0% found this document useful (0 votes)

170 views

STP 595-1976

Uploaded by

Tim Schouw

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

170 views

STP 595-1976

Uploaded by

Tim Schouw

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 344

FATIGUE CRACK GROWTH

UNDER SPECTRUM LOADS

A symposium
presented at the
Seventy-eighth Annual Meeting
AMERICAN SOCIETY FOR
TESTING AND MATERIALS
Montreal, Canada, 23-24 June 1975

ASTM SPECIAL TECHNICAL PUBLICATION 595

R. P. Wei and R. I. Stephens
symposium cochairmen

List price$34.50
04-595000-30

AMERICAN SOCIETY FOR TESTING AND MATERIALS

1916 Race Street, Philadelphia, Pa. 19103

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
9 AMERICAN SOCIETY FOR TESTING AND MATERIALS 1976
Library of Congress Catalog Card Number: 75-32902

NOTE
The Society is not responsible, as a body,
for the statements and opinions
advanced in this publication.

Printed in Tallahassee, Fla.

May 1976

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Foreword
The Symposium on Fatigue Crack Growth Under Spectrum Loads was held on
23-24 June 1975 at the Seventy-eighth Annual Meeting of the American Society
for Testing and Materials in Montreal, Quebec, Canada. Committee E-24 on
Fracture Testing of Metals and Committee E-9 on Fatigue sponsored the
symposium. R. P. Wei, Lehigh University, and R. I. Stephens, The University of
Iowa, served as the symposium cochairmen. Serving as members of the
Symposium Organizing/Program Committee and as session chairmen were J.M.
Barsom, U. S. Steel Corp.; W. G. Clark, Jr., Westinghouse R & D Center; N. E.
Dowling, Westinghouse R & D Center; C. M. Hudson, NASA Langley Research
Center; E.K. Walker, Lockheed California Co.; and Howard Wood, AFFDL/
FBR, Wright-Patterson AFB.

Fatigue at Elevated Temperatures, STP 520 (1973), $45.50,04-520000-30

Probabilistic Aspects of Fatigue, STP 511 (1972), $19.75, 04.511000-30

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
A Note of Appreciation
to Reviewers
This publication is made possible by the authors and, also, the unheralded
efforts of the reviewers. This body of technical experts whose dedication,
sacrifice of time and effort, and collective wisdom in reviewing the papers must
be acknowledged. The quality level of ASTM publications is a direct function of
their respected opinions. On behalf of ASTM we acknowledge with appreciation
their contribution.

A S T M Committee on Publications

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions
Editorial Staff
Jane B. Wheeler,Managing Editor
Helen M. Hoersch,Associate Editor
Charlotte E. DeFranco, Senior Assistant Editor
Ellen J. McGlinchey,Assistant Editor

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Contents
Introduction
Observations on the Prediction of Fatigue Crack Growth Propa.
gation Under Variable.Amplitude Loading--J. SCHIJVE 3
Loads in Service 4
Differences Between Crack Growth Studies 6
Definition and Measurement of Interaction Effects 8
Interaction Effects in Tests with Overloads or Step Load-
ing 10
Interaction Effects in Tests with Program Loading, Ran-
dom Loading, or Flight-Simulation Loading 13
Prediction Methods for Variable-Amplitude Loading 16
Conclusions 19

SIMPLE SPECTRA: EFFECT OF LOADING VARIABLES

Fatigue Crack Growth with Negative Stress Ratio Following

Single Overloads in 2024-T3 and 7075-T6 Aluminum
Alloys---R. I. STEPHENS, D. K. CHEN, AND
B. W. HOM 27
Material and Experimental Procedures 28
Test Results 29
Discussion of Results 36
Summary and Conclusions 37
Effect of Single Overload/Underload Cycles on Fatigue Crack
Propagation--W. X. ALZOS, A. C. SKAT, JR., AND
B. M. HILLBERRY 41
Test Program 42
Experimental Procedure 43
Data ReductionwNumerical Differentiation 45
Test Results 49
Comparison with Crack Closure 53
Conclusions 56
Discussion 58

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Effects of Rest Time on Fatigue Crack Retardation and Observa-
tions of Crack Closure--W. N. SHARPE, JR.,
D. M. CORBLY, AND A. F. GRANDT, JR. 61
Retardation Measurements 62
Surface Measurement of Opening Loads 66
Ultrasonic Measurements 69
Conclusions 73
Discussion 75
Mechanisms of Overload Retardation During Fatigue Crack
Propagation--P. J. BERNARD, T. C. LINDLEY, AND
C. E. RICHARDS 78
Types of Single Overload Effect 80
Experimental 81
Results 82
Discussion 91
Conclusions 95
Discussion 96

SIMPLE SPECTRA: ENVIRONMENTALEFFECTS

AND MODELING

Spike Overload and Humidity Effects on Fatigue Crack Delay in

A17075-T651--OTTO BUCK, J. D. FRANDSEN, AND
H. L. MARCUS 101
Experimental Procedures 102
Results 102
Discussion 106
Conclusions 111
Influences of Chemical and Thermal Environments on Delay in a
Ti-6AI-4V Alloy--T. T. SHIH AND R. P. WEI 113
Material and Experimental Work 115
Results and Discussions 116
Summary 123
Discussion 124
Effect of Various Programmed Overloads on the Threshold for
High-Frequency Fatigue Crack Growthm
S. W. HOPKINS, C. A. RAU, G. R. LEVERANT,
AND A. YUEN 125
Experimental Procedure 126
Experimental Results and Discussion 129
Metallography and Fractography 135
General Discussion 137
Conclusions 139
Discussion 140

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
A Model for Fatigue Crack Growth Delay Under Two-Level
Block Loads--O. A. ADETIFA, C. V. B. GOWDA, AND
T. H. TOPPER 142
Previous Quantitative Models 143
Composite Stress Intensity Parameter 144
Crack Arrest Condition 15l
Discussion 151
Summary and Conclusions 153
Discussion 155
Mathematical Modeling of Crack Growth Interaction Effects---
P. D. BELL ANDA. WOLFMAN 157
Results and Discussion 158
Concluding Remarks 170
Discussion 171
Experimental Results and a Hypothesis for Fatigue Crack Prop-
agation Under Variable.Amplltude Loadingu
G. H. JACOBY, H. NOWACK, AND
H. T. M. VAN LIPZIG 172
Experimental Crack Propagation Behavior and Discussion 174
Conclusions 182

COMPLEX SPECTRA: LOAD DEFINITION, MODELING,

AND SERVICE SIMULATION

Effect of Spectrum Type on Fatigue Crack Growth Life--

J. A. REIMAN, M. A. LANDY, AND M. P. KAPLAN 187
Mission Profile Definition 188
Spectrum Development 189
Analytical Model Verification 192
Results and Discussion 193
Conclusions 200

Stress Spectrums for Short-Span Steel Bridges--

K. H. KLIPPSTEINAND C. G. SCHILLING 203
Field Measurements 204
Control Tapes for Fatigue Tests 211
Conclusions 215
Fatigue Crack Growth Under Variable-Amplitude Loading in
Various Bridge Steels--J. M. BARSOM 217
Materials and Experimental Work 218
Results and Discussion 222
Summary 226
Discussion 233
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Equivalent Constant-Amplitude Concept for Crack Growth
Under Spectrum Loading--WOLF ELBER 236
Analysis 237
Testing 240
Results and Discussion 243
Concluding Remarks 247
Discussion 247
Crack Growth in Ti-8AI-IMo-IV with Real.Time and Acceler-
ated Flight-by-FUght Loading--L. A. IMIG 251
Experimental Considerations 252
Crack-Growth Analysis 255
Results and Discussion 259
Concluding Remarks 263

LIFE PREDICTION AND APPLICATIONS

Prediction of Fatigue Crack Growth Under Irregular

LoadinguD. V. NELSON AND H. O. FUCHS 267
Input Data for Predictions 268
Crack Growth Rate Relations 271
Prediction of Crack Growth for Irregular Loadings 273
Comparison of Predictions with Test Results for Irregular
Loadings 278
Discussion 283
Conclusions 284
Discussion 287
Structural Reliability Prediction Method Considering Crack
Growth and Residual Strength--S. R. VARANASI AND
I. C. WHITTAKER 292
Analytical Development 293
Crack Growth Model 293
Residual Strength Model 297
Structural Reliability Model 298
Results 298
Conclusions 304
Spectrum Crack Growth Prediction Method Based on Crack Sur-
face Displacement and Contact Analyses--H. D. DILL
AND C. R. SAFF 306
Crack Surface Displacement Analyses 307
Contact Stress Analyses 309
Constant-Amplitude Crack Growth 313
Crack Growth Following High Loads 313
Blocked Spectrum Crack Growth 315

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Conclusions 315
Discussion 317
Spectrum Fatigue Crack Growth in Lugs--L. F. IMPELLIZZERI
AND D. L. RICH 320
Test Program 321
Analytical Techniques 326
Summary 334
Glossary 337

Introduction

The importance of load interactions in variable-amplitude loading on the

accurate prediction of fatigue lives or inspection intervals or both of engineering
structures has been recognized for some time. A number of symposia have been
held over the past 20 years (many of them under the auspices of the American
Society for Testing and Materials (ASTM)). Most of these symposia were
directed to the problems of fatigue behavior of smooth and mildly notched
specimens, and of structural components. In more recent years, fatigue crack
growth has begun to receive increasing attention both in research and in
structural design. Fracture mechanics has emerged and matured as an important
tool for design analyses and for studying fatigue crack growth during the past
decade.
As a result of these developments and of well-publicized problems with several
aerospace and highway structures, a considerable amount of effort has been and
is now being devoted to the understanding of load interaction effects in fatigue,
and to the development of rational procedures for predicting fatigue behavior
under spectrum (service) loads. It was clear that a major symposium was needed
to review and assess current technology and understanding in this important
area, to provide a forum for the exchange of ideas, and to help define problem
areas and directions for new research. In recognition of this need, independent
planning for such a symposium was initiated in early 1973 within ASTM
Committee E-24 on Fracture Testing of Metals and Committee E-9 on Fatigue.
The symposium became a joint venture between ASTM Committees E-24 and
E-9 in late 1973.
Because of the complexity of the problem, it was decided to restrict the
symposium only to those topics that are related to fatigue crack growth so that
each topic could be treated in some depth. The primary objective of this
symposium was to bring the current state of the art and information into focus
such that reliable predictive procedures can be identified or developed or both
for use in the design of engineering structures. In support of this primary
objective, the following topics were considered:
1. Review and assessment of current technology.
2. Description of service loading and environmental conditions.
3. Characterization of load spectrum for use in testing and in design.
4. Crack growth under simple load spectra, including the influence of
service/test environment.
5. Crack growth under complex/random load spectra, including the influence
of service/test environment.
6. Mechanism(s) and modeling.
]
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 by ASTM
Downloaded/printed by Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
2 FATIGUE CRACK GROWTH

7. Predictive methods and correlations with service performance.

8. Identification of test procedures and of need for standardization.
This volume details the proceedings of the Symposium on Fatigue Crack
Growth Under Spectrum Loads and provides a reasonable representation of the
current state of the art and thinking. It is organized into four separate sections,
with an introductory overview paper by J. Schijve:
I. Simple Spectra-Effect of Loading Variables
II. Simple Spectra-Environmental Effects and Modeling
III. Complex Spectra-Load Definition, Modeling, and Service Simulation
IV. Life Prediction and Applications
The various topics listed are covered in each of these sections and in the
overview paper by Dr. Schijve. The volume serves to reemphasize the highly
complex nature of fatigue crack growth under spectrum loads, and the diversity
of views on the subject. Load interaction effects result from complex processes
that occur at or near the crack tip and are difficult to assess analytically or
experimentally. The difficulty is compounded by a range of combined load and
environment (chemical and thermal) interactions. Nevertheless, by recognizing
the difference between pragmatism and fundamental understanding, significant
progress can be made. By realizing the importance of various loading and
environmental variables, rational life prediction procedures can be developed for
specific applications. By avoiding broad generalizations, development of under-
standing can proceed in an orderly fashion. The contributions of this symposium
are to be viewed in light of these comments, and represent a good beginning.
The information contained in this volume will be of use to designers, materials
scientists and test engineers, and structural and reliability engineers who are
concerned with this important problem. It is hoped that interactions between
the various disciplines involved will be promoted and that this volume will serve
as a starting point and provide impetus for rapid advance in the field of fatigue
crack growth under spectrum loads.
The success of a symposium and the publication of its proceedings depend on
the tireless efforts of many people. The contributions of the authors, the
reviewers, the members of the Symposium Organizing/Program Committee, and
Jane B. Wheeler and her staff are gratefully acknowledged. Donald Wisdom, who
encouraged and assisted in the initial planning of this symposium, and who
through untimely death could not see it come to fruition, is warmly remembered
by us all.

R. P. Wei,
Lehigh University, Bethlehem, Pa.; ASTM
Committee E-24, symposium cochairman.

R. I. Stephens,
The University of Iowa, Iowa City, Iowa;
ASTM Committee E-9, symposium
cochairman.

Observations on the Prediction of

Fatigue Crack Growth Propagation
Under Variable-Amplitude Loading

REFERENCE: Schijve, J., "Observations on the Prediction of Fatigue Crack Growth

Propagation Under Variable-Amplitude Loading," Fatigue Crack Growth Under
Spectrum Loads, ASTM STP 595, American Society for Testing and Materials, 1976,
pp. 3-23.
ABSTRACT: The paper starts with a discussion on loads in service, after which a
survey is given of various types of variable-amplitude loading as applied in test
programs. The various phenomenological aspects of fatigue damage associated with
fatigue cracks are indicated. Interaction effects between cycles of different
magnitudes are defined. Methods for measuring interaction effects, examples of
interaction effects, and possible explanations are reviewed. This includes both tests
with simple types of variable-amplitude loading (overloads and step loading) and
more complex load-time histories (program loading, random load, and flight-
simulation loading). New evidence on crack closure is presented. Various types of
prediction methods are discussed. The paper is primarily a survey of the present
knowledge, with an analysis of the consequences for prediction techniques.
KEY WORDS: crack propagation, fatigue (materials), loads (forces), predictions

For an operator of a machine or a structure, cracks are of little interest as long

as they cannot be detected by available inspection techniques. However, as soon
as detection is possible, the situation is different. If crack growth is sufficiently
slow, routine inspections can be adopted to prevent failures in service. In aircraft
structures this has led to the well-known fail-safe philosophy. The crack
propagation curve in Fig. 1 illustrates this point. Obviously, the time available

CRACK I
LENGTH
af MINIMUM CRACK SIZE
TO CAUSE FAILURE
I

(30
ESCAPI, NG DETECTION
. ~> TIME
PERIOD FOR ( FLYING HOURS)
CRACK DETECTION
FIG. 1-Limitations of fail-safety.

I Professor, Department of Aeronautical Engineering, Delft University of Technology,

Delft, The Netherlands.
3
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
4 FATIGUE CRACK GROWTH

for crack detection depends on both a0 and af. Since the growth of a crack is
usually an accelerating process, the time is less dependent on af and more
dependent on ao. This emphasizes the significance of inspection techniques. For
a long time, the arbitrary choice was a0 = % in. (12.6 mm), but more optimistic
values, say a few millimetres, appear to be justified, provided the crack location
is accurately known. The time available for crack detection is clearly related to
the crack propagation curve (Fig. 1), and the problem of predicting this curve is
the leading question of this paper.
Crack growth is dependent on numerous variables (see Table 1 which lists the
main groups). This paper attempts to survey the various aspects of predicting

TABLE 1 - Variablesaffecting crack growth.

l load-time history
Variables outside the structure
chemical environment
type of structure
Variables inside the structure
lmaterial

crack growth curves, which should include a discussion of many variables.

However, the main topic will be the effect of the load-time history including a
few comments on environmental effects. It is well known, especially for aircraft
structures, that crack growth can be retarded and temporarily stopped by
adopting specific design features. Fail-safe design procedures, however, will not
be discussed, nor will material selection for slow crack growth. The discussion in
this paper includes: loads in service, present knowledge about crack growth
under variable-amplitude loading, and present prediction methods. These
problems are discussed as they occur in aeronautics, but the situation is similar
to other diciplines.

Loads in Service
The major aspects of service load-time histories include:
1. load occurrences,
2. load sequences,
3. speed of load variations, and
4. environment.
The definition of the varying load is a problem in itself. Two samples are shown
in Fig. 2 to illustrate this point. The first sample is characteristic of an
amplitude-modulated signal [1] ~ with a constant mean and frequency, and the
random feature is in the modulation. The second sample shows a random nature,
which is less easily defined than the first sample. Such samples can be
statistically analyzed with respect to peak loads, load ranges, etc. (counting
2 The italic numbers in brackets refer to the list of references appended to this paper.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
S C H I J V E ON V A R I A B L E - A M P L I T U D E LOADING 5

FIG. 2-Two samples o f service load.time histories: (a) narrow-band random vibrations
[1], and (b) gust loading on a wing.

methods) [2,3]. However, information on the sequence of load occurrences is

lost in such a statistical description.
Investigations on sequence effects have dearly shown their significance.
Results of a comparative study on crack growth with random and programmed
sequences are summarized in Fig. 3 [4]. Although all sequences were statistically

~ : . _ ? . : : ~ : . . ; : . ~ .,:,~ RANDOM , 202/,-T3 ALCLAD

PROGRAMMFD. SHORT PERIOr

~,'t-r r

..................................j PROGRAMMED. LONG PERIOD

10 6 2.10 6 3.106 4.106

CYCLES FOR CRACK GROWTH
FROM 24 mm TO 100 mrn

FIG. 3-Comparison between crack propagation lives under random loading and "statisti-
cally equivalent'" programmed loading [4].
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
6 FATIGUE CRACK GROWTH

equivalent with respect to load occurrences, the crack growth rates were
significantly different. Moreover, the fracture surfaces were also different, even
macroscopically, and this will be discussed later.
The conclusion to be drawn here is that complete knowledge about loads in
service should include information on load sequences. If the loading is a
stationary random process, the sequence appears to be well defined in statistical
terms. However, in many practical situations the loading conditions are
non-stationary. Moreover, mixtures of random loads and deterministic loads
frequently occur. A well-known example is the (non-stationary) random gust
loading on a wing, combined with the deterministic ground-to-air cycles.
An inherent problem of predicting loads in service is the scanty information
on the rarely-occurring very high loads. This feature is disturbing, because these
high loads can have a great effect on crack growth. High tensile loads can
drastically reduce crack growth. Unfortunately, accurate predictions on the
occurrence of such loads in service are difficult to make. This implies a severe
limitation to the practical significance of predicted crack rates.
Service conditions also include the operating environment. If the environment
is affecting crack growth, the phenomenon has to be associated with corrosion
fatigue. The speed of load variations can be significant (effects depending on
time and loading rates). Three groups of environments are:
1. nonaggressive dry environments,
2. water vapor as the most detrimental element, and
3. more aggressive electrolytic conditions.
Nonaggressive environments do not frequently occur in aerospace conditions,
although they are relevant to space structures and to aircraft at very high
altitudes. However, the second group is applicable in many cases. The third
group encompasses wet environments.
The detrimental effect of water vapor on crack growth in aluminum alloys is
well known. Present knowledge appears to indicate the constant effect of water
vapor on many service conditions as well as on testing in the laboratory.
Sufficient water vapor is available for the same maximum detrimental effect.
Comparative flight-simulation tests at 10 Hz, 1 Hz, and 0.1 Hz have shown the
same crack rates in 2024-T3 Alclad and 7075.T6 clad material [5]. This
simplified picture is a most desirable result in view of the relevance of laboratory
results for service conditions. It should be clearly recognized, however, that the
picture is only wishful thinking unless it can be backed up by physical
understanding. Moreover, for more agressive wet environments, the time scale
will certainly be significant. An outstanding example of this is the growth of
cracks in marine environments, which would apply to off-shore structures. It
then becomes extremely difficult to obtain relevant information from high-speed
laboratory tests.

