close and Ritchie [2] and Pippan et al. [3].
Crack Closure as Revealed by Ultrasonic Technique

During the crack initiation and initial crack growth an initially increasing differ-
ence between the crack lengths measured at maximum and minimum loads

FIG. 17—Fatigue crack growth from a notch as determined with the ultrasonic tech-
nique at max and minimum load for HSLA steel.
within a cycle is observed (Fig. 17). Obviously, this indicates the building-up of
crack closure. On the contrary, no differences in crack length and crack closure
behavior between the midsection and near-surface measurements could be
observed. The reason for this is ascribed to the smallness of the plane stress
region as compared to the sensor size. However, when comparing the ultrasonic
crack closure results with the crack closure curves measured with the compliance
techniques, a lower Kop is derived from the first technique. It is expected that this
is due to the fact that the ultrasonic measurements concentrate on the midsection
where the crack length is usually somewhat larger than its mean value.
Following a 100 % tensile overload, a basically similar overall crack growth
behavior is measured by both techniques. On the contrary, the difference
between the inner and near-surface crack lengths is increased. This occurs
because the plane stress region in the near-surface region of the specimen is now
significantly enlarged. Therefore, not only the mentioned crack length differen-
ces develop but now it also becomes possible to measure a locally different crack
closure behavior. Crack closure in the specimen midsection is observed at a sig-
nificantly lower Kop than in the near-surface region (Fig. 18), which indicates
that the crack closure effect is more pronounced in the plane stress near-surface
region as compared to the plane strain region in the inner part of the specimen.
Summary and Conclusions

From the results of our investigations the following conclusions can be drawn:
1. Whereas the ferritic-pearlitic steel exhibits a strong influence of the load
ratio on the threshold value, this effect is largely reduced in the investi-
gated austenitic and martensitic steels. A unique threshold of about
3 MPa m1=2 is measured at R ¼ 0.8.

SCHAPER, doi:10.1520/JAI103968 251
FIG. 18—Crack closure behaviour in the middle and near-surface region of a ferritic-
pearlitic HSLA steel.
2. The influence of loading conditions (load ratio and environment) is

tightly connected with the occurrence and intensity of crack closure
which, in turn, is largely dependent on microstructure of the material
(strength level, microstructural unit size, and oxidation tendency). How-
ever, despite the overall correlations that were found, these effects are,
in various cases, not quantitatively accounted for on the basis of the
DKeff approach which, as a simplified mechanistic interpretation,
ignores the gradual nature of crack closure. Thus, many inconsistencies
are related to the definition and measurement of Kop.
3. Crack closure is not a prerequisite for the existence of a threshold value,
i.e., there is a non-zero effective threshold for all of the materials investi-
gated. Furthermore, there is evidence that intrinsic mechanisms, which
determine the microstructural crack path (grain size effects and micro-
structural strain localization), along with local mean stress play a direct

role in determining the damage evolution at the fatigue crack tip and,
therefore, the crack growth resistance.
4. The Kop is roughly independent of crack length for DK ¼ constant cycling in
the Paris region. Correspondingly, the ratio Kop/Kmax is found to decrease at
R ¼ constant with increasing Kmax, i.e., the ratio U ¼ DKeff =DK increases.
5. A dual-type closure behavior develops after the pre-overloading of
notched samples and after intermittent overloads in crack propagation
tests. The lower opening point is completely eliminated following over-
loads near the general yield.
6. The dynamic compliance resonance vibration technique enables crack
propagation and crack closure measurements without separate crack
length or compliance measuring instrumentation. The technique proved
to be a versatile and highly sensitive means for analyzing the fatigue crack
behavior, even in aggressive environments and at high temperatures.
7. The developed ultrasonic technique uses a signal frequency of 1000 Hz.
Thus, it enables not only crack length measurements, but also crack closure
evaluation during fatigue experiments without interruption of a crack growth
test. Furthermore, some spatial resolution with respect to growth and closure
behavior under plane stress near-surface conditions as compared to the inner
plane strain region may be achieved. A disadvantage of this technique is the
time consuming and complicated evaluation of the diffraction patterns.
References
[1] McClung, R. C., “The Influence of Applied Stress, Crack Length, and Stress Inten-
sity Factor on Crack Closure,” Metall. Trans. A, Vol. 224, 1991, pp. 1559–1571.
[2] Ward-Close, C. M. and Ritchie, R. O., “Mechanics of Crack Closure,” ASTM STP
982, J. C. Newman, Jr. and W. Elber, Eds., ASTM International, West Consho-
hocken, PA, 1988, pp. 93–111.
[3] Pippan, R., Plöchl, L., and Klanner, F., “Threshold of Fatigue Growth,” Materialprü-
fung, Vol. 35, 1993, pp. 333–338.
[4] Tanaka, Y. and Soya, I., “Fracture Mechanics, Perspectives and Directions,” ASTM
STP 1020, R. P. Wei and R. P. Gangloff, Eds., ASTM International, West Consho-
hocken, PA, 1989, pp. 514–529.
[5] Cui, W. C., J. Mater. Sci. Technol., Vol. 7, 2002, pp. 43–56.
[6] Kujawski, D., “Parametric Study on the Variability of Open Load Determination,”
[7] Nowell, D., “Techniques for Experimental Measurement of Fatigue Crack Closure,”
Appl. Mech. Mater., Vols. 7–8, 2007, pp. 3–9.
[8] Chen, D. L., Weiss, B., and Stickler, R., “Contribution of the Cyclic Loading Portion
Below the Opening Load to Fatigue Crack Growth,” Mater. Sci. Eng., A, Vol. 208,
1996, pp. 181–187.
[9] Schlät, F., “Dynamic Compliance Measurement—A Proposed New and Efficient
Method to Investigate Crack Nucleation and Propagation Phenomena,” Int. J. Frac-
ture, Vol. 19, 1982, pp. R37–R40.
[10] Schlät, F. and Schaper, M., Publ. Tech. Univ. Miskolc, Vol. C39, 1983, pp. 173–187.
[11] Schlät, F. and Schaper, M., Publ. Tech. Univ. Miskolc, Vol. C38, 1983, pp. 157–173.
[12] Schaper, M., Böhm, A., Schlät, F., and Tkatch, A., 10th Congress of Materials Test-
ing, E. Csoboly, Ed., Vol. 2, GTE, Budapest, 1991, pp. 556–571.

SCHAPER, doi:10.1520/JAI103968 253
[13] Schaper, M., and Böhm, A., “ECF 10, Structural Integrity: Experiments, Models
and Applications,” Proceedings of the 10th European Congress on Fracture, K. H.
Schwalbe and C. Berger, Eds., EMAS, 1994, pp. 1451–1461.
[14] Che, M. C., 1996, Ph.D. thesis, Univ. of Technology, Dresden.
[15] Sarma, V. S., Jaeger, G., and Koethe, A., “On the Comparison of Crack Closure
Evaluation Using Dynamic and Static Compliance Measurements,” Int. J. Fatigue,
Vol. 23, 2001, pp. 741–745.
[16] Sarma, V. S., Padmanabhan, K. A., Jaeger, G., Koethe, A., and Schaper, M., “On
the Fatigue Threshold Behaviour of Two Ferrite-Pearlite Microalloyed Steels,” Z.
für Metallkd., Vol. 91, 2001, pp. 581–584.
[17] Sankaran, S., Sarma, V. S., Padmanabhan, K. A., Jaeger, G., and Koethe, A., “High
Cycle Fatigue Behaviour of a Multiphase Microalloyed Medium Carbon Steel: A
Comparison Between Ferrite—Pearlite and Tempered Martensite Micro-
structures,” Mater. Sci. Eng., A, Vol. 362, 2003, pp. 249–256.
[18] ASTM E647-11, 2011, “Standard Test Method for Measurement of Fatigue Crack
Growth Rate,” Annual Book of ASTM Standards, Vol. 03.01, ASTM International,
[19] McEvily, A. J. and Yang, Z., “The Nature of the Two Opening Levels Following
an Overload in Fatigue Crack Growth,” Metall. Trans. A, Vol. 21A, 1990, pp. 2717–2727.
[20] Liaw, P. K., Leax, T. R., and Donald, J. K., “Fracture Mechanics, Perspectives and
Directions,” ASTM STP 1020, R. P. Wei and R. P. Gangloff, Eds., ASTM Interna-
tional, West Conshohocken, PA, 1989, pp. 581–604.
[21] Böhm, A., 2003, Ph.D. thesis, Univ. of Technology Dresden.
In Memoriam
With deep sadness we learned that our dear colleague Michael Schaper, materials
scientist at the University of Technology Dresden, Germany, passed away on Janu-
ary 10th 2012. His scientific career focused on researching the physical principles of
fracture mechanics, its industrial application, as well as advancing the academic
education of young scientists. Michael Schaper was awarded his diploma in physics
from the Ernst-Moritz-Arndt University Greifswald, Germany, in 1966. In 1971 he
gained his PhD degree from the former GDR Academy of Sciences for investiga-
tions on the plastic behaviour of pure bcc metals at low temperature. From 1972 to
1993 he conducted research on the strength and fracture behaviour of iron basic
alloys and steels at the Central Institute for Solid State Physics and Materials Scien-
ces Dresden. From 1992 to 1993 he worked as a visiting professor for materials sci-
ence at the University of Kassel. Between 1994 and 2007 he held the professorship
for materials reliability at the University of Technology Dresden. Michael Schaper’s
scientifi c interest was focused on the development of new physics-based methods
which allow a broad characterisation of the processes near the crack tip of dynami-
cally loaded ferritic, martensitic, and austenitic steels and also include environmen-
tal effects. This approach led to a comprehensive understanding of threshold stress
intensity and near-threshold crack propagation phenomena. Michael Schaper was
an admired and respected colleague who will be greatly missed. We would like to
convey our deepest condolences and sympathy to his family and friends.
Dresden, Wolfgang Pompe,
January 2012 Hartmut Worch


doi:10.1520/JAI104215
Benjamin S. Adair,1 W. Steven Johnson,2

Stephen D. Antolovich,2 and Alexander Staroselsky3
Temperature and Load Interaction Effects on

the Fatigue Crack Growth Rate and Fracture
Surface Morphology of IN100 Superalloy
ABSTRACT: A study was conducted to explore some of the load and

temperature interaction effects on the fatigue crack growth rate (FCGR) of
polycrystalline superalloy IN100. Load interaction testing in the form of single
overloads was performed at 316 C and 649 C. Temperature interaction test-
ing was performed by cycling between 316 C and 649 C in blocks of 1, 10,
and 100 cycles. After compiling a database of constant temperature, con-
stant amplitude FCGR for IN100, fatigue crack growth predictions assuming
no load or temperature interactions were made. Experimental fatigue crack
propagation data were then compared with these predictions to assess inter-
action effects. The fracture mechanisms observed during interaction testing
using a scanning electron microscope were compared with the mechanisms
present during constant temperature, constant amplitude testing. Overload
interaction testing led to full crack retardation at 2.0 overloads for both
316 C and 649 C testing. Overloading by 1.6 at both temperatures led
to retarded crack growth, whereas 1.3 overloads at 649 C created
Manuscript received July 16, 2011; accepted for publication February 1, 2012; published
online April 2012.
1
Graduate Research Assistant in the George W. Woodruff School of Mechanical
Engineering, Georgia Institute of Technology, Atlanta, GA 30332 (Corresponding
author), e-mail: benadair@gatech.edu
2
Professor, School of Materials Science and Engineering and George W. Woodruff
School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA 30332.
3
Staff Engineer, Pratt & Whitney, East Hartford, CT 06108.
in Anaheim, CA.
Cite as: Adair, B. S., Johnson, W. S., Antolovich, S. D. and Staroselsky, A., “Temperature
and Load Interaction Effects on the Fatigue Crack Growth Rate and Fracture Surface
Morphology of IN100 Superalloy,” J. ASTM Intl., Vol. 9, No. 4. doi:10.1520/JAI104215.
Copyright V
254

ADAIR ET AL., doi:10.1520/JAI104215 255
accelerated crack growth and at 316 C the crack growth was retarded. One
block alternating temperature interaction testing grew significantly faster than
the non-interaction prediction, while 10 block alternating temperature interac-
tion testing also grew faster but not to the same extent. One hundred block
alternating testing grew slower than non-interaction predictions. Possible
explanations for the interaction effects responsible for the observed crack
growth acceleration and retardation are discussed.
KEYWORDS: IN100 superalloy, fatigue crack growth rate, thermo-mechani-
cal fatigue, fractography, load interactions, temperature interactions, fracture
mechanisms
Introduction
There exists a desire to design advanced aircraft engine components to operate
at higher temperatures and stresses in order to achieve higher thrust, increased
efficiency, and lower pollution. This creates a hostile environment for the tur-
bine engine components that are subjected to large numbers of these thermo-
mechanical loading cycles. The service life of structural components is governed
by different modes of degradation and failure such as fatigue, excessive defor-
mation, yielding, creep, corrosion, and so on. One dominant mode of failure is
due to fatigue, and structural components have to be designed so that they can
adequately endure the fatigue loading during their service life. Therefore, an
understanding of fatigue interactions is needed in order to reliably predict the
lifetime of aircraft engine components and, in particular, turbine disks. Fatigue
life estimations for such components are typically performed by assuming a
flaw size and determining remaining life based on crack growth predictions.
The initial flaw may correspond to the smallest detectable flaw size, or to an ini-
tial quality defect such as an inclusion cluster. The crack growth rate and subse-
quently the remaining component life are defined by the methods of fracture
mechanics. The fatigue crack propagation strongly depends on the loading and
temperature spectrum as well as of materials properties. The model predictions
are to be calibrated against coupon test data and verified by comparison with
known field experience. Thus, it is very important to develop the test methods
that mimic actual turbine conditions. The goal of this paper is to present funda-
mental thermo-mechanical fatigue (TMF) crack growth data that identifies
some temperature and load interaction effects that need to be included in any
successful TMF crack growth life prediction model.
Numerous fatigue crack growth studies on a variety of materials have
reported on the underlying mechanisms responsible for load interaction effects,
as illustrated in extensive literature reviews [1–3]. In particular there have been
several studies on the effect of overloads on the fatigue crack growth rate in
superalloys [4–8]. When it comes to temperature interaction effects most work
has been concerned with using two types of idealized TMF cycles: in-phase (IP)
and out-of-phase (OP) [9–11]. Studies using nontraditional types of TMF
cycles are of particular interest to this research. For example, Cailletaud and
Chaboche used block temperature changes to look at the cyclic viscoplastic

behavior of IN100 [12]. Presently there is no known published literature that

investigates the interaction effect of alternating temperature blocks on fatigue
crack propagation in superalloys.
This investigation focuses on the load and temperature interaction effects
on the fatigue crack growth of IN100. Load interaction testing in the form of
repeated single 1.3 , 1.6 , and 2.0 overloads was performed at 316 C and
649 C at a frequency of 0.33 Hz. Constant amplitude, load controlled tempera-
ture interaction testing was performed by cycling between 316 C and 649 C in
blocks of 1, 10, and 100 cycles at a loading frequency of 0.33 Hz. These speci-
mens were then analyzed under a scanning electron microscope (SEM) to deter-
mine crack growth mechanisms.
Crack Growth Mechanisms and Modeling

The thermo-mechanical loading cycles that turbine disks are exposed to are
composed of many different stress and temperature ranges. In these compli-
cated spectra the prior loading and temperature history plays a major role in
determining the current crack growth rate. Understanding these complex inter-
actions is very important to correctly predict crack growth whether using a
crack closure or yield zone model.
Both models are concerned with crack tip plasticity that occurs when the
yield stress of the material is exceeded [3]. The difference is that crack closure
is concerned with the size of the plastic wake behind the crack tip and the yield
zone approach is concerned with the plastic yield zone size in front of the
crack tip. In both of these cases the plastic zone size is a direct function of
the stress at the crack tip and material yield stress. In turn the yield stress of
the material is a function of the temperature. In addition to crack tip plastic-
ity, superalloy crack growth is very dependent on changes in the precipitate
microstructure and oxygen diffusion at the crack tip. At temperature, oxygen
diffusion can occur very rapidly due to the highly stressed state of the crack
tip. Microstructural precipitate evolution also occurs at temperature, albeit
more slowly than diffusion. Load and temperature interaction effects are cre-
ated when all these factors come into play during TMF cycling. These load
and temperature interaction effects on the fatigue crack growth can be sepa-
rated into two different categories; crack growth acceleration or crack growth
retardation.
Acceleration is when there is more crack growth for a given cycle than
would be produced under isothermal constant amplitude conditions. Load
induced acceleration occurs when an overload is applied after baseline cy-
cling, due to differences in relative plastic zone size. Acceleration resulting
from temperature can occur during low temperature cycling after cycling at
a higher temperature causes a thermally affected zone due to oxygen
diffusion.
Retardation is when there is less crack growth for a given cycle than would
be produced under isothermal constant amplitude conditions. Retardation
resulting from loading occurs after an overload is applied and the subsequent
growth of the baseline cycling is hindered, due to differences in relative plastic

zone size. Temperature induced retardation can occur during low temperature
cycling after cycling at a higher temperature causes changes in the precipitate
microstructure.
The following discussed test results will try to identify conditions and mech-
anisms that will cause either crack growth acceleration or retardation in the
IN100 superalloy.
Experimental Method
Materials
IN100 is a powder metallurgy (P/M) superalloy, developed in the early 1960s,
commonly used for components, such as turbine disks, spacers, and seals,
operating at intermediate temperatures of 300 C to 700 C [13]. In order to
most accurately assess fatigue crack growth rates for engineering components
the specimens used in this study were cut radially from a jet engine turbine
disk, Fig. 1. This disk was heat treated with a solutioning treatment of 1143 C,
(below the gamma prime solvus temperature of 1192 C) then cooled and oil
quenched. Solutioning was followed by a two step aging heat treatment first at
982 C for an hour then forced air cooled, then held at 732 C for 8 h then air
cooled. The chemical composition of the IN100 disk evaluated can be seen in
Table 1.
The two principal phases in IN100 are the c0 , consisting of primary, second-
ary and tertiary precipitates and the solid solution c matrix; creating a relative
volume fraction of 60:40 as seen in Fig. 2. As can be seen in Fig. 2 the fine IN100
grain size is approximately 4 lm with primary c0 sizes of 2 lm. The grain size
was quantified using the Average Grain Intercept (AGI) Method.
Specimen Design
For this investigation the Single Edge Notch Tension (SENT) specimen configu-
ration was chosen due to its ease of crack measurement and ability to be
gripped with water cooled hydraulic wedge grips. This water cooled gripping
arrangement allowed for more rapid cooling than a pin loaded arrangement.
The test specimen size was chosen to be 203.2 mm long, 38.1 mm wide, and
2.54 mm thick, as can be seen in Fig. 1. Most fatigue crack growth data found in
literature is for specimens with a significantly larger thickness of 6.35 mm and
thicker, with very little for specimens as thin as 2.54 mm [14]. For quick fatigue
crack initiation, electro-discharge machining was used to create an 11.4 mm
long notch with a 0.152 mm root radius.
Experimental Procedures
Fatigue crack growth testing was done in accordance with ASTM Standard
E647 [15]. A 100 kN capacity servo-hydraulic load frame was used to apply con-
stant amplitude sinusoidal fatigue cycles. Specimens were gripped approxi-
mately 50.8 mm on each end by pyramid teeth wedge grips 44.5 mm wide.

FIG. 1—Single Edge Notch Tension (SENT) specimen dimensions and specimen orien-
tation relative to the disk from which the specimens were cut.
Nominal grip pressure was set at 16,550 kPa. Specimens were precracked at 20 Hz
at the lowest temperature at which they were tested (either 316 C or 649 C).
A 3.5 kW induction heater along with a K type thermocouple was used to
maintain temperatures ranging from 316 C to 649 C. Shown in Fig. 3 an unique
coil design made out of 4.8 mm copper tubing had an 11.4 mm tall crack view-
ing window with 3 turns above and below to provide uniform temperature
TABLE 1—Chemical composition (in % wt.) of IN100 disk evaluated.
Al B C Co Cr Mo Ti V Zr Ni
IN100 4.90 0.02 0.07 18.20 12.10 3.22 4.20 0.70 0.07 56.52

FIG. 2—IN100 microstructure. Note the combination of large primary c0 particles and c
grains.
FIG. 3—Specimen in test rig showing unique coil design and cooling fans.

distribution. Temperature measurement using an optical pyrometer showed a

temperature variation of no more than 6 5 C across the crack plane. The wedge
grips were water cooled. Also shown in Fig. 3, twin fans mounted on either side
of the specimen were used to control cooling.
Crack length measurement was accomplished optically using a QM-100,
Questar Microscope mounted on a three-axis Remote Measurement System
platform with linear encoders, shown in Fig. 4. This microscope provides mag-
nification between 50 and 300 for optimal crack length measurement. An
Edmund Optics USB 2.0 charge-coupled device (CCD) camera with the ability
to take still images and record video was attached to the microscope. To aid in
crack length resolution, laser etchings spaced every 0.254 mm on the surface of
the specimens normal to the direction of crack growth were implemented.
FIG. 4—TMF test rig with MTS load frame and Questar Microscope visible.

Data Analysis
For each test, crack length and number of cycles was recorded approximately
every 500 cycles. This cycle interval was flexible, being larger at small DK’s and
decreasing as DK increased. Using wedge grips the specimens were essentially
loaded in a uniform displacement condition [16]. The stress intensity factor so-
lution used for the uniform displacement SENT specimen is as follows [17]:
a a 2 a 3 a 4
pffiffiffiffiffiffi
K ¼ S pa 1:126 0:504 þ 10:473 48:17 þ 112:87
W W W W
a 5 a 6
1:24:63 þ 5:327 (1)
W W
where:
S is the far field stress and
W is the specimen width.
This solution is for a specimen height to width ratio of 1.33. The specimen
height between the grips was 50.8 mm.
Because the Single Edge Notch specimen that was used in this research was
not the same as that commonly found in ASTM standards we needed to deter-
mine the geometric correction factors for our geometry. The ASTM SEN speci-
men is pin loaded but we wanted to use water cooled wedge grips to allow for
quicker heat dissipation upon cooling. There is, of course, a very big difference
in boundary conditions between the fixed wedge gripes and a pin loading. Prof.
Jim Newman, Jr. of Mississippi State University did the calculations of the
Stress Intensity geometric correction factors for us, using our unique width to
length between the grips ratio.

The results of load and temperature interaction testing will be presented and
discussed in this section. All interaction testing was compared against non-
interaction model predictions to provide a measure of acceleration and retarda-
tion effects. Scanning electron microscopy was used to observe the crack
growth mechanisms in the presence of load and temperature interactions.
Non-Interaction Model
A simple Paris-type numerical model, using isothermal constant amplitude test
data, was created to calculate crack growth cycle by cycle using a straight accu-
mulation of da/dN based upon the current DK assuming no interaction between
applied stresses or temperatures. The Paris constants were derived from DK ver-
sus da/dN data that was developed at four temperatures (22, 316, 482, and
649 C) at a frequency of 0.33 Hz and at R ¼ 0.1 [17]. The data and the associated
Paris constants are shown in Fig. 5. All isothermal constant amplitude test data
was acquired in the Paris regime. For this reason all interaction testing was also
performed in the Paris regime. Comparison of the crack growth prediction

FIG. 5—Paris equation fitted fatigue crack growth rates for temperatures ranging from
22 C to 649 C at a frequency of 0.33 Hz and R ratio of 0.1.
using the non-interaction model to actual experimental data will provide a mea-
sure of acceleration or retardation that can be attributed to load and tempera-
ture interactions.
Load Non-Interaction Modeling

The non-interaction modeling was used to predict the effect of applying 1.3 ,
1.6 , and 2.0 overloads every 800 cycles at 316 C and 649 C. The 316 C pre-
dictions are shown in Fig. 6. For reference, the constant amplitude (no over-
load) crack growth at 316 C is also shown. The predicted crack growth curves
are all very close together. Upon closer inspection in Fig. 6, the crack growth is
slightly faster due to the overloads, with crack growth rate increasing as the size
of the overload is increased as would be expected from a non-interaction model.

FIG. 6—Non-interaction model prediction for 1.3 , 1.6 , and 2.0 overloads applied
at 316 C every 800 cycles.
This same trend was seen in the 649 C non-interaction overload prediction and
as such no figure is shown for these estimates.
Load Interaction Testing

Load interaction testing in the form of single overloads was performed using
two specimens, with one tested at 316 C and the other at 649 C. Both specimens
were tested at 0.33 Hz and R ratio of 0.1. Yield zone calculations showed that it
would take approximately 800 cycles for the crack to grow out of a
1.6 overload monotonic plastic zone. Using this calculation, overloads of
2.0 , 1.6 , and 1.3 were applied every 800 cycles. It was anticipated that this
would allow the observation of fully retarded crack growth for 2.0 overloads,
crack growth that just exits the zone of influence for 1.6 overloads, and crack
growth that quickly exits the zone of influence for 1.3 overloads. Overloads
were applied in order of decreasing size to maximize specimen life by delaying
fast fracture. Applying overloads in this order it was expected that there would
be significant load interaction effects, but due to a limited number of specimens
this was unavoidable.
After precracking both specimens, 2.0 overloads were applied every 800
cycles for 125,000 cycles as denoted by point A, seen in Fig. 7. It was found with
the aid of fractography that 2.0 overload interaction testing led to full crack
retardation of the 1.0 cycles for both 316 C and 649 C testing. The only
growth seen during 2.0 overload testing was attributed to growth from the
overload cycles.