Differences Between Crack Growth Studies

The first impression from the literature is the overwhelming variety of
different types of variable-amplitude loading. For an analysis of present
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
$CHIJVE ON VARIABLE-AMPLITUDE LOADING 7

knowledge, the various types of loading have been classified in a number of

groups in Table 2. The variety of fatigue loads is partly, but not fully, illustrated
by the second column of the table. The variety of investigations on crack growth

TABLE 2 - Types of variable-amplitude loading with main variables.

I single overload
repeated overloads
Overloads blocks of overloads
magnitude of overloads(includingR)
sequence in overloadcycles
sequence of steps (hi-lo or lo-hi)
Step loading
magnitude of steps (includingR)
sequence of amplitudes
Programmed block loading
lsize of blocks
distribution function of amplitudes
spectral density function (narrow band, broad band)
Random loading
lcrest factor (clipping ratio)
irregularity factor
l distribution function of load cycles
sequence of flights
Flight-simulation loading
sequence of loads in flight
maximum load in the test

is significantly larger because, in addition to the loading history, there are several
additional variables. It is sufficient to mention:
1. type of specimen;
2. material;
3. loading system (tension, bending, etc.);
4. loading rate (frequency); and
5. environmental conditions.
The effect of the type of specimen is generally covered by relevant K-values.
However, differences between thick and thin specimens are significant (plane
strain/plane stress). The other variables further contribute to a large variety of
investigations reported in the literature. It then becomes possible that the
relative amount of duplication is rather limited, despite an impressive number of
publications.
The goals of investigations can also be grouped into different categories. Three
general goals frequently recognized from the literature are:
1. to increase the fundamental understanding of the crack growth mechanism
under variable-amplitude loading,
2. to check crack growth prediction models, and
3. to generate data from which useful empirical trends might be derived.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
8 FATIGUE CRACK GROWTH

Investigations on overload effects and step loading are usually aimed at the first
goal. Conclusions of papers suggest some degree of understanding of a
mechanism. Investigations on program loading, random loading, or flight-
simulation are more directed to the second and the third goals. Relevant papers
suggest that a technical problem is treated in a practical way.

Definition and Measurement of Interaction Effects

9It would be very convenient for the prediction of crack growth if the growth
was a simple addition process of crack length increments (Aa) in each load cycle

a = a 0 + ~ Aa i (1)
This process should be understood to be simple if An was dependent on the
momentary size of the crack, a, only and independent of the history of the
preceding crack growth. Unfortunately, this is not true as a consequence of
so-called interaction effects. The crack growth increment, Aa, in a certain load
cycle will be a function of [6] :
1. the crack geometry being present before the cycle started,
2. the condition of the crack tip material, and
3. the magnitude of the load cycle.
These arguments are further illustrated in Table 3.
The arguments compiled in Table 3 imply that Aa will be a function of the
preceding cyclic load history. Similarly, a stress cycle will affect crack growth

TABLE 3 - Aa in a certain load cycle will depend on.

Amount of cracking crack length

shape of crack front

Crack front orientation tensile mode

shear mode
mixed modes I crack
geometry
Crack tip blunting shape of crack tip
blunted/sharpened fatigue
damage caused
Crack closure plastic deformation by the
in wake of crack preceding
load-history
(Cyclic) strain hardening distribution in crack~ condition
1
Residual stress and strain~tip zone ~ of material
at tip of crack/
Magnitude of load cycle ~S, R ~ external conditions of
Environment and frequency ! present load cycle

increments in subsequent cycles. These effects are referred to as "interaction

effects." Several examples will be shown later on.
Initially, interaction effects were measured from crack propagation curves as
obtained by visual observations of the tip of a crack at the surface of the
specimen. Examples for a thin and a thick specimen are given in Fig. 4 [7]. It

1l.,
CRACKLENGTH
(mml MATERIAL:2024-T3
/ B =THICKNESS(mml

Speok=196 126./,1
lO
~
-

-Smox--lt,7121.3)
8 B:10 '""~"'"-Srnin = 98114.2)
M N Im2l ksi)
6
6O 80 lOO 120 140 xl0 ~
C~LES
FIG. 4-The effect of sheet thickness on crack growth delays [7].

should be pointed out that the tip of a crack cannot always be located
accurately. Especially after an overload, the more extensive plastic deformation
at the tip of a crack will give some extra blurring of the surface image. Moreover,
it is generally recognized now that the two-dimensional picture of the surface
does not give sufficient information about the three-dimensional phenomenon
occurring in the material. This was borne out by fractographic observations
which showed that the curvature of the crack front could change by varying the
fatigue load (for example, forming of "tongues" by peak loads). The topography
could also vary (amount of shear lips).
Another fractographic refinement was due to measurements of striation
spacings. It essentially added to a more detailed picture of interaction effects.
Local accelerations or retardations of crack growth can hardly be observed from
crack growth curves, but striations can reveal such effects [8].
A different approach is associated with crack closure measurements [9,1.0].
Such measurements give indirect indications about plastic deformations left in
the wake of the crack (see Fig. 5). Unfortunately, more information on the
changes of the fatigue damage in the material can hardly be obtained. It is
difficult, if not impossible, to measure crack tip blunting and resharpening. The
same applies to the cyclic strain-hardening or softening in the crack tip zone and
the related residual stress field.

FIG. 5-Plastic deformation in the wake of a crack [real size). Crack viewed through a
window. The upper light part is the reflection of a fluorescent tube lamp.

Interaction Effects in Tests with Overloads or Step Loading

Originally, most information was obtained on ahiminium alloys. Later on,
similar tests were conducted on titanimn alloys [11-20] and high-strength steels
[11,12,15,21-29]. More or less similar trends were observed. Several test series
were also performed on mild steel and other non-aircraft steels [30-39]. The
behavior of these materials is not fully similar, but crack growth delays were
found as well. Apparently, similar results may be expected for materials with a
similar (cyclic) strain behavior. For most aeronautical materials this implies a
fairly high Oo.2/ou ratio, a limited ductility, and a rapid cyclic strain hardening.
Mild steel does not conform to this picture, which should be related to its
characteristic yielding and strain aging behavior.
The more prominent observations on interaction effects in tests with overloads
or step loading are summarized in the following:
1. Positive overloads introduce significant crack growth delays
[6,12,14-20,22,27,29,34,40-50]. In general, longer delays are obtained
by:
(a) increasing the magnitude of the overload,
(b) repeating the overload during the crack propagation life, and
(c) application of blocks of overloads instead of single overloads.
The retarded growth need not immediately follow the overloads. Some
further growth may be required before the crack rate decreases
[29,34,43,45-47,50]. Sometimes a small initial acceleration was even
observed. This delayed retardation was clearly verified by observations of
striation spacings.
2. The crack extension caused by the overloads themselves is larger than
expected from constant-amplitude tests [49,51]. This acceleration usually
requires fractography also.
3. Negative overloads have a relatively small detrimental effect on crack
growth [20,40,52]. However, a negative overload added immediately after
positive overloads can significantly reduce the crack growth delay of the
latter ones. If the negative overload precedes the positive overload, the
reduction of the delay is much smaller. There is an apparent sequence
effect of the overload cycles [12,20,40,41,43,49].
4. In step loading, a hi-lo sequence produces qualitatively similar results as
overload cycles [9,10,40,41,45,46,49,50,53-55]. Once again, delayed
retardation was observed. Interaction effects after a lo-hi sequence are
hardly detected from macroscopic crack growth observations. However,
more accurate measurements and striations do reveal locally accelerated
crack growth [21,25,45,46,50].
5. Delays clearly depend on the ductility of the material. If the ductility of an
alloy is controlled by heat treatment, a lower yield strength will produce
longer delays [27,29,54].
Originally, the explanation of crack growth delay was based on residual
stresses in the crack tip zone [40,54]. Later it turned out that this view was too
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions author
SCHIJVE ON VARIABLE-AMPLITUDE LOADING 11

simple. Crack closure, discovered by Elber [9,10], explained both retardations

and accelerations by the mechanism. Elber pointed out that crack growth
retardation did not occur immediately after an overload. The crack had to
penetrate into the plastic zone created by the overload before crack closure
could become effective (delayed retardation). This was amply confirmed by the
work of others.
Some exploratory tests were recently carried out [7] to study crack closure
and delays in relation to specimen thickness. The crack-opening stress, Sop (that
is, the stress at which crack closure is removed) was measured in constant-
amplitude tests for three different thicknesses. A small crack opening displace-
ment (COD) meter was used for this purpose. Results in Fig. 6 show that Sop is
lower for a larger thickness. In the thicker specimen the plastic zone size will be
smaller (approximately plane strain) than in the thinner specimen (approxi-
mately plane stress). Similarly, the elongation in the plastic zone is smaller and
the residual deformation in the wake of the crack will be smaller. This explains
the lower crack opening stress. Adding one overload cycle gives a significant
reduction of Sop immediately after the overload cycle (see Fig. 6). This has to

Sop 9 BEFOREOVERLOAD
(MNIn~) 60 ~
j- m AFTEROVERLOAD

,0 I .Srnax=l/,7 (21,3)
2(] ~vv,wv,w,v~,Smin= gB(14.2)
MNIm2(kst)
:2 ;, 6 8 10 MATERIAL:2024-T3
THICKNESS(turn) SPECIMENWIDTH:I00mm
CENTRALCRACK: o=10turn
FIG. 6-Effect of sheet thickness and overload on crack closure [7].

be expected, because the plasticity ahead of the crack opens the wake of the
crack. The results of these overloads on subsequent crack growth are shown in
Fig. 4. The crack growth delay is twice as long in the thinner specimen. In view
of more crack closure occurring in the thinner specimen, this result had to be
expected. The investigation is still in progress, but some data on measurements
of Sop during the delay period for a thin specimen can already be given here. As
shown by Fig. 7, Sop < Smin before the overload is applied. Immediately after
the overload, Sop is reduced still further. This would allow an accelerated
growth for a small number of cycles, which was reported to occur in the
literature [43]. However, shortly afterwards, Sop is significantly raised beyond
Smin . Later, Sop again decreases and Sef f increases. After Sop = Srnin, the
retardation of the growth has vanished. In Fig. 4 the crack length increment
affected by the overload is approximately equal to its estimated plastic zone
size. In Fig. 7 this increment is larger than the plastic zone. As pointed out by
Van Lipzig and Nowack [50], the retardation due to crack closure can very well
be effective beyond the overload plastic zone.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
12 FATIGUE CRACK GROWTH

DELAY
33 kc
CRAI~ f
LENGTH /
(mini 18
/ / II
,0 , / II
14 / III

12 1 II
2C 40 60 {80xl03C'h'CLES
~!:::'~":::::~:::::::~: '""::::::::e::"::":":~ii::i! ~ ':::':::::::::":::~ :~ "~ S max
SOp 140 ~i :i~.-'.,:::.:-
:.:.:~:':'::
!::::;:c
i.":-.:.-'~'~!.!~~:;!i~"~t9!~i ~ ~ : ~
,::::::':::'::
'::.:-:::. ::::::::::::::::::::::
(MN/~) 120 ~.................... ~:::-':~'~::~.~::::::"!:

8o iI
I
6o

4O
FIG. 7-Crack-opening stress be]ore and during delay period [material and stress levels, see
Fig. 6) [V.

Results on crack closure in flight-simulation tests, discussed later on, also

confirm that crack closure is a real phenomenon. Surprisingly enough, its
existence was overlooked for a long time, but recently, the first attempts to
incorporate crack closure into calculations were published [17,56-58]. The
more rigorous calculations made by Newman and Armen [58] produced
promising results, but such calculations are still rather expensive. More progress,
however, may be expected. The question to be raised here is whether interaction
effects might be due to crack closure alone. Keeping in mind the fatigue damage
picture outlined in the previous section (Table 3), this assumption appears to be
somewhat too optimistic. It would be surprising if the cyclic straining of the
material in the crack tip zone, the crack tip geometry, and the crack front
orientation would be fully irrelevant. Interaction effects as described in Ref 49
cannot be explained completely by crack closure. The crack extension during an
overload (for instance, tongues [29,42,49,59]) is too large to be due to a low
Sop. The conditioning of the material in the crack tip zone during the preceding
low-amplitude cycles may be another contributing factor. Incompatible crack
front orientation [49] may further add to the observed behavior. More research
is needed to unravel the complex phenomena occurring during variable-
amplitude loading. For further studies on this issue a mandatory requirement is
to include measurements on crack closure. Tests with load sequences avoiding
the occurrence of crack closure should also be enlightening, as shown by Shih
and Wei [17]. It should be recognized, however, that highly localized crack tip
closure might not be detected by "macro crack closure measurements."

Interaction Effects in Tests with Program Loading, Random Loading, or

Flight-Simulation Loading
During a complex load sequence, it is more difficult to observe the local
interaction effects separately. The overall effect, however, can easily be deduced
from macroscopic measurements of the crack growth. The predicted crack rate
without interaction effects is obtained from the predicted crack growth curve

a = ao + ~; ~ , (1)

with Aa i = crack extension in one cycle = daMN = fR (AK) (2)

It is generally observed that the actual crack rates are considerably different
from the predicted values. More detailed observations were obtained by
fractographic techniques and crack closure measurements. A summary of the
main results is given in the following.
(a) Crack rates derived from crack growth curves were usually found to be
significantly lower than the values predicted by Eqs 1 and 2
[5,15,24,26,28,41,59-63]. Values two to eight times lower were fre-
quently found. Apparently, retardation effects are predominating the
possibilities for acceleration effects.
(b) In program tests, similar sequence effects were found as observed in step
loading [4,8,16,19,25,26]. Specifically, a retarded crack growth after a
drop of the stress amplitude was clearly observed. In tests with a lo-hi-lo
sequence, the crack rate in the descending part was lower than in the
ascending part of the program.
(c) Nowack [64] studied crack growth under random loading with a constant
Srms, but a step wise change of Sm. A decrease of S m caused retarded
crack growth.
(d) The effect of high loads is similar to the effect of overloads in
constant-amplitude loading. In flight-simulation tests, it is well established
that the application of rarely occuring very high loads can decrease the
crack rate significantly. Truncation of these high loads to lower levels
gives faster growth [6,13,23,26,61,6.5,66]. Overloads applied to a
structure (for instance fail-safe loads) can drastically reduce subsequent
crack growth [66].
(e) Comparison of program loading and random loading has revealed
significant sequence effects [4]. These effects were not restricted to crack
rates (see Fig. 3) but also applied to the topography of the fracture
surface. Macroscopically, the roughness of the fracture surface and the
transition from the tensile mode to the shear mode were completely
different for the two types of loading, in spite of the same load spectrum
applying to both.
(f) Crack closure was recently shown to occur during flight-simulation tests
[67,68]. A sample of the loads in a flight with severe gusts is shown in
Fig. 8. Crack closure measurements made by a COD meter during a

FIG. 8-Load history during a flight-simulation test/66,67]. Flight with severe gust loads.

severe flight are shown in Fig. 9. The measurements were made only for
larger stress amplitudes of the flight. When the load passed the mean stress
in flight, small horizontal shifts were given to the recorder to separate the
loops of the various loads. Two interesting observations can be made.
First, five maximum peak loads occurring in the same flight do not have
the same effect. The nonlinearity at the top of the first one is mainly due
to crack extension and crack tip plasticity. Apparently, the other four

FIG. 9-COD measurements during a severe flight o f a flight-simulation test/66,67].

peak loads produce smaller contributions. This is in agreement with results

from a more elementary study [47,49]. The second observation is related
to crack closure which is easily observed from the nonlinearities in the
bottom part of the recording. Arrows indicate the stress levels below
which the crack is partly closed (Sop) . The first high load widens the
crack and thus causes a similar decrease of Sop as reported in the previous
section. After further crack growth during subsequent flights with milder
gust spectra, the crack opening stress was again restored to a higher level.
It appears that the observations of the tests with more complex load histories
are qualitatively in good agreement with the results of the more simple types of
loading, discussed in the previous section. The agreement concerns the effect of
high loads, delays, sequence effects, and crack closure. In view of all this
evidence, it is opportune to reconsider the occurrence of both crack growth
delaying and acceleration effects during complex load-time histories. A useful
comment was recently made by Katcher [15]. He started from the observation
that crack growth retardation after a high load requires a number of cycles
before it becomes effective (delayed retardation). He then pointed out that a
reapplication of a high load, before the retardation could become effective, may
considerably reduce the delay effect. A similar suggestion can be made for
acceleration effects. Accelerations were observed after lo-hi sequences, occurring
immediately after the lo-hi transition. This is in agreement with the crack closure
concept, although it need not be the only explanation. Anyhow, the acceleration
requires some prehistory of lower-amplitude cycles. Consequently, high-load
cycles which have not seen a prehistory of lower-amplitude cycles, will not be
associated with the same acceleration effect. This was illustrated before (see
subsection (e) and Fig. 9).
The conclusion now is that accelerations and delays in a complex load history
are very sensitive to the sequence of the various loads. The assumption that
random sequences and equivalent programmed sequences could give the same
overall crack rate has to be considered wishful thinking.
There is another conclusion to be drawn from the present knowledge. During
variable-amplitude loading, accelerations occur during the high loads. Retarda-
tions occur during the low-amplitude cycles. The result may well be that the
major part of the crack extension will occur during the high-amplitude cycles.
Crack growth observations during flight-simulation tests and fractography appear
to confirm this view [59,65]. This appears to be important for the mutually
related effects of environment and frequency. Environmental effects during
crack growth (corrosion fatigue) were amply shown in the literature to be large
for low AK values and relatively small during high AK values. If it is true that
the major contribution in ~ Aa is coming from the higfier load-amplitude cycles
a relatively small environmental effect should be expected. This is now being
investigated at the National Aerospace Laboratory, Amsterdam (NLR). Flight-
simulation tests are carried out on 2 and 10-ram sheet specimens (width 160
ram) of 2024-T3 and 7075-T6 material at 15 Hz. Some preliminary results for
2-mm specimens of 2024-T3 are shown in Table 4.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
16 FATIGUECRACK GROWTH
TABLE 4 - Preliminaryresultsof 2-ram2024-T3specimens.
Crack growth life
Environment (a = 10 mm to failure) Ratio

Dry air (H20 ( 2 0 ppm) 23 120 flights 1.3

Lab ~ir (RH 50%, 20 to 25~ 18 210 flights 1
Salt water 7 845 flights 0.43

Life in dry air is only 30 percent longer than in wet air, whereas
constant-amplitude data [69] would suggest it to be about three times longer.
The effect of salt water is probably not much different from the expected ratio
derived from constant-amplitude data. The problem of environmental effects
under service loading requires further clarification [19]. The present consider-
ations, however, already emphasize the need for realistic testing if realistic
answers are to be found.

Prediction Methods for Variable-Amplitude Loading

After the conclusive proof of the usefulness of AK for correlating constant-
amplitude crack rate data, it was all but natural that extensions to variable-
amplitude loading were proposed. Available propositions can be classified in five
groups:
1. non4nteraction,
2. interaction based on Keff,
3. equivalent K-concept,
4. characteristic K-concept, and
5. empirical trends.
The non-interaction method simply assumes that An in any load cycle is
dependent on the applicable K-value pertaining to that cycle. As discussed
before, this is physically incorrect and leads to overconservative crack rate
estimates.
Interaction methods based on Kef f also start from the idea that the crack rate
is uniquely related to z2xK and R values. In addition, attempts are made to
account for the effect of crack tip plasticity in preceding load cycles on the real
values of K (Keff) and R. A notable example is the Willenburg model [70]. More
comments on this model and the Wheeler model [11 ] are given later on.
The equivalent K method was proposed for random loading [60,62, 71]. The
basic assumption is that an equivalent ZkK can be indicated, which under
constant-amplitude (CA) loading will give the same crack rate as the random
loading.

[da[dN = f(lkKeq)random toading] = [da/dN = f(lkK)CA loading] (3)

The root mean square (rms) value of the random loading was proposed for this
purpose: /XKeq =/XKrm s . There are no theoretical reasons to see that the basic
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
$CHIJVE ON VARIABLE-AMPLITUDE LOADING |7

assumption is plausible. Checks in the literature have shown systematic

deviations [60,62, 71].
The characteristic K-concept was proposed by Paris [72] for stationary
random loading. Contrary to the equivalent K-method, a relationship to
constant-amplitude data was abandoned. The basic idea is that the random
variations of the stress in the crack tip zone are fully described by AKrms
daMN =fR (ZkKrms) (4)
with AKrm s = C Srms%/~ (5)
Available evidence in the literature [60, 71, 73] appears to confirm the validity of
this approach. It should not be overlooked that the relationship in Eq 4 will be a
function of a stress ratio, which was defined by S.H. Smith [60] as 7 =
Sm/Srm s . Two more important variables of random loading are the spectral
density function and the crest factor (clipping ratio). The first one appears to
have a small effect on the crack rate [60,71], although the evidence is still
limited. It is difficult to indicate theoretical expectations on this topic. The
effect of the crest factor is practically unknown, but a significant effect should
be expected in view of results obtained in flight-simulation tests [6,61].
The promising results obtained for random loading have promoted a similar
approach to flight-simulation loading [5, 74]. A characteristic K-value can also
be defined easily for this type of loading. Unfortunately, the test results
indicated that a relation similar to Eq 4 did not hold. In a theoretical analysis of
this observation, it was shown that Eq 4 should be replaced by
daMN = fR (,ilK, dK/da) (6)
In constant-amplitude tests and random load tests, the effect of dK/da turned
out to be negligible, but for flight-simulation loading this did no longer apply. A
more extensive discussion is given in Refs 5 and 74.

The Willenborg Model/26,69] and the WheelerModel [ 11]

Both the Willenborg and Wheeler models were proposed to explain crack
growth delays induced by high loads. Both employ plastic zone sizes as indicated
in Fig. 10, however, the concepts are different. Wheeler related the retardation

PLASTICZONECAUSED
BY OVERLOAD
PLASTICZONE OF CYCLE"i"

. { , ~ ~ ~ PLASTICZONEREQUIRED
'~OL _~.rploLI I -J " TO FULLY ELIMI NATE
~L rp i~ ]
r , - "7 DELAY EFFECT OF OVERLOAD
oi -[
r-
~,' ]
-r

FIG. l O-Plastic zone size concepts in the models o f Wheeler and Willenborg.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
18 FATIGUE CRACK GROWTH

to rpi/X. Willenborg made a different assumption about the effective stress as

affected by the plastic deformation of the overload. As a result, z~gef f and Ref f
are calculated and used to obtain the applicable da/dN values.
Both models were checked by various authors [14,15,26,62,63,75], but a
systematic agreement with test results was rarely found. The advantage of the
WiUenborg model was said to be that no empirical material constants were
required. However, in order to improve the reliability of the Willenborg model a
material constant was again introduced [29,62]. Actually, the number of
variables of complex load-time histories is rather large, and it would be surprising
if a single empirical constant would be sufficient to account for all variables.
This can be expected only if the model itself is physically correct. Unfortu-
nately, this does not appear to be true. The Willenborg model and the Wheeler
model consider plasticity in the crack tip zone only with a simple assumption
about the plastic zone size. A reversed plastic zone as discussed by Rice [76] is
not included. Both models predict maximum retardation immediately after an
overload, while the retardation is assumed to pass off as soon as rpi = h (Fig.
I0). These features were shown to be incorrect, which can easily be explained by
crack closure. The models exclude the possibility of accelerated growth, which is
also a real phenomenon.
In Ref 68 requirements for a more realistic model were indicated. The
complexity of the problem became clearly evident. The first approach would be
to incorporate crack closure, but, it cannot be ruled out that this will not be
enough to make a model sufficiently reliable for prediction problems. Extensive
research is still required to reach this goal.

Empirical approach
If fundamentally correct laws are not available, it is a practical solution
to look for a systematic empirical rule. Attempts were made to start from delays
as observed in tests with overloads. Delay factors were derived from such
tests, which were then translated to more complex loading programs
[12,14,18,19,48]. The more complex programs were still rather simple as
compared to service load-time histories. Since these propositions start from
delays only, ignoring accelerations, it is doubtful whether it will ever lead to a
useful rule. Checking empirical rules by unpractical types of loading is not a
logical approach. Similarly, it is also illogical to disprove an empirical rule by
tests with unpractical types of loading.
The prediction method proposed by Habibie [77-79] for application to
service load-time histories is starting from crack growth retardations as observed
in flight-simulation tests. Apparently, this is a more practical approach. Habibie
employs the K factor to arrive at crack growth retardation formulas. Originally
he adopted eight material constants, but this number was later reduced. His
predictions for flight-simulation test results are quite good, but the method was
less successful in predicting the trends for program tests. For an empirical
method, the latter result need not be a disadvantage if the prediction for realistic
load sequences is considered.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
SCHIJVE ON VARIABLE-AMPLITUDE LOADING 19

Multi- VariableRegression Analysis

Habibie was still employing the K-factor to support his method by physical
arguments. A more rigorous empirical approach is to start directly from
empirical data and to look for mathematical relations which best represent the
trends. This leads to multi-variable regression analysis of test data. An example
of this approach was recently suggested for flight-simulation loading by
Simpkins, Neulieb, and Golden [80]. The regression function proposed is
N = D " (O2) a " ( S l g ) b . (-ff)e. (1.z-)d (7)
where D is a constant, o2 is the variance of the load spectrum in flight, Slg is the
average mean stress in flight, K is the average number of cycles per flight, and Z
is the average of Smin of the ground-to-air cycle divided by $1 e. The equation
accounts for four independent variables which are supposed to characterize the
severity of the load history. The five constants D, a, b, e, and d are determined
by a regression analysis applied to empirical data. Equation 7 applies to the
fatigue life, N, under flight loading. A similar approach was proposed for random
loading [81]. Applications to crack propagation although not yet made so far,
could occur in the same way. The problem is to indicate the independent
variables, which sufficiently characterize the load-time history.
Incidentally, a rigorous analysis is still hampered by insufficient available test
data. For this purpose a systematic test program was proposed earlier [61]. The
purpose should be to obtain systematic data on the effect of the more important
variables of realistic load-time histories.

Conclusions
1. The problem of predicting crack growth rates in service cannot be solved
without a thorough knowledge of the load-time histories occurring in service. A
statistical distribution function of peak loads is insufficient, and knowledge
about load sequences is essential.
2. During crack growth under variable-amplitude loading, significant inter-
action effects will occur. Both crack growth retardations and accelerations have
been amply demonstrated. Relevant evidence is available from tests with simple
types of loading (overloads, step loading), but similar trends were observed in
tests with more complex load sequences (random loading, program loading, and
flight-simulation loading).
3. Several mechanisms can contribute to interaction effects, and it is difficult
to separate the contributions of each. Actually, the picture of damage
accumulation during crack growth under variable.amplitude loading is very
complex. However, it has been shown that crack closure gives a significant
contribution, and new evidence is reported in this paper. It is strongly
recommended that empirical studies on crack growth under variable-amplitude
loading should include crack closure measurements.
4. During complex load-time histories, high-amplitude cycles contribute more
crack extension than the non-interaction concept will predict, whereas low-
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
20 FATIGUE CRACK GROWTH

amplitude cycles contribute less. This observation could be significant for

environmental effects.
5. Since interaction effects are very sensitive to the sequence of the loads the
"equivalence" of complex load-time histories and simplified histories (for
example, program loading) is illusory. Actually, realistic answers from tests can
only be expected if realistic load-time histories are simulated.
6. There is an increasing activity reported in the literature to develop
prediction techniques. Unfortunately, available crack growth models are too
simple, and reliable predictions cannot be expected with any certainty.
Observations on crack closure as affected by overloads in relation to material
thickness and yield strength strongly emphasize the controlling influence of
plastic deformation on fatigue crack growth. Therefore, it should be attempted
to include crack closure in a fatigue model, although this alone may not be
sufficient.
7. For random loading the applicability of AK rms is promising. Information
on the effect of the crest factor (clipping ratio) is insufficient.
8. For the time being, a prediction method based on multi-variable regression
analysis could provide useful information for practical problems. However,
sufficient empirical data is not yet available. Empirical investigations to s this
gap should be recommended.

References
[1] Flillbery, B. M. in Effects of Environment and Complex Load History on Fatigue
Life, ASTM STP 462, American Society for Testing and Materials, 1970, pp.
167-183.
[2] V. Dijk, G. M. in Advanced Approaches to Fatigue Evaluation, NASA SP-309,
1972, pp. 565-598.
[3] Schijve, J. in Fatigue of Aircraft Structures, Pergamon, 1963, pp. 115-149.
[4] Schijve, J., Engineering Fracture Mechanics, Vol. 5, 1973, pp. 269-280.
[5] Schijve, J., Jacobs, F. A., and Tromp, P. J., "Fatigue Crack Growth in Aluminium
Alloy Sheet Material under Flight-Simulation Loading. Effects of Design Stress
Level and Loading Frequency," NLR TR 72018, Amsterdam, 1972; also in
AGARD-CP-118, 1972.
[6] Schijve, J., "The Accumulation of Fatigue Damage in Aircraft Materials and
Structures," AGARDograph No. 157, 1972.
[ 7] Recent test results of W. J. Arkema, Aerospace Department, Delft University of
Technology, The Netherlands.
[8] McMiUan, J. C. and PeUoux, R. M. N., in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, pp. 505"--532.
[9] Elher, W., Engineering Fracture Mechanics, Vol. 2, 1970, pp. 37--45.
[10] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[11 ] Wheeler, O. E., Journal of Basic Engineering, Transactions, ASME, March 1972, pp.
181-186.
[12] Porter, Th. R., Engineering Fracture Mechanics, Vol. 4, 1972. pp. 717-736.
[13] Schiitz, W. and Zenner, H., "Crack Growth in Titanium Alloy Sheets under
Flight-Simulation Loading" (in German), IABG Report TF-224, Ottobrunn, Jan.
1972.
[14] Jones, R. E., Engineering Fracture Mechanics, Vol. 5, 1973, pp. 585--604.
[15] Katcher, M., Engineering Fracture Mechanics, Vol. 5, 1973, pp. 793-818.

[16] Wei, R. P., Shih, T. T,, and Fitz Gerald, J. H., "Load Interaction Effects on Fatigue
Crack Growth in Ti-6AI-4V Alloy," NASA Report CR 2239, 1973.
117] Shih, T. T. and Wei, R. P., "A Study of Crack Closure in Fatigue," NASA Report
CR-2319, Oct. 1973.
[18] Wei, R. P. and Shih, T. T., International Journal of Fracture, Vol. 10, 1974, pp.
77-85.
[19] Shih, T. T. and Wei, R. P. in Prospects of Fracture Mechanics, Noordhoff, Leyden,
1974, pp. 231-248.
120] I-Isu,T. M. and Lassiter, L. W., "Effects of Compressive Overloads on Fatigue Crack
Growth," AIAA Paper No. 74-365, April 1974.
[21] Matthews, W. T., Baratta, F. 1., and Driscoll, G. W., International Journal of
Fracture Mechanics, Vol. 7, 1971, pp. 224-228.
[22] Feddersen, Ch. E., "Fatigue Crack Propagation in D 6 AC Steel Plate for Several
Flight Loading Profiles in Dry Air and JP-4 Fuel Environments," AFML-TP-72-20,
Air Force Materials Laboratory, Jan. 1972.
[23J Davidson, J. R., Engineering Fracture Mechanics, Vol. 4, 1972, pp. 777-788.
[24] Keays, R. H., "Numerical Evaluation of Wheeler's Model of Fatigue Crack
Propagation for Programmed Load Spectra," ARL/SM 736, Melbourne, April 1972.
[25] Ryan, N. E., "The Influence of Stress Intensity History on Fatigue Crack Growth,"
ARL/Met. 92, Melbourne, June 1973.
[26] Wood, H. A. in Fatigue Life Prediction for Aircraft Structures and Materials,
AGARD Lecture Series No. 62, May 1973.
[27] Gallagher, J. P., Stalnaker, H. D., and Rudd, J. L., "A Spectrum Truncation and
Damage Tolerance Study Associated with the C-5A Outboard Pylon Aft Tress
Lugs," AFFDL-TR-74-5, Air Force Flight Dynamics Laboratory, May 1974.
[28} Petrak, G. J., Engineering Fracture Mechanics, Vol. 6, 1974, pp. 725-733.
[29] Gallagher, J. P. and Hughes, T. F., "Influence of Yield Strength on Overload
Affected Crack Growth Behavior in 4340 Steel," AFFDL-TR-74-27, Air Force
Flight Dynamics Laboratory, July 1974.
[30] Griffiths, J. R., Mogford, 1. L., and Richards, C. E., Metal Science Journal, Vol. 5,
1971, pp. 150-154.
[3l] Von Stubendorff, A., "'Fatigue Crack Growth in Notched Stee! Plates Loaded in
Tension," Ph.D. thesis, Delft University of Technology, Dec. 1971.
[32] Klesnil, M. and Lukas, P., Engineering Fracture Mechanics, Vol. 4, 1972, pp.
77-92.
[33] Lukas, P. and Klesnil, M., "Transient Effects in Fatigue Crack Propagation,"
Institute of Physical Metallurgy, Czechoslovak Academy of Sciences, Brno, Jan.
1975.
134] Rice, R. L, and Stephens, R. I. in Progress in Flaw Growth and Fracture Toughness
Testing, ASTM STP 536, American Society for Testing and Materials, 1973, pp.
95-114.
[35] Gardner, F. H. and Stephens, R. i. in Fracture Toughness and Slow Stable Cracking,
ASTM STP 559, American Society for Testing and Materials, 1974, pp. 225-244.
[36] Barsom, J. M. in Progress in Flaw Growth and Fracture Toughness Testing, ASTM
STP 536, American Society for Testing and Materials, 1973, pp. 147-167.
[37~ Priddle, E. K., "Fatigue Crack Propagation due to Random Amplitude Vibrations:
Prediction Using Constant Amplitude Data," Central Electricity Generating Board,
Report RD/B/N2784, Nov. 1973.
[38] Priddle, E. K., "Fatigue Crack Propagation Due to Random Amplitude Vibrations:
the Influence of Stress Intensity Mean on Propagation Rate," Central Electricity
Generating Board, RD/B/N 3097, Sept. 1974.
[391 Lindley, T. C. and Richards, C. E., Materials Science and Engineering, Vol. 14,
1974, pp. 281-293.
[401 Schijve, J. in Advances in Aeronautical Sciences, Vol. 3, Pergamon, 1962, pp.
387---408.
[41] Schijve, J. and Broek, D.,Aircraft Engineering, Vol. 34, 1962, pp. 314-316,
[42] Hudson, C. M. and Hardrath, H. F., "Investigation of the Effects of Variable-
Amplitude Loadings on Fatigue Crack Propagation Patterns," NASA, TN D-1803,
Aug. 1963.