FIG. 7—Experimental load interaction data for 1.3 , 1.6 , and 2.0 overloads applied
every 800 cycles.
At point A the overload ratio was changed to 1.6 , knowing that the
2.0 overload zone would influence the initial crack growth rate. Between point
A and point B the 2.0 overload zone was still retarding most of the 1.0 cycles
and probably retarding the 1.6 overload cycles to some extent. At point B the
crack emerged from the 2.0 overload zone of influence. This 1.6 overload re-
tarded crack growth rate was faster than the 2.0 overload testing. This faster
growth rate can be attributed to full growth of some of the 1.0 (baseline) cycles
that are applied after the 1.6 overload zone of influence is cracked through.
At point C the 1.6 overloads were replaced with 1.3 overloads. The 649 C
1.3 overloads quickly grew out of the 1.6 zone of influence, whereas the
316 C 1.3 overloads remained affected for about another 10,000 cycles. It was
found that 1.3 overloads at 649 C created accelerated crack growth when
compared with the non-interaction prediction at the same temperature. This is
a result of the overloads growing much more than they typically would in a con-
stant amplitude situation, plus only a little retardation takes place, so most of
the 1.0 cycles also contribute to growth. This low amount of retardation is a
result of a faster da/dN, due to increased crack tip embrittlement at higher tem-
peratures, which allows the crack growth through the 1.3 overload plastic
zone to be much faster. However, the 1.3 overloads at 316 C were found to
retard the crack growth rate when compared with the non-interaction predic-
tion at the same temperature. The importance of being able to properly model
load interactions can clearly be seen in Fig. 7, where the 316 C and 649 C non-
interaction models severely under predict the total fatigue life.

During the course of overload testing an interesting multiple overload crack

closure phenomenon was observed. A visualization of this process can be seen
in Fig. 8. Precracking was performed with 1.0 (baseline) cycles until a crack
length of 2.54 mm was obtained. This is illustrated by the region between point
1 and point 2 in Fig. 8. Immediately following the precracking the crack growth
rate was accelerated due to the application of the first 2.0 overload; after this
first overload the crack growth rate was found to steadily decrease upon appli-
cation of further overload cycles. This phenomenon can be attributed to the sta-
bilized Kopening created during the 1.0 precracking, shown as KA in Fig. 8.
Upon additional application of overload cycles Kopening slowly increased leading
to a decreasing DKeff. Once the 2.0 overload crack growth stabilized with a
Kopening of KB, illustrated as point 3, da/dN then continued to increase with
increasing DK. Perhaps this phenomenon is more pronounced at higher over-
loads such as in the 2.0 overload test sincebecause all 1.0 cycles quickly
enter full retardation and the subsequent 2.0 overloads acted like multiple
overloads, as depicted in Fig. 8.
Fractographs of 316 C Specimen

Overload testing was performed on a specimen at 316 C, at an R ratio of 0.1 and
a frequency of 0.33 Hz. Three different overloads in sequences of 2.0 , 1.6 ,
and 1.3 were applied every 800 cycles to this specimen. SEM fractographs for
this specimen are shown in Figs. 9 through 11.
In Fig. 9, showing 2.0 overload cycling, it can be seen that the predomi-
nant mechanism for fatigue crack growth was ductile rupture caused by the
2.0 cycles. Each band in the micrograph corresponds to the application of a
FIG. 8—Variation in crack closure stress intensity factor with variation in load level.

FIG. 9—SEM micrograph of specimen tested at 316 C and 0.33 Hz showing 2.0 overload
cycles. Crack growth was found to only occur due to the overload cycles.
2.0 overload while the 800 baseline cycles between overloads produced no
crack growth. This ductile rupture caused by overload cycles continues into the
1.6 overload cycles, seen in Fig. 10. At 1.6 overload testing the crack growth
was predominantly due to the overload cycles with a little growth from the
1.0 cycles. The crack growth from each 1.6 overload cycle was found to be
larger than predicted with the non-interaction model due to acceleration but
the 800 baseline cycles were severely retarded for overall retarded crack growth
when compared with the non-interaction model prediction.
Figure 11 shows the 1.3 overload fracture surface at 316 C. For relatively
low temperatures oxidation is minimal in superalloys. As such, the failure
mechanism is essentially transgranular in nature being a mixture of fatigue
striations and normal rupture. As a result of the lack of environmentally
enhanced crack tip embrittlement at 316 C the crack growth acceleration due to
the 1.3 overloads was overshadowed by the retardation of the 800 baseline
cycles for an overall crack growth rate that was slightly slower than what is seen
during constant amplitude testing.
Fractographs of 649 C Specimen

Next the test performed at a temperature of 649 C, at an R ratio of 0.1 and a fre-
quency of 0.33 Hz will be examined. Three different overloads consisting of 2.0 ,
1.6 , and 1.3 cycles in that order were applied every 800 cycles to this speci-
men. SEM fractographs for this specimen can be seen in Figs. 12 through 14.

and 1.0 cycles. Notice the voids caused by areas of incomplete consolidation.
FIG. 11—SEM micrograph of specimen tested at 316 C and 0.33 Hz showing

1.3 overload and 1.0 cycles. Transgranular fracture surface showing patches of fa-
tigue striations and other less defined features.


2.0 overload cycles. Notice the additional intergranular effect of increased temperature
in comparison to the 316 C test.
The effect of increased temperature can clearly be seen in Fig. 12 where

the 2.0 overload fracture surface is composed of ductile rupture with an
intergranular component. At 649 C more crack tip plasticity that would tend
to retard crack growth is counteracted by the higher crack growth rate caused
by increased crack tip embrittlement due to oxidation. This results in a crack
growth rate that is very similar to the 2.0 overload rate seen at the lower
temperature of 316 C. As in the 316 C case the crack grew only during
the 2.0 overload cycles and this ductile rupture fracture surface is indicative
of that.
The 1.6 overload cycles can clearly be seen in Fig. 13. This initial 1.6
overload crack growth was retarded by the overload plastic zones created by the
2.0 testing. As a result of this the 1.6 crack growth was initially only caused
by the 1.6 overloads. Once the 1.6 crack growth was out of the influence of
the 2.0 overload zone the growth became a mix of 1.6 and 1.0 cycles.
Taken near the end of 1.6 overload cycling, Fig. 13 shows clearly the intergra-
nular crack growth from the 1.0 cycles and the intergranular and ductile rup-
ture crack growth due to the 1.6 cycles.
Figure 14 reveals the predominantly intergranular fracture present during
1.3 overload cycling. The 1.3 overload cycles were not large enough to create
ductile rupture but were large enough to accelerate the crack growth. Notice the
grain boundary tearing normal to the direction of crack growth, indicative of
fast crack growth at higher DK’s.


1.6 overload and 1.0 cycles. The mix of ductile rupture due to overload cycles and
intergranular fatigue crack growth from the 1.0 cycles can clearly be seen.
and 1.0 cycles. Notice the grain boundary tearing normal to crack growth.

Temperature Non-Interaction Modeling

In order to assess the effect of changing temperature on crack growth under
constant amplitude loading cycles, non-interaction modeling for the case of
changing temperature between 316 C and 649 C was carried out. In particular
the effect of cycling between 316 C and 649 C every 1, 10, and 100 cycles was
investigated. The results can be seen in Fig. 15. Also shown in the figure is iso-
thermal crack growth data for the 316 C and 649 C experiments, from which
the Paris constants were extracted. As can be seen in Fig. 15, when temperature
interaction effects are ignored the 1, 10, and 100 alternating cycle block growth
rates are the same and fall between the isothermal crack growth data.
Temperature Interaction Testing

Turbine disks see significant fluctuations in temperature during service. In
order to quantify the effect that higher temperature crack growth has on lower
temperature crack growth, temperature interaction testing was carried out. It is
felt that the environmental and other associated thermal effects on IN100 at
316 C are minimal. However, there is a time dependent effect at 649 C: oxida-
tion and material evolution can take place rather rapidly at the crack tip due the
high temperature and stress state. These tests are aimed at assessing how the
cyclic time at 649 C affects the subsequent crack growth behavior at 316 C.
Temperature interaction testing was performed on one specimen by cycling
between 316 C and 649 C in blocks of 1, 10, and 100 cycles at 0.33 Hz and
R ¼ 0.1. While heating and cooling the specimen between temperature blocks
FIG. 15—Non-interaction model prediction for 1, 10, and 100 cycle blocks alternating
between 316 C and 649 C.

the load was held at the minimum cycle load. After precracking, 10 cycle alter-
nating temperature blocks were applied for a total of 17,500 cycles. When 10
cycle testing was finished, 100 cycle alternating temperature blocks were
applied for 16,500 cycles. Alternating temperature every cycle was performed
last for 2300 cycles because it would take fewer cycles to get a decent amount of
crack growth data due to DK being fairly large.
Alternating the temperature every cycle caused the fatigue crack growth
rate to be substantially faster than the non-interaction prediction as seen in
Fig. 16. Ten block alternating temperature interaction testing also grew faster
than the non-interaction prediction but not as rapidly as the 1 block alternat-
ing test. This can be seen in Fig. 17. However, 100 block alternating testing,
shown in Fig. 18, showed a different trend; the crack grew slower than the
non-interaction prediction.
Although the principal goal of this work was to demonstrate possible inter-
action effects, some potential explanations may be offered based on possible
microstructural changes, changes in the deformation mechanism and oxide-
induced closure. For alternating temperature every cycle, it was shown via SEM
fractography that crack growth at 316 C was accelerated due to crack tip
embrittlement caused by cycling at 649 C. One could hypothesize that a ther-
mally affected zone (TAZ) caused by oxygen diffusion but without macroscopic
oxide formation was easily cracked at 316 C, thus accelerating crack growth. As
the number of cycles at each temperature was increased to 10, the crack growth
rate at 316 C was again accelerated, however not to the extent that it was for
FIG. 16—Experimental data and non-interaction model prediction for 1 cycle tempera-
ture interactions between 316 C and 649 C. The experimental data grew much faster
than predicted by ignoring interaction effects.

FIG. 17—Experimental data and non-interaction model prediction for 10 cycle temper-
ature interactions between 316 C and 649 C. The experimental data grew slightly faster
FIG. 18—Experimental data and non-interaction model prediction for 100 cycle tem-
perature interactions between 316 C and 649 C. The experimental data grew slower

changing the temperature every cycle. In fact for this case, the crack growth
rate was only slightly faster than what was predicted by the non-interaction
model. It would thus appear that a counter veiling mechanism (or mechanisms)
was being introduced. This view is strengthened by examining the crack growth
rate when the temperature was changed every 100 cycles. In this case the crack
growth rate was even lower, and less than predicted by the non-interaction
model. The continuous decrease in the crack growth rate with increasing num-
ber of cycles at temperature could be due to (a) changes in the precipitate
microstructure (i.e., coarsening due to more time at temperature) and attendant
changes in the deformation mechanism and/or (b) retardation due to oxide-
induced closure. For case (a) it is possible that some coarsening of the structure
produced a more damage tolerant microstructure. For case (b) a thicker oxide
layer would give more closure and a slower crack rate. Both of these potential
mechanisms operate so as to reduce the crack growth rate. Although both
mechanisms appear to be reasonable explanations, more work needs to be done
to determine the validity and the magnitude of these effects.
Fractographs of Temperature Interaction Specimen

The effect of 649 C crack growth on the 316 C crack growth for the 1 and 10
cycle block tests could clearly be seen, as more intergranular crack growth was
FIG. 19—SEM micrograph of temperature interaction specimen at a DK of 32 MPaHm

in the 649 C crack growth zone. Notice the highly intergranular fatigue surface formed
during 649 C cycling.

present at 316 C than found in isothermal testing. SEM fractographs for the
alternating 100 cycle test can be seen in Figs. 19 through 21. In looking at
Figs. 19 and 20 it can be observed that the crack growth at 316 C was unmis-
takably affected by prior crack growth at 649 C. The 316 C fatigue surface is
more intergranular than what is normally found under isothermal conditions.
This phenomenon can still be seen even as DK increases, which would tend to
promote an environment that is more favorable to transgranular fatigue crack
growth.
From Fig. 21, it is seen that there is a roughly 10 size ratio between the
width of the large band and the width of the small band. The large band is fa-
tigue crack growth attributed to 649 C while the dark smaller band can be
attributed to 316 C fatigue crack growth. From isothermal testing it is known
that the da/dN ratio between those two temperatures at any given DK is
approximately 4 with 649 C isothermal testing having the larger da/dN. Know-
ing that the alternating 100 cycle testing grew slower than what was predicted
by the non-interaction model, it can be surmised that the 316 C crack growth
was 2.5 smaller than the same 100 cycles at a constant temperature. The
316 C cycles quickly grew through the TAZ created at 649 C but the growth
was then slowed by changes in the precipitate microstructure and attendant
changes in the deformation mechanism and/or retardation due to oxide-
induced closure.
FIG. 20—SEM micrograph of temperature interaction specimen at a DK of 32 MPaHm

in the 316 C crack growth zone. Notice the intergranular component to the fatigue sur-
face formed during 316 C cycling.

FIG. 21—SEM micrograph of temperature interaction specimen at a DK of 32 MPaHm.

The crack growth during the 649 C cycling was found to be 10 times larger than the
crack growth at 316 C cycling for the 100 alternating cycle temperature interaction test.
Conclusions
The purpose of this research was to investigate the load and temperature inter-
action effects on the fatigue crack growth rate (FCGR) of polycrystalline super-
alloy IN100. Load interaction testing in the form of 1.3 , 1.6 , and
2.0 single overloads with 800 baseline cycles in between was performed at
316 C and 649 C. Temperature interaction testing was performed by cycling
between 316 C and 649 C in alternating blocks of 1, 10, and 100 cycles. All test-
ing was performed at 0.33 Hz and an R ratio of 0.1. Experimental results were
compared to non-interaction crack growth predictions to determine first order
interaction effects. The fracture surfaces were then examined using scanning
electron microscopy to better understand the crack surface morphology and
determine crack growth mechanisms. The primary conclusions from this work
are as follows:
Overload interaction testing led to full crack retardation at 2.0 overloads
for both 316 C and 649 C testing. The only growth seen during
2.0 overload testing was attributed to growth from the overload cycles.
1.6 overloading at both 316 C and 649 C led to retarded crack growth
that consisted of growth from both the 1.6 and 1.0 cycles.
It was found that 1.3 overloads at 649 C created accelerated crack
growth when compared with constant amplitude data at the same tem-
perature. The 1.3 overloads at 316 C were found to minimally retard

the crack growth when compared with constant amplitude data at the
same temperature.
During alternating temperature cycling it was shown that at 1 and 10
alternating cycles, crack growth at 316 C was accelerated due to crack
tip embrittlement caused by 649 C cycling. At 100 alternating cycles the
crack tip propagating at 316 C quickly grew through the thermally
affected zone but then grew slower than expected.
It was postulated that the decrease in the crack growth rate with
increasing number of cycles at temperature could be due to changes in
the precipitate microstructure and attendant changes in the deforma-
tion mechanism and/or retardation due to oxide-induced closure.
There is significant impact in changing temperature on the resulting
crack growth for IN100. In this case, the time at high temperature had
substantial influence on the lower temperature fatigue crack growth
rate. This must be accounted for in any TMF crack growth life predic-
tion model.
There are significant load interaction (both retardation and accelera-
tions) effects present in IN100 under TMF conditions. This also must be
accounted for in TMF crack growth prediction models
The above conclusions indicate that accounting for fatigue behavior
where loads and temperatures are changing simultaneously cannot be
done using a simple additive approach. Instead, our results indicate
that a physics-based approach in which true mechanism interactions is
required for life prediction.
Acknowledgments
The writers would like to extend their gratitude to Pratt & Whitney for funding
this research and also to Dr. James C. Newman, Jr. of Mississippi State Univer-
sity for his help with the SENT boundary element solution.
References
[1] Skorupa, M., “Load Interaction Effects During Fatigue Crack Growth under Vari-
able Amplitude Loading—A Literature Review. Part I: Empirical Trends,” Fatigue
Fract. Eng. Mater. Struct., Vol. 21, 1998, pp. 987–1006.
[2] Skorupa, M., “Load Interaction Effects During Fatigue Crack Growth under Vari-
able Amplitude Loading—A Literature Review. Part II: Qualitative Interpretation,”
Fatigue Fract. Eng. Mater. Struct., Vol. 22, 1999, pp. 905–926.
[3] Chang, J. B. and Hudson, C. M., “Methods and Models for Predicting Fatigue Crack
Growth Under Random Loading,” ASTM Spec. Tech. Publ., 748, 1981.
[4] Larsen, J. M., Rosenberger, A. H., Hartman, G. A., Russ, S. M., and John, R., The
Role of Spectrum Loading in Damage-Tolerance Life-Management of Fracture
Critical Turbine Engine Components, Defense Technical Information Center, 2003,
http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADP014134.
[5] Macha, D. E., “Fatigue Crack Growth Retardation Behavior of IN-100 at Elevated
Temperature,” Eng. Fract. Mech., Vol. 12, 1979, pp. 1–11.

[6] Larsen, J. M., Schwartz, B. J., Annis, C. G., and Air Force Materials Laboratory, Cu-
mulative Damage Fracture Mechanics under Engine Spectra, Wright-Patterson Air
Force Base, OH: Air Force Materials Laboratory, AFML-TR-79-4159, 1980.
[7] Nicholas, T., Haritos, G. K., Hastie, Jr., R. L., and Harms, K., “Effects of Overloads
on Sustained-Load Crack Growth in a Nickel-Base Superalloy. Part II.
Experiments,” Theor. Appl. Fract. Mech., Vol. 16, 1991, pp. 51–62.
[8] Gemma, A. E., “Hold-Time Effect of a Single Overload on Crack Retardation at Ele-
vated Temperature,” Eng. Fract. Mech., Vol. 11, 1979, pp. 763–774.
[9] Kanesund, J., Moverare, J. J., and Johansson, S., “Deformation and Damage Mech-
anisms in IN792 During Thermomechanical Fatigue,” Mater. Sci. Eng. A, 528, 2011,
pp. 4658–4668.
[10] Zhang, J. X., Harada, H., Ro, Y., Koizumi, Y., and Kobayashi, T.,
“Thermomechanical Fatigue Mechanism in a Modern Single Crystal Nickel Base
Superalloy TMS-82,” Acta Mater., Vol. 56, 2008, pp. 2975–2987.
[11] Jung, A. and Schnell, A., “Crack Growth in a Coated Gas Turbine Superalloy Under
Thermo-Mechanical Fatigue,” Int. J. Fatigue, Vol. 30, No. 2, 2008, pp. 286–291.
[12] Cailletaud, G. and Chaboche, J. L., “Macroscopic Description of the Microstruc-
tural Changes Induced by Varying Temperature: Example of IN100 Cyclic Behav-
iour,” Proceedings—Computer Networking Symposium,Pergamon Press, Oxford,
England, Vol. 2, 1980, pp. 23–32.
[13] Wusatowska-Sarnek, A. M., Blackburn, M. J., and Aindow, M., “c0 Precipitation
Kinetics in P/M IN100,” Mater. Sci. Forum, Vol. 426–432:, 2003, pp. 767–772.
[14] Skinn, D. A., Gallagher, J. P., Berens, A. P., Huber, P. D., Smith, J., and Dayton Uni-
versity Ohio Research Institute, Damage Tolerant Design Handbook, Vol. 2, Chap. 5,
Defense Technical Information Center, Ft. Belvoir, VA, 1994.
[15] ASTM E647, 2008, “Standard Test Method for Measurement of Fatigue Crack
Growth Rate,” Annual Book of ASTM Standards, Vol. 03.01, ASTM International,
West Conshohocken, PA.
[16] Telesman, J. and Ghosn, L. J., “Fatigue Crack Growth Behavior of PWA 1484 Single
Crystal Superalloy at Elevated Temperatures,” Proceedings of the International Gas
Turbine and Aeroengine Congress and Exposition, Houston, TX, June 5–8, 1995,
ASME, New York, 1995.
[17] Adair, B., 2010, “Thermo-Mechanical Fatigue Crack Growth of a Polycrystalline
Superalloy,” MS Thesis, Georgia Institute of Technology, Atlanta, GA.


doi:10.1520/JAI104293
Mauro Filippini,1 Stefano Beretta,2 Luca Patriarca,2

Giuseppe Pasquero,3 and Silvia Sabbadini3
Fatigue Sensitivity to Small Defects of a

Gamma–Titanium–Aluminide Alloy
ABSTRACT: The fatigue properties of a Ti-48Al-2Cr-2Nb alloy obtained by

electron-beam melting (EBM) with a patented process has been examined
by conducting high cycle fatigue tests performed at different R ratios at room
temperature. Fatigue-crack propagation tests have been performed for the
purpose of characterizing the fatigue-crack growth rate and threshold of the
material. Additionally, specimens with artificially introduced defects have
been fatigue tested with the objective of studying the growth behavior of
small cracks. Artificial defects with different sizes have been generated in
the gauge section of the specimens by electron-discharge machining (EDM).
After EDM defects are produced, the specimens are pre-cracked in cyclic
compression, so that small cracks can be generated at the root of the EDM
starter defects. Fatigue tests are conducted by applying the staircase tech-
nique with the number of cycles of censored test (runout) fixed at 107 cycles.
By employing the Murakami model for the calculation of the range of stress
intensity factor, the threshold stress intensity factor range dependence on the
loading ratio R and on the defect size is evaluated, highlighting the relevant
parameters that govern the specific mechanisms of failure of the novel c–TiAl
alloy studied in the present work.
KEYWORDS: gamma titanium aluminides, high-cycle fatigue, fatigue-crack
propagation, compression pre-cracking, short-crack behavior.
Manuscript received August 26, 2011; accepted for publication February 16, 2012;
published online May 2012.
1
Politecnico di Milano, Dipartimento di Meccanica, Via La Masa 1, 20156 Milano, Italy,
e-mail: mauro.filippini@polimi.it
2
Politecnico di Milano, Dipartimento di Meccanica, Via La Masa 1, 20156 Milano, Italy.
3
Avio S.p.A., Via I Maggio 99, 10040 Rivalta di Torino, Italy.
Cite as: Filippini, M., Beretta, S., Patriarca, L., Pasquero, G. and Sabbadini, S., “Fatigue
Sensitivity to Small Defects of a Gamma–Titanium–Aluminide Alloy,” J. ASTM Intl., Vol.
9, No. 5. doi:10.1520/JAI104293.
Copyright V
278

FILIPPINI ET AL., doi:10.1520/JAI104293 279
Introduction
Gamma–titanium–aluminide-based alloys have become an important contender
for structural applications in the aircraft industry to replace current nickel-
based superalloys as the material of choice for low-pressure turbine blades
[1,2]. The advantages achieved by the use of c–TiAl intermetallics are principally
their low density (3.9–4.2 g=cm3 as a function of their composition), high spe-
cific yield strength, high specific stiffness, substantial resistance to oxidation,
and good creep properties up to high temperatures. In particular, the lower den-
sity will contribute to significant engine weight savings and reduce stresses on
rotating components, such as low-pressure turbine blades [3]. Although such
materials appear very promising for the turbine engine industry, optimizing the
performance improvements requires more advanced approaches to accurately
predict fatigue life. Therefore, there is a need to understand and address the
specific fatigue properties of these materials to assure adequate reliability of
these alloys in structural applications [4]. Additionally, their intrinsic brittleness
at low temperatures is a matter of concern for application in the highly loaded
parts of gas turbine engines. Moreover, it is difficult to obtain a component pro-
duced with c–TiAl intermetallics with exactly the composition and microstruc-
ture desired. A further difficulty is that, for the typical aeroengine applications,
the material must have an extremely low oxygen content, preferably much lower
than 1500 ppm.
Electron-beam melting (EBM) is a type of additive manufacturing for metal
parts. It is often classified as a rapid manufacturing method. The technology
manufactures parts by melting metal powder layer by layer with an electron
beam in a high vacuum. Using EBM technology, the process of material produc-
tion operates under high-vacuum conditions, thereby reducing the risk of oxida-
tion in the material of the final components. EBM technology for “layer-by-layer”
productions offers several advantages with respect to other competing technolo-
gies and it is possible to operate at temperatures closer to the melting points of
the intermetallic alloys [5]. In the EBM process, components are produced with-
out vaporization of the powders of the initial material and the powders are made
of an intermetallic alloy based on titanium and aluminium with the same chemi-
cal composition as the final intermetallic alloy with which the components are
produced.
In the present study, the fatigue properties of a Ti-48Al-2Cr-2Nb alloy
obtained by electron-beam melting (EBM) has been examined by conducting
high-cycle fatigue tests performed at different R ratios at room temperature.
Additionally, fatigue-crack-growth (FCG) tests have been conducted by means
of the compression pre-cracking, constant amplitude (CPCA) test methodology,
to characterize the fatigue-crack behavior for the material under investigation.
Finally, a set of specimens with artificially introduced defects has been used to
conduct fatigue endurance tests (up to 107 cycles) with the objective of studying
the growth behavior of small cracks. The aim of this paper is to establish the
threshold stress intensity factor range dependence on the loading ratio R and
on the defects size, highlighting the relevant parameters that govern the specific
mechanisms of failure of the studied c–TiAl alloy.

Material and Specimen Geometry
Material
The gamma–titanium-aluminide (c–TiAl) Ti-48Al-2Cr-2Nb alloy studied in this
work was produced according to a patented process [6]. The material was pro-
duced by focused electron-beam melting in a high-vacuum condition using an
EBM A2 machine manufactured by ARCAM AB (Sweden). The EBM material
was hot isostatically pressed (HIPed) at 1260 C under a pressure of 1700 bar for
4 h. A heat treatment (TT), to be performed after HIP, was set up to obtain the
optimal duplex microstructure (2 h at 1320 C)[7]. Final microstructure after
heat treatment is shown in Fig. 1. Material has been produced in the form of
near net shape specimens and final specimens geometry was manufactured by
conventional machining with carefully selected cutting parameters for remov-
ing the machining allowance.
Specimens
For the tests conducted in the present work, three different types of specimens
have been produced. A set of 30 unnotched specimens suitable for high-cycle-fa-
tigue testing have been produced with the geometry shown in Fig. 2(a). Prior to
fatigue testing, the surface of the specimens has been pre-oxidized, by furnace
treatment in air for 20 h at a temperature of 650 C [8]. Also, a smaller set of six
specimens suitable for crack propagation testing have been produced with the
geometry shown in Fig. 2(b), designed according to ASTM E647-08 [9].
FIG. 1—Microstructure of the c–TiAl alloy after electron beam melting (EBM) and heat
treatment.