[43] McMillan, J. C, and Hertzberg, R. W. in Electron Fractography, ASTM STP 436,

American Society for Testing and Materials, 1968, pp. 89-123.
[44] Hudson, C. M. and Raju, K. N., "Investigation of Fatigue Crack Growth Under
Simple Variable-Amplitude Loading," NASA, TN-D-5702, 1970.
[45] Von Euw, E. F. J., Hertzberg, R. W., and Roberts, R. in Stress Analysis and Growth
of Cracks, ASTM STP 513, American Society for Testing and Materials, 1972, pp.
230-259.
{46] Trebules, V. W., Roberts, R., and Hertzberg, R. W. in Progress in Flaw Growth and
Fracture Toughness Testing, ASTM STP 536, American Society for Testing and
Materials, 1973, pp. 115-146.
[47] Corbty, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973,
pp. 479--497.
148] Probst, E. P. and HiUberry, B. M., "Fatigue Crack Delay and Arrest Due to Single
Peak Tensile Overloads," AIAA Paper No. 73-325, 1973.
[49] Schijve,J., Engineering Fracture Mechanics, Vol. 6, 1974, pp. 245-252.
[50] Van Lipzig, H. T. M. and Nowack, H., "Fatigue Crack Growth and Residual
Strength of Light Metals under Multi-Level Loading" (in German), paper of the
lnstitut fdr Werkstoff-Forschung (DFVLR), Porz-Wahn,Oct. 1974.
[51] Swift, T., "The Application of Fracture Mechanics to the Design of Damage-
Tolerant Stiffened Aircraft Structure," Douglas Paper 5981, 1972.
[52] Schijve, J. and De Rijk, P., "The Effect of Ground-to-Air Cycles on the Fatigue
Crack Propagation in 2024-T3 Alclad Sheet Material," NLR-TR M 2148,
Amsterdam 1965.
[53] Christensen, R. H. in Metal Fatigue, McGraw-Hill, 1959, p. 376.
[54] Hudson, C. M. and Hardrath, H. F., "Effects of Changing Stress Amplitude on the
Rate of Fatigue-Crack Propagation in Two Aluminium Alloys," NASA, TN D-960,
1961.
[55] Plunkett, R. and Viswanathan, N., Journal of Basic Engineering, Transactions,
ASME, March 1967, pp. 55"-68.
[56] Crews, J. H. and White, N. H., "Fatigue Crack Growth from a Circular Hole With
and Without High Prior Loading," NASA, TN-D-6899, Sept. 1972.
[57] Hanel, J. J., "Crack Growth Prediction for Programmed Loading, Including Crack
Closure as Based on the Dugdale Barenblatt Model" (in German), paper of the
Technische Hochschule Darmstadt, Oct. 1973.
[58] Newman, J. C. and Armen, H., "Elastic-Plastic Analysis of a Propagating Crack
under Cyclic Loading." AIAA Paper No. 74-366, April 1974.
[59] Brussat, T. R. in Damage Tolerance in Aircraft Structures, ASTM STP 486,
American Society for Testing and Materials, 1971, pp. 122-143.
[60] Smith, S. H. in Acoustical Fatigue in Aerospace Structures, Syracuse University
Press, 1965, pp. 331-360.
[61] Schijve,J., TheAeronauticalJournal, Vol. 74, 1970, pp. 517-532.
[62] Gallagher, J. P. and Stalnaker, H. D., "Methods for Analyzing Fatigue Crack
Growth Rate Behavior Associated with Flight-by-Flight Loading," AIAA Paper No.
74-367. April 1974.
[63] Sayer, R. B., "Non-linear Effects of Spectrum Loading on Fatigue Crack Growth in
Transport Wings," AIAA Paper No. 74-984, Aug. 1974.
[64] Nowack, H. and Mukherjee, B. in Fail-Safe Aircraft Structures, RAE TR 73183,
Vol. 1, Farnborough, 1974.
[65] Impellizzeri, L. F. in Effects of Environment and Complex Load History on Fatigue
Life, ASTM STP 462, American Society for Testing and Materials, 1970, pp.
40-68.
[66] Schijve, ]. in Advanced Approaches to Fatigue Evaluation, NASA, SP-309, 1972,
pp. 253-273.
[67] Schijve, J., Jacobs, F. A., and Tromp, P. J., "Flight-Simulation Tests on Notched
Elements," NLR TR 74033, Amsterdam, Feb. 1974.
[681 Schijve, J., "'Fatigue Crack Growth under Variable-Amplitude Loading," Delft
University of Technology, Report VTH-181, April 1974.
[69] Hartman, A. and Schijve, J., Engineering Fracture Mechanics, Vol. I, 1970, pp.
615--631.

[70] Willenborg, J. D., Engle, R. M., and Wood, H. A., "A Crack Growth Retardation
Model Using an Effective Stress Concept," AFFDL-TM-FBR-71-1, Air Force Flight
Dynamics Laboratory, 1971.
[71 ] Paris, P. C. in Fatigue, an Interdisciplinary Approach, Syracuse University Press,
1964, pp. 107-132.
[72] Paris, P. C., "The Growth of Cracks due to Variations in Load," Ph.D. thesis,
Lehigh University, 1962.
[ 73] Swanson, S. R., Cicci, F., and Hoppe, W. in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, pp. 312-362.
[74] Schijve, J. in Fail-Safe Aircraft Structures, Proceedings, ICAF Symposium London,
July 1973, RAE TR 73183, Vol. 1, March 1974.
[75] Engle, R. M. and Rudd, J. L., "Analysis of Crack Propagation under Variable
Amplitude Loading Using the Willenborg Retardation Model," AIAA Paper No.
74-369, April 1974.
[76] Rice, J. R. in Fatigue Crack Propagation, ASTM STP 415, American Society for
Testing and Materials, 1967, pp. 247-309.
[77] Habibie, B. J., "A Prediction Method for Crack Growth under Random Loading.
Comparison with Test Results" (in German), Paper No. 71-111, DGLR Conference,
Baden-Baden, Oct. 1971.
[ 78] Habibie, B. J., "On the Integration Method of Crack Propagation in Elastic Plastic
Material under Operational Loads and Plane Stresses," paper of 3rd Conference on
Fracture, Munich, April 1973.
[ 79] Habibie, B. J., ZeitschriftfdrFlugwissenschaften, Vol. 4, 1974, pp. 120-127.
[80] Simpkins, D., Neulieb, R. L., and Golden, D. J., Journal of Aircraft, Vol. 11, 1974,
pp. 563---570.
181] Bussa, S. L., Sheth, N. J., and Swanson, S. R., Materials Research and Standards,
Vol. 12, 1972, pp. 31-43.

Fatigue Crack Growth with Negative

Stress Ratio Following Single
Overloads in 2024-T3 and 7075-T6
Aluminum Alloys

REFERENCE: Stephens, R. I., Chert, D. K., and Horn, B. W., "Fatigue Crack
Growth with Negative Stress Ratio Following Single Overloads in 2024-T3 and
7075-1"6 Aluminum Alloys," Fatigue Crack Growth Under Spectrum Loads, ASTM
STP 595, American Society for Testing and Materials, 1976, pp. 27-40.
ABSTRACT: Modified pre-cracked compact specimens of 2024-T3 and 7075-T6
aluminum alloys were subjected to four different overload patterns followed by
subsequent constant-amplitude steady-state loading with R =Pg.min/Pg~maxequal to
0, -1/2, -1, and -2. The overload patterns were tension, compression-tension,
tension-compression, and compression. Cyclic loading with negative stress ratio, R,
drastically reduced crack-growth retardation. The higher the negative R ratio the
greater the reduction in retardation. Overload ratios, OLR =Phmax/Pg,max, ranging
from 2.0 to 3.0 were used. For compression overloads, the OLRranged from -2.0 to
-4.0. High compression overloading was detrimental and dependent upon subsequent
R ratio loading. Substantial fracture surface abrasion near the mid-thickness occurred
for higher negative R ratios. Striations were not readily found in this region, however,
they were quite evident near the edges, which indicated crack closure was greater
near the mid-thickness. The 2024-T3 gave better crack growth life than 7075-T6 in
some loading conditions, while the opposite was true for other loadings. The results
indicate negative R ratio must be considered in retardation models and that
retardation life cannot be modeled based solely on overload plastic zone sizes.

KEY WORDS: fatigue (materials), crack propagation, fracturing, residual stress,

loads (forces), plastic zone, stress ratio, aluminum alloys

Fatigue crack-growth retardation following high tensile excursions has recently

received widespread interest. In most instances cyclic loading following a high
tensile excursion was also in the tensile region and resulted in appreciable
crack-growth retardation and even crack arrest. Representative crack.growth
retardation with zero or positive R = /~ following high tensile
excursions have been reported in 2024-T3 aluminum [1] 2 7075-T6 aluminum
[2], Ti-6A1-4V [3], austenitic manganese steel [4], cold-rolled low carbon steel
[5], hot-rolled low carbon steel [6], and 4340 steel [7]. These materials
represent both cyclic strain hardening and cyclic strain softening behavior.

I Professor and research assistants, respectively, Materials Engineering Division, The

University of Iowa, Iowa City, Iowa 52242.
2 The italic numbers in brackets refer to the list of references appended to this paper.
27
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions auth
28 FATIGUE CRACK GROWTH

Crack-growth retardation has been attributed to both compressive residual

stresses ahead of the crack tip and to crack closure. Stephens et al, however, in a
brief current research report [8] showed fatigue crack-growth retardation
following a high single tensile excursion could be substantially reduced and even
eliminated if the subsequent load ratio, R = P~min/Ps was negative. The
present paper provides an in-depth extension of the original brief report and
indicates the importance of load ratio R when considering fatigue crack-growth
retardation.

Material and Experimental Procedures

The 2024-T3 and 7075-T6 aluminum plate alloys with yield strengths of 355
and 558 MN/m 2 , respectively, in the rolled direction were used in this research.
Modified compact specimens shown in Fig. la were cut from the 10. I-ram-thick

' IlllVl TIME ~0 vvwvt ~, t

_ [P~mi.
(3.4--
----,,4 101.6 (4.0) I.,-
b 0

TIME
.

d e f

FIG. l-Modified compact specimen and loading spectrum: (a) modified compact
specimen, mm (in.); (b) no overload (reference); (c) tension overload (7'); (d) compression-
tension overload (C-T); (e) tension-compression overload (T-C); and (f) compression
overload (C).

2024-T3 and 9.5-mm-thick 7075-T6 aluminum plates. The six holes provided
facility to fasten specimens to a spherical monoball gripping system which
allowed both tensile and compressive loadings. The tests were performed in an
88 960 N closed-loop electrohydraulic test system with ram loading transferred
through the spherical monoball bearings which prevented external bending
moments. The loading direction was parallel to the rolled direction. The crack
length, a, was measured from the centerline of the axial ram loading. Specimen
side grooves were used to ensure crack growth path direction and resulted in net
thicknesses of 9.15 mm for 2024-T3 and 8.9 mm for 7075-T6. The side grooves
were polished in order to better monitor crack growth which was measured
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
STEPHENS ET AL ON NEGATIVE STRESS RATIO 29

optically using x36 magnification and stroboscopic lighting. A least-reading

0.25-ram scale was taped directly adjacent to the side groove. Fatigue
crack-growth tests were run at room temperature with relative humidity between
68 and 74 percent.
The loading spectrum incorporated is shown in Fig. lb t h r u f a n d consisted of
four different overload patterns followed by constant load cycling with R =
P~min/l~ = 0, -1/2, -1, and -2 for each overload pattern. The four
overload patterns were tension (T) (Fig. lc), compression-tension (C-T) (Fig.
ld), tension-compression (T-C) (Fig. le), and compression only (C) (Fig. 1r
When both tension and compression occurred in the overload cycle, PhmaxWaS
always equal to -Phmin 9 No-overload tests (Fig. lb), used as reference, were also
obtained for each R ratio. The machined V-notch was pre-fatigue sharpened in
all specimens with R = 0 to a crack length of 25.4 mm with the maximum load
equal to Ps = 8580 N for 2024-T3 and 8310 N for 7075-T6. This resulted in
a final stress intensity of 19.7 MNm -3~2 . Overload patterns were then applied at
0.5 Hz with overload ratio (OLR) = Phmax/e~max equal to 2.0, 2.3, 2.5, or 3.0.
Compression overloads only were applied with OLR = -2, -2.3, -2.5, -3.0, and
-4.0. Subsequent steady-state constant-load cycling always had P~max equal to
the pre-overload value. P~min was established from the particular R value for
each specimen. Thus, if no crack extension occurred during overload, then each
test for both materials would begin with identical initial K~max --- 19.7
MNm -a/2 . Crack tension during overloading, however, did occur in many tests
and is described later.
The maximum overload stress intensity for OLR = 2.0, 2.3, 2.5, and 3.0 was
39.4 MNm -3/2 , 45.3 MNm -3/2 , 49.2 MNm -3/2 , and 59.1 MNm -3/2 ,
respectively. Monotonic fracture toughness tests using the same specimen
geometry, maximum load, and initial crack length of 25.4 mm resulted in
non-valid average fracture toughness, Ke, of 70.6 MNm -3/2 for 2024-T3 and
51.9 MNm -3/2 for 7075-T6. Two tests were performed for each material. OLR
equal to 3.0 was therefore not possible for 7075-T6 since fracture would occur
during the overload. The ratio of the maximum overload stress intensity, K hmax
to Ke, ranged from 0.56 to 0.835 for 2024-T3 and 0.76 to 0.95 for 7075-T6.
Steady-state constant-load amplitude following overloading was at 10 to 20 Hz
depending upon the R ratio. The larger negative R ratios required lower
frequencies in order to maintain proper frequency response. As the crack grew
to about 40 ram, the frequency had to be lowered to about 5 Hz for the larger
negative R ratios. Total fatigue crack growth life following overloads was taken
from ao equal to 25.4 mm to fracture. If crack growth following an overload did
not occur in 300 000 cycles then complete crack arrest was assumed and the test
terminated. A total of 60 tests were performed with 2024-T3 and 52 tests with
7075-T6.

Test Results
A summary of all crack growth life following single overloads is shown in
Table 1 for both 2024-T3 and 7075-T6. Some duplicate tests were run and both
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions autho
t.o
0
TABLE 1 - Summary o f crack growth lives
"1"1

--4
2024-T3 7075-T6
C
m
t")
20
Loading OL~RR >
Patterns ~ 0 -1/2 -1 -2 0 -1/2 -1 -2 t'%

Xl
No overload no 38 800 33 700 29 700 29 500 no 21 600 22 400 22 300 22 300 o
(Fig. lb) overload 36 100 30 700 27 700 overload 22 800
--4
"-r
Tension 2.0 85 200 39 400 36 100 31 300 2.0 25 900 a 27 000 26500
overload 38 000 "'" 22 200
(Fig. lc) 2.3 ... 2.3 222 700 119 900 64 500 26 700
2.5 >300 000 135 700 46600 30900 2.5 >300 000 253 400 174 000 51 400
26 600 40 000
3.0 %/ >300 000 187 600 45 800 3.0 . . . . . . . . . . . .

Compression- 2.0 50 600 29 900 26 800 28 800 2.0 . . . . . . . . . . . .

tension 30 200
overload 2.3 . . . 2.3 119 400 106 000 59 400 27 100
(Fig. ld) 2.5 271 100 58 900 40 600 29 300 2.5 278 000 230 000 172 000 52 600
3.0 ~/ >300 000 133 200 43 400 3.0 . . . . . . . . . . . .

Tension- 2.0 36 900 30 500 28 400 28 500 2.0 . . . . . . . . . . . .

compression
overload 2.3 . . . 2.3 33 900 30 500 28 300 26 400
(Fig. le) 2.5 34 200 30 600 25 400 28 500 2.5 57 600 37 000 36 800 36 300
3.0 30 900 29 400 23 900 25 800 3.0 . . . . . . . . . . . .

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Compression -2.0 34 0 0 0 27 7 0 0 26 9 0 0 27 4 0 0 -2.0
overload
(Fig. If) -2.3 . . . . . . . . . . . . -2.3 21 0 0 0 21 700 24 100 24 6 0 0
-2.5 31 0 0 0 24 8 0 0 25 800 23 0 0 0 -2.5 17 4 0 0 18 800 19 800 21 4 0 0
-3.0 25 700 21 0 0 0 21 4 0 0 20 6 0 0 -3.0
--4.0 25 500 18 4 0 0 18 4 0 0 14 200 -4.0 15 8 0 0 18 6 0 0 19 500 21 5 0 0

a Average of seven tests9

01
--t
m
"r
m
Z
ol
m
"-I

I'-
o
Z
Z
rln

--t

m
01
-q

m
01

..-n

.-t

values are included in the table. Normal scatter existed within these duplicate
tests. No-overload constant-amplitude reference tests are included for both
materials. For 2024-T3, a decrease in constant-amplitude life of approximately
25 percent occurred as R increased from 0 to -2. A general conclusion should
not be made that under constant-amplitude loading increased negative R ratio
decreases 2024-T3 life, because these data only include the upper portion of the
sigmoidal log da/dN versus log zSd~ curve due to their high crack growth rates.
The constant amplitude 7075-T6 crack growth life was essentially constant as R
varied from 0 to -2. This data was also in the upper portion of the sigmoidal
crack-growth rate curve. For these identical constant-amplitude no-overloading
tests, 2024-T3 had greater life than 7075-T6 and, thus, lower crack growth rates
for all R ratios.
The OLR = 2.0 for 7075-T6 was applied to just tension overloading since a
negligible amount of crack-growth retardation or crack life increase occurred.
The effect of negative R ratio for a given overload pattern and material can be
determined by traversing across a row in Table 1. The effect of overload ratio for
a given R ratio can be obtained by scanning a column within a given overload
pattern section. A check mark in Table 1 implies crack arrest would occur for
the specific loading conditions. Tests labeled >300 000 indicate crack arrest
occurred.
Representative crack length versus applied cycles following a specific overload
pattern and overload ratio with R ratio as the parameter are shown in Figs. 2 thru
4. Figure 2 indicates that crack-growth life in 2024-T3, for a tensile overload

t~ 45' R=-2
E

-r- 4 0 -
I.-
(9
Z
Lz.I
9J 35-

,,,=X
30-

25 ~
I 1 l I
o ib 2'0 30 go 50 60 70 do
APPLIED CYCLES - N (103)

FIG. 2-Crack growth following single tensile overload in 2024-T3 aluminum, OLR = 2. O.

ratio of 2.0, is drastically reduced for R = -1/2, -1, and -2 compared with life
at R = 0. Figure 3 indicates the same result for 7075-T6 with a tension overload
ratio of 2.5. Compression-tension overloading for 2024-T3 with an overload
ratio equal to 2.5 illustrates similar reduction in crack-growth life in Fig. 4. A
general conclusion from these representative curves along with scanning the rows
in Table 1 is that increasing the value of R from 0 to - 2 following a given
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
STEPHENS ET AL ON NEGATIVE STRESS RATIO 33

NO OVERLOAD
50-

45-
E

/
R=-
"r 4 0 -
I--
O
Z
~ 35-

~ 30-

)
;'5- r iV'
~o 40 r~o T~o 21o 2~o 2so
APPLIED CYCLES- N ( IOs)

FIG. 3-Crack growth following single tensile overload in 7075-T6 aluminum, OLR = 2.5.

50-

F: 45.
EI
a 40.
-r
I.--
z
~ 35.
v

a: 3 0
0

25 84

o ib ~ io 4:0 io 6o
APPLIED CYCLES- N (I0 a)
F I G . 4-Crack growth following single compression-tension overload in 2024-T3 alumi-
num, OLR =2.5.

tension or compression-tension overload causes drastic reduction in crack-growth

retardation and hence total crack-growth life. Under constant-load amplitude
cycling following overloads, retardation is best characterized by the constant
crack-growth region (straight line) of the a versus N curve. It is apparent that
with higher overload ratios, complete crack arrest occurred for R = 0 whereas
only small increases in crack-growth life resulted with negative R ratios.
In order to better visualize the drastic effects of negative R ratio following
tension or compression-tension overloads, the data from Table 1 for these two
loading patterns has been normalized and plotted against R in Fig. 5. The
normalized life was obtained by dividing the total life to fracture for a given
overload and R ratio by the no-overload constant-amplitude life at the same R
ratio. This figure indicates that with R = -2, very little benefit results from the
overloads even with the highest overloads. Figure 5 also indicates that
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
34 F A T I G U E C R A C K GROWTH

2024-T3 7075 - T 6
1:5- 13-
! rt
,~w I F r, OLR= 3 . 0 T, OLR = 2.5
q,, ll-
,,,."5 ' / - C-T, OLR = 3.0 nr..J
l/d
C-T, OLR =2.5
tg~-

ao
woe
JV

Z
7~

3-
WIlE
N~) 7"

3-
~ 2.3

-C-~ .

I I-
0 -v2 J -~ 0 -1/z -1 -2
R RATIO R RATIO

FIG. 5-Normalized overload crack growth life versus R ratio, tension (T) and com.
pression.tension (C-T) overloads.

compression-tension overload life was slightly less than tension-only overload life
for most conditions. This implies a high compression overload preceding a high
tension overload should not be completely neglected, but its effect is not large.
The tension-compression overload pattern, Fig. le, for 2024-T3 caused either
little change in crack-growth life or decreased life. For all R ratios, the higher the
overload ratio the smaller the life. However, for 7075-T6 the tension-
compression overload was beneficial in all cases with the higher the overload
ratio the greater the life. The increase in life was substantially less than tension
or compression-tension overloads, however, and as R went from 0 to -2 this
increase became very small.
Compression overloads, Fig. l f, ranging from -2.0 to -4.0 resulted in either no
affect on crack-growth life or decrease in life for both materials. In general, the
higher compression overload ratios caused greater decrease in life. The decrease
in life was due to higher crack-growth rates (accelerated crack growth)
immediately following the compressive overload. Life decreases ranged from 0 to
50 percent. For 2024-T3 increasing the negative R ratio caused greater decreases
in life, whereas with 7075-T6 increasing the negative R ratio tended to eliminate
the detrimental effects of the compression overload. Thus, 2024-T3 and
7075-T6 behaved in somewhat opposite manners for the compression and
tension-compression overload patterns.
The general effect of overload pattern on crack-growth behavior for a given
OLR and R is illustrated in Fig. 6. It is seen that tension or compression.tension
overloads can cause appreciable retardation while compression or tension-
compression overloads showed somewhat the same behavior as the no-overload
constant-amplitude loading.
Partial macroscopic fracture surfaces of specimens subjected to tensile
overloads are shown in Fig. 7. Crack extension during overload is easily seen as
the very dark area. For both materials, the higher the overload ratio the greater
the macroscopic crack extension during overload. The 7075-T6 shows crack

50-
~ ' - COMPRESSION

E,50
E ER OAO E f COMPRESSION---~ I I
' tl ~ TENSION \J 1 (
9I" 4 0 -
l-
z
I,d
.d 3 5 -

/==/ ,/ '-TENSION- #," j~

30- 2J/ a/ COMPRESSION/ /" \
j' / OVERLOAD / / "-TENSION
~ _ . j J OVERLOAD
25
o ,b 2b 3'0 4'o ~o do rb
APPLIED CYCLES-N(IOs)
FIG. 6-Crack growth following different overload patterns in 7075-T6 aluminum; OLR =
2.3, R = - 1 .

FIG. 7-Macroscopic fracture surfaces of tensile overload specimens as dependent upon

overload ratio: (a) 2024.T3 and (b) 7075-T6.

tunneling at the OLR = 2.3 and 2.5, and some crack tunneling also occurred
with the 2024-T3 for OLR = 3.0. Crack extension during overload ranged from
approximately 0.1 to 1.5 mm for 2024-T3 and 0.2 to 4.0 mm for 7075-T6. The
ratio of ghmax/ K e was between 0.56 and 0.835 for 2024-T3 and 0.76 and 0.95
for 7075-T6. However, Khmax was the same for each material for a given OLR.
The large crack extension that occurred with the higher overload ratios are
undesirable, however, crack tip plasticity and crack closure more than offset the
increase in crack length.
In general~ overload crack extension for a given OLR was similar for tension,
compression-tension, and tension-compression overloading. Fatigue crack-growth
surfaces remained flat following overloads due to specimen side grooves.
The effect of negative R ratio on macroscopic fracture appearance for both
materials is indicated in Fig. 8 for 2024-T3. As R was increased from 0 to -2,

FIG. 8-Macroscopic fracture surfaces as dependent upon negative R ratio, 2024-7"3

aluminum: (a) compression-tension overload and (b) compression overload.

substantial abrasion resulted in the mid-thickness region following the overload.

This abrasion region is darker and increased in magnitude as the negative R value
increased. This indicates that crack closure was most predominant on the inner
surfaces. The abrasion was evident for all four loading patterns, however, the
greatest abrasion, for a given R ratio, occurred with compression and
tension-compression overloads. It is also shown in Fig. 8b that no crack
extension occurred under the compressive overloads which was consistent for
both materials for all OLR.

Discussion of Results
Crack-growth retardation or acceleration occurred within a small crack-growth
region immediately following the overload pattern. That is, the overall effect on
total life was controlled primarily in the overload plastic zone region
immediately following the overload. Reversed plane stress plastic zones formed
from the tensile portion of an overload can be approximated as one-fourth the
monotonic plane stress plastic zone, 2ry, [9] where

and
Khmax : e h m a x ~ f ( a / W )

according to ASTM Test for Plane-Strain Fracture Toughness of Metallic

Materials (E 399-74). Values of the reversed plane stress plastic zone varied from
1.1 to 2.4 mm for 2024-T3 and 0.4 to 0.62 mm for 7075-T6. This plastic zone
size, however, can be drastically altered by subsequent cycling, particularly with
negative R ratio. Thus, the significance of the reversed stress plastic zone is
substantially reduced under subsequent negative R cycling. It is also substan-
tially reduced under tension-compression overload patterns.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
STEPHENS ET AL ON NEGATIVE STRESS RATIO 37

Delayed retardation following overloads could not be depicted from macro-

scopic observations of a versus N curves. This can be attributed to the small
effective crack-growth region where retardation occurred, crack tunneling
effects, least scale reading of 0.25 mm, and difficulties of observing crack tips
during negative R cycling. Scanning electron fractographs shown in Fig. 9,
however, do show delayed retardation for 2024-T3 following a tensile OLR =
2.5 with R = -1/2. This is indicated in Fig. 9a by fine striation spacings before
overloading, coarser spacing immediately after the overload region, and eventual
return to finer spacing in Fig. 9b. The overload band is indicated in Fig. 9a.
In general, fatigue crack striations could easily be depicted with the scanning
electron microscope in the pre-overload region for all R values. They were also
easily found after overloading for R = 0 and -1/2. However, with the higher
negative R ratios, it became more difficult to find striations. This was
particularly true in the abrasion regions indicated previously in Fig. 8. A
representative scanning electron fractograph of this region for R = -2 is shown
in Fig. 10a, and the lack of clear distinct striations is evident. Figure 10b shows
distinct striations near the edge of this same specimen.
Under constant-amplitude loading the use of the log.log linear relationship,
da/dN = A(AK) n has been predominant. AK has been taken as the positive stress
intensity range which implies the negative or compression portion of constant-
amplitude loading and has either no effect or negligible effect on crack-growth
life. The experimental results described, however, indicate that under spectrum
loading conditions the negative portion of a loading spectrum cannot be
neglected if possible retardation is to be considered in the crack-growth-life
analysis.
The Wheeler [10] and Willenborg [11] retardation models, and extensions of
these models, have received the widest application to crack-growth life under
spectrum loading. As pointed out by Wood [12] however, these models still
have appreciable need for improvement and incorporation of new behavior as it
becomes known, such as these negative R effects.

Summary and Conclusions

1. Negative R-ratio cycling following tension or compression-tension overload-
ing drastically reduced crack-growth retardation in both 2024-T3 and 7075-T6
aluminum. As the negative R ratio was increased, retardation decreased. An R
ratio of just -1/2 was sufficient to eliminate from 10 to 80 percent of the total
life found with R = 0. However, even with R = -2 the tension or
compression-tension overloads were either slightly beneficial or had no effect on
crack-growth life.
2. Tension-compression or compression overloading in 2024-T3 was detri-
mental or caused no change in life. For 7075-T6, tension-compression
overloading was somewhat beneficial, however, compression overloads were
either detrimental or had no effect. Compression overloading with overload
ratios o f - 2 . 0 to -4.0 reduced life from 0 to 50 percent.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
..q

C
m
7J

7~
7o
0

FIG. 9 - S E M fractographs indicating delayed retardation following tensile overload in 2024-T3 aluminum; O L R = 2.5, R = - 1 / 2 . (a)
overload region and (b) O. 13 m m after overload.

0
Z
Z
m
~3
>
-4
m

7~
m

FIG. I O - S E M f r a c t o g r a p h s in 7 0 7 5 - T 6 a l u m i n u m f o l l o w i n g c o m p r e s s i o n - t e n s i o n o v e r l o a d , R = - 2 : (a} m i d - t h i c k n e s s a n d (b) n e a r e d g e .

tad

3. For a given R ratio, the higher tension or compression-tension overload

ratios caused the greatest crack-growth retardation.
4. Compression-tension overloading caused slightly less retardation than
tension overloading and, thus, a high compression loading preceding a high
tension loading should not be completely neglected. Compression overloads
following tension overloads were extremely detrimental.
5. Crack extension during overloads (except for compression-only overloads)
ranged from 0.1 to 1.5 m m in 2024-T3 and 0.2 to 4.0 mm in 7075-T6. Overload
crack-tip tunneling was most evident in 7075-T6.
6. Appreciable fracture surface abrasion occurred in the mid-thickness region
following overloads as the negative R ratio increased. Striations were not readily
found in these regions, however, they were quite evident near the edges for all R
ratios. This implies greater mid-thickness crack closure following overloads.
7. Under identical loading conditions 2024-T3 had better life than 7075-T6 in
some loading patterns and poorer life in others. Thus, overload plastic zone sizes
do not completely control fatigue crack-growth behavior.
8. Negative R ratio loading following overloads must be considered in any
mathematical model involving crack-growth retardation life predictions.

Acknowledgment
The authors would like to thank Ronald Kelsey, Richard Feild, and Mick
Christensen of Alcoa Corporation for providing the material.

References
[1 ] vonEuw, E. F. J., Hertzberg, R. W., and Roberts, R. in Stress Analysis and Growth
of Cracks, ASTM STP 513, American Society for Testing and Materials, 1972, pp.
230-259.
[2] Corbly, D. M- and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 19.73,
pp. 479--497.
[3] Wei, R. P. and Shih, T. T., International Journal of Fracture Mechanics, Vol. 10,
No. 1, 1974, pp. 77-85.
[41 Rice, R. C. and Stephens, R. I. in Progress in Flaw Growth and Fracture Toughness
Testing, ASTM STP 536, American Society for Testing and Materials, 1973, pp.
95-114.
[5] Vargas,L. G. and Stephens, R. I. in Proceedings, Third International Conference on
Fracture, Munich, Germany, Vol. 6, April, 1973, p. V-325.
[6] Sheets, E. C., "Predicting Fatigue Crack Propagation in Steel Plates Under the
Effects of Periodic Overloads", MS thesis, The University of Iowa, Dec. 1973.
[ 7] Gallagher, J. P. and Hughes, T. F., "The Influence of Yield Strength on Overload
Affected Fatigue Crack Growth Behavior in 4340 Steel," AFFDL-TR-74-27, Air
Force Flight Dynamics Laboratory, Wright-Patterson AFB, Ohio, 1974.
[8] Stephens, R. I., McBurney, G. W., and Oliphant, L. J., International Journal of
Fracture, Vol. 10, No. 4, 1974, pp. 587--589.
[9] Rice, J. R., Fatigue Crack Propagation, ASTM STP 415, American Society for
Testing and Materials, 1967, pp. 247-309.
[101 Wheeler, O. E., "Spectrum Loading and Crack Growth," ASME, Paper No.
7 I-Met-X, American Society of Mechanical Engineers, 1971.
[11 ] Willenborg, J., Engle, R. M., and Wood, H. A., "A Crack Growth Retardation Model
Using an Effective Stress Concept," Air Force Flight Dynamics Laboratory,
Technical Memorandum 71-1-FBR, Wright-Patterson AFB, Ohio, Jan. 1971.
[12] Wood, H. A. in Fatigue Life Prediction for Aircraft Structures and Materials,
AGARD Lecture Series No. 62, NATO, May, 1973, pp. 8-1 to 8-31.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
W. X. Alzos, 1 A. C. Skat, Jr., 2 and B. M. Hillberry 2

Effect of Single Overload/Underload

Cycles on Fatigue Crack Propagation

REFERENCE: Alzos, W. X., Skat, A. C., Jr., and Hillberry, B. M., "Effect of Single
Overload/Underload Cycles on Fatigue Crack Propagation," Fatigue Crack Growth
Under Spectrum Loads, ASTM STP 595, American Society for Testing and Materials,
1976, pp. 41-60.

ABSTRACT: Tests, represented in a three-dimensional matrix, were performed to

investigate the effects of a single overload/underload sequence on the fatigue
crack-growth rate in 2024-T3 aluminum alloy. Three different stress intensity ratios
were studied to investigate the delay phenomenon: (1) overload level to maximum
cyclic level, Kol]Kmax; (2) underload level to overload level, Kul/Kol; and (3)
minimum cyclic level to overload level, Kmin/Kol. These ratios ranged from 1.6 to
3.0, -1.0 to +0.3, and 0.11 to 0.3, respectively. All tests had the same overload stress
intensity level. Sufficient data were recorded to accurately determine the crack-
growth rate through the overload affected zone. The crack length versus number of
cycles data for each test were represented by a spline function which was then
analytically differentiated to obtain the growth rate. The number of delay cycles was
found to be directly related to the minimum growth rate following the overload/
underload sequence. The ratios Kol[Km~x and Kui/Kol were shown to be of
particular significance to the delay while the ratio Kmin/Kol was shown to be of less
significance. The results were correlated with an extended crack closure concept.
From this the maximum value of the opening stress intensity level following the
overload/underload sequence can be determined and used to predict the number of
delay cycles.