FIG. 2—Shape and dimensions of (a) unnotched specimens employed for uniaxial
fatigue testing, and (b) fatigue-crack-propagation testing.
Additionally, a set of 40 specimens with a gauge diameter of 8 mm has been

produced and two types of surface artificial defects in the form of tiny rectangu-
lar micro-slots have been carefully produced in the mid-section of the speci-
mens by EDM, as shown in Fig. 3. Small artificial defects with dimensions of
pffiffiffiffiffiffiffiffiffiffi
500 100 lm ( area ¼ 220 lm) (not shown in Fig. 3), and larger artificial
defects with dimensions of 1500 300 lm ( area ¼ 644 lm) have been intro-
duced in the specimens, Fig. 3.
Test Methods
Fatigue Testing with Plain Specimens

Fatigue tests with plain axial specimens have been carried out at room tempera-
ture (RT) by employing the Rumul Testronic test system available at the laborato-
ries of the Dipartimento di Meccanica of the Politecnico di Milano. Fatigue tests
have been conducted by applying the staircase technique [10] and the number of
cycles of censored test (runout) has been fixed at 107 cycles. Tests have been car-
ried out with three different loading ratios: (i) R ¼ rmin=rmax ¼ 0 (zero to tension),
(ii) R ¼ rmin=rmax ¼ 0.6, and (iii) R ¼ rmin=rmax ¼ 1 (pure alternating stress).
Fatigue-Crack-Growth Testing
Fatigue-crack-growth tests have been carried out in a servo-hydraulic MTS 810
testing machine and the crack length has been monitored by COD gage. Addi-
tional cross-check, even without continuous measurement capabilities, was pro-
vided by a traveling microscope for direct eye crack evolution observations
during the tests. The fracture mechanics specimens have been pre-cracked in
cyclic compression [11–13]. Compression pre-cracking has been applied to the
FCG specimens by employing a specially designed gripping device, as shown in
Fig. 4. To avoid bending because of misalignment, instead of using the conven-
tional clevis and pin assembly, the cyclic loading in compression is applied to
the small C(T) specimens along the load line by means of two opposing cylindri-
cal surfaces with a radius of 80 mm acting on the outer surface of the speci-
mens. The positioning of the specimens prior to the compression pre-cracking

FIG. 3—Geometry of specimens (8-mm gauge diameter) for assessing defect sensitivity
in short crack fatigue testing (a); nominal shape of artificial defect of area ¼ 644 lm
(b); and SEM picture of artificial defects produced by EDM (c).
is ensured by lateral flat surfaces, whereas pins with clearance of about 1 mm

are inserted in these special clevises only for safety purposes. During the com-
pression pre-cracking procedures, it has been observed that a minimum (com-
pressive) force is sufficient to avoid the lateral displacement of the specimens.

FIG. 4—Schematic drawing of the device for applying the compressive loading during
the compression pre-cracking of small fatigue-crack-growth specimens.
By doing so, the crack is generated at the notch tip similar to the crack initiation
in cyclic tension but, by compression pre-cracking, the crack growth then
decreases progressively until it stops propagating [11,12]. By pre-cracking speci-
mens in cyclic compression, the effects of crack closure at the beginning of the
actual crack growth test are nearly cancelled. For starting a crack in cyclic com-
pression by small load amplitudes, and thus preventing damage to the speci-
mens during the pre-cracking phase inadvertently, the wire EDM starter notch
was sharpened by a razor blade polishing technique [13]. The effect of this tech-
nique and the initial crack, obtained by the compression pre-cracking proce-
dure, is shown in Figs. 5(b)–5(d).
Additionally, finite-element analyses with a non-linear elastic–plastic mate-
rial model have been carried out to verify the effect of a sharpened notch on the

FIG. 5—EDM notch in FCG compact specimens (a); view of the sharpened notch after
application of the razor blade polishing technique (b); and view of a pre-crack generated
by compression pre-cracking out of the sharpened notch (c).

length of the plastic region ahead of the crack tip. In general, the sharper the
notch is, the smaller the load needed to initiate a pre-crack at the notch root,
and the smaller the needed length of the pre-crack (about two times that of the
notch radius) to avoid notch effects. It has been demonstrated that the pre-
crack driving force is related to the magnitude of the residual stress field estab-
lished during the first compressive cycle. The definition of the initial plastic
extension is strictly related to the dimensions of the initial pre-crack. In Fig. 6,
the stress field distributions ahead of the starter notch in terms of von Mises
FIG. 6—FE results in terms of von Mises stresses in the region of the starter notch:
original EDM notch (a); and after razor blade polishing technique (b).

stress at the maximum compressive loads for the original notched specimen
region, Fig. 6(a), and for the razor blade notched specimen, Fig. 6(b), are
shown. It was estimated that, by reducing the load by a factor of about 3.4, the
extension of the initial plastic zone size of the razor notched specimens does
not change significantly with respect to the case of the original notches.
After compression pre-cracking, to determine the DKth and the long crack
propagation behavior, FCG tests at room temperature have been carried out at
constant R ¼ Kmin=Kmax ratio (R ¼ 0.05 and R ¼ 0.6) by increasing the load am-
plitude in small steps until the threshold value for a long crack is reached.
Fatigue Testing with Artificial Defects

To generate small cracks at the root of the EDM artificial defects, all specimens
with artificial defects have been submitted to a pre-cracking procedure consist-
ing of fatigue loading in cycling compression for a number of cycles up to 107
cycles. This procedure ensures that fatigue cracks are generated at the root of
the EDM notch, by keeping at minimum compressive residual stresses at crack
tip. After pre-cracking, all specimens have been pre-oxidized by furnace treat-
ment in air for 20 h at a temperature of 650 C, as in the case of unnotched speci-
mens. Finally, fatigue tests have been performed according to the staircase
7
pffiffiffiffiffiffiffiffiffiffi(with runout fixed at 10 cycles) at R ¼ 0 and R ¼ 0.6, with defects
procedure
with area equal to 220 lm and 644 lm, respectively.
Experimental Results
Fatigue Tests with Plain Specimens

The fatigue test results of the set carried out with plain specimens at R ¼ 0,
R ¼ 0.6, and R ¼ 1 are shown in Fig. 7(a). In all the tests, it has been observed
that, independently of the R ratio, the Wöshler curves are extremely flat. This
means that a small variation in the applied stress amplitude can lead to substan-
tial differences in the number of cycles to failure. The HCF test results obtained
in the test campaign can be condensed in a single (Haigh) diagram, Fig. 7(b). By
comparing the tests results at different loading ratios, it can be observed that
the fatigue endurance strength, in terms of maximum stress, is nearly independ-
ent of the loading ratio. In the diagram of Fig. 7(b), the experimentally obtained
fatigue endurance strength values lie just below the dashed curve, representing
the equation:
rmax ¼ rm þ ra ¼ UTS (1)
where rm and ra represent the mean and alternating stress, respectively,

whereas UTS is the ultimate tensile strength, obtained in earlier monotonic
tests. This specific behavior has been observed also in the course of the test cam-
paign, with nearly all specimen failing in the case the applied (maximum) stress
was near or above the ultimate tensile strength, with no specimen failing
(within 107 cycles) for maximum stress equal or below 320 MPa, irrespective of

FIG. 7—Wöhler diagram of HCF test results at R ¼ 0.6 (a); and Haigh diagram for the
HCF tests with plain specimens (b).

FIG. 8—Typical failure initiation site found in fatigue tests: specimen failed after
3.2 106 cycles (R ¼ 0; Dr ¼ 340 MPa): the relatively dark area can be associated to an
initial defect of about area ¼ 150 lm.
the loading ratio R. The fracture surfaces analyzed by SEM reveals that fatigue
failures originate from lamellas that, because of their unfavorable direction
with respect to that of loading, determine a translamellar initial fracture, with
the appearance of a decohesion of weak lamellar grains, Fig. 8. These micro-
structural features have an average area of about 22,000 lm, projected in the
direction
pffiffiffiffiffiffiffiffiffiffi normal to that of loading, corresponding to an equivalent crack size of
area ¼ 150 lm.
Fatigue-Crack-Growth Behavior
After compression pre-cracking procedure was adopted for generating initial
cracks, to avoid sudden fracture upon loading, fatigue-crack-growth testing
required a suitable procedure, by increasing the load amplitude in small steps
until the threshold value of the long crack was reached. Fatigue thresholds,
DKth, were defined as the applied stress-intensity range corresponding to growth
rates below 109 m=cycle. In the FCG tests, a coherent behavior was observed:
for the tests conducted at R ¼ 0.05, near threshold crack growth was observed
for DK about 6 MPa m1=2, Fig. 9(a), whereas for the tests at R ¼ 0.6, DKth is about
4 MPa m1=2, Fig. 9(b).
The critical Kmax value, corresponding to specimen failure, falls in the range
10.5–11.5 MPa m1=2, independently of the applied R ¼ Kmin=Kmax ratio. The FCG
rate curves are shown in Fig. 10. The threshold values determined here are in
accordance with those reported in the literature for the duplex microstructure
of c–TiAl alloys [14,15]. It can be observed that the available DK range for crack
growth is rather narrow, because of the relatively limited difference between

FIG. 9—Typical experimental result obtained by increasing the load amplitudes in

small steps at constant R ratio shown here as change of crack extension versus number
of cycles: at R ¼ 0.05 (a); and R ¼ 0.6 (b).

FIG. 10—Fatigue-crack-growth rate curves in terms of: DK, range of stress intensity
factor (a); and max stress intensity factor in a loading cycle Kmax (b).
DKth and Kmax, resulting in high value of the slope. However, it must be
observed that the c–TiAl produced with the patented EBM process offer supe-
rior FCG characteristics respect to those of TiAl alloys obtained by more con-
ventional processes [15].

Fatigue Tests with Artificial Defects

By employing the Murakami model [16] for the assessment of the range of
stress intensity factor (surface defects) as DK ¼ 0.65Dr (p area)1=2, the thresh-
old corresponding to the endurance strength for each R ratio can be evaluated.
In the plots of Fig. 11, for loading ratios of R ¼ 0 and R ¼ 0.6, respectively,
FIG. 11—Kitagawa diagrams for: loading ratio R ¼ 0 (a); and loading ratio R ¼ 0.6 (b).

endurance strength stress ranges are given as a function of the equivalent defect
size area.
In the case of runout specimens, tested at stress amplitudes just below the
estimated endurance strength, slowly propagating small cracks emanating from
the notch have been observed at both R ratios, as shown in Fig. 12.
Because it’s been observed that the fatigue failures in unnotched specimens
were foundpin correspondence of peculiar microstructural features with a typi-
ffiffiffiffiffiffiffiffiffiffi
cal size of area ¼ 50 lm, the modification of the El-Haddad model by Tanaka
et al. [17] have been applied in the form:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u pffiffiffiffiffiffiffiffiffiffi0 FCG 2
u area pffiffiffiffiffiffiffiffiffiffi0 1 DKth
it
Drth ¼ Dre pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi00 pffiffiffiffiffiffiffiffiffiffi with area0 ¼
area þ area0 areai p 0:65Drie
pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi
¼ area0 þ areai
where Drie represents the fatigue endurance strength obtained in the fatigue
pffiffiffiffiffiffiffiffiffiffiffi
tests with plain specimens and an inherent defect areai of 150 lm has been
taken into account both for loading ratio R ¼ 0 and R ¼ 0.6.
If a smaller volume of material would be stressed up to the threshold level,
the probability of activating an inherent “microstructural” feature with size
pffiffiffiffiffiffiffiffiffiffiffi
areai is likely to become lower, i.e., the initial active defect size may be smaller
for smaller stressed material volume. Theoretically, as the Kitagawa diagrams
in Fig. 11 reveal, there is a possibility to observe an increased fatigue strength of
FIG. 12—Slowly propagating fatigue cracks are observed in runout specimens

(Dr ¼ 300 MPa, R ¼ 0; 107 cycles without specimen failure).

the material if a non-uniformly distributed loading would be applied, as in the

case of bending loading of thin sections. Additionally, it may be noted that the
size of inherent microstructural features of 150 lm made in the present study is
in the range of the colony dimensions [18].
Conclusions
A potential disadvantage of cast and PM c–TiAl alloys, in terms of component
design, is their limited fatigue-crack-growth resistance and damage tolerance
compared to nickel-based superalloys, and the relatively poor fatigue endurance
strength because of the negative superposition of the effect of rather limited
available range of DK for stable crack growth and the presence of defects, like
pores and non-metallic inclusions. In general, as in the case of duplex titanium
aluminides, there is a small difference between the fatigue threshold stress-in-
tensity-range of long cracks and the apparent fracture toughness, leading to
shortened lifetimes for small changes in applied stress, should the fatigue
threshold be exceeded. On the other hand, in the case of the Ti-48Al-2Cr-2Nb
alloy examined in this work, the advantage of the c–TiAl produced by the EBM
process [6] is that typical defects of cast or PM materials can be avoided and
higher fatigue strength with respect to competing technologies can be obtained.
From the observation of the test results, the following conclusions may be
drawn: the mechanism of fatigue failure of c–TiAl studied in this work does not
seem to be governed by Kmax. only, as it might be assumed from the fatigue tests
with unnotched specimens at different R ratios, Fig. 6(b); on the contrary, the
fatigue tests with artificial defects show that DKth for defects larger than 100 lm
can be described very accurately by a modified El-Haddad relationship; the val-
ues of the threshold stress-intensity factor range depend on the loading ratio R.
Even if the benefit of the EBM process for c-TiAl studied in this work looks
promising for structural applications, further development work needs to be
carried out to take full advantage of the strength-to-weight ratio of gamma tita-
nium aluminides.
References
[1] Winstone, M. R., Partridge, A., and Brooks, J. W., “The Contribution of Advanced
High-Temperature Materials to Future Aero-Engines,” Proc. Inst. Mech. Eng= L-J
Mater., Vol. 215, 2001, pp. 63–73.
[2] Dimiduk, D. M., “Gamma Titanium Aluminide Alloys—An Assessment within the
Competition of Aerospace Structural Materials,” Mater. Sci. Eng. A-Struct., Vol.
263, No. 2, 1999, pp. 281–288.
[3] Bartolotta, P., Barrett, J., Kelly, T., and Smashey, R., “The Use of Cast Ti-48Al-2Cr-
2Nb in Jet Engines,” JOM, J. Min. Met. Mater. Soc., Vol. 49, No. 5, 1997, pp. 48–50, 76.
[4] Henaff, G. and Gloanec, A.-L., “Fatigue Properties of TiAl Alloys,” Intermetallics,
Vol. 13, No. 5, 2005, pp. 543–558.
[5] Murr, L. E., Gaytan, S. M., Ceylan, A., Martinez, E., Martinez, J. L., Hernandez, D.
H., Machado, B. I., Ramirez, D. A., Medina, F., Collins, S., and Wicker, R. B.,

“Characterization of Titanium Aluminide Alloy Components Fabricated by Additive

Manufacturing Using Electron Beam Melting,” Acta Mater., Vol. 58, No. 5, 2010,
pp. 1887–1894.
[6] Andersson, L.-E. and Larsson, M., “Device and Arrangement for Producing a
Three-Dimensional Object,” Patent No. WO 01=81031 A1, International Application
Number PCT=SE01=00932, World Intellectual Property Organisation, 2001.
[7] Biamino, S., Penna, A., Ackelid, U., Sabbadini, S., Tassa, O., Fino, P., Pavese, M.,
Gennaro, P., and Badini, C., “Electron Beam Melting of Ti-48Al-2Cr-2Nb Alloy:
Microstructure and Mechanical Properties Investigation,” Intermetallics, Vol. 19,
No. 6, 2011, pp. 776–781.
[8] Wu, X., Huang, A., Hu, D., and Loretto, M. H., “Oxidation-Induced Embrittlement
of TiAl Alloys,” Intermetallics, Vol. 17, No. 7, 2009, pp. 540–552.
[9] ASTM E647-08, 2008, “Standard Test Method for Measurement of Fatigue Crack
West Conshohocken, PA.
[10] ISO12107, 2003, “Metallic Materials-Fatigue Testing-Statistical Planning and Anal-
ysis of Data,” International Organisation for Standardisation, Geneva, Switzerland.
[11] Forth, S., Newman, J. C., and Forman, R., “On Generating Fatigue Crack Growth
Thresholds,” Int. J. Fatigue, Vol. 25, No. 1, 2003, pp. 9–15.
[12] Newman, J. C. and Yamada, Y., “Compression Precracking Methods to Generate
Near-Threshold Fatigue-Crack-Growth-Rate Data,” Int. J. Fatigue, Vol. 32, No. 6,
2010, pp. 879–885.
[13] Pippan, R., Hageneder, P., Knabl, W., Clemens, H., Hebesberger, T., and Tabernig,
B., “Fatigue Threshold and Crack Propagation in Gamma-TiAl Sheets,” Intermetal-
lics, Vol.9, 2001, pp. 89–96.
[14] Campbell, J., Rao, K., and Ritchie, R., “The Effect of Microstructure on Fracture
Toughness and Fatigue Crack Growth Behavior in Gamma-Titanium Aluminide
Based Intermetallics,” Metall. Mater. Trans. B, Vol. 30, No. 3, 1999, pp. 563–577.
[15] Gloanec, A.-L., Henaff, G., Bertheau, D., Belaygue, P., and Grange, M., “Fatigue
Crack Growth Behaviour of a Gamma-Titanium-Aluminide Alloy Prepared by Cast-
ing and Powder Metallurgy,” Scripta Mater., 2003, Vol. 49, pp. 825–830.
[16] Murakami, Y., and Metal Fatigue: Effect of Small Defects and Nonmetallic Inclu-
sions, Elsevier, Oxford, 2002.
[17] Tanaka, K., Nakai, Y., and Yamashita, M., “Fatigue Growth Threshold of Small
Cracks,” Int. J. Fract., Vol. 17, No. 5, 1981, pp. 519–533.
[18] Voice, W. E., Henderson, M. B, Shelton, E. F. J., and Wu, X. H., “Gamma Titanium
Aluminide, TNB,” Intermetallics, Vol. 13, No. 9, 2005, pp. 959–964.


doi:10.1520/JAI104005
David T. Rusk1 and Robert E. Taylor2
Investigation of Load Control Errors for

Spectrum Fatigue Testing at High Frequencies
ABSTRACT: Ultra-high cycle fatigue (gigacycle) tests have shown that a

true fatigue endurance limit does not exist for most metallic materials used in
high cycle applications. These findings have significant implications for rotor-
craft dynamic structural components that have traditionally been designed
using endurance limit, stress-life methods. Unfortunately, the gigacycle fa-
tigue test results generated to date cannot be easily applied to rotorcraft
component design, because the interaction of different crack nucleating
mechanisms under variable amplitude loading is not well understood, and
must be studied using spectrum fatigue tests carried out to very long lives. To
quantify the effect of spectrum load control errors on the rate of fatigue dam-
age accumulation for a standard servo-hydraulic fatigue test machine, a dam-
age ratio parameter for crack initiation is calculated using the high cycle
fatigue portion of the Coffin-Manson strain-life curve with rainflow cycle
counting. The fatigue damage parameter allows the level of controller error to
be assessed as a function of test frequency, peak load levels, and test spec-
trum complexity. Also, different command feedback compensation schemes
are tested to determine the range of control error that can be expected for a
given set of test parameters. A sensitivity study was performed on the com-
mand signal response errors as a function of changes in the slope of the
mean damage curve beyond 106 cycles. The test results demonstrate that
rotorcraft spectrum fatigue tests can be performed at frequencies up to
online March 2012.
1
Aerospace Engineer, Structures Division, Bldg. 2187 Suite 2340A, NAVAIRSYSCOMHQ,
48110 Shaw Rd. Unit 5, Patuxent River, MD 20670-1906.
2
Test Technician, Code 4.3.4.1, Bldg. 2188, NAVAIRSYSCOMHQ, 48066 Shaw Rd. Unit 5,
Patuxent River, MD 20670-1908.
in Anaheim, CA.
Cite as: Rusk, D. T. and Taylor, R. E., “Investigation of Load Control Errors for Spectrum
Fatigue Testing at High Frequencies,” J. ASTM Intl., Vol. 9, No. 3. doi:10.1520/JAI104005.
Copyright V
295

150 Hz with a level of nominal damage accumulation error less that 1 %

under most conditions. When electrical measurement system error is
accounted for, the damage accumulation error is less than 2 % under most
conditions. The results also show that it is important to consider the shape of
the stress-life curves in the gigacycle fatigue range when performing spec-
trum tests out to long fatigue lives.
KEYWORDS: Gigacycle fatigue, fatigue testing, crack initiation, variable
amplitude loading, spectrum loading, servo-hydraulic control error
Introduction
Application of ultra-high cycle fatigue (gigacycle) testing capabilities has
recently shown that a true fatigue endurance limit (infinite life) does not exist
for most metallic materials used in high cycle applications. In fact, such
research has shown that for many materials, the critical crack nucleating mech-
anisms may change as a very large number of low-amplitude cycles are accumu-
lated that are below the traditionally assumed endurance limit. Crack initiation
at the specimen surface is the typical failure mechanism in nearly all ductile me-
tallic materials that are fatigue tested to a 107 cycle runout limit. Bathias has
shown that for several types of high-strength spring steels, the crack initiation
location changes from surface to subsurface beyond 107 cycles to failure [1].
Murakami has analyzed these subsurface “fish eye” fractures on Cr-Mo steel
[2]. Shiozawa et al. have found similar behaviors in other high-strength steels
[3]. Other researchers have reported similar transitions in crack initiation
mechanisms at long lives, for a variety of steel types. In aluminum, the crack ini-
tiation location has been found to remain at the surface of the test specimen,
even at very high numbers of fatigue cycles (>107). However, the dominant fail-
ure mechanism changes from inclusion crack nucleation to nucleation from
slip-band formation at fatigue cycles beyond 107. Marines et al. provides test
data and analyses that illustrate the influence of these mechanisms on the shape
of the traditional stress-life (S-N) curve for 2024-T3 aluminum [4]. Changes in
crack nucleation mechanisms at gigacycle fatigue lives are less clear for tita-
nium. Bathias et al. gave a brief description of the gigacycle fatigue behavior of
titanium alloys [5]. The results to date appear to be highly dependent on the
heat treatment used and the resulting microstructural variation, due to the ab-
sence of large inclusions or porosity.
These cumulative findings have significant implications for rotorcraft
dynamic structural components such as rotor heads, pitch links, and so on,
which have traditionally been designed using endurance limit, stress-life meth-
ods [6–8]. Unfortunately, the gigacycle fatigue test results generated to date can-
not be easily applied to rotorcraft component design, because the interaction of
different crack nucleating mechanisms under variable amplitude loading is not
well understood. A major limitation to developing this understanding is the
inability to conduct variable-amplitude fatigue tests using representative rotor-
craft loading histories to very long lives. Current gigacycle fatigue tests use pie-
zoelectric resonant test frames that have the capability to generate very high

RUSK AND TAYLOR, doi:10.1520/JAI104005 297
test frequencies (20–30 kHz), but are limited to tests of constant-amplitude load
blocks with variable mean stresses, or purely random noise type loading.
Stanzl-Tschegg and Mayer have generated gigacycle fatigue results on 2024-
T351 aluminum for a Gaussian random loading distribution [9]. Pöting et al.
have developed a method to approximate variable amplitude loading on a reso-
nant test frame by translating the spectrum frequency content into a “beat-like”
loading sequence that can be run on a resonant test frame [10]. However, these
approaches only approximate the level of damage accumulation in the original
spectrum, because they do not follow the prescribed loading sequence exactly.
This limitation makes it difficult to investigate the influence of specific load
interaction effects on materials with competing damage mechanisms in the
gigacycle fatigue range. To overcome this and other limitations, a research pro-
ject was initiated to investigate the potential for existing servo-hydraulic test
frame technologies to run variable amplitude fatigue tests at speeds signifi-
cantly higher than what the current standard practice is in the testing commu-
nity. Recent improvements in servo-hydraulic test frame technologies, such as
high-frequency servo-valves, low friction actuators, and command feedback
compensation schemes have made it possible to conduct constant amplitude fa-
tigue tests at frequencies up to 1000 Hz [11]. The ability of such equipment to
follow a predefined, variable amplitude rotorcraft fatigue spectrum while pro-
viding reasonably close control was investigated, and is documented in the
following.
Test Apparatus
The test frame used for these experiments is an MTS 810 High-Frequency Test
System,3 which can be configured for use with standard or voice-coil servo-
valves. Only standard servo-valves were used in this investigation, in both single
and dual servo configurations. A 25 kN (5.5 kip) load cell was used with a
22.7 kN (5 kip) calibration range. Lightweight aluminum hydraulic flat-wedge
grips were used for all tests performed. Circular wedge blocks for the grips were
fabricated in-house. The test specimens used for this investigation were manu-
factured to conform to ASTM E606-04 recommendations for uniform gage fa-
tigue test specimen geometry [12]. The test specimens have a 6.35 mm (0.25 in.)
gage diameter, a 19 mm (075 in.) gage length, and a 12.7 mm (0.5 in.) grip diam-
eter. A 25.4 mm (1.0 in.) diameter spacer, made of linen-phenolic composite, is
press-fit onto the specimen grip sections to isolate the specimen from contact
with the hydraulic grip wedges. This has proven to significantly reduce instan-
ces of premature failure in the specimen grip sections at large numbers of accu-
mulated test cycles. 2024-T351 aluminum alloy test specimens were used for the
tests. An MTS FlexTest SE digital controller was used to control the test frame,
which was enhanced to provide a 6000 samples/s data rate. All tests were con-
ducted in closed loop force control.
3)
MTS Systems Corp., Eden Prarie, MN USA.