KEY WORDS: crack propagation, fatigue (materials), loads (forces), stress ratio,
stress cycle, overload, underload

The load interaction effects on fatigue crack propagation due to variable-

amplitude loading has been a subject of investigation for many years and will
undoubtedly continue to be so for some time. The inability to adequately
predict these interaction effects has prompted recent interest in studying the
interaction due to simple load patterns and its influence on fatigue crack
propagation. Of particular interest is the decrease in growth rate (delay effect or
crack retardation) which normally follows a high overload or a reduction in the
load level. As pointed out by Schijve [1] 3 and Gallagher and Stalnaker [2] as

1 Mechanical engineer, Engineering Service Division, E. I. DuPont de Nemours and

Company, Wilmington, Del. 19898; formerly, graduate research assistant, Purdue University,
West Lafayette, Ind. 47907.
Graduate research assistant and professor, respectively, School of Mechanical Engineer-
ing, Purdue University, West Lafayette, Ind. 47907.
3 The italic numbers in brackets refer to the list of references appended to this paper.
41
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
42 FATIGUE CRACK GROWTH

well as many others, this delay can have a significant influence on the fatigue life
of a structure.
The crack-growth reduction resulting from a single overload preceded and
followed by the same level of cyclic loading has been studied by several
investigators including Von Euw [3], Jonas and Wei [4], Wei and Shih [5,6],
Probst and Hillberry [7], Gardner and Stephens [8], Trebules et al [9], Corbly
and Packman [10], and Himmelein and Hillberry [11]. Trebules et al [9],
Corbly and Packman [10], Gardner and Stephens [8], Wei and Shih [5], and
Hsu and Lassiter [12] also examined the effect of single periodic and multiple
overloads. Petrak and Gallagher [13] examined the effect of yield strength level
on the retardation behavior. Probst and Hillberry [7] observed complete crack
arrest for an overload sufficiently higher than the steady-state loading and
established an arrest/delay boundary. The boundary was found to be a function
of the stress ratio by Himmelein and Hillberry [11]. Jonas and Wei [4], Corbly
and Packman [I0], and Stephens et al [14] report that the delay following an
overload is reduced if the overload is followed by an underload before the
constant-amplitude cycling is resumed. Several techniques have been proposed to
describe the delay phenomenon [2,15-19]. Crack closure [20] has also been
used to explain the delay phenomenon [9,21,22].
Crack-growth retardation is known to be related to many factors. Following
the overload there is generally a rapid decrease in the growth rate to a minimum
level followed by a gradual increase back to the steady-state growth rate. Since
this minimum growth rate is very slow, it significantly affects the total number
of delay cycles. Small variations can result in considerable variation in the
resulting number of delay cycles. This study [23] was undertaken to investigate
the crack-growth rate behavior through the interaction zone following the
overload/underload sequence on otherwise steady-state load cycling and to
determine the correlation of the minimum growth rate with the applied load
conditions and number of delay cycles.

Test Program
The test program was designed so that the effects of each of the loading
parameters in the overload/underload sequence would be isolated. The test
matrix was defined in terms of the three stress ratios (the corresponding ranges
are given in parentheses)
Ruo = KuffKol (-1.00 to 0.30)
Qol = Koi/Kmax (1.6 to 2.2)
R m = Kmin]Kol (0.11 to 0.30)
The overload/underload test sequence is shown in Fig. 1. The level of Kot was
the same for all tests, and therefore varying each of the preceding ratios one at a
time corresponded to varying Kup Kmax , or Kmin, respectively, with the
remaining stress intensity levels constant.

KOL

KMAx '

KMIN
K
W
UL
tim~
KoL ~I~
QOL = 1 ~ RM " KOL

KUL KNIN
~o - Ko-
~ RF Ka~
FIG. 1-Overload/underloadsequenceand definitionof variables.

Previous results [3,11] have indicated that the delay behavior is different for
overloads which create plane stress as opposed to plane strain conditions. For
this investigation, the overload level (which was the same for all tests) was
chosen to create plane stress conditions. Kol was selected such that the plastic
zone diameter, 2ry, was greater than the specimen thickness, that is

2ry = l IK~ ]2 > 2.54 mm (1)

~L Oysj
For each test, the specimen was cycled to establish steady-state growth,
overloaded into a state of plane stress and then cycled again at the pre-overload
level.
The principal tests in the test matrix are illustrated in Figs. 2 and 3 which
show the stress intensity levels for each test. Tests in addition to these were also
run and are included in Table 1. It can be seen in Figs. 2 and 3 that each row
on either plane isolates the Ruo variable. Each column isolates the effect of
either Qol or R m (also Kmax or Kmin) depending on the plane. It should be
pointed out that the middle row in each of the two figures represents the same
set of tests.

Experimental Procedure
The material used for this investigation was 2024-T3 aluminum alloy from the
same stock as that used by Himmelein and Hillberry [11]. The specimen
geometry was a center crack panel (558.8 by 152.4 by 2.54 mm thick) of the
same dimensions as used by Probst and Hillberry [7] and Himmelein and
Hillberry [11].

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
44 FATIGUE CRACK GROWTH
Stress
stty
-1.00 -0,50 0.01 0.11 0.22 0.30 HI~-312
I
36.61
22.88
1
1.6 8.05
! , ,

V v
0~374-03
~18.31
-36.61

' 36.61
20,34
1.8
8,05

/ V v , 0,37
4.03

-18.31
-36.61

9 36,61

16.64
2.2
8"054.03
/ / 0.37

V -18.31
36.61

RH - 0.22

FIG. 2-Qol versus Ruo plane of test matrix; R m constant.

Stress
luteustty
-1.00 -0.50 0.01 0.11 0.22 0.30 ~-3/2
36.6:
20, 3b,

0.11
4.03
0.37
-18.31
-36.61

36.61
20.34
0.22 8.05
4.03

V v 0.37
-18.31
-36.61

36.61
0.3C
M ~ ~v~M~ ^~A M~ 20.34
10.99
8.05
1 V V v, .4"030.37
-18.31
-36.61

FIG. 3-R m versus Ruo plane of test matrix; Qol constant.

The material had the following tensile properties; yield strength = 392
MN/m 2, ultimate strength = 476 MN/m 2 , and percent elongation = 14.2
percent. The test specimens were polished to a mirror finish in the vicinity of the

crack path to facilitate optical observation of the crack tip during crack-growth
measurement. The specimens were loaded parallel to the direction of rolling of
the material with a closed-loop, electrohydraulic test system. Lightweight
aluminum compression guides lined with 1.5 mm felt were used to support the
specimen when the compressive underloads were applied.
Since the scope of this study strictly involved the effects of loading on fatigue
crack propagation, care was taken to control as many other variables as possible.
All tests were subjected to nearly identical environmental conditions of
dessicated air and room temperature. All fatigue cycling was done at 20 Hz
except for Tests 1, 2, and 3, which were cycled at 10 Hz due to the rapid crack
propagation. All overload and undefload cycles were applied at 0.02 Hz. In order
to have a basis for comparison between tests in the test matrix, the stress
intensity factors were held at quasi-constant values throughout each test. This
was accomplished by changing the load required to produce a specified K for
every 5 percent increase in crack length. This ensured that the actual value of K
was within 3 percent of the desired value. Care was taken in each test to
establish equilibrium before the application of the overload/underload sequence.
The steady-state condition was achieved by propagating the fatigue crack at least
6 mm (over twice the diameter of the theoretical overload plastic zone of each
test) at the quasi-constant stress intensity values specified for the test. The
overload/undedoad sequence was then applied when the crack had propagated 5
percent past the point of the previous load shed. Cycling was continuous for
both pre-overload and post-overload fatigue cycling to negate the possibility of
time or unloading to zero load affecting subsequent crack growth.
The crack length was measured optically with a xl00 microscope mounted on
a two-directional traverse. The traverse had a resolution of 0.001 mm in the
horizontal direction and a direct digital display. The number of cycles was
recorded with a printer connected to a counter. Data were actually recorded by
advancing the optical system the specified increment (0.01, 0.02, or 0.05 mm
depending on growth rate). When the crack had grown this incremented
distance, the trigger button to the printer was pressed, recording the number of
cycles.

Data Reduction-Numerical Differentiation

One of the objectives of this study was to characterize the growth-rate

behavior through the overload affected zone following the overload/underload
sequence. To obtain this growth rate, da/dN, it was necessary to record
sufficient a versus N data and then differentiate these data. Several different
techniques have been used for the numerical differentiation of experimental
data. Incremental differentiation methods tend to amplify any small-scale
variation in the acutal data. As an alternative, the spline method was selected in
which a series of cubic polynomials were fit to the experimental data and then
analytically differentiated to obtain the crack-growth rate. This method,

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions author
TABLE 1 - Test results.
--4
ol d a I p o s t . ol
da pre.
da I] A~* max /~DXper. N~redict C
Test gmax Z~K "~ ss d"Nlss d'~ rain Kop 113
t'3
No. (MN/m -a/2 ) (MN/m-3/2) Rf Qol Ruo Rm (ram/cycle) (ram/cycle) (ram/cycle) (mm) (MN/m -3/2 ) (cycles) (cycles) ~0
O
7~
1 36.61 12.45 0.65 1.00 0.11 0.65 6.55 x 10 - 4 6 . 6 0 x 10 - 4 6.60 x 10 -4 0 27.34 0 650 G3
2 28.17 12.63 0.55 1.30 0.11 0.42 5 . 1 8 x 10 - 4 5 . 3 8 x 10 - 4 9.27 x 10 -s 1.55 22.60 6 800 3 470
O
3 22.88 12.63 0.45 1.60 0.11 0.28 3 . 5 3 x 10 - 4 3 . 5 8 x 10 - 4 9 . 6 3 x 10 - 6 2.01 19.16 22 000 21 7 4 0
-4
4 19.27 12.63 0.34 1.90 0.11 0.18 3 . 0 2 x 10 - 4 2 . 5 4 x 10 - 4 4.72 x 10 -6 3.51 16.70 121 0 0 0 247 800 -r
5 16.64 12.63 0.24 2.20 0.11 0.11 1.66 x 10 - 4 0 >16.64 oo oo
6 22.88 14.83 0.35 1.60 --1.00 0.22 1.03 x 10 - 3 9.02"x 10 -4 2.45 x 10 -4 2.0 89 --15.72 5 850 8 330
7 22.88 14.83 0.35 1.60 -0.50 0.22 6.30 x 10 -4 6 . 7 6 x 10 - 4 1.28 x 10 - 4 2.62 16.83 10 6 0 0 7 960
8 22.88 14.83 0.35 1.60 0.01 0.22 5.82 • 10 -4 4 . 5 2 x 10 - 4 7 . 4 4 x 10 - s 1.88 17.62 11 4 0 0 14 100
9 22.88 14.83 0.35 1.60 0.11 0.22 4.55 • 10 -4 5 . 0 8 x 10 - 4 5 . 3 8 x 10 - s 2.36 18.04 18 4 0 0 20 700
10 22.88 14.83 0.35 1.60 0.22 0.22 4 . 7 8 x 10 - 4 4 . 5 2 x 10 - 4 1.34 x 10 - s 2.36 19.51 44 000 40 200
11 20.34 12.29 0.40 1.80 --1.00 0.22 3.53 x 10-4 4 . 0 1 x 10 - 4 1.09 x 10 - 4 1.88 14.52 11 8 0 0 16 6 0 0
12 20.34 12.29 0.40 1.80 -0.50 0.22 4 . 9 3 x 10 - 4 3 . 5 8 x 10 - 4 1.02 x 10 - 4 2.36 14.63 14 400 18 2 0 0
13 20.34 12.29 0.40 1.80 0.01 0.22 3.15 x 1 0 - 4 2.54 x 10 -4 2.37 x 10 -s 3.23 16.43 42 600 50 200
14 20.34 12.29 0.40 1.80 0.11 0.22 2 . 9 5 x 10 - 4 2.54 • 10 -4 6.48 x 10 -6 2.59 17.55 53 000 98 900
15 20,34 12.29 0.40 1.80 0.22 0.22 2.84 x 10 -4 3.05 x 10 -4 2.72 x 10 -6 3.05 18.11 142 000 395 3 0 0
16 16.64 8.59 0.48 2.20 -1.00 0.22 1.21 x 10 - 4 1.27 x 10 - 4 2 . 6 7 x 10 - s 2.79 12.61 52 000 43 400
17 16.64 8.59 0.48 2.20 -0.50 0.22 1.66 x 1 0 - 4 2 . 0 2 x 10 - 4 4 . 8 8 x 10 - 6 4.32 14.05 127 000 79 4 0 0
18 16.64 8.59 0.48 2.20 0.01 0.22 1.24 x 1 0 - 4 9. . 0 . . . >16.64 =o oo
19 16.64 8.59 0.48 2.20 0.11 0.22 1.51 x 1 0 - 4 . . . 0 . . . >16.64 oo oo
20 16.64 8.59 0.48 2.20 0.22 0.22 1.22 • 10 - 4 0 ~16.64 oo oo
21 20.34 16.31 0.20 1.80 -1.00 0.11 5 . 5 4 x 10 - 4 4.52"x i0 -4 1.36 • 10 - 4 ;.98 --14.19 I0 300 13 8 0 0
22 20,34 16.31 0.20 1.80 --0.05 0.11 7.87 x 1 0 - 4 3.58 x 10 -4 1 . 1 2 x 10 - 4 3.18 14.50 17 6 0 0 15 8 0 0
23 20.34 16,31 0.20 1.80 0.01 0.11 5.89 x 10 -4 4 . 6 2 x 10 - 4 8 . 3 6 x 10 - 6 1.73 17.36 44 500 43 700
24 20.34 16.31 0.20 1.80 0.11 0.I1 5.13 x 10 -4 5 . 0 8 x 10 - 4 8.99 x 10 -6 3.48 17.30 63 700 84 600

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
30 20.34 9.36 0.54 1.80 --1.00 0.30 1.93 x 10 - 4 1.84 • 10 - 4 2.69 x 10 - s 2.36 16.30 25 0 0 0 19 2 0 0
31 20.34 9.36 0.54 1.80 --0.50 0.30 2.44 x 10 - 4 2.02 x 10 - a 6.07 • 10 - s 1.65 15.36 17 6 0 0 20 3 0 0
32 20.34 9.36 0.54 1.80 0.01 0.30 1,69 x 10 - 4 1,60 • 10 - 4 1.50 • 10 - 5 1.50 16.87 52 800 55 700
33 20.34 9.36 0.54 1.80 0.11 0.30 1.60 • 10 - 4 1.61 x 10 - a 7,14 • 10 - 6 3.00 17.47 83 200 111 5 0 0
34 20.34 9.36 0.54 1.80 0.22 0.30 1.61 • 10 - 4 1.80 x 10 - 4 2.62 x 10 - 6 3.12 18.14 197 000 469 800
35 20.34 9.36 0.54 1:80 0.30 0.30 1,80 x 10 - a . . , 0 . . . ~>20.34 oo oo
16-2.4 15.26 7.21 12.31 124 0 0 0 54 500 3>
g.-
0.53 2.40 --I.00 0.22 9.35 x 10 - s 7.16 x 10 - s 7.98 x 10 - 6 2.87
16-2.6 14.08 6.03 0.57 2.60 --1.00 0.22 3.86 x 10 - s 6.68 x 10 - s 3.23 x 10 - 6 2.92 l'q
11,74 260 0 0 0 58 0 0 0
17-2.4 15.26 7.21 O
0.53 2.40 ---0.50 0.22 9.07 x 10 - s 5.44 x 10 - s 1.41 x 10 - s 3.48 11.83 126 000 143 0 0 0 co
ill
3A/D-.4a 12,20 8.55 0.30 3.00 ~0.40 0.10 1.31 x 10 - 4 . . . 0 0 ~>12.20 oo 387 0 0 0 --t
3A/D-.4b 12.20 8.55 0.30 3.00 -0.40 0.10 1.02 x 10 - 4 0 0 ~12.20 oo 387 0 0 0
3A/D-.5 12,20 8.55 0.30 3,00 -0.50 0.10 1.25 x 10 - 4 7.16'x i0 -s 2.67 x 10 - 6 3.81 9.99 593 0 0 0 204 700 f-.
O
R-I 2 20,34 12.29 0.40 1.80 --0.50 0.22 1.02 x 10 - 3 2.54 x 10 - 4 1.96 x 10 - 6 2.26 . .. 160 5 0 0 . . . z
R-13 20,34 12.29 0,40 1.80 0,01 0.22 2.67 x 10 - 4 2.37 x 10 - 4 3.86 x 10 - 6 2.54 . .. 169000 ... co
Z
i'-
N O T E - K o l = 36,61 M N / m - 3 / 2 for all tests. O v e r l o a d p r e c e e d e d u n d e r l o a d e x c e p t in Tests R-12 a n d R-13. m
O
a T e s t s 25 to 29 are for the same set o f tests as Tests 11 to 15. <
m
:37
f-,
O

C
Z
I1-1
:13
i--
0

0
-(
0
r-
m

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
48 FATIGUECRACK GROWTH
frequently used to differentiate experimental data [24,25], provides a smoothed
crack-growth rate curve.
To apply the spline method, third order polynominals were fit to three
intervals of the entire post-overload a versus N data of a given test. The values of
the polynomials and their first two derivatives were matched at the knots (points
where the polynomials join). The computer technique of de Boor and Rice
[26,27] was modified to allow the first interior knot to be fixed at a specified
position. This first knot was located after the initial deceleration of the crack
where the a versus N curve started to level out. It was found for all tests that this
knot could be located at a point equal to 5 percent of the calculated plastic zone
size. The computer routine then optimally located the second knot and
determined the coefficients of the three polynomials which minimized the
least-square error over the entire range of the data.
Figure 4 shows a typical fit of the spline function to a versus N data, and Fig.
5 shows the corresponding analytically differentiated curve, that is, da/dN versus
a. The spline method of differentiation was compared to a linear, seven-point
least-squares movable-strip method [2,28]. The resulting growth rate from this
method is also shown in Fig. 5 for the same data. To compare the two methods,
the da/dN from the movable-strip method was numerically integrated and
compared with the original data. This is shown in Fig. 4. With the spline method,
int.egration of da/dN will simply reproduce the original spline function since the
derivation is obtained analytically. As can be seen from Figs. 4 and 5, the spline

ta 4

mnvable
2 strip " ~

0 ! I I I J I t
0 60,000 120,0O0 180,000
N
F I G . 4-Spline function and integrated da/dN from the movable strip fit o f a versus N data
for Test 15.

~176
,~ lo "4
|
Z movable

O
'V
10-5

10-6

0 2 4 6
0 (ram)
FIG. 5-Comparison of da/dN obtained from spline function and from the movable strip
method from a versus N data for Test 15.

method provides a good representation of the a versus N data and a smoothed

da/dN curve. An analytical expression is also obtained for an entire data set.
One of the important characteristics of the delay behavior is the minimum
growth rate following the overload. Therefore, the minimum growth rate for
each test as determined from the spline method was compared with the data
graphically. In some of the tests the spline method produced a slightly lower
value for the minimum growth rate. For these tests (10, 17, 23, 32, 33, and 34)
the graphical values were used.