For the standard, laboratory air uniaxial fatigue test of interest here, the re-
sultant testing error of primary significance is the difference between the level
of loading commanded at each peak and valley in the load spectrum, and the
peak-valley (P-V) loads that the test specimen is actually subjected to by the test
frame. If other testing conditions are of interest such as high temperature, time
dependent, and/or environmental effects, this may not be strictly true, and addi-
tional test response errors, such as phase lag, may need to be monitored. Mod-
ern servo-hydraulic test frame controllers have incorporated command
feedback compensators to reduce the amount of error in P-V values for dynamic
testing. The compensators compare the command with the corresponding load
or strain sensor feedback to ensure that the command is fully applied to the
specimen. If the sensor feedback indicates that the specimen is not reaching the
target P-V levels, the compensator dynamically adjusts the gain in the command
signal to minimize the P-V error within the constraints of the control system.
The method of gain adjustment, and the level of P-V error reduction experimen-
tally achieved, are highly dependent on the compensation scheme and the test
parameters being used.
Two different command feedback compensation schemes embedded in the
controller software were investigated for load response accuracy and stability.
These were adaptive inverse control (AIC) and arbitrary end-level compensation
(ALC), as described by Soderling et al. [13]. The two schemes were chosen from
a range of possible options because they are well suited to controlling random
load histories to minimize P-V errors. AIC uses an inverse linear model of the
test system, the parameters of which are continuously updated to minimize
errors as the test progresses. This scheme will simultaneously compensate for
amplitude and phase, but is limited to linear test system behavior. ALC is an
adaptive compensation technique that uses a matrix of to-from end level ranges
to calculate the adjustment in gain required to minimize P-V errors for each
range pair in the matrix. The compensation matrix is recalculated after each
pass of the spectrum, so several spectrum passes are required before the peak-
valley errors converge to stable values. Unlike AIC, ALC will work for both linear
and non-linear test systems. For each combination of test spectrum, test fre-
quency, and compensation scheme, the controller and compensation gains
were adjusted to optimize the real-time waveform shape and P-V response that
was output to a stand-alone digital oscilloscope.
Test Load Spectrum

To generate test results relevant to rotorcraft structural applications, the Helix
standardized fatigue loading sequence for helicopter rotors was used [14,15].
This spectrum was developed to approximate the fatigue loading history of a
hinged rotor blade subjected to a mix of generic mission profiles categorized as
training, transport, anti-submarine warfare (ASW), and search and rescue
(SAR). Each mission profile has three possible mission durations: 0.75, 2.25 and
3.75 h. The full Helix spectrum consists of 140 flights of all of the mission pro-
files and durations ordered in a random sequence, with some missions and
durations repeated more than others. The full spectrum represents 190.5 flight

hours, with 2 132 024 cycles. For a compensation scheme that recalculates gain
adjustments after each spectrum pass, long spectrum files can result in a large
number of cycles being accumulated on the test specimen before the P-V errors
converge to stable values. To the reduce the duration of testing necessary to
achieve stable error values, a single flight from the full Helix spectrum was
selected for use in this investigation. Flight 21 is a transport flight with a dura-
tion of 0.75 h that contains 8295 cycles. The flight consists of blocks of mostly
constant-amplitude loading at positive R ratios, interspersed with periodic over-
loads (Fig. 1). The beginning and end of each flight has a 20 % compressive load-
ing from blade droop when the rotor is not turning. This load defines the
ground-air-ground (GAG) cycle for each flight.
For the purposes of enabling a detailed comparison of individual P-V errors
within a spectrum load sequence, a simplified spectrum block loading sequence
(Block_Helix2) was defined based on the characteristics of Helix Flight 21. This
spectrum has a repeatable pattern of overloads and underloads embedded in
each loading block, which improved the ability to examine the dynamic
response of the test setup to small and large perturbations from steady state
conditions. The spectrum definition is listed in Table 1 and plotted in Fig. 2.
Test Response Errors

There are many sources of testing variability and uncertainty that contribute to
errors in the test response achieved as compared to the test response desired.
ASTM E1942-98 categorizes the sources of these errors as: the mechanical test
FIG. 1—Flight 21 from the Helix standard fatigue rotor loading sequence.

TABLE 1—Normalized peak-valley (P-V) ranges for Block_Helix2 spectrum (11/137 total
cycles).
P-V Range Cycles Repeat Cycles Block
0.2 –
0.92 to 0.52 100
0.96 to 0.48 2 10X A
0.8 to 0.4 8
0.84 to 0.36 1
0.8 to 0.4 4
0.84 to 0.36 1 267X B
0.92 to 0.36 500 1X C
0.8 to 0.2 100
0.7 to 0.3 2 20X D
1.0 to 0.4 100 1X E
0.8 to 0.4 8
0.84 to 0.36 1
0.8 to 0.4 4
0.84 to 0.36 1 267X F
0.2 –
FIG. 2—Block_Helix2 spectrum.

frame and its components, the electrical measurement system, and the com-
puter processing of data [16]. The test frame and components comprise the larg-
est source of test response error in most cases, but this error is the most difficult
to quantify. The ability of the test frame control system to provide the fidelity of
command signals necessary to achieve a desired response depends in part on
the level of sophistication of the plant model embedded in the controller.
Non-linearities, hysteresis, and dynamic effects in the testing system that may
be insignificant at lower test frequencies may become significant at higher fre-
quencies, if they are not properly modeled in the control system logic. Plummer
gives a brief description of the sources of some of these non-linearities for a sim-
ple servo-hydraulic test frame [17]. For all tests discussed here, peak loads were
below the proportional limit of the material, so test specimen non-linearity
would consist only of the slight hysteresis from loading and unloading in the lin-
ear elastic range. This investigation did not attempt to quantify or attribute the
sources of non-linearities, hysteresis, or dynamic effects in any other part of the
test frame. The recorded load response as compared to the spectrum input sig-
nal was the principal method used to assess the relative error in the test frame
and components.
The resonant frequencies of the test specimen, load train, and test frame
can cause significant difficulties with controllability if the frequency of testing is
sufficiently close to any one of the resonant frequencies. To investigate this pos-
sibility, a frequency sweep from 100 Hz to 300 Hz was performed with the test
setup described previously. The sweep was run using constant amplitude, sine
wave loading (R ¼ 0.1) with P-V compensation enabled. The test results showed
that the test system as configured was able to provide stable control with a con-
sistent level of P-V error for all frequencies tested. No control response instabil-
ities or anomalies were detected in these tests.
ASTM E1942-98 gives instructions for assessing the level of error in the elec-
trical measurement system. The sources of these errors are categorized as: sig-
nal conditioning bandwidth, data rate, noise level, and phase shift and data
skew. For the tests performed in this investigation, no signal conditioning was
applied to the load cell signal output, so there are no errors associated with
bandwidth limitations. Data rate errors are dependent on the waveform type
used in the test. For all of the tests described here, sinusoidal waveforms were
used. The basic data rate of the test controller as configured for these tests is
approximately 6000 samples/s. The actual data rate of the system was verified
by writing the load cell signal response to a file at the maximum data rate avail-
able in the control software, along with the elapsed time for each data point.
This output confirmed an actual data rate of 6145 samples/s on average. Spec-
trum tests were performed at a range of frequencies from 100–180 Hz. For the
maximum basic data rate, the maximum errors in simple P-V detection in a sine
wave signal can be calculated from the formula given in Ref. [16], and range
from 0.13 %–0.42 %. Tests at speeds beyond 125 Hz will have P-V error values
greater than the 0.2 % recommended in Ref. [16]. These error values represent
the maximum possible error that will be experienced in P-V detection, but the
actual error for any given peak or valley will be a random number from zero to
the maximum possible value. For simulation purposes, the probability

distribution of these errors needs to be defined so that the impact of the error
on the final test results can be quantified. The schematic for characterizing P-V
errors in a sine wave signal is described in Ref. [16], and shown in Fig. 3. The
time interval (t) between the actual cycle peak and the closest recorded data
point can be any value between zero and ts/2, where ts is the time interval
between data samples. This interval can be modeled as a continuous uniform
random number distribution
t Unifð0; ts =2Þ (1)
The random error in peak load response (es) can be simulated by randomly sam-
pling from the uniform distribution and transforming into a sine wave ampli-
tude by Eq (2), where the absolute value of the error is expressed as a fraction of
the load response amplitude
jes j ¼ cosð2ptÞ (2)
The distribution of the random P-V errors can be closely approximated by a

beta distribution, where the distribution parameters a and b are a function of
the data sampling rate and the cyclic frequency
jes j Betaða; bÞ 0 jes j jesmax j (3)
The absolute P-V detection errors are always added to the measured peak val-
ues, and subtracted from the measured valley values, and have the effect of
increasing the measured P-V range for every load segment in a spectrum. Note
that these error calculations are for simple peak-picking based on the maximum
and minimum values of the basic data. The exact peak-picking algorithm for
sine waveforms that is embedded in the test controller is not known, and more
sophisticated methods such as sinusoidal or quadratic least squares, as investi-
gated by McKeighan et al. [18], may be used, which could reduce the range of
errors calculated here.
FIG. 3—Data sampling error in sine wave, from ASTM E1942-98 [16].

Noise in the load transducer signal was measured with the test specimen at
a constant zero load and at a tensile load of 6.4 kN (1400 lb), at the maximum
basic data rate of the system. Mean noise level (ln) values were 0.026 N to
0.214 N (0.00586 lb to 0.0481 lb), and standard deviation (rn) values were
3.71 N to 3.75 N (0.835 lb to 0.842 lb) over 10 s of collected data. The noise level
as a percentage of the load cell calibration range of 22.2 kN (5 kip) was
0.0168 %. Because the load transducer noise (en) is the result of a random pro-
cess, the noise can be modeled as a Gaussian distribution, and is simply added
to the measured P-V values for simulation
en Normðln ; rn Þ 1 en 1 (4)
For the tests described here, phase shift and data skew are only an issue when
comparing the recorded load response to the spectrum command sequence.
Using AIC compensation, the phase lag between the command input signal and
load response was only a few degrees at the frequencies tested, while for ALC
compensation, the phase lag between command and response at the highest fre-
quencies was as much as 216 . To accommodate these phase shifts, the output
P-V data from the spectrum command and load response were written to sepa-
rate files to ensure that independent P-V triggers were being used for each data
channel.
Damage Accumulation Errors

From the previous discussion on test response errors, it is expected that the
cycle-by-cycle P-V error may vary substantially depending on the load spectrum
content, the compensation scheme used and the individual cycles preceding
and following the current cycle. However, the influence that these cycle-by-cycle
errors have on the final results of a spectrum loaded fatigue test are difficult to
determine based solely on range, amplitude, or mean stress errors in the meas-
ured loading response. A parameter based on the ratio of fatigue damage accu-
mulated versus the fatigue damage expected would allow the calculation of the
relative error in experimental fatigue lives for any test, given a representative
sample of the load response history data for that test. A similar type of metric
for variable-amplitude crack growth testing was developed by McKeighan et al.
[18] and Donald and George [19], based on a Paris law relationship for crack
growth rate as a function of stress intensity factor range [20]. For constant-
amplitude HCF tests, the stress-life (S-N) curve has traditionally been used to
assess the influence of test parameters on resulting fatigue life. The typical S-N
curve shape often assumes an endurance limit at long fatigue lives [21], and
cycles accumulated below this stress level are assumed to cause no fatigue
damage in the test. For fatigue tests under spectrum loading at very long lives,
an S-N curve with an assumed endurance limit is no longer adequate, and more
complex methods must be used.
Marines et al. produced an S-N curve for 2024-T3 aluminum in [4] that
demonstrates the influence of competing damage mechanisms on long-life fa-
tigue behavior (Fig. 4), where Mode A failures originate from surface-breaking

FIG. 4—2024-T3 aluminum alloy S-N curve (R ¼ 0.1), from Marines et al. [4].
inclusions, and Mode B failures originate from persistent slip bands at the sur-
face. The resulting S-N curve shows a distinct bifurcation between 106 and 107
cycles that could not be easily modeled using a single S-N curve. The approach
taken here was to use the HCF portion of the Coffin-Manson strain-life curve,
which does not assume the presence of an endurance limit [22]
ra ¼ rf0 ð2Nf Þb (5)
where:
ra is the stress amplitude,
r0 f is the fatigue strength coefficient,
2Nf is the reversals to failure, and
b is the fatigue strength exponent.
Coefficients for 2024-T351 aluminum were taken from Dowling [22] for ma-
terial in the non-prestrained condition. This basic curve was used to describe
the mean S-N behavior at lives less than 106 cycles. Beyond 106 cycles, different
slopes were used to approximate the mean S-N behaviors in the gigacycle fa-
tigue range. The S-N curve slope in the gigacycle range may not be readily avail-
able for most materials, so it may have to be assumed based on whatever
published data is available for similar materials. Bathias gives a range of fatigue
strength reduction of 100–200 MPa from 106 to 109 cycles in aluminum alloys
[5]. For this study, a range of possible fatigue strength reduction levels between
106 to 109 cycles was modeled to determine the sensitivity of the damage
accumulation errors to the S-N curve slope in the gigacycle region. These curves
are shown in Fig. 5. The baseline S-N curve has a stress amplitude of 185 MPa
(26.9 ksi) at 106 cycles, which decreases by 103 MPa (15.0 ksi) at 109 cycles. The
fatigue strength for the short-life S-N curve decreases by 165 MPa (24 ksi) from

FIG. 5—Gigacycle S-N curves for 2024-T351 aluminum (R ¼ 1).
106 to 109 cycles, and the fatigue strength for the long-life S-N curve decreases
by 50 MPa (7.3 ksi) from 106 to 109 cycles. These three curves cover a wide range
of possible fatigue behavior in the gigacycle range, due to changing fatigue
damage mechanisms or a change in damage accumulation rate of a single
mechanism.
For variable-amplitude loading, there are several ways to define a cycle.
ASTM E1049-85 lists several different methods for cycle counting in fatigue
analysis and testing [23]. Of these, rainflow counting has proven to be the most
accurate in assessing the rate of damage accumulation on spectrum loaded test
specimens. For the tests performed in this program, stabilized load response
histories for a minimum of 10 spectrum passes were recorded during testing
and saved for post-processing. Individual complete spectrum passes were rain-
flow cycle counted using a modified version of the algorithm outlined by Glinka
and Kam [24]. This algorithm is more robust that the method outlined in ASTM
E1049-85 because it does not require that the spectrum block be rearranged to
start with the maximum peak value. The influence of mean stresses on the level
of damage accumulation for each closed rainflow-counted cycle was accounted
for using the Smith-Watson-Topper (SWT) correction method [25]. Dowling
et al. have shown that SWT is more accurate in determining equivalent fully
reversed stress amplitudes than the modified Goodman approach traditionally
used in the rotorcraft industry [26]. The damage fractions for the mean stress
corrected rainflow cycles in each spectrum pass are interpolated from the S-N
curves shown in Fig. 5, and the cumulative damage fraction for a complete

spectrum pass is summed using the Palmgren-Miner linear damage accumula-

tion rule [27]
n
Di ¼ (6)
Nf
where:
n is the number of cycles at a defined stress level,
Nf is the cycles to failure, and
Di is the ith damage fraction for an individual load cycle.
The validity of linear damage accumulation for the case of competing crack
nucleation mechanisms has not been established by research, but is assumed
here because the objective is to provide a relative measure of damage accumula-
tion error compared to a target value, and not to predict the absolute value of
damage accumulation for a particular test specimen. The damage ratio parame-
ter (C) is then defined as the ratio of the cumulative damage fractions for one
spectrum pass of the recorded load history response (subscript R) versus the
target load history (subscript T)
RDiR
C¼ (7)
RDiT
Test Results
Tests were performed using AIC and ALC compensation schemes, under both
constant frequency and constant load rate control. The target testing frequen-
cies, loading rates, and load levels were varied to determine the system response
and damage accumulation errors over a range of parameters. Initial compari-
sons were made using the Block_Helix2 spectrum to establish a baseline of
expected errors for different test setups. These tests were performed at a peak
tensile load of 6.23 kN (1.44 kip), giving a stress of 197 MPa (28.5 ksi) in the test
specimen gage section. Total displacement range for the load train was approxi-
mately 0.230 mm (0.009 in.). Comparing the cycle-by-cycle P-V errors of the
spectrum input and load response provides a cumulative measure of the errors
inherent in the test frame and controller for the configuration being tested. As
testing progressed, several locations in the Block_Helix2 spectrum were ana-
lyzed to determine the test system response to small and large perturbations
from constant-amplitude conditions. In Fig. 6, load response to the two cycle
perturbation in Block A is plotted along with the target P-V values for the ALC
compensation scheme. Also, the relative error in P-V range for each loading seg-
ment is plotted. The segment that starts the perturbation cycles shows an under-
shoot from the target peak value. The valley value undershoots the target for the
cycle after the perturbation cycles before reaching the target value in subse-
quent cycles. The mean of the P-V range errors is 0.027 % in the constant ampli-
tude portion of the block, with a standard deviation of 0.60 %. However, in the
segments adjacent to the perturbation, the P-V range error increases to

FIG. 6—Loading perturbation response in Block A of Block_Helix2 spectrum, ALC

compensation.
1.95 %. This is slightly beyond three standard deviations of the constant-

amplitude error. The same Block A perturbation is shown in Fig. 7 for AIC com-
pensation. The mean of the P-V range errors is þ1.25 % in the constant ampli-
tude portion of the block, with a standard deviation of 0.50 %. This indicates
that there is a stable and consistent overshoot in P-V loads from the target val-
ues. Unlike the ALC results, the perturbation cycles do not cause any significant
over or undershoots from the target P-V values. In the cycle subsequent to the
perturbation, the P-V range error increases to 2.4 %, but returns to the steady
state range in the subsequent cycles. This is within three standard deviations of
the constant-amplitude error.
The transition from Block A to Block B in the Block_Helix2 spectrum con-
sists of a step change in mean load level with overload cycles mixed in. In Fig. 8,
the load response in this transition region is plotted for ALC compensation. The
load response to the two perturbation cycles at the end of Block A results in sim-
ilar P-V range error to what was shown in Fig. 6, but the segment at the transi-
tion between the blocks show a significant undershoot from the target peak

FIG. 7—Loading perturbation response in Block A of Block_Helix2 spectrum, AIC

compensation.
value, resulting in a maximum range error of 6.1 %. The periodic overload

cycles in Block B also increase the range errors from what was measured in
Block A, because it takes a few cycles for the controller to establish a steady-
state response before the next overload cycle occurs. The resulting maximum
P-V range errors are over 3 % in Block B. The load response for AIC compensa-
tion in the transition region is plotted in Fig. 9. This response has similar error
characteristics to ALC, but with a positive bias in the error range. However, the
maximum P-V range error of the block transition cycle is only 4.0 %.
The GAG cycle in the Block_Helix2 spectrum presents the greatest challenge
to achieving close control for the compensation schemes tested, because of the
large compressive loading cycle between mean tensile load level step changes.
The load response for the GAG region under ALC compensation is shown in
Fig. 10. ALC does an excellent job of hitting the target values for the compres-
sive valley and subsequent peak, but the previous cycle misses the target peak
value by a wide margin. The resulting P-V range error for this cycle goes from
þ9 % to 66 %. The error for the subsequent GAG segment is 0.13 %. The error

FIG. 8—Loading response in Block A to B transition of Block_Helix2 spectrum, ALC

compensation.
in the cycle after the GAG segment is also large, with a maximum value of
20 %. The load response for the GAG region under AIC compensation is shown
in Fig. 11. Here, the difficulties are with the three cycles after the compressive
GAG cycle. AIC overshoots the peak and undershoots the valley following the
compressive valley, and takes a few cycles to return to a steady-state range of
error. The resulting P-V range error for this cycle goes from þ10 % to 13 %.
Damage accumulation error results for the set of tests using AIC compensa-
tion under constant frequency control are shown in Table 2. Errors are calcu-
lated for the 55th and 63rd pass through the spectrum for each test run. This
allows the evaluation of the stability and convergence of the error as the test
progresses. The AIC results show that the lowest level of damage accumulation
error occurs at a frequency of 140 Hz, with the errors increasing significantly at
150 Hz. The 140 Hz errors were 1 % for the short-life S-N curve, 3 % for the base-
line S-N curve and 8 %–15 % for the long-life S-N curve. The significant increase
in damage accumulation error for the long-life S-N curve is due to the P-V range
errors of the largest rainflow counted cycles in the spectrum, including the GAG

FIG. 9—Loading response in Block A to B transition of Block_Helix2 spectrum, AIC

compensation.
cycle at the beginning and end of the spectrum block. These few large cycles
have mean fatigue lives that are several orders of magnitude shorter than the
smaller cycles that make up the bulk of the spectrum block. An S-N curve with a
shallow slope at long life will weigh the large cycle damage much more heavily
in proportion to the small cycles, in comparison to an S-N curve with a steep
slope where the large and small cycles are weighted more equally. As a result,
spectrum tests of materials with shallow slope S-N curves will be much more
sensitive to P-V range error in the large cycles than materials with steeper
sloped S-N curves. The modest increase in damage accumulation error at fre-
quencies lower than 140 Hz in the AIC tests is likely due to differences in the
tuning parameter setup for each test run. Additional fine tuning may reduce the
error somewhat from that shown in the table.
Results for the set of Block_Helix2 tests using ALC compensation under
constant frequency control are shown in Table 3. The damage accumulation
errors for these tests are two orders of magnitude lower than the AIC results for
the short-life and baseline S-N curves. The errors for the long-life S-N curve are

FIG. 10—Loading response to GAG cycle in Block_Helix2 spectrum, ALC

compensation.
similar to the AIC results. For the ALC tests, test frequencies of 170 Hz are
achieved before the errors start to increase significantly. Also, the errors are
generally decreasing as more spectrum passes are accumulated in the 150 Hz
and slower tests. This indicates that the compensator is continuing to optimize
the controller gains to minimize P-V error in subsequent passes. Test results
were also generated for the Block_Helix2 spectrum using AIC and ALC compen-
sation under constant load rate control, with the ALC error results shown in Ta-
ble 4 for comparison. The damage accumulation errors are of the same order of
magnitude as the AIC results under constant frequency control, but are much
higher than achieved for ALC with constant frequency control. The error
increases significantly at test frequencies above 160 Hz. Test results for AIC
compensation under constant load rate control showed errors significantly
greater than what was measured for AIC under constant frequency control, and
therefore are not listed here.
All previously discussed tests were run at a stress of 197 MPa (28.5 ksi) in
the test specimen gage section. To determine the sensitivity of the damage accu-
mulation error response to peak load levels, a sweep of peak load levels were

FIG. 11—Loading response to GAG cycle in Block_Helix2 spectrum, AIC

compensation.
tested to cover the range of elastic stresses that a typical fatigue test would
cover. In addition, these tests were performed using the Helix Flight 21 spec-
trum, to assess how the damage accumulation errors might change due to a
spectrum with more random variation and complexity. These tests were run
TABLE 2—Damage accumulation errors for varying test frequency, Block_Helix2 spectrum,
constant frequency control, AIC compensation.
Short life S-N Baseline S-N Long life S-N
55 Passes 63 Passes 55 Passes 63 Passes 55 Passes 63 Passes

Frequency (%) (%) (%) (%) (%) (%)
120 Hz 1.93 1.99 5.69 6.01 45.1 40.4

130 Hz 1.51 1.67 4.53 4.88 10.9 16.8
140 Hz 1.04 1.04 2.93 2.96 8.04 14.8
150 Hz 2.32 1.95 6.60 4.89 72.2 37.4

constant frequency control, ALC compensation.

Frequency (%) (%) (%) (%) (%) (%)
120 Hz 0.059 0.051 0.090 0.085 5.54 7.09

140 Hz 0.003 0.010 0.089 0.080 11.4 12.9
150 Hz 0.096 0.068 0.010 0.068 17.8 11.1
160 Hz 0.091 0.065 0.059 0.18 7.76 27.7
165 Hz 0.020 0.026 0.014 0.15 8.80 9.78
170 Hz 0.64 0.85 0.94 1.21 9.32 14.4
175 Hz 4.35 5.07 7.81 9.10 10.0 0.064
180 Hz 3.29 3.30 4.8 4.84 29.5 12.5
using only ALC compensation under constant 150 Hz frequency control, with
the results shown in Table 5. The damage accumulation errors are similar to,
but somewhat greater than, those that were calculated for the Block_Helix2
spectrum at 150 Hz. For the Helix Flight21 spectrum, the errors are less than
0.2 % for the short-life S-N curve, less than 0.6 % for the baseline S-N curve, and
between 0.2 %–20 % for the long-life S-N curve. The error for the long-life S-N
curve generally decreases as the peak stress level in the test is increased because
all of the spectrum cycles are shifted up the S-N curve, reducing the impact of
P-V range error in the largest cycles.
The influence of electrical measurement error on the estimate of damage
accumulation error can be determined by propagating the electrical measure-
ment errors through the damage accumulation calculations described previ-
ously. The electrical measurement error is a random variation of P-V values for
every cycle in the load history, so a Monte Carlo simulation was used to sample
the distributions in Eqs. 3 and 4, and modify the recorded values for the Helix
constant load rate, ALC compensation.

Frequency (%) (%) (%) (%) (%) (%)
120 Hz 7.31 7.53 9.14 9.32 7.39 11.7

140 Hz 0.53 0.47 0.23 0.12 10.5 17.7
150 Hz 1.67 1.65 1.26 1.26 19.0 14.0
160 Hz 0.84 1.42 0.52 1.43 19.9 187
165 Hz 8.15 8.05 7.28 7.07 11.9 9.93
170 Hz 11.1 11.6 13.5 14.4 41.4 49.7

TABLE 5—Damage accumulation errors for varying maximum tensile stress, Helix Flight
21 Mod1 spectrum, 150 Hz constant frequency, ALC compensation.
Maximum 74 Passes 85 Passes 74 Passes 85 Passes 74 Passes 85 Passes

stress (MPa) (%) (%) (%) (%) (%) (%)
138 0.09 0.10 0.50 0.46 19.8 9.57

197 0.001 0.02 0.10 0.13 2.01 11.4
276 0.10 0.10 0.55 0.55 5.27 4.67
345 0.17 0.14 0.55 0.49 3.29 3.15
Flight 21 spectrum test. The results of this simulation are listed in Table 6 for
the test performed at 197 MPa (28.5 ksi) peak stress and 150 Hz. For each S-N
curve type and spectrum pass, 100 simulations were run to provide estimates of
the mean and 90 % confidence bounds on the damage accumulation ratio. The
results show that the mean damage accumulation error increases 0.5 % for the
short-life S-N curve, 1.5 % for the baseline S-N curve, and 2.5 %–3 % for the
long-life S-N curve. This increase was expected because including the P-V detec-
tion error increases the range and amplitude of every cycle in the load history.
The 90 % confidence intervals for the short-life and baseline S-N curves are
60.01 % and 60.035 % respectively, which are quite small given the range of
scatter in electrical measurement error. For the long-life S-N curve, the confi-
dence interval is much greater at 64 % because a small number of large ampli-
tude cycles contribute most of the damage, so the results will be much more
sensitive to significant variation in the P-V values of these few cycles. In the
short-life and baseline S-N curves, errors in individual P-V values are averaged
out over a much larger number of damaging cycles, so the net result is a tight
confidence interval.
TABLE 6—Propagation of electrical measurement system error into damage accumulation

error for Helix Flight 21 Mod1 Spectrum, 150 Hz, 197 MPa peak stress, ALC compensation.
74th Spectrum pass 85th Spectrum pass
90% 90% 90% 90%

Confidence Confidence Confidence Confidence
lower Mean upper lower Mean upper
S-N curve bound (%) (%) bound (%) bound (%) (%) bound (%)
Short life – 0.001 – – 0.02 –

Short life þ EM error 0.534 0.544 0.554 0.512 0.522 0.531
Baseline – 0.10 – – 0.13 –
Baseline þ EM error 1.52 1.56 1.59 1.55 1.58 1.61
Long life – 2.01 – – 11.4 –
Long life þ EM error 1.59 4.60 9.13 11.1 14.5 19.2

Conclusions
The tests described in this paper show that it is possible to conduct rotorcraft
spectrum fatigue tests under close control at frequencies far higher than what is
typically achieved in the standard spectrum fatigue tests performed in industry.
These results demonstrate that tests can be performed at frequencies up to
150 Hz with a nominal level of damage accumulation error less that 1 % under
most conditions. When electrical measurement system error is accounted for,
the damage accumulation error is less than 2 % under most conditions. The
results also show that it is important to consider the shape of the constant-
amplitude S-N curves in the gigacycle fatigue range when performing spectrum
tests out to long fatigue lives. Materials with shallow slope S-N curves will be
much more sensitive to P-V range errors in large amplitude cycles than materi-
als with steeper sloped S-N curves, which can significantly increase the damage
accumulation error and scatter in a given test. For the test frame controller
compensation schemes investigated here, ALC under constant frequency con-
trol provides the lowest damage accumulation error for all of the configurations
and load spectra tested here. These results are specific to the load spectra used
in this investigation. Results for other types of variable-amplitude spectrum
may be significantly different depending on the range of load and displacement
amplitudes and the amount of random variation present in the loading
sequence. Standard considerations regarding accelerated life testing must also
be taken into account when performing such tests. The influence of higher test
frequencies on material damage mechanisms and damage accumulation should
also be characterized as part of any rigorous test program, especially when
atmospheric or environmental effects such as corrosion may be present. Addi-
tional time or temperature-dependent test factors may change the damage
mechanisms and damage accumulation behavior enough to measurably alter
the long-life test results, even when the test control errors due to these addi-
tional factors are fully accounted for in the test protocols.
References
[1] Bathias, C., “Gigacycle Fatigue of High Strength Steels Prediction and Mecha-
nisms,” Fracture Mechanics: Applications and Challenges, 13th European Conference
on Fracture, Vol. 26, San Sebastian, Spain, European Structural Integrity Society,
2000, pp. 163–171.
[2] Murakami, Y., Metal Fatigue: Effects of Small Defects and Nonmetallic Inclusions,
Elsevier, Oxford, 2002, pp. 273–303.
[3] Shiozawa, K., and Lu, L., “Internal Fatigue Failure Mechanism of High Strength
Steels in Gigacycle Regime,” Key Eng. Mater., Vol. 378–379, 2008, pp. 65–80.
[4] Marines, I., Bin, X., and Bathias, C., “An Understanding of Very High Cycle Fatigue
of Metals,” Int. J. Fatigue, Vol. 25, 2003, pp. 1101–1107.
[5] Bathias, C., and Paris, P. C., Gigacycle Fatigue in Mechanical Practice, Marcel Dek-
ker, New York, 2005.
[6] Dickson, B., Roesch, J., Adams, D., and Krasnowski, B., “Rotorcraft Fatigue and
Damage Tolerance,” 25th European Rotorcraft Forum, Sept. 14-16, Rome, Italy, No.
N10, 1999.

[7] Viswanathan, S. P., Tata, V., Boorla, R., McLeod, G., and Slack, J., “A Statistical
Analysis to Assess the Reliability of a Rotorcraft Component in Fatigue,” 43rd An-
nual Forum, May 18-20, St. Louis, MO, American Helicopter Society, 1987.
[8] Thompson, A. E., and Adams, D. O., “A Computational Method for the Determina-
tion of Structural Reliability of Helicopter Dynamic Components,” 46th Annual
Forum, May 21-23, Washington, DC, American Helicopter Society, 1990.
[9] Stanzl-Tschegg, S., and Mayer, H., “Fatigue and Fatigue Crack Growth of Alumi-
num Alloys at Very High Numbers of Cycles,” Int. J. Fatigue, Vol. 23, 2001, pp.
S231–S237.
[10] Pöting, S., Traupe, M., Hug, J., and Zenner, H., “Variable Amplitude Loading on a
Resonance Test Facility,” J. ASTM Intl., Vol. 1, No. 10, 2004, pp. 67–80.
[11] Morgan, J. M., and Milligan, W. M., “A 1kHz Servohydraulic Fatigue Testing Sys-
tem,” High Cycle Fatigue of Structural Materials, The Minerals, Metals and Materials
Society (TMS) - American Institute of Mining, Metallurgical and Petroleum Engi-
neers (AIME), Warrendale, PA, 1997, pp. 305–312.
[12] ASTM E606-04e1, 2010, “Standard Practice for Strain-Controlled Fatigue Testing,”
Annual Book of ASTM Standards, Vol. 3.01, ASTM International, West Consho-
hocken, PA, pp. 611–626.
[13] Soderling, S., Sharp, M., and Leser, C., “Servo Controller Compensation Methods,
Selection of the Correct Technique for Test Applications,” VII International Mobil-
ity Technology Conference & Exhibit, Sao Paulo, Brazil, SAE Tech. Pap. 1999-01-
3000, SAE International, Warrendale, PA, 1999.
[14] Edwards, P. R., and Darts, J., “Standardised Fatigue Loading Sequences for Heli-
copter Rotors, Helix and Felix; Part 1: Background and Fatigue Evaluation,” NLR
TR 84043 U, National Aerospace Laboratory NLR, Amsterdam, Netherlands.
[15] Edwards, P. R., and Darts, J., “Standardised Fatigue Loading Sequences for Heli-
copter Rotors, Helix and Felix; Part 2: Final Definition of Helix and Felix,” NLR TR
84043 U, National Aerospace Laboratory NLR, Netherlands.
[16] ASTM E1942-98: Standard Guide for Evaluating Data Acquisition Systems Used in
Cyclic Fatigue and Fracture Mechanics Testing, Annual Book of ASTM Standards,
Vol. 3.01, ASTM International, West Conshohocken, PA, 2010, pp. 1186–1197.
[17] Plummer, A. R., “Control Techniques for Structural Testing: A Review,” Proc. Inst.
Mech. Eng., IMechE Conf., Part I: J. Syst. Control Eng., Vol. 221, 2007, pp. 139–169.
[18] McKeighan, P. C., Fess, II., F. E., Petit, M., and Campbell, F. S., “Quantifying the
Magnitude and Effect of Loading Errors During Fatigue Crack Growth Testing
Under Constant and Variable Amplitude Loading,” Applications of Automation
Technology in Fatigue and Fracture Testing and Analysis: Vol. 4, ASTM STP 1411,
A.A. Braun, P.C. McKeighan, A.M. Nicolson and R.D. Lohr, Eds., ASTM Interna-
tional, West Conshohocken, PA, 2002, p. 146.
[19] Donald, J. K., and George, K., “Variable Amplitude Fatigue Crack Growth Using
Digital Signal Processing Technology,” J. ASTM Intl., Vol. 1, No. 9, pp. 53–66.
[20] Paris, P. C., and Erdogan, F., “A Critical Analysis of Crack Propagation Laws,”
ASME J. Basic Eng., Vol. D85, 1963, pp. 528–534.
[21] Metallic Materials Properties Development and Standardization (MMPDS), MMPDS-
04, Federal Aviation Administration, Washington, DC, 2008.
[22] Dowling, N. E., Mechanical Behavior of Materials, 2nd Ed., Prentice Hall, NJ, 1999.
[23] ASTM E1049-85: Standard Practices for Cycle Counting in Fatigue Analysis, An-
nual Book of ASTM Standards, Vol. 3.01, ASTM International, West Conshohocken,
PA, 2010, pp. 710–718.
[24] Glinka, G., and Kam, J. C. P., “Rainflow Counting Algorithm for Very Long Stress
Histories,” Int. J. Fatigue, Vol. 9, No. 3, 1987, pp. 223–228.

[25] Smith, K. N., Watson, P., and Topper, T. H., “A Stress-Strain Function for the Fa-
tigue of Metals,” J. Mater., Vol. 5, No. 4, 1970, pp. 767–778.
[26] Dowling, N. E., Arcari, A., Calhoun, C. A., and Moore, D. C., “Strain-Based Fatigue
for High-Strength Aluminum Alloys,” 65th Annual Forum, May 27-29. Grapevine,
TX, American Helicopter Society, 2009.
[27] Miner, M. A., “Cumulative Damage in Fatigue,” J. Appl. Mech., Vol. 12 Trans.
ASME, Vol. 67, 1945, pp. A159–164.

FRACTURE MECHANICS

doi:10.1520/JAI103924
T. Fongsamootr1 and S. Bernard2
FEM Analysis of a DCP Implant on a Human

Femoral Bone With a Fracture Gap
ABSTRACT: Our research aims to determine the optimal screw configuration

of a dynamic compressive plate (DCP) implant on a human femoral bone. The
number of screws and the positioning are sensitive parameters of DCP
implant stress repartition. Several previous studies have assessed the influ-
ence of thescrew configuration of a DCP implant. Using a realistic geometry
of a human left femur and the finite element method (FEM), the calculations in
those papers were based on a safe femoral bone. This study evaluates the
influence of the application of a simulated fracture gap in the diaphyseal part
on the stress repartition of the bone, plate, and screws. The main purpose is
to complete the existing studies in order to provide surgeons with information
on an optimal prosthesis screw configuration. The plate and screws were
modeled and assembled on a cracked femoral bone. The hip region of the fe-
mur was loaded with vertical and horizontal forces. The femoral bone was cut
into two parts because of the gap: the top part, close to thehip, and the bottom
part, close to the knee. The FEM analysis shows that the stresses in screws
located in the top part of the femoral bone had significantly increased,
whereas the stresses on the plate and the bone had been reduced.
KEYWORDS: fracture gap, femur, DCP implant, finite element method,
stress
Introduction
Femoral fracture is a frequent injury that usually involves an expensive and
essential surgical procedure. A simple or severe fall is often the cause of this
fracture. Osteoporosis makes bones weak and more likely to fracture. Anyone
can develop osteoporosis, but it is more common in older woman. As many as
Manuscript received April 17, 2011; accepted for publication November 1, 2011;
published online November 2011.
1
Dept. of Mechanical Engineering, Chiang Mai Univ. (CMU), Chiang Mai, Thailand.
2
Institut Français de Mécanique Avancée (IFMA), Clermont-Ferrand, France.
Cite as: Fongsamootr, T. and Bernard, S., “FEM Analysis of a DCP Implant on a Human
Femoral Bone With a Fracture Gap,” J. ASTM Intl., Vol. 9, No. 1. doi:10.1520/JAI103924.
Copyright V
321

half of all women and one-quarter of men older than 50 will break a bone due to
osteoporosis. Femur fractures are a major cause of morbidity, often leading to
death among the elderly. In younger patients, femoral fractures are usually the
result of high-energy physical trauma such as a car crash or sport injuries.
There are different types of femoral fractures. The most common is the
proximal fracture, located in the hip region; this can be either an intertrochan-
teric or a femoral neck fracture (Figs. 1 and 2). Both of these types of fractures
account for 90% of the proximal femoral fractures occurring in elderly patients.
This study will focus on short oblique and transverse fractures of the femur
(Fig. 3). Previous studies on these fractures have been conducted by Cheung
et al. [1], Lestviboonchai et al. [2], and others [3–6].
Some fractures do not require the placement of prosthesis. A simple reduc-
tion, minimal handling of bone fragments, and the natural ability of the bone to
repair itself is enough to obtain complete healing. But some fractures require
the application of an implant. To achieve fast healing using biological osteosyn-
thesis, bone fragments must stay attached to their soft tissues so as not to dis-
rupt the blood supply. In severe cases, combinations of implants are used to
maintain alignment of the fracture, to stabilize it, and to rebuild the bony col-
umn. The application of a combination of prosthesis systems involves invasive
surgery and generally leads to disruption of the blood supply; thus indicating a
longer healing period. Because of the obvious difference between intertrochan-
teric or femoral neck fractures and short oblique or transverse fractures, several
types of prosthesis have been developed: e.g., the intramedullary nail and the
dynamic compressive plate (DCP). To determine which implant to apply, sur-
geons use radiography, direct sight, and clinical trials. This method can lead to
prosthesis failure in some cases. Our study will focus on one of these systems:
the dynamic compressive plate (DCP) that is attached to the bone with screws
(Fig.4). A DCP implant is usually used to heal short oblique and transverse
FIG. 1—Femoral neck fracture.

FONGSAMOOTR AND BERNARD, doi:10.1520/JAI103924 323
FIG. 2—Intertrochanteric fracture.
fractures of the femur, and can also be used in conjunction with an intramedul-
lary nail for intertrochanteric or femoral neck fractures. Applying the patient’s
weight on the injured leg can lead to failure of the plate or screws if the bone is
not completely reconstructed.
FIG. 3—Radiographies of diaphyseal fractures.

FIG. 4—Radiographies of diaphyseal femoral fracture in a 7-month-old dog following a

reduction and fixation of a DCP implant.
Previous finite element method (FEM) studies conducted on the DCP sys-
tem by Elkholy [7], Lestviboonchai et al. [2], and Fongsamootr et al. [8] have
shown that the screw parameter was assumed to be the most influential ele-
ment. However, those studies were performed using a nonfracture femur for
the FEM calculus. Our study aims to complete the results obtained by Fongsa-
mootr et al. [8]. The main difference is the application of a simulated fracture
gap in the diaphyseal part of the femur. The analysis of the results and a com-
parison with those of Fongsamootr et al. [8] will provide additional information
to identify the best screw configuration in order to avoid prosthesis failure.
Material and Methods
Bone Material Properties

Bone is the structural support of our body and exists in a variety of shapes with
complex internal and external structure, allowing them to be lightweight yet
strong. Human bones are living tissues that constantly evolve. Their heterogene-
ous structure is composed of two main forms of bone tissues: cortical bone and
cancellous bone.
Cortical bone, also known as compact bone, is dense and forms the surface
of bones, contributing 80% of the weight of a human skeleton. It is extremely
hard, and is formed of multiple stacked layers with few gaps. Its main functions
are to support the body, protect organs, provide leverage for movement, and (to-
gether with cancellous bone) store minerals.

Trabecular or cancellous bone is spongy and makes up the bulk of the inte-
rior of most bones, including the vertebrae. Cancellous bone is a type of osseous
tissue, with a low density and strength but very high surface area, that fills the
inner cavity of long bones. The external layer of cancellous bone contains red
bone marrow, where the production of blood cellular components (known as
hematopoiesis) takes place. Cancellous bone is also where most of the arteries
and veins of bone organs are found.
Bone behavior is also assumed to be anisotropic; Reilly and Burstein eval-
uated the five elastic moduli of bone. Thanks to the three-dimensional images
generated by computed tomography scans (commonly called CT scans) we can
now achieve a good bone model. Indeed, it provides a density field of the bone
and can generate a cuboid mesh with very precise element size (see Lang et al.
[9]). Then an elastic model can be applied using the relations of Garcı́a et al. [10]

2014q2:5 for q 1:2g=cm3
EðMPaÞ ¼
1763q3:2 for q 1:2g=cm3
and

0:2 for q 1:2g=cm3
t¼
0:32 for q 1:2g=cm3
where:
E ¼ Young’s modulus,
q ¼ bone density, and
t ¼ Poisson’s ratio.
These linear relations can easily be computed, but the data on bone density
and a CT scan are needed. As shown in Table 1 [1,2,7,8,10–17] many other linear
behaviors can be found in the literature that are not based on bone density, and
which provide different values for E, t, and ry (tensile yield stress).
To ensure reasonable duration calculus, and because the nonlinear or
orthotropic models based on CT scan computations are not available for free
use, our study will be linear, elastic, and isotropic.
Bone CAD Model

For purposes of simplification, cancellous bone is not addressed in this study. As
shown by Fongsamootr et al. [18] this part has little influence on the model behav-
ior, and moreover it generates contact problems in the analysis. However, unlike
Fongsamootr et al. [18], the femur used for FEM calculus in the present study is
not a safe one; it has a simulated horizontal fracture gap in the diaphyseal part.
The CAD representation of the femur for our study was downloaded from
the “standardized femur” homepage at the International Society of Biome-
chanics Finite Element Repository managed by the Instituti Ortopedici Rizzoli
in Bologna, Italy. This geometry has been used in a large number of experimen-
tal biomechanical studies, and to validate a number of FEM studies. A list of
publications using this geometry can be found on the homepage [17].

TABLE 1—Elastic models proposed in the literature.
Cortical Bone Cancellous Bone
References Date E (GPa) t ry (MPa) E (MPa) t ry (MPa)
[1] Bayraktar et al. 2004 19.9 6 1.8 … 107.9 6 12.3 - … …

[4] Cheung et al. 2004 10 … … 206 … …
J_ID: DOI: Date: 15-June-12
[6] Dong et al. 2004 16.61 6 1.83 0.37 6 0.3 … … … …

[8] Elkholy 1995 16 … … … … …
[9] Fongsamootr et al. 2005 15 0.3 200 500 0.3 50
[10] Garcı́a et al. 2002 f(q)* f(q)* … f(q)* f(q)* …
Stage:
[12] Keyak et al. 2001 f(q)* 0.4 … f(q)* 0.4 …

[14] Kotha et al. 2005 19.3 6 2.9 … 100.5 6 19.1 … … …
[16] Lestviboonchai et al. 2005 10 0.3 … … … …
[17] Margolis et al. 2004 22.125 6 6.24 … … … … …
Page: 326
[18] Ng et al. 2004 10 0.29 … 100 0.29 …

[19] Wirtz et al. 2003 f(q)* 0.3 150 f(q)* 0.12 3-15
This study 2006 16 0.29 200 350 0.29 20
Note: In these studies mechanical properties are assumed to be a function of bone density.
Total Pages: 17

The femur used for this study was a left femur. As shown by Margolis et al.
[15] the variations in mechanical properties we have chosen to focus on
(Young’s elastic modulus and Poisson’s ratio) are not significantly related to the
side of the body and can be considered as being symmetric. (Indeed, the differ-
ence between left and right Young’s modulus was evaluated at 0.9% and will not
be considered in this study.)
Volumes were recreated from the surfaces in the Initial Graphics Exchange
Specification (IGES) file using SolidWorks Office Premium 2006 SP0.0 (Solid-
Works, Concord, MA). This resulted in a two-part solid assembly, with one part
assumed to be the cortical bone and the other to be the cancellous bone. The
cancellous bone was used only to generate a hole in the cortical bone, because
the cancellous bone was not considered in our study.
To create the fracture gap, we began with a nonfracture femur model from
which the cancellous bone had already been removed. First we cut the femur
model into two parts—the top part (close to the hip) and the bottom part (close
to the knee)—to obtain two independent models; these were then reassembled
to obtain the complete model (Fig. 5).
Plate and Screws

The DCP chosen was a centered-holes version (Fig. 6), as was the one used by
Fongsamootr et al. [18], with a thickness of 5 mm, a width of 15 mm, and a
length of 200 mm. The distance between each hole was 10 6 1.5 mm.
The diameter of the screws was 2 mm, that may appear very small in com-
parison to some previous studies (e.g., Elkholy [7] who used screws with
diameters between 4 and 5.24 mm). But the screws were the same size used
by Fongsamootr et al. [8,18]—and in any case their size is of no real impor-
tance because of the linear elastic model chosen. The length of each screw
was fixed at 34 mm (without the head), as in Fongsamootr et al. [8,18], in
order to be long enough to cross through the bone from one side to the other.
The head of the screw was modeled with a spherical countersink, allowing a
sphere-sphere contact between the plate’s holes and screws. The plate and
screws were made of stainless steel, whose mechanical properties are reported
in Table 2.
Contact Conditions
The contact conditions were defined as the bound between bones and screws
and between screws and plate, with the plate kept free in relation to the outside
surface of the cortical bone.
In contrast to the study by Fongsamootr et al. [18], a fracture gap was
applied in the model. The gap size was fixed at 0 mm; therefore the top and
bottom parts remained in contact. The experimental study led by Lestviboon-
chai et al. [2] showed that the friction coefficient between bone to bone is
0.3. The contact conditions between the two plane surfaces of the top and
bottom parts in contact were defined as no penetration with a friction coeffi-
cient of 0.3.

FIG. 5—CAD model (using SolidWorks) of a left femur with a fracture gap of 0 mm
with DCP and two screws.
FIG. 6—DCP chosen for this study, with centered holes numbering.

TABLE 2—Mechanical properties of steel used in surgery.
E (GPa) t ry (MPa)
195 0.3 1000
Forces and Boundary Conditions

Many studies have been conducted to determine how to apply the forces. To
simulate body weight and muscle forces, Elkholy [7] defined a one-point force,
while Keyak et al. [13,19,20] used surface pressure (or multinode forces), and
Duda et al. [21] used several lines of pressure.
Bergmann et al. [22] showed that postural attitude and patient activity
(slow walking, fast walking, climbing stairs, etc.) are highly influential parame-
ters to assess the forces applied around the hip contact. Cheung et al. [1] have
also taken gait loading into account in their report.
Bergmann et al. [22] identified an average loading of the femoral bone in
normal walking of around 238% of body weight. In the present study this was
defined as the constant resulting force that is applied on the femur, as follows:
( qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
F2X þ F2Y ¼ ð2:38BWÞ2 (1)
3:3FX ¼ 0:2FY

FX ¼ 111:44N
) (2)
FY ¼ 1838:77N
where:
BW ¼ body weight (774 N ¼ mean weight of a male body, according to a
national measuring campaign of the French population in 2006),
FX ¼ horizontal component of the total applied force, and
FY ¼ vertical component of the total applied force.
Ratio (1) is taken from the study by Elkholy [7], that also used a two punc-
tual force model.
The force applied in this study is the same as used by Fongsamootr et al.
[18], a two punctual force model. The result was in accordance with Berg-
mann’s and Elkholy’s ratios. The two forces applied on a small surface at the
top of the bone model represent the weight that the body exerts on the femur
(Fig. 7). The forces due to muscle were disregarded.
The constraints were classically defined (as in most of the literature) in the
foot area of the femur (the condyle region near the knee), and were assumed to
be built-in conditions with zero degree of freedom displacement allowed. In
order to allow the application of this boundary condition to the cortical bone,
the model was cut in the condylar region. This condition ensures the stability of
the load-restraint system and allows the matrix decomposition algorithm to
converge.