Test Results

Constant-Amplitude Test Results

A total of ten constant stress amplitude tests were run with five stress ratios,
Rf, ranging from 0.01 to 0.55. By keeping the load amplitude constant in these
tests, Rf remained constant, but the stress intensity level and also the AK level
increased with increasing crack length. Data was collected every 0.01 mm of
growth. Spline functions were fit to the data from which da/dN values were
determined for each of the tests. Because of the density of the data, the ten tests
provided over 1700 data points for da/dN. The resulting growth rate data were

fit to Elber's form of the growth rate equation using a least-squares method
which gave

da = C(AKeff) n ram/cycle

6"4 = 1 . 2 2 x 1 0 -7

n = 3.86

~eff = ~ ( 0 . 5 + 0.4Rf)

Overload/Underload Test Results

For each of the tests, the following experimental variables were determined:
(1) the number of delay cycles, No; (2) the overload affected crack length
increment, ~a*; (3) the minimum growth rate following the overload/underload
sequence, (da/dN)] rain; (4) pre and post-overload crack-growth rates; and (5)
the coefficients for the spline function describing the a versus N data. N D is the
number of cycles following the overload before the crack growth returns to a
steady-state rate, and Aa* is the interval over which N o is observed. Both
quantities were determined graphically from plots of the data. The results are
presented in Table 1. Tables 2 and 3 give the values of (da/dN) I rain and N o in
the matrix form as functions of the test variables Qoi, Ruo, and R m . These
results are for the corresponding test conditions shown in Figs. 2 and 3.
Examination of the results of Table 2 show an increase in N o and a
corresponding decrease in (da/dN) l rain as either Ruo or Qol is increased with
R m held constant. A trend is less pronounced for R m as columns of Table 3
reveal. These results indicate that R m has less influence on the delay
phenomenon than the two parameters Ruo and Qop
The calculated value of the load interaction zone, Zol = 2ry, from Eq 1 was
2.77 mm and was the same for all of the tests since Koj was the same. This can
be compared to the experimental results, Aa*, given in Table 1. The average
value for ~ka* was 2.61 mm.
Tests 1 to 5 are in a plane of constant Ruo and illustrate the effects of
changing only the mean level of the fatigue cycling, that is, Rf. Tests R-12 and
R-13 were performed to determine the effect of reversing the load sequence by
applying the underload before the overload. These results, when compared with
the same test conditions having no underload (Test 15) indicate that any effect
of the underload is essentially eliminated by the subsequent overload. Tests
16-2.4 to 3A/D-.5 in Table 1 were run with larger values of Qol and negative
values of Ruo to more accurately establish trends in that region of the Qol versus
Ruo plane.
Observation during data collection and examination of the shape of the
growth rate curve following the overload indicated that the minimum growth
rates were directly related to the total number of delay cycles. The correlation

4 Here, C is defined as a constant and differs from that defined in the Glossary.

~>
Ruo F-
N
0
Lgol -1.00 ~.50 0.01 0.11 0.30 0.30 m
--4
~>
r-
O
atd3.~t 2.45 x 10 - 4 1.28 x 10 - 4 7 . 4 4 x 10 - s 5 . 3 8 x 10 - 5 1.34 x 10 - s z
I rnin
1.6 Z
ND 5 850 10 6 0 0 11 4 0 0 18 4 0 0 44 000 I"
m
O
<
m
d.~a I 1.09 x 10 - 4 1.02 x 10 - 4 2.37 x 10 - s 6 . 4 8 x 10 - 6 2 . 7 2 x 10 - 6
c/Jr I rain r--
1.8 0

ND 11 8 0 0 14 4 0 0 42 600 53 0 0 0 142 0 0 0 O

Z
0
m
d..~a [ 2 . 6 7 • 10 - 5 4 . 8 8 x 10 - 6 0 0 0
dN Irain r-"
2.2 0
ND 52 000 127 0 0 0 co oo oo
.<
t~
r
m

O'1

"I'1
.-I
T A B L E 3 - Minimum growth rate and number of delay cycles for Rm/Ruo plane of test matrix. c
I1"1
t")
Ruo
O
7~
Rm --1.0 ~).5 0.01 0.11 0.22 0.30 ~D
0
---I
"I"
da { 1.36 x 10 - 4 1.12 x 1 0 - 4 8.36 x 10 - 6 8 . 9 9 x 10 - 6
d'N I rain
0.11
ND 10 3 0 0 17 6 0 0 44 500 63 700

dd~ rain 1.09 x 10 - 4 1.02 x 1 0 - 4 2.37 x 10 - s 6 . 4 8 x 10 - 6 2 . 7 2 x 10 - 6

0.22
ND 11 8 0 0 14 4 0 0 42 600 53 000 142 0 0 0

da 2.69 x 10 - s 6.07 x 10 -5 1.50 x 10 - 5 7.14 • 10 - 6 2 . 6 2 x 10 - 6 0

d'N rain
0.30
ND 25 0 0 0 17 6 0 0 52 800 83 2 0 0 197 0 0 0 oo

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
ALZOS ET AL ON SINGLE OVERLOAD/UNDERLOADCYCLES 53
of the minimum growth rate with the delay cycles is illustrated in Figure 6.
Therefore, if (da/dN)]min is known, N D can be determined. It should be
recalled, however, that Ko~ was the same for all of the tests in this investigation.
This correlation should be investigated for other levels of Kol and also other
metals.

10-3
\

\k \\N~\ \

i10"4 \\ FACTOR OF 2

Z ~0 \
"2\ e\

I0 -6 I I I t . t , ,,~
iooo Io,ooo ioo,o0o I,ooo,ooo
No
FIG. 6-Minimum growth rate following overload/underload cycle versus the number of
delay cycles.

Comparison with CrackClosure

Elber's crack closure concept has been used to qualitatively explain the delay
behavior following an overload condition. However, by knowing the crack-
growth rate through the overload affected zone it is possible to provide a more
quantitative comparison of crack closure with the delay behavior. For crack
closure [20], the effective stress intensity range is

AKeff = gma x - Kop (3)

where Kop is the stress intensity at which the crack opens. At the minimum
growth rate following the overload, Kop is a maximum and Eq 2 becomes

da 1 C(Kmax max n
d N rain = -g~ ) (4)

In Eq 4, C and n were determined from the constant-amplitude test results,

Kmax from the test conditions and (da/dN)l rain from the test results.
Therefore, this equation could be solved for K~opax following the overload [29].

These results are given in Table 1. (This is not restricted to the point where
da/dN is a minimum since Kop can be determined throughout the overload
affected zone [23] .) For constant-amplitude loading, Elber defined

U = Kmax - K~ (5)
Kmax - Kmin
For 2024-T3 aluminum alloy, he found
U = 0.5 + 0.4Rf (6)

At the minimum growth rate following an overload/underload cycle, Uol can be

defined, analogous to Eq 5,

Uol _- Kol - Kmax

--op (7)
Kol - Kul
Since K max can be determined from the experimental results as just described,
--op
Uol can be determined. It is expected that Uol would be related to the loading
conditions as well as environmental and material factors, that is

(Jot = f(Qot, Ruo, Rm, R f, Kol . . . . . ) (8)

If crack closure, as presented for constant-amplitude loading, describes the
results, then this function would be, analogous to Eq 6,

Uol = 0.5 + 0.4 Ruo (9)

Examination of the data, however, suggested that this equation should be of the
form

Uo~ = C, + C2Ruo + C3R,,o ~ + C,,Qol (10)

Using the results from the tests that did not arrest, a multiple linear regression
analysis was performed to determine the Ci constants, From the regression
analysis,

Cl = 0.408
6'2 = 0.367
(73 =0.117
c,.= 0.075 (11)

By combining Eqs 7 and 10 and solving for Kmpax gives

Ko%ax =Xol [1 -(1 - R u o ) (C, + C2Ruo + C3Ruo 2 + C4Qol)] (12)
The values for Komp ax calculated from Eq 12 are compared with the
experimentally determined values (obtained using (da/dN) ] min and Eq 4) in
Fig. 7. The calculated maximum opening stress intensity value correlates almost
exclusively within 5 percent of the measured value.

20 /

~0.16
1,r0

12
,/

/o
I0
I0 12 14 16 18 20
EXPERIMENTAL ~pl~ (MNm.5/2)

FIG. 7-Comparison of the predicted and experimentally determined values of the

maximum opening stress intensity.

With CsR m added to Eq 10, the regression analysis showed that R m was
statistically insignificant.
The effects of the test parameters on the minimum growth rate can be more
readily illustrated by substituting Kompax from Eq 12 into Eq 4 and plotting the
results for constant values of (da/dN) I rain Figure 8 shows several level curves
"

of constant (da/d.N) lmi n as functions of Qol and Ruo. In this study, if the
growth was less than 0.01 mm in 106 cycles, the test was stopped and the crack
was assumed to have arrested. Therefore, a minimum growth rate of 10 - s
mm/cycle which is the experimental resolution for the data, can be considered
arrest. The curve in Fig. 8 with this growth rate is then an arrest/delay boundary.
It should be pointed out that this arrest/delay boundary is based entirely on
finite growth test results, however, it compares favorably with the experimen-
tally determined arrest/delay boundary found by Himmelein and Hillberry [11 ]
which was for an overload with no underload, that is, Ru~, = Rr~ [23]. The
arrest~delay boundary defined by Elber's crack closure equation (Eqs 5 and 6) is
also shown in Figure 8. In evaluating the validity of the correlation of the results
of this study with crack closure, it is important to keep in mind the number of
tests and the distribution of the loading conditions of these tests. Of the 24 tests
used in determining the coefficients of Eq 10, only three tests had Qol > 2.2.
Furthermore, eleven of the tests were performed in the small region of 0 < Ruo

~t 3.2' 'Qo-

~ 2.8

5 x 10-5 ~ . . . . . ~

j J i.6
IxlO " 4 ~ "~-

1.2 """-..
1I I I ~..
-I.0 -0.5 0.0 0.5 Ruo
FIG. 8-Curves of constant da/dN [ rain for R m = 0.22 and Kol = 36,61 MN/m 3/2.

< 0.3 and 1.6 <_ Qot <- 1.8. Even with this very unequal distribution of the data,
the correlation is exceptionally good.
In Figure 6, the correlation of the minimum growth rate and the number of
delay cycles was shown. This, along with Eq 4, suggests that an average growth
rate 2 r y / N D would correlate with Kmax - --opKmax, that is,

2re =A(Kmax-Kmax)
--up ,
b (13)

The results showed a straight line on a log-log plot and Constants A and b were
then determined from a least-squares fit. Using the resulting equation, the
number of delay cycles were calculated. These predicted number of delay cycles
are compared with the experimental results in Table 1. Comparison with the
experimental results show the prediction to be nearly within a factor of two for
all tests. It should be mentioned, however, that for these tests 2ry was the same.

Conclusions

1. Utilizing a three-dimensional test matrix approach, the effects of each of

the selected loading parameters could be isolated and studied.
2. The number of delay cycles correlates directly with the minimum growth
rate following the overload/underload sequence.
3. The extension of the crack closure theory to include the effects of the
underload accurately predicts values of -/(max
-up which may be used to calculate
the minimum growth rate following the overload/underload cycle and the
corresponding numbers of delay cycles.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
ALZOS ET AL ON SINGLE OVERLOAD/UNDERLOAD CYCLES 57

4. The method of cubic splines may be used to represent the a versus N data
and for determining da/dN.

Acknowledgments

The authors wish to express their appreciation to the Air Force Flight
Dynamics Laboratory, Wright-Patterson Air Force Base, for their support, under
Contract F33615-73-C-3056, of this investigation and to the Aluminum
Company of America for providing the test samples.

References
[1] Schijve,J., "The Prediction of Fatigue Crack Propagation Under Service Load-Time
Histories," National Aerospace Laboratory, NLR, The Netherlands, NLR MP
73016U, 1973.
[2] Gallagher, J. P. and Stainacker, H. D., "Methods for Analyzing Fatigue Crack
Growth Rate Behavior Associated with Flight by Flight Loading," AIAA Paper No.
74-367, April 1974.
[3] Von Euw, E. F. J., "Effect of Overload Cycle(s) on Subsequent Fatigue Crack
Propagation in 2024-T3 Aluminum Alloy," Ph.D. thesis, Lehigh University, 1971.
[4] Jonas, O. and Wei, R. P., International Journal of Fracture Mechanics, Vol. 7, 1971,
pp. 116-118.
[5] Wei, R. P. and Shih, T. T., International Journal of Fracture Mechanics, Vol. 10,
1974, pp. 77-85.
[6] Wei,R. P., Shih, T. T., and Fitzgerald, J. H., "Load Interaction Effects on Fatigue
Crack Growth in Ti-6AI-4VTitanium Alloy," NASA CR2239, 1973.
[ 7] Probst, E. P. and Hillberry, B. M., AIAA Journal, Vol. 12, 1974, pp. 330-335.
[8J Gardner, F. H. and Stephens, R. I. in Fracture Toughness and Slow-Stable Cracking,
ASTM STP 559, American Society for Testing and Materials, 1973, pp. 225-244.
[9] Trebules, V. W., Roberts, R., and Hertzberg, R. W. in Progress in Flaw Growth and
Fracture Toughness Testing, ASTM STP 536, 1973, pp. 115-146.
[10] Corbly, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973,
pp. 479--497.
[11] Himmelein, M. K. and Hillberry, B. M., "The Effect of Stress Ratio and Overload
Ratio on Fatigue Crack Delay and Arrest Behavior Due to Single Peak Overloads,"
Proceedings of the Eighth National Symposium on Fracture Mechanics (1974),
Mechanics of Crack Growth, A STM STP 590, to be published.
[12] Hsu, T. M. and Lassiter, L. W., "Effects of CompressiveOverloads on Fatigue Crack
Growth," AIAA Paper No. 74-365, April 1974.
[13] Petrak, G. J. and Gallagher, J. P., "Predictions of the Effect of Yield Strength on
Fatigue Crack Growth Retardation in HP-9Ni-4Co-3OC Steel," accepted for
publication in Journal of Engineering Materials and Technology, ASME.
[14] Stephens, R. 1., McBurney, G. W., and Oliphant, L. J., "Fatigue Crack Growth with
Negative R Ratio Following Tensile Overloads," International Journal of Fracture,
Vol. 10, 1974.
[15] Porter, T. R., Engineering Fracture Mechanics, Vol. 4, 1972, pp. 717-736.
[16] Wheeler,O. E., Journal of Basic Engineering, Vol. 94, 1972, pp. 181-186.
[17] Willenborg,J., Engle, R. M., and Wood, H. A,, "A Crack Growth Retardation Model
Usingan Effective Stress Concept," AFFOL-TM 71-1-FBR, 1971.
[18] Gray, T. D. and Gallagher, J. P., "Predicting Fatigue Crack Retardation Following a
Single Overload Using a Modified Wheeler Model," Proceedings of the Eighth
National Symposium on Fracture Mechanics (1974), Mechanics of Crack Growth,
ASTM STP 590, to be published.
[19] Gallagher, J. P., "A Generalized Development of Yield Zone Models," AFFDL-TM-
FBR-74-28, Air Force Flight Dynamics Laboratory, Wright-Patterson Air Force
Base, Ohio, Jan. 1974.
[20] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM, STP 486 American
Society for Testing and Materials, 1971, pp. 230-242.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
58 FATIGUE CRACK GROWTH

[21 ] Newman, J. C., "Finite Element Analysis of Fatigue Crack Propagation-Including

the Effects of Crack Closure," Ph.D. thesis, Virginia Polytechnic Institute and State
University, 1974.
[22] Sharpe, W. M. and Grandt, A. F., "A Preliminary Study of Fatigue Crack
Retardation Using Laser lnterferometry to Measure Crack Surface Displacements,"
proceedings of the Eighth National Symposium on Fracture Mechanics (1974),
Mechanics of Crack Growth, ASTM STP 590, to be published.
[23] Alzos, W. X., "The Effects of Single Overload/Underload Cycles on Fatigue Crack
Propagation," M.S. thesis, Purdue University, May 1975.
[24] Wold, S., "Spline Functions in Data Analysis," Research Group for Chemometrics,
Institute of Chemistry, Umea University, S 901 87, Umea, Sweden.
[25] Ahlberg, J. H., Nilson, N. H., and Walsh, J. L., The Theory of Splines and Their
Applications, Academic Press, New York, 1967.
[26] de Boor, C. and Rice, J. R., "Least Squares Cubic Spline Approximation: I-Fixed
Knots," Purdue University Computing Center Publications, 1968.
[27] de Boor, C. and Rice, J. R., "Least Squares Cubic Spline Approximation:
lI-Variable Knots," Purdue University Computing Center Publication, 1968.
[28] Clark, W. G., Jr., and Hudak, S. J., "Variabihty in Fatigue Crack Growth Rate
Testing," ASTM E-24.04.01 Task Group Report, 18 Sept. 1974.
[29] James, L. A and Anderson W. E., Engineering Fracture Mechanics, Vol. 1, 1963,
pp. 565--568.