FIG. 7—Forces and boundaries applied on the femur. (a) The surface in white on which
the forces are applied represents the hole of the intramedullary channel. (b) The bottom
surface is fixed.
Mesh and Solving

A free mesh of the femur using tetrahedral 10-node elements was computed
using the SolidWorks mesher (Fig. 8). The second-order shapes function of
these elements ensures a mesh that will be close to the bone’s boundary
FIG. 8—Mesh was computed on SolidWorks. There were between 60 941 and 68 322 ele-
ments with an average size of 3.27 mm.

surfaces. Hexahedral elements are known to be more accurate than tetrahedral

ones, but these elements are not available in SolidWorks. Also, an Ansys or Aba-
qus finite element mesher would have been more precise using hexahedral ele-
ments and would have provided a more accurate mesh, but the complexity of
our model did not enable us to use them.
The number of elements varied between 60 941 and 68 322 depending on
the number of screws used in the model. The average size of the elements was
3.27 mm.
The static direct sparse solver of SolidWorks was used to solve the calculus
on a computer (Intel Pentium 1.6 GHz, 640 MB RAM). The calculation time was
approximately 45 min.
Results
Fongsamootr [8] and Elkholy [7] have shown that the number of screws and
their positioning may have an influence on the stress distribution of the parts.
The study by Fongsamootr et al. [18] evaluated these effects on a safe femur
bone. To complete their results, under the same conditions we tested a femur
with a fracture gap of 0 mm instead of a safe femur.
The analysis was performed using the maximum von Mises stress. No plas-
ticity criterion was set, and the materials were assumed to be perfectly elastic
even over their yield stress limit.
As in the survey of Fongsamootr et al. [18], ten models were created with
four different numbers of screws (two, four, six, and eight screws). For each
number of screws, one or several patterns were analyzed, as shown in Table 3
(the numbering convention is the one given in Fig. 6).
The main consequence of the presence of a fracture gap was to raise the
stresses in the screws located in the top part of the femur bone; this was the
case for each pattern. The stresses in these screws nearly doubled or tripled in
some cases (Fig. 9).
On the contrary, the stresses in the screws of the bottom half of the prosthe-
sis system were not significantly modified, except for screw 1 whose stresses
were reduced. We were unable to detect a clear difference from the “no crack”
model for the bottom part screws. Without a fracture gap, the extreme screws
(1 and 8) were by far the most stressed, as shown by Fongsamootr et al. [18].
With the presence of a fracture gap, the stresses in screw 1 were close to the
stresses of the other screws in the bottom half of the prosthesis system. The
stresses in screw 8 can reach high values—more than 1600 MPa in patterns 4.1
and 6.2, for example. Screw 8 was still the most stressed screw, but the stress
was not as obvious as in the case without a gap. Because of the crack, the
stresses in the plate were significantly reduced, except for pattern 6.2 where we
can observe a peak value due to local effects.
The stresses in the cortical bone (top and bottom parts) were divided by two
in comparison with the analysis of Fongsamootr et al. [18] of a nonfracture fe-
mur bone. However, as in this study the influence of screw number and their
positioning had only a minimal effect on cortical stress values (Figs. 10 and 11).
Contrary to Fongsamootr et al. [18], we did not observe a peak value for pattern

TABLE 3—Maximum Von Mises stress (MPa) in parts in relation to the patterns.
Top Bottom
Number Cortical Cortical
of Screws Pattern Screw 1 Screw 2 Screw 3 Screw 4 Screw 5 Screw 6 Screw 7 Screw 8 Plate Bone Bone
2 2.1 744.3 1436 447 175.9 146.4

4 4.1 674.5 829.6 1301 1630 408 197.5 149.8

4.2 551.4 721.1 1400 1312 403.3 170.5 128.8
4.3 850.2 849.8 1033 1316 221.3 172.3 130.7
Stage:
4.4 655.2 749.4 1398 1222 173.7 170.3 161.1

4.5 659.6 727.4 1180 1169 174.3 155.4 151.9
6 6.1 643.9 541.8 600.9 1145 1268 1454 263.7 164.8 134.4
6.2 526.2 603 727.8 1177 1325 1615 670.8 250.4 141.7
Page: 332
6.3 595.9 679 758.8 1170 1107 1239 178.8 233.4 126
8 8.1 530.8 602.7 574.4 654.4 934.1 1237 1142 1292 237.2 168.8 126.7
Total Pages: 17

FIG. 9—Maximum von Mises stress in parts in relation to the number of screws and the chosen pattern in a safe femur, as determined
by Fongsamootr et al. [22]. The peak value for pattern 4-4 shows that a damaging local effect can occur if the screws at the two
Stage:
extremities are removed. The average and max values of the stress in all screws has been plotted, instead of individual screw values,
for easier reading.
Page: 333
Total Pages: 17
FIG. 10—Maximum von Mises stress in parts in relation to the number of screws and the chosen pattern in a cracked femur. The

application of a fracture gap in our CAD model led to an increase of the stress in the screws. The average and max values of the stress
in all screws has been plotted, instead of individual screw values, for easier reading.
Stage:
FIG. 11—Maximum von Mises stress in parts in relation to the number of screws and the chosen pattern in a cracked femur. The
application of a fracture gap in our CAD model led to an increase of the stress in the screws. The average and max values of the stress
Page: 334
in all screws has been plotted, instead of individual screw values, for easier reading.
Total Pages: 17

4.4 when the fracture gap was applied. However we did observe very high maxi-
mum stress value in screws, and a peak value for the plate and the cortical
bone, for pattern 6.2.
The average stress value in screws, as shown in Fig. 9, decreases as the num-
ber of screws increases. This interesting phenomenon was quite predictable,
but was less significant than in cases without a gap. This gives rise to the idea
that different screw diameters can be used in the same prosthesis system.
As shown by Fongsamootr et al. [18] in their study on a nonfracture femur,
the extreme screws are the most stressed. They concluded that large diameters
could be chosen for the extreme up and down screws, while small diameters
could easily be used for the middle screws in order to reduce the implant’s
weight and to improve its resistance. The stress analysis of our study pointed
out a significant rise of the stresses in the top part screws. Taking these results
into account, all the top part screws should have their diameters increased.
Finally, the involvement of a fracture gap induces local effects in the top
half of the prosthesis system that are highly damaging, particularly to the top
part screws (5, 6, 7, and 8). To ensure the reliability of the prosthesis system the
results from both studies, with and without a gap, should be taken into
consideration.
Discussion and Conclusion

The determination of the prosthesis parameters is essential and complex. A
wrong choice can lead to stress concentration in the implant and can eventually
cause system failure. The application of a fracture gap in the FEM model has
revealed some important consequences, such as a substantial increase of the
stresses in the top part screws. The analysis has also shown that a peak value
could appear in the plate or in the cortical bone in some screw configurations
(for example pattern 6.2). Consequently, a screw configuration that generates a
significant rise in stress should be avoided, because it can lead to the failure of
the screws or the plate.
To reduce the weight of the implant and to improve its resistance, it might
be desirable to use screws with different diameters, as Fongsamootr et al. [18]
suggest. A study could be conducted to determine the ideal diameter for each
screw in the model using optimization. Elkholy [7] has already applied optimi-
zation on the plate-screw-bone model, but did not attempt to choose different
diameters for the screws in the same implant.
Material properties have a great influence in every finite elements analysis.
Research by Keyak et al. [13,19,20] and Bayraktar et al. [11] has helped improve
the accuracy of information regarding material properties. As the elastic models
have shown their limits for bones, the use of a nonlinear, heterogenic and/or
orthotropic law could be suitable research tracks for future studies.
As described in Bone Material Properties, the bone properties are very scat-
tered because of the variation among the specimens, and it is neither physically
nor ethically acceptable to eliminate the scatter completely. To take into
account this variability, reliability studies can provide interesting results: con-
sidering; for example, E or BW as random variables distributed by a normal

law. In order to determine the sensitivity of all design parameters, the Monte
Carlo reliability method can be used, as in the study by Ng et al. [16] on cervical
bones.
FEM and stress analysis are an inexpensive way to provide information to
surgeons. Nevertheless the FEM models involved a number of simplifying
assumptions that may have reduced the robustness of the models and the accu-
racy of the results. For example, the torque applied to the screws, and the fric-
tion between the plate and the bone, were not taken into account. The bone was
considered to be isotropic, whereas it is assumed to be anisotropic. This study is
linear and elastic. All these simplifications may have an influence on the results.
This points out the necessity of using a more powerful finite element solver, or
the need to pursue further studies that take those parameters into account.
Another solution could be to work on a CT scan-created model that could easily
handle the nonlinear, heterogenic and/or orthotropic laws.
Acknowledgments
The authors would like to thank Thailand Research Fund (TRF) for support.
References
[1] Cheung, G., Zalzal, P., Bhandari, M., Spelt, J. K., and Papini, M., “Finite Element
Analysis of a Femoral Retrograde Intramedullary Nail Subject to Gait Loading,”
Med. Eng. Phys., Vol. 26, 2004, pp. 93–108.
[2] Lestviboonchai, T., Manonukul, A., and Rhodkwan, S., “Numerical Investigation of
Dynamic Compression Plate Attached on Fracture Human Femur Subjected to
Static Loading,” 19th Conference of Mechanical Engineering Network of Thailand,
Phuket, Thailand, Oct 19–21, 2005, pp. 233–238.
[3] Beale, B., “Orthopedic Clinical Techniques Femur Fracture Repair,” Clin. Techniq.
Small Animal Practice, Vol. 19, 2004, pp. 134–150.
[4] Cordey, J., Borgeaud, M., and Perren, S. M., “Force Transfer Between the Plate and
the Bone: Relative Importance of the Bending Stiffness of the Screws Friction
Between Plate and Bone,” Injury, Vol. 31, 2000, pp. S3-C21–28.
[5] Wirtz, D. C., Pandorf, T., Portheine, F., Radermacher, K., Schiffers, N., Prescher,
A., Weichert, D., and Niethard, F. U., “Concept and Development of an Orthotropic
FE Model of the Proximal Femur,” J. Biomechan., Vol. 36, 2003, pp. 289–293.
[6] Wirtz, D. C., Schiffers, N., Pandorf, T., Radermacher, K., Weichert, D., and Forst,
R., “Critical Evaluation of Known Bone Material Properties to Realize Anisotropic
FE-Simulation of the Proximal Femur,” J. Biomechan., Vol. 33, 2000, pp.
1325–1330.
[7] Elkholy, A. H., “Design Optimization of the Hip Nail-Plate-Screws Implant,” Com-
put. Methods Programs Biomed., Vol. 48, 1995, pp. 221–227.
[8] Fongsamootr, T., Latourte, F., and Blanche, E., “FEM Analysis of a Plate-Screw
Implant of the Femoral Human Bone,” Project Report, Chiang Mai Univ., 2005.
[9] Lang, T. F., Keyak, J. H., Heitz, M. W., Augat, P., Lu, Y., Mathur, A., and Genant,
H. K., “Volumetric Quantitative Computed Tomography of the Proximal Femur:
Precision and Relation to Bone Strength,” Bone, Vol. 21, 1997, pp. 101–108.
[10] Garcı́a, J. M., Doblaré, M., and Cegoñino, J., “Bone Remodelling Simulation: A
Tool for Implant Design,” Comput. Mater. Sci., Vol. 25, 2002, pp. 100–114.

[11] Bayraktar, H. H., Morgan, E. F., Niebur, G. L., Morris, G. E., Wong, E. K., and
Keaveny, T. M., “Comparison of the Elastic and Yield Properties of Human Femoral
Trabecular and Cortical Bone Tissue,” J. Biomechan., Vol. 37, 2004, pp. 27–35.
[12] Dong, X. N., and Guo, X. E., “The Dependence of Transversely Isotropic Elasticity
of Human Femoral Cortical Bone on Porosity,” J. Biomechan., Vol. 37, 2004, pp.
1281–1287.
[13] Keyak, J. H., Rossi, S. A., Jones, K. A., Les, C. M., and Skinner, H. B., “Prediction of
Fracture Location in the Proximal Femur Using Finite Element Models,” Med. Eng.
Phys., Vol. 23, 2001, pp. 657–664.
[14] Kotha, S. P., and Guzelsu, N., “Tensile Behavior of Cortical Bone: Dependence of
Organic Matrix Material Properties on Bone Mineral Content,” J. Biomechan., Vol.
40, 2007, pp. 36–45.
[15] Margolis, D. S., Lien, Y. H. H., Lai, L. W., and Szivek, J. A., “Bilateral Symmetry of
Biomechanical Properties in Mouse Femora,” Med. Eng. Phys., Vol. 26, 2004, pp.
349–353.
[16] Ng, H. W., Teo, E. C., and Lee, V. S., “Statistical Factorial Analysis on the Material
Property Sensitivity of the Mechanical Responses of the C4–C6 Under Compres-
sion, Anterior and Posterior Shear,” J. Biomechan., Vol. 37, 2004, pp. 771–777.
[17] Papini, M., “Third Generation Composite Femur,” available at http://www.tecno.
ior.it/VRLAB/researchers/repository/BEL_repository.html (Last Accessed in July
2007).
[18] Fongsamootr, T., and Pottier, T., “FEM Analysis of a DCP Implant on a Human
Femoral Bone,” Society for the Advancement of Material and Process Engineering
(SAMPE) Conference, Baltimore MD, May 18–21, 2009.
[19] Keyak, J. H., “Improved Prediction of Proximal Femoral Fracture Load Using Non-
linear Finite Element Models,” Med. Eng. Phys., Vol. 23, 2001, pp. 165–173.
[20] Keyak, J. H., and Rossi, S. A., “Prediction of Femoral Fracture Load Using Finite
Element Models: An Examination of Stress- and Strain-Based Failure Theories,” J.
Biomechan., Vol. 33, 2000, pp. 209–214.
[21] Duda, G. N., Heller, M., Albinger, J., Schulz, O., Schneider, E., and Claes, L.,
“Influence of Muscle Forces on Femoral Strain Distribution,” J. Biomechan., Vol.
31, 1998, pp. 841–846.
[22] Bergmann, G., Deuretzbacher, G., Heller, M., Graichen, F., Rohlmann, A., Strauss,
J., and Duda, G.N., “Hip Contact Forces and Gait Patterns From Routine
Activities,” J. Biomechan., Vol. 34, 2001, pp. 859–871.


doi:10.1520/JAI103962
Zhaoyu Jin1 and Xin Wang2
Point Load Weight Functions for

Semi-Elliptical Cracks in Finite
Thickness Plate
ABSTRACT: This paper presents the application of the weight function

method for the calculation of stress intensity factors for surface semi-elliptical
cracks in finite thickness plates subjected to arbitrary two-dimensional stress
fields. A new general mathematical form of point load weight function has
been formulated by taking advantage of the knowledge of a few specific
weight functions for two-dimensional planar cracks available in the literature
and certain properties of the weight function in general. The existence of the
generalized form of the weight function simplifies the determination of a spe-
cific weight function for specific crack configurations. The determination of a
specific weight function is reduced to the determination of the parameters of
the generalized weight function expression. These unknown parameters can
be determined from reference stress intensity factor solutions. This method is
used to derive the weight functions for semi-elliptical surface cracks in finite
thickness plates. The derived weight functions are then validated against
stress intensity factor solutions for several linear and non-linear two-dimen-
sional stress distributions. The derived weight functions are particularly use-
ful for the fatigue crack growth analysis of planar surface cracks subjected to
fluctuating nonlinear stress fields resulting from surface treatment (shot
peening), stress concentration, or welding (residual stress)
KEYWORDS: semi-elliptical surface crack, weight function, stress intensity
factor, non-linear stress distribution
Manuscript received May 10, 2011; accepted for publication September 20, 2011;
published online October 2011.
1
Dept. of Mechanical and Aerospace Engineering, Carleton Univ., Ottawa, Ontario K1S 5B6,
Canada.
2
Dept. of Mechanical and Aerospace Engineering, Carleton Univ., Ottawa, Ontario K1S 5B6,
Canada. (Corresponding author), email: xwang@mae.carleton.ca
Cite as: Jin, Z. and Wang, X., “Point Load Weight Functions for
Semi-Elliptical Cracks in Finite Thickness Plate,” J. ASTM Intl., Vol. 9, No. 2. doi:10.1520/
JAI103962.
Copyright V
338
ID: mohameda Time: 21:10 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120235/APPFile/AI-STP#120235

JIN AND WANG, doi:10.1520/JAI103962 339
Introduction
The stress intensity factor was introduced by Irwin as a measure of the strength
of the singularity. It is important in determining crack-tip stress fields, and
many different methods have been devised for obtaining it. However, it is diffi-
cult to calculate stress intensity factors for defective components subjected to a
complex stress distribution. Normally, most stress intensity calculating meth-
ods require a separate analysis of each load and geometry configuration, such
as the finite element analysis (FEA). Bueckner introduced the weight function
method to analyze two-dimensional elastic cracks in 1970 [1]. Rice’s research
[2,3] also developed the weight function method for two-dimensional and three-
dimensional elastic cracks. The advantage of the weight function is that it only
depends on cracked geometry. Once the weight function is known for a given
cracked geometry, the stress intensity factor due to any load system applied to
the body can be determined by using the same weight function. Acquiring the
accurate solution of the weight function is the key to the successful use of the
weight function method. The methods for obtaining the weight functions for
one-dimensional cracks have been well developed [4–6]. However, for two-
dimensional cracks, the methods of obtaining the weigh functions are not as
well developed, and are the topic of ongoing research; see [7], for example.
Semi-elliptical surface cracks are among the most common flaws in engi-
neering structural components (Fig. 1). Accurate stress intensity factors of these
surface cracks are needed for reliable predictions of fatigue crack growth rates
and for fracture assessment. The most common method to analyze surface
FIG. 1—Geometry and coordinate system of semi-elliptical surface cracks in the finite
thickness plate.

cracks is the finite element method. Stress intensity factor results for semi-
elliptical surface cracks in finite thickness plates were obtained by Raju and
Newman [8,9] and Shiratori et al. [10]. However, only remote tension and bend-
ing loading were analyzed in [8,9]. Meanwhile, Shiratori et al. only applied con-
stant, linear, parabolic, or cubic stress distributions on the crack face [10].
Weight functions for one-dimensional stress variations were developed for sur-
face cracks in [11,12]. However, these weight functions are only applicable for
one-dimensional stress variations (through the plate thickness). In engineering
applications, the stress distributions can be two-dimensional, it is therefore nec-
essary to develop point load weight functions which enable the calculation of
stress intensity factors under arbitrary two-dimensional stress distributions.
In [7], a new general mathematical form of point load weight function has
been formulated by taking advantage of the knowledge of a few specific weight
functions for two-dimensional planar cracks available in the literature and cer-
tain properties of the weight function in general. The determination of a specific
weight function is then reduced to the determination of the parameters of the
generalized weight function expression. These unknown parameters can be
determined from reference stress intensity factor solutions. In this paper, this
method is extended to derive the weight functions for semi-elliptical surface
cracks in finite thickness plates. The derived weight functions are then validated
against stress intensity factor solutions for several linear and non-linear two-
dimensional stress distributions.
Approximate Point Load Weight Functions
Theoretical Background
The weight function technique for calculating stress intensity factors is based
on the principle of superposition. For one-dimensional cracks, it can be shown
[1] that the stress intensity factor for a cracked body (Fig. 2(a)) subjected to the
external loading system S is the same as the stress intensity factor in a geometri-
cally identical body (Fig. 2(c)) with the local stress field r(x) applied to the crack
faces. The local stress field r(x) induced in the prospective crack plane by pre-
load S is determined from an uncracked body (Fig. 2(b)). The stress intensity
factor for a cracked body with loading applied to the crack surface can be calcu-
lated by integrating the product of the weight function mðx; aÞ and the stress dis-
tribution rðxÞ in the crack plane
ða
K¼ rðxÞmðx; aÞdx (1)
0
where a is the crack length. The weight function m(x, a) depends only on the ge-
ometry of the crack and the cracked body. Once the weight function has been
determined, the stress intensity factor for this geometry can be obtained from
Eq 1 for any stress distribution, r(x). Mathematically, the weight function
m(x, a) is the Green’s function for the present boundary value problem scaled
with respect to the crack dimension a. It represents the stress intensity factor at

FIG. 2—Weight function for one-dimensional cracks; (a), (b), and (c).
the crack tip for a pair of unit point loads acting on the surface at the location x.
The method for determining weight functions m(x, a) are well established; see
[4–6].
For a two-dimensional crack, the stress intensity factors vary along the
crack front, as shown in Fig. 3. The counterpart to Eq 1 for two-dimensional
cracks is a double integral over the crack surface
ðð
KðP0 Þ ¼ rðx; yÞmðx; y; P0 ÞdS (2)
FIG. 3—Two-dimensional crack under two-dimensional stress distribution.

where mðx; y; P0 Þ is the point load weight function. It represents the stress inten-
sity factor at a general point P0 on the crack front for a pair of unit point loads
acting on the crack surface at pointðx; yÞ, and rðx; yÞ is a general two-
dimensional stress distribution, as shown in Fig. 3.
If the stress distribution r(x, y) is one-dimensional, for example, only a
function of x, then Eq 2 can be simplified to
ða ð ða
0
KðP Þ ¼ rðxÞ mðx; y; P Þdy dx ¼ rðxÞMðx; P0 Þdx
0
(3)
0 0
where M(x; P0 ) represents the stress intensity factor at point P0 for unit line load
at position x as shown in Fig. 3, and a is the crack depth in the x-direction. In
other words, M(x; P0 ) is the line-load weight function for two-dimensional
cracks.
For any one-dimensional or two-dimensional cracks, if the weight functions
m(x, a), m(x, y; P0 ), or M(x; P0 ) are obtained, the stress intensity factors for other
loading conditions can be calculated using Eqs 1, 2 or 3.
For semi-elliptical surface cracks in finite thickness plates, the weight func-
tions M(x; P0 ) have been derived in [11,12]. However, these weight functions are
only applicable for one-dimensional stress variations rðxÞ, in accordance with
Eq 3. There are no solutions of the general point load weight function m(x, y; P0 )
available that can handle two-dimensional stress variations using Eq 2.
Formula of Point Load Weight Function for Semi-Elliptical Surface Cracks

In a recent work, Wang and Glinka [7] have proposed a weight function form
for embedded elliptical cracks. Consider the embedded elliptical crack in Fig. 4;
the point load weight function form was suggested as the series expansion form
pffiffiffiffiffi " #
2s Xn
rðuÞ t
0
mðx; y; P Þ ¼ 3=2 2 1 þ Mi ðh; aÞ 1 ; n ¼ 1; 2; 3… (4)
p q i¼1
RðuÞ
where s is the shortest distant from the load point P to the boundary of the crack
front, and q is the distance between the load point P and the point P0 under con-
sideration, as shown in Fig. 4. Note here that h is the angle related to point P0 ,
and u is the angle associated with the load point P(x, y). The radii r(u;) and R(u;)
are shown in Fig. 4. The weight function is expressed through the coefficients of
the expansion, i.e., parameters Mi(h, a). These parameters (M-factors) are func-
tions of the location of the crack front (through angle h) and the aspect ratio of
the ellipse, a ¼ a=c. The determination of point load weight functions for a par-
ticular crack is now simplified to the determination of these M-factors. In [7],
this general expression in Eq 4 was used to derive point load weight functions
for embedded elliptical cracks in both infinite and semi-finite thickness plates. In
addition, it was also found that only one term expansion (n = 1) of Eq 4 is needed
to provide excellent approximations for these weight functions [7].
In this paper, the weight function formula for an elliptical crack is applied
to derive the weight function of a semi-elliptical surface crack in the thickness

FIG. 4—Weight function for an embedded elliptical crack.
plate. The geometry and the coordinate system used to analyze the plate con-
taining a semi-elliptical surface crack are shown in Fig. 1. The weight function
will depend on the aspect ratio of the ellipse, a and the thickness of the plate, t.
Since the surface crack geometry is not an embedded crack geometry
(Fig. 5(a)), a symmetric half is added (Fig. 5(b)). Here, we are dealing with open-
ing mode loading conditions; therefore, one virtual symmetric load was added
to account for the crack mouth effect, see Fig. 5(b). The virtual load point is
symmetric about the general load point P. The weight function expression for
the surface crack then becomes
pffiffiffiffiffi
0 2s a rðuÞ
mðx; y; P Þ ¼ 3=2 2 1 þ M h; a; 1
p q t RðuÞ
pffiffiffiffiffiffi
2s0 a rðuÞ
þ 3=2 02 1 þ M h; a; 1
p q t RðuÞ
pffiffiffiffiffi pffiffiffiffiffiffi !
2s 2s 0 a rðuÞ
¼ þ 1 þ M h; a; 1 (5)
p3=2 q2 p3=2 q02 t RðuÞ
where s0 is the shortest distant from the virtual point load to the boundary of
the crack front, and q0 is the distance between the virtual point load and point
P0 , as shown in Fig. 5(b), due to the symmetry, s’ = s. In addition, parameter M
is also the function of a=t to reflect the finite thickness effect.

FIG. 5—Weight function for semi-elliptical surface cracks; (a) and (b).
Note that in Eq 5 one term expansion of Eq 4 is used (n ¼ 1). Our analyses

have indicated that with one term, i.e., n ¼ 1, we can approximate the point load
weight functions with good accuracy. This provides a good compromise
between the accuracy and complexity of the solutions.
Determination of Weight Function Parameters

Knowing the general weight function form, Eq 5, the derivation of the weight
function for a particular semi-elliptical surface crack is now reduced to the deri-
vation of the parameters M(h, a, a=t) along the entire crack front.

The parameter M(h, a, a=t) can be determined using Eq 2, provided that one
reference stress intensity factor solution Kr is known. The stress distribution
expression and the general weight function expression in Eq 5 can be substi-
tuted for r(x, y) and m(x, y; P0 ) into Eq 2. This leads to the equation for the deter-
mination of the unknown parameters M(h, a, a=t)
ðð pffiffiffiffiffi pffiffiffiffiffiffi !
0 2s 2s0 a rðuÞ
Kr ðP Þ ¼ rr ðx; yÞ 3=2 2 þ 3=2 02 1 þ M h; a; 1 dS (6)
p q p q t RðuÞ
After integration, Eq 6 can be used to solve for M(h, a, a=t). Note that this calcu-
lation needs to be carried out at any point along the crack front to obtain the
corresponding M(h, a, a=t).
Point Weight Function for Semi-Elliptical Surface Cracks

in Finite Thickness Plate
In this section, Eq 6 is applied to derive weight functions for semi-elliptical
cracks in a finite thickness plate. The M-factors, M(h, a, a=t), are determined for
each crack configuration. The derived weight functions are then validated using
stress intensity factor solutions for other loading conditions. Although the cur-
rent method can be used to derive weight functions for any point along the
crack front, since the stress intensity factors at the deepest A and surface points
B (i.e., P0 corresponding to h ¼ p=2 and h ¼ 0; see Figs. 1 and 5(a)) are the most
important values for engineering applications, the weight functions are derived
and validated for these two points.
For any point P0 along the crack front, it can either be identified by the polar
angle h or the parametric angle / (see Figs. 1 and 5(a)), and they are related sim-
ply by tanh ¼ atan/. In the section titled in the following derivations, the para-
metric angle / is used to represent point P0 . The factor M(h, a, a=t) is solved at
the deepest point (which corresponds to / ¼ p=2), and the surface point (which
corresponds to / ¼ 0).
The parameters Mðh; a; a=tÞ are determined by using one reference stress in-
tensity factor solution. The finite element results of the stress intensity factor
calculated by Shiratori et al. and Wang [10,12] were chosen as a reference and
for the verification of the weight functions. Four types of loading were applied
to the crack surface in each crack geometry, with the following stress
distributions:
xn
rðxÞ ¼ r0 1 (7)
a
where n ¼ 0, 1, 2 or 3, r0 is the nominal stress, and a is the crack depth.

The stress intensity factor results, under uniform stress distributions, will
be used to calculate the M factor. Then the SIFs under linear, parabolic or cubic
stress distributions will be calculated by the weight function method. Finally,
the results will be compared between the finite element method (FEM) and the
weight function method.

Reference Stress Intensity Factor Solutions

For surface-elliptical crack, as shown in Fig. 1, the stress intensity factor for a
uniform stress field is used as a reference solution. The uniform stress is applied
directly onto the crack face
rðxÞ ¼ r0 (8)
Shiratori et al. and Wang [10,12] calculated the stress intensity factor for the
deepest and the surface points with aspect ratios of 0.05, 0.1, 0.2, 0.4, 0.6, and
1.0, and a=t values of 0.2, 0.4, 0.6, and 0.8 by the FEM. The resulting stress in-
tensity factors were normalized as follows:
K
F¼ pffiffiffiffiffiffi (9)
ðr0 pa=EÞ
where F is the boundary correction factor and E is given by

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
E¼ 1:0 þ 1:464a1:65 (10)
The reference stress intensity factors used are summarised in Table 1.
Determination of Weight Functions

By substituting Eq 8 and the reference stress intensity factor results into Eq 6,
an equation with unknown M(h, a, a=t) is established. Numerical integration is
required to solve for M(h, a, t=a).
TABLE 1—Boundary correction factors at deepest point A and surface point B, used as the
reference, taken from [10,12].
a=t ¼ 0.2 a=t ¼ 0.4 a=t ¼ 0.6 a=t ¼ 0.8
Deepest Point, A a ¼ 0.05 1.245 1.74 2.5875 3.6407

a ¼ 0.1 1.2219 1.5953 2.1328 2.5844
a ¼ 0.2 1.162 1.371 1.651 1.787
a ¼ 0.4 1.119 1.216 1.327 1.378
a ¼ 0.6 1.09 1.143 1.206 1.228
a ¼ 1.0 1.047 1.083 1.106 1.107
a=t ¼ 0.2 a=t ¼ 0.4 a=t ¼ 0.6 a=t ¼ 0.8
Surface Point, B a ¼ 0.05 0.3044 0.3655 0.5029 0.7949

a ¼ 0.1 0.414 0.4901 0.6391 0.9712
a ¼ 0.2 0.582 0.688 0.882 1.201
a ¼ 0.4 0.81 0.911 1.06 1.32
a ¼ 0.6 0.954 1.025 1.192 1.366
a ¼ 1.0 1.145 1.22 1.318 1.441

A computer program was developed to perform the numerical integration

based on the standard Gauss-Legendre quadrature technique to solve for
Mðh; a; a=tÞ in MATLAB. Curved eight-node elements were used to discretize the
entire elliptical areas. The accuracy of the integration algorithm was verified by
comparing the prediction of SIF from the exact point load weight function with
the exact stress intensity factor (SIF) for an embedded penny-shaped crack under
uniform tension; both are available from [13]. The differences are within 1 %.
Next, the present approach was used to develop weight functions for a semi-
circular crack in a semi-infinite plate. The numerical results of the stress inten-
sity factor are available from [13]. The stress intensity factors under a uniform
stress field for the semi-circular crack in a semi-infinite body were chosen for a
reference to calculate the weight function M factors. Additionally, the stress in-
tensity factors under linearly varying stress fields were calculated by the derived
weight functions. The difference between the numerical stress intensity factor
solution from [13] and the prediction based on the present weight function along
the whole crack front was within 4 %. The results are shown in Table 2. Through
these verifications, the present methodology is considered to be accurate for the
derivation of point load weight functions for semi-elliptical cracks.
To perform the numerical integrations for elliptical cracks in Eq 6, an elliptical
transformation was used to generate the mesh used for integration. The lines of
elements around the crack tip were elliptic or hyperbolic, so that the intersecting
lines were orthogonal, as required for the evaluation of the stress intensity factors
[8]. Figure 6 shows a typical mesh used in the present calculations for a=c ¼ 0.2.
TABLE 2—The derived weight function M-factors for semi-circular crack in a semi-infinite
body, and stress intensity factors for linear varying loads.
FWF
Weight FFEM (Weight
Function (Numerical Function
2/=p M-factor Method) [13] Method) jFFEM FWF j F100
FEM
%
0.0625 0.4945 0.9298 0.9434 1.46 %

0.125 0.4009 0.8631 0.8793 1.87 %
0.1875 0.3356 0.7997 0.8155 1.97 %
0.25 0.2876 0.7401 0.7533 1.79 %
0.3125 0.2526 0.6846 0.6937 1.34 %
0.375 0.2286 0.6335 0.6377 0.65 %
0.4375 0.2146 0.5871 0.5858 0.22 %
0.5 0.2099 0.5455 0.5387 1.24 %
0.5625 0.2146 0.5088 0.4971 2.31 %
0.625 0.2286 0.4771 0.4612 3.32 %
0.6875 0.2526 0.4503 0.4318 4.11 %
0.75 0.2876 0.4284 0.4092 4.48 %
0.8125 0.3356 0.4114 0.3941 4.22 %
0.875 0.4009 0.3994 0.3873 3.02 %
0.9375 0.4945 0.3921 0.3907 0.38 %

FIG. 6—Typical mesh used for numerical integration (a=c ¼ 0.2).
Based on reference SIFs and Eq 6, the results for the parameters Mðh; a; a=tÞ at
point A and B are obtained. The data are presented in Table 3. The aspect ratios con-
sidered are a=c ¼ 0.05, 0.1, 0.2, 0.4, 0.6, and 1.0, and a=t values of 0.2, 0.4, 0.6, and
0.8, respectively. The results of these M-factors are plotted in Fig. 7. For engineering
applications, the empirical equations are fitted from the data in Table 3 using the
least squares method. The empirical equations of Mðh; a; a=tÞ are given in the Appen-
dix. Comparisons of M-factors from Table 3 and from empirical equations are also
shown in Fig. 7. For the deepest point A, the accuracy of the equations is within 3 %
of the numerical data; for the surface point B, the accuracy is generally within 2 %,
except for a=c ¼ 1, and a=t ¼ 0.6 and 0.8, where the differences are within 7 and
23%, respectively. These empirical equations are readily implemented into the com-
puter program to calculate the stress intensity factors according to Eq 2.
Validation of Weight Functions

Several linear and nonlinear loading cases were applied to the surface of the
semi-elliptical crack to validate the derived weight functions in the form of
Eq 5. The weight function factors were calculated from the empirical formulas,
TABLE 3—Weight function parameter Mðh; a; a=tÞ at points A and B, for a ¼ 0.05, 0.1, 0.2,
0.4, 0.6, and 1 and a=t ¼ 0.2, 0.4, 0.6, and0.8.
a=t ¼ 0.2 a=t ¼ 0.4 a=t ¼ 0.6 a=t ¼ 0.8
Deepest point, A a ¼ 0.05 0.4075 1.6131 3.6771 6.2421

a ¼ 0.1 0.413 1.3728 2.7734 3.9437
a ¼ 0.2 0.3456 0.9431 1.7436 2.1324
a ¼ 0.4 0.2957 0.6124 0.9749 1.1447
a ¼ 0.6 0.1952 0.418 0.457 0.449
a ¼ 1.0 0.0523 0.0943 0.1879 0.192
a=t ¼ 0.2 a=t ¼ 0.4 a=t ¼ 0.6 a=t ¼ 0.8
Surface point, B a ¼ 0.05 2.1099 2.0691 11.4667 31.4384

a ¼ 0.1 0.951 1.6909 6.8636 18.3927
a ¼ 0.2 0.2327 2.125 5.5882 11.2828
a ¼ 0.4 0.7496 1.6894 3.0759 5.4952
a ¼ 0.6 0.6446 1.0984 2.1659 3.2782
a ¼ 1.0 0.3468 0.6522 1.0512 1.5521

FIG. 7—Mðh; a; a=tÞ for aspect ratios considered are a=c = 0.05, 0.1, 0.2, 0.4, 0.6, and
1.0, and a=t values of 0.2, 0.4, 0.6, and 0.8. (a) Deepest point A, and (b) surface point B.
and then were incorporated into Eq 2. The stress intensity factors along the
crack front of a semi-elliptical crack of aspect ratio a ¼ 0.05, 0.1, 0.2, 0.4, 0.6,
and 1.0, and the a=t values of 0.2, 0.4, 0.6, and 0.8 were calculated for the follow-
ing stress fields:
Uniform stress field
rðx; yÞ ¼ r0 (11)

Linear stress field

x
rðx; yÞ ¼ r0 1 (12)
a
Parabolic stress field

x 2
rðx; yÞ ¼ r0 1 (13)
a
Cubic stress field

x 3
rðx; yÞ ¼ r0 1 (14)
a
The boundary correction factors F (following Eq 9) from the weight function

predictions were compared with the numerical solutions from Shiratori et al.
and Wang [10,12]. The results at the deepest points and the surface points are
shown in Figs. 8–15 for uniform, linear, parabolic, and cubic stress fields,
respectively. Note that the uniform stress distribution is the reference case.
At the deepest point A, the prediction from the weight functions and the fi-
nite element data are generally within 6 % for all the loadings, as shown in Figs.
8–11. For the surface point B, the weight function based stress intensity factors
are generally within 6.5 % for all the loadings (Figs. 12–15). Overall, very good
FIG. 8—Comparisons of the weight function based SIF and FEA data for the deepest
point under constant stress distribution.

point under linear stress distribution.
point under parabolic stress distribution.

point under cubic stress distribution.
FIG. 12—Comparisons of the weight function based SIF and FEA data for the surface
point under constant stress distribution.

point under linear stress distribution.
point under parabolic stress distribution.

point under cubic stress distribution.
agreements are achieved. The derived weight function can be used to predict
the SIFs for other complex two-dimensional stress distributions.
Conclusions
The point load weight functions are derived for semi-elliptical cracks in a finite
thickness plate. One reference stress intensity factor solution is used to derive
these weight functions. It is demonstrated that this method gives very accurate
weight functions for a wide range of geometric configurations for semi-elliptical
cracks. The empirical equations of the weight functions are readily implemented
into computer code. The derived weight functions are suitable for calculating
stress intensity factors for semi-elliptical cracks under complex stress distribu-
tions such as nonlinear stress fields resulting from surface treatment (shot peen-
ing), stress concentration, or welding (residual stress). They are particularly
useful for the fatigue crack growth analysis of surface cracks in engineering com-
ponents whose crack shape remains semi-elliptical during the entire fatigue life.
Acknowledgments
The writers gratefully acknowledge the financial support from the Natural Sci-
ences and Engineering Research Council (NSERC) of Canada and the Ontario
Centres of Excellence (OCE). They also are grateful to Prof. G. Glinka for help-
ful discussions on the subject.

APPENDIX
The weight function parameters Mðh; a; a=tÞ for a semi-elliptical surface crack in
a finite thickness plate presented in Table 3 at the deepest and surface points
were fitted into empirical formulas as follows. For the deepest point A, the accu-
racy of the equations is within 3 % of the numerical data; for the surface point
B, the accuracy is generally within 2 % except for a=c ¼ 1, and a=t ¼ 0.6 and 0.8,
where the differences are within 7 and 23 %, respectively.
For the deepest point (which corresponds to / ¼ p=2)
a a 6 a4 a2
M h; a; ¼ B1 þ B2 þ B3 þ B4 (A1)
t t t t
a4 a 3 a2 a a1

B1 ¼ A11 þ A12 þ A13 þ A14 þ A15 þ A16
c c c c c
a4 a 3 a2 a a1

B2 ¼ A21 þ A22 þ A23 þ A24 þ A25 þ A26
c c c c c
a4 a 3 a2 a a1

B3 ¼ A31 þ A32 þ A33 þ A34 þ A35 þ A36
c c c c c
a4 a 3 a2 a a1

B4 ¼ A41 þ A42 þ A43 þ A44 þ A45 þ A46
c c c c c
2 3
6202 13010 9201 2691 8:832 310:9
6 4872 10090 7014 2017 8:932 237 7
A¼6
4 474:8 933:4
7
623 197:4 0:07013 35:9 5
46:87 108:3 89:79 32:45 0:04374 3:845
For the surface point (which corresponds to / ¼ 0)

a a6 a4 a2
M h; a; ¼ C1 þ C2 þ C3 þ C4 (A2)
t t t t
a4 a 3 a2 a ha i1
C1 ¼ D11 þ D12 þ D13 þ D14 þ D15 þ 0:2 þ D16
c c c c c
a4 a 3 a2 a ha i1
C2 ¼ D21 þ D22 þ D23 þ D24 þ D25 þ 0:2 þ D26
c c c c c
a4 a 3 a2 a ha i1

C3 ¼ D31 þ D32 þ D33 þ D34 þ D35 þ 0:2 þ D36
c c c c c
a4 a 3 a2 a ha i1

C4 ¼ D41 þ D42 þ D43 þ D44 þ D45 þ 0:2 þ D46
c c c c c

2 3
13560 31780 27490 11830 635:9 3091
6 12320 29080 25510 11260 643:1 3049 7
6
D¼4 7
416:4 784:4 477:8 124:8 5:049 13:82 5
10:43 15:39 0:7795 12:81 3:149 9:925
References
[1] Bueckner, H. F., “A Novel Principle for the Computation of Stress Intensity
Factors,” ZAMM, Vol. 50, 1970, pp. 529–545.
[2] Rice, J., “Some Remarks on Elastic Crack Tip Field,” Int J. Solids Struct., Vol. 8,
1972, pp. 751–758.
[3] Rice, J., “Weight Function Theory for Three-Dimensional Elastic Crack Analysis,”
Fracture Mechanics: Perspectives and Directions (Twentieth Symposium), ASTM
Spec. Tech. Publ., Vol. 1020, 1989, pp. 29–57.
[4] Wu, X. R. and Carlsson, A. J., Weight Functions and Stress Intensity Factor Solu-
tions, Pergamon Press, Oxford, 1991.
[5] Fett, T. and Munz, D., Stress Intensity Factors and Weight Functions, Computational
Mechanics Publications, Southampton, 1997.
[6] Glinka, G. and Shen, G., “Universal Features of Weight Functions for Cracks in
Mode I,” Eng. Fract. Mech., Vol. 40, 1991, pp. 1135–1146.
[7] X. Wang, X. and Glinka, G., “Determination of Approximate Point Load Weight
Function for Embedded Elliptical Cracks,” Int. J. Fatigue, Vol. 31, 2009, pp.
1816–1827.
[8] Raju, I. S. and Newman, J. C., Jr., “Stress Intensity Factors for a Wide Range of
Semi-Elliptical Surface Cracks in Finite Thickness Plates,” Eng. Fract. Mech., Vol.
11, 1979, pp. 817–829.
[9] Newman, J. C., Jr. and Raju, I. S., “Analysis of Surface Cracks in Finite Plates
Under Tension and Bending Loads,” NASA TP-1578, 1979, Langley Research Cen-
ter, Hampton, VA.
[10] Shiratori, M., Miyoshi, T., and Tanikawa, K., “Analysis of Stress Intensity Factors
for Surface Cracks Subjected to Arbitrarily Distributed Surface Stresses,” Stress In-
tensity Factors Handbook, Vol. 2, Y.Murakami, Ed., Pergamon Press, Oxford, 1978,
pp. 725–727.
[11] Shen, G. and Glinka, G.,“Weight Function for a Surface Semi-Elliptical Crack in a
Finite Thickness Plate,” Theor. Appl. Fract. Mech., Vol. 15, 1991, pp. 247–255.
[12] Wang, X. and Lambert, S. B., “Stress Intensity Factors for Low Aspect Ratio Semi-
Elliptical Surface Cracks in Finite-Thickness Plates Subjected to Nonuniform
Stresses,” Eng. Fract. Mech., Vol. 51, No. 4, 1995, pp. 517–532.
[13] Tada, H., Paris, P. C., and Irwin, G. R., The Stress Analysis of Cracks Handbook, 2nd
ed., Paris Production Incorporated, and Del Research Corporation, St. Louis, MO,
1985.


doi:10.1520/JAI103979
Yasuhito Takashima,1 Mitsuru Ohata,1 Masaru Seto,2

Yoshitomi Okazaki,3 and Fumiyoshi Minami1
Evaluation of Fracture Toughness Test Data

for Multilayer Dissimilar Joint Welds Using
a Weibull Stress Model
ABSTRACT: Fracture behaviour of welded joints with dissimilar weld metals

(WMs) has been investigated in this paper. A low-toughness WM and a
high-strength WM with a moderate toughness were layered alternately in the
thickness direction. Fracture tests were conducted with 3-point bend (3PB)
specimen and tension plate with a through-thickness crack. The 3PB speci-
men showed a multistage fracture. The first fracture was originated from the
low-toughness WM. On the other hand, no multistage fracture occurred in the
tension plate, although the low-toughness WM was responsible for brittle frac-
ture initiation in the same manner as in the 3PB specimen. The fracture tough-
ness of the multilayer dissimilar joint was apparently larger than the welded
joint made with the low toughness WM only. The weakest-link model has been
applied for analyzing those toughness properties of the multilayer dissimilar
joint. The estimated fracture toughness by the weakest-link model was not
necessarily consistent with the toughness data, which was because of the
local stress elevation in the low-toughness WM area close to the high-strength
WM. Because of this, a modified Weibull stress model was used to estimate
fracture toughness values for the multilayer dissimilar joint WM and to account
for an active fracture zone length scale in the multilayer weld joint materials.
KEYWORDS: fracture toughness, Weibull stress, multilayer dissimilar weld
metals, strength mismatch, weakest link model
online March 2012.
1
Graduate School of Engineering, Osaka Univ., Osaka, Japan.
2
Kawasaki Heavy Industries, Ltd., Kobe, Japan.
3
Kobe Steel, Ltd., Kobe, Japan.
Cite as: Takashima, Y., Ohata, M., Seto, M., Okazaki, Y. and Minami, F., “Evaluation of
Fracture Toughness Test Data for Multilayer Dissimilar Joint Welds Using a Weibull
Stress Model,” J. ASTM Intl., Vol. 9, No. 3. doi:10.1520/JAI103979.
Copyright V
357

Introduction
The construction welds of liquid natural gas (LNG) storage tanks are usually by
tungsten inert gas (TIG) or metal inert gas (MIG) welding of 9 % Ni–steel plate
sections. LNG tank construction costs are increased when an austenitic weld
metal containing 70 % Ni is used. Some reduction in welding cost can be
achieved by using MIG welding with a weld metal containing 9 % Ni. However,
during cooldown of single-pass 9 % Ni MIG welds of 9 % Ni steels, austenitic to
ferrite transformations can occur in the weld metal. The transformation to fer-
rite can cause embrittlement phases to form in the weld metal at the nominal
welding temperatures of LNG tanks. The embrittlement can be reduced by sub-
sequent reheat welding passes, but the variability in reheat passes creates weld
zone metal phases that contain unknown percentages of softened and hardened
weld compositions. During storage tank construction conditions, the composi-
tion percentages of the weld metal are not well controlled and depend on local
differences in the welding temperature, the weld reheat temperatures, and the
weld cooldown temperature rates. Thus, multi-pass LNG welds that use 9 % Ni
with reheat passes will have spatial metallurgical and mechanical property het-
erogeneities that are difficult to evaluate and will introduce uncertainty in frac-
ture toughness values of the weld metal.
The uncertainty in fracture toughness values because of localized embrittle-
ment regions has been investigated. Satoh et al. [1–6] performed many fracture
tests and estimated fracture toughness values for cross-bond-type notched
specimens that had random localized embrittled regions in front of the crack.
During the tests, the load-displacement data before fracture initiation were not
affected by the random spatial positions of metallurgical and mechanical
heterogeneities in front of the crack-tip; and the subsequent cleavage fracture
initiation occurred in front of the crack-tip only at spatial positions of the local-
ized embrittlement heterogeneities. For tests performed on specimens with
weld metal heterogeneities, the lower limits for fracture toughness values
approached the fracture toughness value of specimens with only embrittlement
weld metal in front of the crack-tip [2]. Based on these test data, it was con-
cluded that the random size of the embrittlement heterogeneities in front of the
crack-tip determined the scatter in fracture toughness values [5]. Given this
conclusion, Satoh et al. [6] developed a probabilistic method that uses a
weakest-link model to estimate fracture toughness values for welds containing
embrittlement regions.
In the following, differences of fracture toughness values were investigated
for three different weld processes of high-strength steel (HT780) joints. The
weld processes were multi-pass TIG welds with two different filler weld metals,
namely, a general purpose carbon steel–ER70S weld metal, a 9 % Ni weld metal,
and an alternating layered combination weld with these two dissimilar filler
weld metals. Combination ER70S and 9 % Ni-layered welds can have a substan-
tial welding fabrication cost reduction for LNG storage tank construction com-
pared to only austenitic 70 % Ni welds and compared to only 9 % Ni welds,
provided that fracture toughness values of the combination layered welds are
adequate. The investigation will provide both fracture toughness test data and a

TAKASHIMA ET AL., doi:10.1520/JAI103979 359
modified weakest-link model to determine probable fracture toughness values

of alternating layered welds in LNG storage tank constructions.
Experiments and Finite-Element Analysis

A high-strength structural steel of 780 MPa strength class (HT780) with thick-
ness t ¼ 25 mm was welded by TIG welding. Double-V groove geometry was
adopted as shown in Fig. 1. The welding conditions are shown in Table 1. Three
types of welded joint test specimens (M-joint, L-joint, H-joint) were made with
two types of welding electrodes, arbitrary waveform synthesizer A5.18 ER70S-G
and TIG wire for 9 % Ni steel with similarly composed nickel alloy (ER9Ni). The
L-joint specimens were welded with only ER70S-G filler metal and the H-joint
specimens were welded with only 9 % Ni filler metal. The M-joint specimens
were welded with alternating layers of ER70S-G filler metal and 9 % Ni filler
metal. The chemical compositions and mechanical properties of the HT780 steel
and welding electrodes are given in Tables 2 and 3, respectively.
The cross section of M-Joint is shown in Fig. 2. Figure 3 shows the Vickers
hardness distribution along the welded cross sections of L-, H-, and M- joint
specimens. The hardness of weld metal (WM) 9Ni is higher than WM ER70S.
The hardness data of WM ER70S and WM 9Ni in the M-joints are close to the
WM made of the same welding electrode in the L-joint and the H-joint,
respectively.
Tensile tests and V-notch Charpy tests were performed on L-joint and H-joint
WM specimens. The tensile test specimens were cylindrical and were extracted
from the WMs in the length direction as shown in Fig. 4. The Charpy specimens
had a vertical V-notch located at the center of the L- and H-joint WM cross sec-
tion and the notch location is also shown in Fig. 4. Tensile test results are shown
in Fig. 5 and Table 3. The yield stress and tensile strength of WM 9Ni in H-joints
are higher than WM ER70S in L-joints. The Charpy-absorbed energy values for
the ER70S L-joint weld specimens were temperature dependent and between
80 and 20 C the values were significantly less than those for 9 %Ni H-joint
weld specimens (Fig. 6). The WM ER70S has lower toughness than WM 9Ni.
FIG. 1—Groove geometry used.

TABLE 1—Welding conditions.
M-joint L-joint H-joint
Base metal HT780 steel HT780 steel HT780 steel

Welding process TIG TIG TIG
Shielding gas 100%Ar 100%Ar 100%Ar
Welding electrode ER70S-G (/ 1.2 mm) ER70S-G (/ 1.2 mm) ER9Ni (/ 1.2 mm)
ER9Ni (/ 1.2 mm)
Weld heat input, kJ/mm 1.7 2.5 1.7 2.5 1.7 2.5
Fracture tests were conducted with 3-point bend (3PB) specimens and ten-
sion plate specimens with through-thickness cracks. The configuration of these
test specimens is shown in Fig. 7. Both the 3PB specimen and the edge through-
thickness crack plate (ETCP) specimens have through-thickness cracks at the
center of the multilayered dissimilar WMs. For the 3PB specimen, the crack
length a0 including fatigue pre-crack is 25 mm, which is equal to half of the
specimen width W (50 mm). The length of the one-side edge crack in the ETCP
specimens is 15 mm. The width of the ETCP specimens is 100 mm.
The crack-tip opening displacements (CTOD) fracture toughness tests were
conducted in the temperature range from 60 C to 100 C. The cracked panel
tension tests on ETCP specimens were conducted at a test temperature of
80 C, which corresponds to the temperature at which brittle fracture occurred
in the WM with 3PB specimens. The specimens were cooled in a 80 C 6 2 C
test temperature bath for at least 25 min prior to testing.
CTOD, for the 3PB specimens, was calculated from the load and crack
mouth displacement Vg, according to the procedure specified in the BS 7448
Part-2 [7]. CTOD values for the ETCP specimens were calculated by means of
the Dugdale and Bilby, Cottrell, Smith (DBCS) model [8]. In this study, the yield
stress value in Table 3 for ER70S WM was used to calculate CTOD values.
The near crack-tip stress–strain fields in the layered weld materials of the
3PB and the ETCP specimens were numerically evaluated with a general
purpose three-dimensional (3D) finite element (FE) code [ABAQUS Standard
Ver-6.7]. The FE models used in the analysis are shown in Fig. 8. Because of
symmetry, half of the 3PB and one-quarter of ETCP specimens were modelled.
The FE analysis used the eight-node element with eight Gaussian integration
points; and the minimum element size near the crack-tip had dimensions of
0.03 0.03 0.2 (mm).
TABLE 2—Chemical composition of HT780 steel and welding electrodes (wt. %).
C Si Mn P S Others
HT780 steel 0.25 0.30 1.49 0.015 0.007 Mo, Ni, Cr, Ti, B
ER70S-G 0.09 0.73 1.35 0.009 0.010 –
ER9Ni 0.02 0.01 0.38 0.002 0.005 Ni

TABLE 3—Mechanical properties of HT780 steel and welding electrodes.
Yield stress, Tensile strength, Yield-to-tensile Uniform

MPa MPa ratio elongation, %
HT780 steel 734 806 0.91 6.2

WM ER70S 509 605 0.84 11.3
WM 9Ni 732 758 0.97 8.3
Note: gage length (G.L.) ¼ 25 mm (HT780), 32 mm (WM ER70S, WM 9Ni).
In the FE analysis, the following Swift type, power-hardening law was used
for the elastic–plastic material response
¼ Cð1 þ e p =aÞn
r (1)
where r and e p are the equivalent stress (Mises stress) and equivalent plastic
strain, respectively, C is the elastic limit, and n and a are material constants (n
being a strain-hardening coefficient). These mechanical property values for the
different materials were measured in the round bar tension tests. The stress–
strain relationships used in the analysis are shown in Fig. 9. Values for the yield
stress rY, tensile strength rT, and uniform elongation eT (nominal strain at rT)
are shown in Table 4, and the values were measured at 80 C, which corre-
sponded to the fracture test temperature. CTOD values for the 3PB specimens
were calculated by using the procedure BS 7448-Part-2 [7]. CTOD values for the
ETCP specimens were calculated by the following equation based on the DBCS
model [8]
FIG. 2—Cross section of welded joints with dissimilar WMs.

FIG. 3—Vickers hardness distribution in thickness direction of WMs.
dETCP ¼ Vtip =Vedge Vg (2)
Vtip and Vedge are opening displacement at crack-tip and crack-edge, respec-
tively. The Vtip and Vedge are calculated by using the DBCS model. Vg is clip
gauge opening displacement obtained by the experiments.
Fracture Behaviour of Multilayer Dissimilar Joints

Load versus displacement curves were measured to investigate the fracture
toughness behavior of different weld materials. Some load-clip gauge opening
FIG. 4—Extraction of round-bar specimen from WM.

FIG. 5—Nominal stress–nominal strain curves obtained by round-bar tension test.
displacement curves obtained by the CTOD fracture toughness testing are

shown in Fig. 10. The curves in Fig. 10 for multilayered M-joint 3PB specimens
show that multi-stage fractures occurred at 80 C and 100 C, but multi-stage
fracture did not occur for the 60 C test shown in Fig. 10(a). These test results
FIG. 6—Charpy toughness of WM.

FIG. 7—Configuration of test specimens with a crack at the center of WM.

FIG. 8—FE-models used in this study.

FIG. 9—Equivalent stress–equivalent plastic strain relationships of HT780 steel and

WMs used in FE analysis.
contrast with CTOD fracture test results at 80 C for L-joint 3PB specimens
that had only ER70S WM and did not show multi-stage fracture behavior.
The fracture appearance of the 3PB specimen tested at 80 C is shown in
Fig. 11. Cleavage fracture was not found in the WM 9Ni. It appears that cleavage
cracks were initiated and propagated in the ER70S WM layers and were
arrested in the more ductile, high-toughness 9 % Ni WM layers. This implies
that the crack extensions during multi-stage pop-in fracture events originate in
the low-toughness ER70S WM layers, and the ER70S WM layer responds as a
weakest-link layer relative to an adjacent 9 % Ni WM layer.
Critical CTOD values at first fracture initiation in M-joint specimens are
compared with L-joint specimen CTOD values in Fig. 12. The critical CTOD val-
ues measured for L-joint shows a good fit to a Weibull two-parameter distribu-
tion with a slope shape parameter value of 2. On the other hand, the CTOD
values for the M-joint layered weld specimens have a slope shape parameter
value of 3.7, which is larger than L-joint. In Fig. 12, the critical CTOD values for
M-joint weld specimens were larger than those for L-joint weld specimens for
low-fracture toughness specimens. However, for high-fracture toughness speci-
mens, the critical CTOD values of M-joint and L-joint specimens are similar.
TABLE 4—Mechanical properties used in FE-analysis.
Yield stress, MPa Tensile strength, MPa Uniform elongation, %
HT780 791 813 6.2

WM ER70S 573 613 11.3
WM 9Ni 740 766 8.3

FIG. 10—Load-clip gauge opening displacement relationship obtained by CTOD frac-

ture toughness test for M-joint.
Both tension plate specimens, with double-edge though-thickness cracks,

and 3PB specimens with multilayer welds have brittle cleavage cracks in the
ER70S WMs for fracture tests at temperatures of 80 C. However, multi-stage
fracture did not occurred in the tension plate, although the low-toughness WM
was responsible for brittle fracture initiation in the same manner as in the 3PB
specimen. Critical CTOD values for ETCP specimens with M-joint welds are
compared with 3PB specimen of M-joint (Fig. 13). In Fig. 13, critical CTOD val-
ues for the tension plate specimens are consistently higher than those for the
3PB specimens, even at the test temperature of 80 C brittle cleavage fractures
were observed for both types of specimens. These differences in fracture tough-
ness CTOD values are believed to be a result of near crack-tip strain constraint
loss effects for tension plate specimens relative to the 3PB specimens.
Evaluation of Fracture Toughness for Multilayer Dissimilar Joints

The fracture toughness values of the multilayer dissimilar joint (M-joint) were
apparently larger than the welded joint made with the low-toughness WM only
(L-joint). The differences between multi-metal M-joint and single-metal L-joint
weld specimens have been discussed, and the differences in their CTOD fracture
toughness values compared.

FIG. 11—Fracture appearance for 3PB specimen of M-joint.
FIG. 12—Critical CTOD measured by 3PB specimen of M-joint and L-joint.

FIG. 13—Cumulative distribution of critical CTOD for 3PB and ETCP specimens
of M-joint.
Thickness effect on fracture toughness value has been widely investigated

[9–12]. The critical CTOD values obey the Weibull two-parameter distribution
Fðdcr Þ ¼ 1 expfðdcr =d0 Þa g (3)
where, a and d0 are the Weibull-shape parameter and Weibull-scale parameter,

respectively. According to the weakest-link model, the scale parameter d0 shows
a size dependence of the form
d0;L1 ¼ d0;L2 ðL2 =L1 Þ1=a (4)
where, d0,L1 and d0,L2 are the scale parameter of critical CTOD for specimen
thickness L1 and L2, respectively. Equation 4 has been widely used to predict
size effect in fracture toughness values [9–13].
The shape parameter for L-joint weld specimen equals 2. Therefore, the crit-
ical CTOD in M-joint, dcrM-joint could be calculated from that in L-joint, dcrL-joint,
as follows:
dcr Mjoint ¼ dcr Ljoint ðLeff Ljoint =Leff Mjoint Þ1=2 (5)
where LeffM-joint and LeffL-joint are effective lengths for cleavage fracture, which
corresponds to total thickness of WM ER70S in the welded joints. The estimated
dcrM-joint with Eq 5 is shown in Fig. 14. The weakest link prediction with dcrL-joint

FIG. 14—Predicted critical CTOD in 3PB specimen of M-joint based on the weakest-
link model and the Weibull stress criterion from test results obtained with L-joint
specimen.
overestimates the critical CTOD for multilayer dissimilar joints in larger CTOD
levels than 0.05 mm, whereas the predicted value agrees well with experimental
data in the lower toughness level.
The influence of strength mismatch in thickness direction on crack opening
stress has been evaluated with 3D finite element stress analysis. An example of
the stress field predicted by a finite element analysis at spatial points in ER70S
WM for the M-joint weld is approximately the same as that in ER70S WM for
the L-joint as shown in Fig. 15. However, a more detailed spatial resolution of
the stress component normal to the crack plane for the multilayered M-joint
weld compared to the L-joint weld is plotted in Fig. 16, and shows that the aver-
age opening stress is roughly 30 % higher in the 9 % Ni WM relative to the
stress in adjacent layers of ER70S WM. Near the boundary interface of the 9 %
Ni and ER70S WM layers, the opening stress in the ER70S metal is locally higher
by about 15 %, about half the difference between the averaged opening stresses
in the two WMs. Thus, the yield stress mismatch in Table 4 for the two WMs
greatly affects the spatial profile details of the opening stress component at adja-
cent layer interfaces in front of the crack-tip. However, for weakest-link model
applications, where an idealized brittle metal layer is adjacent to a ductile layer,
the affects of the brittle metal layer on the opening stress spatial profile are
expected to be significantly less. Thus, the predictions of an idealized weakest-
link model for the CTOD value of dcrM-joint are not necessarily consistent with the

FIG. 15—Comparison of crack opening stress fields between M-joint and L-joint.
experimental toughness data for multilayered layered welded joints when the
WMs are ductile.
The layered effects of the strength mismatches in the thickness direction on
stress fields was estimated in the evaluations of fracture toughness values for
the welded joints with multilayer dissimilar WMs. An effective fracture Weibull
stress [14] was estimated by a moving volume averaging integration of the crack
FIG. 16—Crack opening stress distribution in thickness direction for M-joint and
L-joint.

opening stress at spatial points in front of the crack-tip. Therefore, it is expected

that the Weibull stress could characterize the volume effect of WM ER70S in
multilayered WMs on fracture toughness values.
An effective fracture Weibull stress was evaluated by integrating a near-tip
stress reff over the fracture process zone Vf in the form
" ð #1=m
1 m
rW ¼ ðreff Þ dV f (6)
V0 Vf
where V0 (¼1 mm3) and m are the reference volume and a material constant,
respectively. Physically, Vf approximates a plastic zone near the crack-tip, and
reff is an effective stress for cleavage fracture in each volume element consider-
ing random orientation of the microcrack. In this analysis, the fracture process
zone was limited to the WM ER70S, on the basis of the fracture behavior of the
multilayered dissimilar joint.
An effective stress, reff, in the above volume integration across multilayered
dissimilar joints is conceptual similar to an effective stress defined in Ref 15.
The value selected for V0 did not affect the transferability analysis of fracture
mechanics test results, and the value of the empirical parameter m was inde-
pendent of the value selected for volume V0. In other studies that used local vol-
ume integral averaging for a Weibull stress parameter [12], the size of unit
volume was 1 mm3, which is also the volume size used in this study. The Wei-
bull stress defined in Eq 6 is considered a stress metric for the fracture driving
force. Therefore, in this research work, it is assumed that the critical Weibull
stress rW,cr is independent of the specimen geometry.
The cumulative distribution of the critical Weibull stress rW.cr at the onset
of brittle fracture is shown in Fig. 17. The Weibull parameters m-value and ru
were determined with the test result of L-joint by using a maximum likelihood
method [12]. The Weibull stress for M-joint was calculated with m ¼ 19.8. It was
assumed that the m-value is independent of the strength mismatch between dis-
similar WMs. The data plotted in Fig. 17 do not show marked differences
between estimated rW.cr valued for L-joint and M-joint test specimens.
The procedure used to calculate CTOD fracture toughness values for the M-
joint from the L-joint weld data is illustrated in Fig. 18. First, a value for the pos-
sible critical Weibull stress is estimated for the L-joint weld material with a
stress ru-parameter value equal to 1719 MPa and m-parameter value equal to
19.8. Next, the possible Weibull stress estimate, rW, is used to calculate a corre-
sponding CTOD value for M-joint weld material; and this is the estimate for the
critical CTOD value for the M-joint multilayered weld material. Predicted criti-
cal CTOD values by the Weibull stress criterion is shown in Fig. 14. The Weibull
stress, taking the active fracture process zone into account, enables accurate
estimation of the fracture toughness value of the multilayer dissimilar joint.
The modified Weibull stress criterion developed to calculate CTOD values
M-joint welds was used to predict critical CTOD values for ETCP specimens and
3PB specimens with multilayered WMs. A comparison of the predicted critical
CTOD values with ETCP data is shown in Fig. 19. These numerical predictions

FIG. 17—Cumulative distribution of critical Weibull stress.
FIG. 18—Procedure of fracture toughness evaluation based on the Weibull stress criterion.

FIG. 19—Predicted critical CTOD of multilayer dissimilar joints by means of the

Weibull stress criterion.
of critical CTOD values agreed reasonably well with the available ETCP experi-
mental data. This agreement establishes that a modified Weibull stress criterion
developed for M-joint and L-joint welds can be applied to describe fracture
toughness for other welding geometries.
Conclusions
In this study of alternative welding methods for LNG storage tanks constructed
with 9 % Ni–steel plate, influences on the fracture toughness CTOD values of dif-
ferent single material WMs, multilayered dissimilar WMs, and weld geometries
have been discussed. Fracture toughness tests were conducted on 3PB and ten-
sion plate specimens for joint welds with single-material and multi-material
welds. Because of weld seam depths, the single-material welds and the multi-
material welds were multi-pass welds. The multi-material and multi-pass welds
had alternating layers of 9 % Ni WM (ER9Ni) and general purpose carbon steel
WM (ER70S-G). The fracture toughness behavior of the multilayereds with dis-
similar weld materials had layer-to-layer strength and failure mismatch charac-
teristics that were not well described with an ideally brittle weakest-link
fracture model.
The 3PB multilayered specimens showed multi-stage fracture characteristic
that progressed from layer to adjacent layer. For these specimens, the first frac-
tures were initiated in the lower toughness WM (ER70S-G) layers. On the other
hand, multi-stage fracture did not occur in the multilayered tension plate

specimens; but fracture initiation occurred first in the lower toughness WM

(ER70S-G) layers in a similar manner to those in the 3PB multilayer specimens.
In general, the fracture toughness characteristics of welded joints made with
multilayered dissimilar WMs were greater than welded joints made with only
the lower-toughness WM (ER70S-G).
The fracture toughness CTOD values estimated with a classical weakest-link
model did not compare well with the fracture toughness data of the multilay-
ered WM for the M-joint specimens. The detailed finite element stress analysis
results for the multilayered welds suggested that the weakest-link model, which
describes well fracture toughness characteristics of materials with localized,
idealized brittle heterogeneous material sub-domains, was not physically con-
sistent because the local stress field in front of the crack-tip was greatly elevated
in the lower-strength, more ductile WM (ER70S-G) layers that were adjacent to
the higher-strength, less ductile WM (ER9Ni) layers. To account for local stress
increases in low-strength weld layers in the M-joint specimens, the weakest-link
model was modified by defining a localized volume-averaged effective stress
metric to be used as an approximate Weibull stress.
Using this volume-averaged stress as a Weibull stress, the differences
between calculated fracture toughness CTOD values relative to fracture tough-
ness test data were reduced for the M-joint specimens. The volume-averaged
Weibull stress seems to account for strength mismatch effects across dissimilar
layers of WM in front of the crack-tip. Furthermore, using the same volume-
averaged stress metric to approximate the Weibull stress, it was shown possible
to predicate fracture toughness values for the tension plate and 3PB multilay-
ered WM specimens that agree well with the available fracture toughness data,
though further research with more data sets will be required to confirm uncer-
tainty in the rank probability.
Acknowledgments
This research was carried out as a part of research activities of “Fundamental
Studies on Technologies for Steel Materials with Enhanced Strength and
Functions” by the Consortium of JRCM (The Japan Research and Development
Center of Metals). Financial support from NEDO (New Energy and Industrial
Technology Development Organization) is gratefully acknowledged. The writers
highly acknowledge our master course student, Mr. Yuki Hirade, for his out-
standing contributions in carefully conducting many difficult experiments.
References
[1] Satoh, K., Toyoda, M., Mutoh, Y., and Doi, S., “Fractographic Study of Fracture
Toughness Evaluation of Welded Joint,” Q. J. Jpn. Weld. Soc., Vol. 49, 1980, pp.
766–772 (in Japanese).
[2] Toyoda, M., Oda, I., and Satoh, K., “Fracture Behaviors of Fracture Toughness
Testing Specimen with Heterogeneity Along Crack Front: Fundamental Study
Using Specimens Extracted from Stainless Clad Steel,” J. Soc. Nav. Archit. Jpn.,
Vol. 148, 1980, pp. 203–211 (in Japanese).

[3] Satoh, K., Toyoda, M., and Minami, F., “Effects of Fracture Controlling Factors on
Cleavage Fracture Initiation in Specimens with Heterogeneity along Crack Front,”
Q. J. Jpn. Weld. Soc., Vol. 50, 1981, pp. 743–749 (in Japanese).
[4] Satoh, K., Toyoda, M., and Minami, F., “Fracture Initiation Toughness of Materials
with Locally Embrittled Region: With Reference to Electron Beam Welds of HT 80
Steel,” J. Soc. Nav. Archit. Jpn., Vol. 153, 1983, pp. 322–328 (in Japanese).
[5] Satoh, K., Toyoda, M., and Minami F., “Probabilistic Analyses on Scatter of Cleav-
age Fracture Toughness of Welds with Heterogeneity along Crack Front,” Q. J. Jpn.
Weld. Soc., Vol. 2, 1984, pp. 440–447 (in Japanese).
[6] Satoh, K., Toyoda, M., and Minami, F., “Prediction of Fracture Toughness at Local-
ized Embrittlement Region in Welds Based on Weakest Link Model,” Q. J. Jpn.
Weld. Soc., Vol. 3, 1985, pp. 82–89 (in Japanese).
[7] BSI, BS7448-Part2, 1997, “Fracture Mechanics Toughness Tests. Method for Deter-
mination of KIc, Critical CTOD and Critical J Values of Welds in Metallic Materials,
London, UK.
[8] Bilby, B. A., Cottrell, A. H., Smith, E., and Swinden, K. H., “Plastic Yielding from
Sharp Notches,” Proc. R. Soc. London, Vol. 279, 1964, pp. 1–9.
[9] Minami, F., Toyoda, M., and Satoh, K., “A Probabilistic Analysis on Thickness
Effect in Fracture Toughness,” Eng. Fract. Mech., Vol. 26, 1987, pp. 433–444.
[10] Wallin, K., “The Size Effect in Results,” Eng. Fract. Mech., Vol. 22, 1985, pp. 149–
163.
[11] Brückner-Foit, A., Ehl, W., Munz, D., and Trolldenier, B., “The Size Effect of Micro-
structural Implications of the Weakest Link Model,” Fatigue Fract. Eng. Mater.
Struct., Vol. 13, 1990, pp. 185–200.
[12] Minami, F., Bruckner-Foit, A., Munz, D., and Trolldenier, B., “Estimation Proce-
dure for the Weibull Parameters Used in the Local Approach,” Int. J. Fract., Vol. 54,
1992, pp. 197–210.
[13] ASTM Standard E1921-11, 2011, “Standard Test Method for Determination of Ref-
erence Temperature, T0, for Ferritic Steels in the Transition Range,” Annual Book
of ASTM Standards, ASTM International, West Conshohocken, PA.
[14] Beremin, F. M., “A Local Criterion for Cleavage Fracture of a Nuclear Pressure Ves-
sel Steel,” Metall. Trans. A, Vol. 14A, 1983, pp. 2277–2287.
[15] Ruggieri, C., Minami, F., Toyoda, M., Hagiwara, Y., and Inoue, T., “Local Approach
to Notch Depth Dependence of CTOD Results,” J. Soc. Nav. Archit. Jpn., Vol. 171,
1992, pp. 493–499.

377
Author Index
A J
Adair, B. S., 254-277 Jin, Z., 338-356

Antolovich, S. D., 254-277 Johnson, W. S., 254-277
B K
Beretta, S., 278-294 Kim, J., 65-86

Bernard, S., 321-337
L
C
Laengler, F., 215-230
Carpinteri, A., 3-19 Lingenfelser, D. J., 87-108
Cordes, T. S., 87-108
M
D
Mao, T., 215-230
Daniewicz, S. R., 136-156 Matos, J. C., 126-135
Dhinakaran, S., 179-196 Minami, F., 357-376
E N
Endo, M., 157-175 Newman, J. C., Jr., 87-108, 109-125

Escuadra, J., 126-135
O
F
Ohata, M., 357-376
Filippini, M., 278-294
Okazaki, Y., 357-376
Fongsamootr, T., 321-337
P
G
Paggi, M., 3-19
González, B., 126-135
Pasquero, G., 278-294
H Patriarca, L., 278-294
Potirniche, G. P., 197-214
Hill, M. R., 65-86 Prakash, R. V., 179-196
I R
Ismonov, S., 136-156 Rusk, D. T., 295-317

378
S Toribio, J., 126-135
Sabbadini, S., 278-294 W

Schaper, M. K., 231-253
Scholz, A., 215-230 Wang, X., 338-356
Seto, M., 357-376
Shaw, J. W., 87-108 Y
Staroselsky, A., 254-277
Sunder, R., 20-64 Yamada, Y., 109-125
Yanase, K., 157-175
T
Z
Takashima, Y., 357-376
Taylor, R. E., 295-317 Ziegler, B. M., 87-108
379
Subject Index
3 fatigue crack growth simulation,
136-156
3.5 %, 179-196 fatigue crack propagation, 126-135
fatigue property charts, 3-19
C fatigue testing, 295-317
fatigue-crack propagation, 278-294
cold working process, 136-156
femur, 321-337
compression precracking, 109-125,
finite element method, 321-337
278-294
finite-element analysis, 215-230
contact pressure, 65-86
fractography, 254-277
corrosion fatigue, 179-196
fracture gap, 321-337
crack closure, 20-64, 87-108, 109-125,
fracture mechanisms, 254-277
179-196, 231-253
fracture toughness, 357-376
crack front aspect ratio, 126-135
frequency shedding, 179-196
crack incubation, 197-214
crack initiation, 295-317
G
cracked cylinder, 126-135
cracks, 87-108, 109-125 gamma titanium aluminides,
creep, 197-214 278-294
creep-fatigue life assessment, 215-230 Gigacycle fatigue, 295-317
Cr-Mo-V steel, 197-214
H
D
high-cycle fatigue, 278-294
DCP implant, 321-337
dimensional analysis, 3-19 I
dimensionless SIF, 126-135
Dugdale model, 157-175 IN100 superalloy, 254-277
E K
elastoplasticity, 215-230 Kmax effect, 109-125
F L
fatigue crack closure, 65-86 lab air, 179-196

fatigue crack growth, 3-19, 20-64, LEFM, 157-175
65-86, 87-108, 109-125, 231-253 load interactions, 254-277
fatigue crack growth rate, 254-277 load ratio, 109-125
380
M servo-hydraulic control error,

295-317
multilayer dissimilar weld metals, short cracks, 3-19
357-376 short-crack behavior., 278-294
small fatigue crack, 157-175
N S-N curves, 3-19
spectrum loading, 295-317
NaCl solution, 179-196 steel, 87-108, 231-253
Ni–Mn–Cr steel, 179-196 strength mismatch, 357-376
Ni-resist D5S, 215-230 stress, 321-337
non-linear stress distribution, stress intensity factor, 87-108,
338-356 338-356
notch effect, 157-175 strip-yield model, 197-214
numerical modeling, 126-135
T
O
temperature interactions, 254-277
On-line crack compliance technique,
thermo-mechanical fatigue, 215-230,
136-156
254-277
threshold, 109-125
P
turbine housing, 215-230
plasticity, 87-108
V
R
variable amplitude loading, 20-64,
residual stress, 20-64, 65-86 295-317
residual stress intensity factors,
136-156 W
S weakest link model, 357-376

Weibull stress, 357-376
semi-elliptical surface crack, 338-356 weight function, 338-356
Daniewicz | Belsick | Gdoutos
Selected Technical Papers
STP 1546
Fatigue and
Fracture
Mechanics:
JAI • Fatigue and Fracture Mechanics: 38th Volume

38th Volume
JAI Guest Editors:

Steven R. Daniewicz
Emmanuel E. Gdoutos
STP 1546
www.astm.org
ISBN: 978-0-8031-7532-7
Stock #: STP1546

STP 1546-2012

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

STP 1546-2012

Uploaded by

Copyright:

Available Formats

Daniewicz | Belsick | Gdoutos

Journal of ASTM International

JAI • Fatigue and Fracture Mechanics: 38th Volume

JAI Guest Editors:

JAI Guest Editors:

Printed in the U.S.A.

ASTM Stock #: STP1546

Printed in Bay Shore, NY

Fatigue Crack Growth

High Temperature, High Frequency, and Environmental Effects

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379

Reprinted from JAI, Vol. 8, No. 10

Alberto Carpinteri1 and Marco Paggi1

Dimensional Analysis and Fractal Modeling

ABSTRACT: In the present paper, generalized Paris and Wöhler equations

ID: kumarva Time: 12:35 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

4 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS

ID: kumarva Time: 12:35 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 5

ID: kumarva Time: 12:35 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

6 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS

ID: kumarva Time: 12:35 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 7

In the present paper, we extend the dimensional analysis approach pioneer-

Generalized Cumulative Fatigue Damage Formulation

N ¼ Fðry ; KIC ; Drfl ; DKth ; E; Dr; x; h; a; d; 1 RÞ (1)

where W is a nondimensional function. The dimensionless number r2y h=KIC 2

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

8 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS

that Pi is nonessential for the representation of the physical phenomenon. In

The exponents ai cannot be determined from considerations of dimensional analy-

1 Drny ¼ N1 Drnfl ¼ NDrn ¼ k (4)

Generalized Fatigue Crack Propagation Formulation

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 9

where the exponent b1 and the nondimensional function U cannot be deter-

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

10 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS

! b2 b3 b4

The experimentally observed deviations from the simplest power-law regime

This generalized mathematical representation encompasses several improved

The Anomalous Crack-Size Dependency of The Paris’ Law:

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 11

and dG is a parameter related to the fractal dimension of the invasive rough

This model was referred to as monofractal approach to size effect on fatigue

Equation 13 has also related consequences on the crack-size dependency of the

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

12 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS

FIG. 4—Multifractal scaling law for the Paris’ law parameter C.

FIG. 5—Multifractal scaling law for the fatigue threshold.

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 13

Analytical Correlations Between the Fatigue Properties

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

14 JAI STP 1546 ON FATIGUE AND FRACTURE MECHANICS

or, in a bilogarithmic form

Equation 18 establishes a correspondence between DKth , KIC and m in the long-

that can be rewritten as follows

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

CARPINTERI AND PAGGI, doi:10.1520/JAI104105 15

FIG. 7—Interpretation of the material property charts adapted from [24].

ID: kumarva Time: 12:36 I Path: //xinchnasjn/AIP/3b2/STP#/Vol01546/120219/APPFile/AI-STP#120219

where the exponent b1 and the nondimensional function U cannot be deter-