DISCUSSION
R. P. Wei 1 and T. T. Shih ~ (written discussion)-The authors are to be
complimented for introducing factorial design into their experiments on load
interaction effects in fatigue crack growth. Unfortunately, they did not follow
this approach to its logical conclusion. The authors' statement regarding
prediction and its implications in the context of Fig. 7 and Eqs 4 to 12 appears
to us to be inappropriate and misleading. One gets the mistaken impression that
Fig. 7 represents a comparison between the predictions of an analytically derived
model and experimental results. In actuality, the authors had simply assumed
the validity and applicability of a "crack closure" model to account for load
interaction effects (see Eq 4), and had made an ad hoc assumption regarding the
functional form of the load interaction parameter (defined to be Uol; Eqs 7 and
10). Incidentally, the authors could have dealt with --op K"max directly without
having to invoke Uol ). The authors then used a multiple linear regression analysis
to determine the unknown coefficients Ci in the assumed equation (Eq 10 or 12)
so as to obtain a least-error fit to the values of K~opax computed from their
experimental data, using Eq 4. As such, Fig. 7 depicts only the closeness of this
fit, and does not represent a comparison between prediction and experimenta-
tion, and a test of the model. Although there is merit in the authors' approach,
one needs to remain objective in its implementation.
B. M. Hillberry (author's closure)-I agree that the word prediction in Figure 7

i Department of Mechanical Engineering and Mechanics, Lehigh University, Bethlehem,

Pa. 18015.
2 Ebasco Services, Inc., New York, N. Y.

is possibly misleading and I wish to thank Drs. Wei and Shih for pointing this
out.
In response to the remaining comments I find it particularly interesting that
they were able to so clearly describe what we actually did in analyzing our data.
Unfortunately their conclusions did not arrive at the logical approach. In
actuality the authors did not assume the validity or applicability of any model in
this study. The results were however compared with several models to see if the
model described the observed behavior. There was no apparent correlation with
either the Wheeler model [16] or the Willenborg model [17]. In comparing the
crack closure model with the results, the only assumption made was that the
overload/undefload cycle establishes the residual stress field in the vicinity of the
crack tip and is not affected by subsequent cycling. This in turn establishes the
crack opening stress intensity level throughout the overload affected region. The
functional form for U given in Eq 5 was defined by Elber and for the
overload/underload cycle is, by definition, Uol as given in Eq 7. As stated in the
text the functional form for (Jot given in Eq 10 was predicted by the data, and
no ad hoc assumption as to the form of Eq 10 was made as Drs. Wei and Shih
suggest. Plotting Uol versus Qol showed a linear relationship between Uol and
Qol for Ruo constant. This accounts for the Qo~ term of Eq 10. Plotting Uol
versus Ruo showed a definite second order relationship between Uol and Ruo
for Qol constant. This accounts for the quadratic form in Ruo contained in Eq
10. By finding the slope, m, of the Uol versus Qol curve, the data collapsed to
nearly a single line on a plot of (/Jot - mQol versus Ruo ). From this, it was
clearly evident that Uol was related to Ruo and Qol as given in Eq 10 and the
crack closure did in fact describe the behavior due to the overload/underload
cycle. Possibly, we should have quit at this point to avoid the possibility of
confusing our readers. However, since the functional relation between Uo~ and
Qol and Ruo was known (not assumed), the constants were determined using a
regression analysis. This in turn provides an equation for calculating Uoi which
can be used to calculate Ko~ax . This along with Eq 13 can then be used to
predict .delay behavior as is done in Fig. 8. It is interesting to note that in
comparing (Jot from Eqs 9 and 10, there is very close agreement for 0 _<Ruo <
0.5 which is the region over which Eq 9 would be expected to be applicable.
In the recent studies of delay behavior due to overloads, there has been very
little success in correlation of the results. Several investigators have qualitatively
described their results using the crack closure concept. The results presented
here, however, provide a quantitative correlation (within 5 percent, Fig. 7). It is
also interesting to note that in subsequent studies Eqs 10 and 13 have provided
excellent prediction of delay behavior. 3 ,4

Skat, A. C., Jr., "Evaluation of Extended Crack Closure in Fatigue Crack Delay
Prediction for Single Overload/Underload Sequences, M. S. thesis, Purdue University, May
1975.
4 Crandall, G. M., "Residual Stress Intensity Parameters for Predicting Delay in Fatigue
Crack Propagation," M. S. thesis, Purdue University,Dec. 1975.

As I have just described, our objectives and procedures for analyzing the data
were not the same as that suggested by Drs. Wei and Shih. The procedure they
suggest is an elementary approach which can be very misleading. In this study,
the regression analysis was used only to find the magnitude of the constants for
an equation whose functional form was predicted by the data. As they point out
one should remain objective in their approach.

Effects of Rest Time on Fatigue Crack

Retardation and Observations of
Crack Closure

REFERENCE: Sharpe, W. N., Jr., Corbly, D. M., and Grandt, A. F., Jr., "Effects of
Rest Time on Fatigue Crack Retardation and Observations of Crack Closure,"
Fatigue Crack Growth Under Spectrum Loads, ASTM STP 595, American Society for
Testing and Materials, 1976, pp. 61-77.

ABSTRACT: Fatigue cracks grown in compact tension specimens of 2024-T851

aluminum at AK ,~ 6.93 MN/m 3/2 (R = 0.1) were subjected to single-peak overloads
of 17.31 MN/m 3/" . After the overload, the specimens were held at zero load for
periods of 3 min, 1 h, and 20 or more hours before recycling at AK = 6.93
MN/m 3/2 . Increasing the rest time had the effect of slightly reducing the crack
retardation as measured on the specimen surface.
The load at which the fatigue crack faces fully separate was measured on the
specimen surface by laser interferometric techniques and in the specimen interior by
through-transmission ultrasonic methods. Crack opening loads measured on the
specimen surface were found to increase with application of an overload and then
decrease to the original value when the specimen was allowed to rest at zero load,
while opening loads measured through the sample by the ultrasonic method did not
vary significantly with peak loads or rest times. Varying specimen thickness between
0.64 and 2.54 cm had little effect on surface measured retardation or opening loads.
The relationship between applied load and crack surface displacement as measured by
ultrasonics varied significantly with specimen thickness.

KEY WORDS: crack propagation, fatigue (materials), loads (forces), stress cycle,
retarding, ultrasonics

It is well known that tensile peak overloads may significantly delay subsequent
constant-amplitude fatigue crack propagation in many materials. Since real
structures are usually subjected to complex load histories, the ability to predict
accurate crack-growth lives under realistic service conditions is of major
engineering interest. Thus, several mechanisms have been proposed to account
for load history effects on crack growth, including crack closure [I] ,3 effective
stress concepts [2-4], and crack tip blunting [5].
Although these models differ in their explanation for fatigue crack retarda-
tion, all share the view that crack tip plasticity is the controlling parameter. The

] Professor, College of Engineering, Michigan State University, East Lansing, Mich. 48842.
Research metallurgist and materials research engineer, respectively, Metals Behavior
Branch, Metals & Ceramics Division, Air Force Materials Laboratory, Wright-Patterson AFB,
Ohio 45433.
3 The italic numbers in brackets refer to the list of references appended to this paper.
61
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
62 FATIGUE CRACK GROWTH

differences between mechanisms arise in how the plasticity affects crack growth.
They agree that any variable which would alter the amount of crack tip yielding
would be expected to have an influence on subsequent crack extension. Factors
which have been considered by previous investigators include the size and sign of
the overload, the number and sequence of overloads, the mean stress level, the
influence of material properties, and the effects of temperature.
This paper examines two other variables which could influence fatigue crack
retardation: rest periods and specimen thickness. Although residual stresses
could be expected to relax out with time, the authors are aware of only limited
data on the effect of hold times [6]. Thus, a series of experiments are described
in which a tensile overload was applied and the specimen allowed to rest at zero
load prior to resuming the original cyclic loading. Since the size of the plastic
zone is larger for conditions of plane stress than for plane strain, these tests were
repeated for several specimen thicknesses.
An optical interference method was employed to measure the opening load
required to separate the crack faces in order to interpret the results in terms of
the closure model. As discussed in previous work [7,8], this technique provides
an accurate and sensitive magnification of the crack opening displacements at
the specimen surface. In an attempt to "look" into the interior of the aluminum
specimens to determine if internal crack behavior is similar to that on the
specimen surface, corresponding measurements were made by ultrasonic
transmission. These observations tend to support other evidence [9,10] that
surface measurements are not necessarily indicative of the internal crack
behavior.

Retardation Measurements
Procedure
Compact tension specimens (with specimen width W = 5.08 cm and
height-to-width ratio of 0.6) were machined from 2024-T851 aluminum in
thicknesses of 2.54, 1.27, and 0.64 cm. Fatigue cracks approximately 3-ram long
were grown from machined starter notches (a/W = 0.25) in these specimens by
cyclic loading at large values of ~ = Kmax -Kmi n , with a gradual reduction to
Kmax = 7.7MN/m 3/2 andR = Kmin / g m a x = 0.1. These values ofKma x and R
were chosen for baseline conditions because the crack growth rate was fast
enough for reasonable test times and AK was small enough to allow overloads
that did not cause fracture. All subsequent testing was done with crack lengths
in the range 0.3 < a/W < 0.7 where a is the total crack length. By shedding load
in predetermined increments as the crack extended, it was possible to keep the
baseline AK constant to within +5 percent.
All tests were performed in a closed-loop electrohydraulic testing machine
with the baseline fatigue cycling run at 40 Hz. The overload was a load pulse
that rose linearly to a maximum value of 17.31 MN/m 3/2 in 0.1 s, stayed at that
value for 60 s, and unloaded linearly in 0.1 s. This overload magnitude was
chosen because it gave appreciable, but not excessive, retardation (on the order
of 30 000 cycles). The particular pulse shape was chosen because it was
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
SHARPE ET A L ON EFFECTS OF REST TIME 63

controllable and reproducible on the test machine. Following the overload, the
specimen was allowed to rest at zero load in the test machine for the specified
period of time.
Crack lengths were measured during the fatigue testing with the aid of a small
plastic scale taped to the specimen surface parallel to the crack. By examining
the crack tip and scale through a microscope, it was possible to resolve 0,051
mm of crack extension. For each rest-time data point, the crack was grown
approximately 3 mm under baseline conditions, overloaded, allowed to rest at
zero load, and then subjected to baseline loading conditions until it started to
grow again. The number of cycles of baseline loading required to restart the
crack growth is the number of retardation cycles, N D .
Wei et al [6] noted that interaction effects due to sequential applications of
single peak overloads were negligible when the crack had grown approximately
ten plastic zone sizes between overloads. The maximum plastic zone size for
experiments conducted here (using the relationship, r y = ~1 (ghmax/Oys):2, with
ghmax = 17.31 MN/m 3/2 and oy s = 448 MN/m ~) is 0.238 mm. Therefore,
interaction effects should be negligible for these experiments. Moreover, care
was taken to ensure that steady-state growth rates were reestablished well before
the application of a subsequent overload. In addition, in order to avoid any bias
in the results due to possible load/rest-time interaction effects, the test sequence
was randomly mixed. The results in Tables ] and 2 are listed in order of
increasing rest times solely for the purpose of clarifying data presentation.

Results and Discussion

Figure 1 is a plot of a representative crack-growth curve. The crack was grown
from Points A to B under the baseline loading conditions (Krnax = 7.7
MN/m 3/: and R = 0.1). At Point B, the 17.31 MN/m 3/2 overload was applied

0.20
E

0.15
/ 4
u) :i
,,e =E
U o/* i
_z
o
0.10 Z
141
z oc .J
kl
.J o
2 = =(.)
o
o o
n,, U
o 0.05 0
0
0

0 I I I I I !
0 20 40 60 80 I00 120 140
KILOCYCLES

FIG. l-Typical crack-growth curve. Overload o f 2.25 Kma x was applied at Point B, rest
time was 3 rain.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
64 FATIGUE CRACK GROWTH

TABLE 1 - Summary of retardation data and effective stress intensity ratio (U)
measurements by optical techniques.

ND b ND b U U
Specimen Test Conditiona (direct) (slope) (IDG) (P--6)

1
(2.54 cm thick) baseline . . . . . . 0.83 0.78
3 rain 0.72 0.78
growth 36"I~ 321~ 0.78 0.81
2
(2.54 cm thick) baseline . . . . . . 0.81 0.75
3 min 0.61 0.67
growth 4"8"~[ 413 i( 0.78 0.72
1h 0.67 0.72
growth 3"2"K i6" k 0.72 0.75
70 h 0.78 0.75
growth 3"2"k ~6 k 0.83 0.78
3
(2.54 cm thick) baseline . . . . . . 0.83 0.78
3 rain 0.69 0.81
growth 3"2"K i6 k 0.83 0.83
1h 0.78 0.86
growth 4"4"K 35" k 0.72 0.83
4
(2.54 cm thick) baseline . . . . . . 0.86 0.78
3 rain 0.72 0.81
growth 3"8"~ ~ k o.81 o.81
1h 0.75 0.81
growth 28"I~ 21" K 0.83 0.86
66 h O.83 0.81
growth I'4'K 12 K 0.83 0.89
5
(1.27 cm thick) baseline . . . . . . 0.86 0.86
3 rain 0.75 0.83
growth i8" K 2i K 0.86 0.86
1h 0.78 0.83
growth i6"k 29"k 0.83 0.86
24 h 0.83 0.83
growth 2"8"K 26"k 0.89 083
6
(0.64 cm thick) baseline . . . . . . 0.83 0.86
3 rain 0.75 0.78
growth 3"ik ~6 k 0.81 0.86
1h 0.78 0.78
growth 3"4"I~ i7" i( 0.83 0.86
20 h 0.83 0.86
growth i4"k ii k 0.83 0.86

a The time given in this column is the time between application of overload and
commencement of cycling at the original baseline conditions.
b In cycles.

TABLE 2 - Summary of retardation data and effective stress intensity ratio (U)
measurements by ultrasonic techniques.

U
(ultrasonic)
NDb (sl
)ffpe
N'~b Before After
Specimen Test Conditiona (direct) Overload Overload

7
(2.54 cm thick) 3 min +growth 33 K 26 K 0.76 0.78
1 h +growth 32 K 23 K 0.79 0.79
1 h +growth 34 K 29 K 0.80 0.80
1 h +growth 40 K 32 K 0.76 0.76
1 h +growth 40 K 19 K 0.73 0.76
24 h +growth 30 K 25 K 0.84 0.84

8
(1.27 cm thick) 3 min +growth 34 K 29 K ..
3 min +growth 30 K 22 K ..
3 min +growth 34 K 26 K ..
1 h +growth 27 K 24 K ..
24 h + growth 22 K 20 K ..
24 h +growth 24 K 22 K ..
24 h + growth 28 K 25 K ..

9
(0.64 cm thick) 3 rain +growth 34 K 25 K 09 0.73
3 rain +growth 25 K 10 K 09 0.73
1 h +growth 25 K 17 K 0.66 0.72
1 h +growth 34 K 25 K 0.71 0.74
1 h +growth 30 K 25 K 0.72 0.79
24 h + growth 30 K 26 K 0.64 0.64
24 h +growth 25 K 20 K 0.66 0.74

a The time given in this column is the time between application of overload and
commencement of cycling at the original baseline conditions9
b
In cycles9

and the baseline loading was restarted after the appropriate delay (in this case 3
min). Crack growth was not visible on the specimen surface until Point C. The
crack then grew very rapidly until Point D, at which time the previous baseline
growth rate was reestablished. Since the crack length was not uniform through
the specimen thickness, leading somewhat in the interior, it may be possible that
the crack really started growing earlier inside the specimen than observed on the
surface9 If this is the case, it is more appropriate to extrapolate the line DE back
to Point F and take the distance BF as N D instead of the distance BC.
Tables 1 and 2 present the experimental results in detail. The two values of
N D were obtained by taking the point at which the crack growth was visible on
the surface (direct) or by extrapolating the growth curve (slope). The number of
delay cycles is plotted against the rest time between overload and restart of
baseline cycling in Fig. 2. The values plotted there are average values with the

I SLOPEDIRiCT
;z I0 9 2 54cm thick
9 I 27 cm thick
9 0.64 cm fhick

0 3 minutes I hour 1d o y
Rest - Time

FIG. 2-Number of cycles delay versus rest time.

vertical bars representing a best estimate of one standard deviation. There is a

tendency for the amount of retardation to decrease with increasing rest time,
although examination of Tables 1 and 2 shows a few tests where the opposite
effect was observed. There is no great difference in the behavior because of
thickness, although the 2.54-cm specimens usually exhibited delays that were
slightly longer than the thinner ones. The decrease in delay cycles after long rest
times at zero load is not very large in this material under these test conditions.
Although, as expected, the slope method gives smaller N D values than the direct
procedure, results obtained by the two methods do not differ greatly.

Surface Measurement of Opening Loads

Procedure
The load at which the crack became completely open on the specimen surface
was measured by laser interferometry in a manner described earlier [8]. Briefly,
two shallow reflective grooves were impressed on either side of the crack
extending ahead of the crack tip. A laser beam incident on the grooves generates
two interference fringe patterns, one on either side of the beam. The fringes are
basically straight when the crack is closed, and the fringes emanating from the
grooves astride the crack become displaced as the crack opens. The load required
to produce measurable fringe displacement over the entire crack length is
defined here as the opening load.
The opening load was measured by photographing both fringe patterns at each
load increment. These photographs were referenced to the zero load photograph,
and the fringes compared to determine the opening load. The opening load was
thus determined to lie within a certain range of load. The data are presented in
terms of U (U = AKeff/AK), the effective stress intensity ratio. Because of the
incremental nature of the measurements, U may be underestimated by
approximately 0.06.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
$HARPE ET AL ON EFFECTS OF REST TIME 67

The interferometric displacement measurement technique permits measure-

ments of the crack displacement at any point along the crack. As an additional
measure of U, crack surface displacements were plotted as a function of load for
a position 1 mm from the crack tip, and U was determined by the departure
from linearity of the load-displacement curve in a manner described by Elber
[1]. The fringe motion technique for finding U is referred to as the IDG
method, and the load-displacement technique referred to as the P - 6 method.
Values of U found by both methods are presented in Table 1.

R e s u l t s a n d Discussion

Figure 3 shows the variation of the effective stress intensity ratio U with rest
time. The vertical bars on the 2.54-cm-thick specimens denote extreme values.
The original data for Fig. 3 are given in Table 1. U for the baseline fatigue

1.0

0.8

~.E Q6

:;I:; o.,
q
P-S
9
9
IDG
0
i"l
2 5 4 CM THICK
I 27 CM THICK
9 ~ 0 64 CM THICK
0.2

BEFORE 3 MINUTES I HOUR IDAY AFTERGROWTH

OVER LOAD
REST -- TIME
FIG. 3 - E f f e c t i v e stress intensity ratio versus rest time. Vertical bars represent extreme
values.

growth is approximately 0.84, meaning that 84 percent of the applied load is

effective in growing the crack. After an overload plus a 3-min rest, the value of U
drops slightly; for longer rest periods it tends to return to its original value.
There is little difference in the behavior for specimens of different thicknesses or
between the two techniques for measuring U.
The closure model can be used, with the data of Fig. 3, to predict the amount
of delay due to retardation. The Elber model states that the growth rate is given
by
d a / d N = A (AKeff )n (1)
or

d a / d N = A (M(,) n U n (2)
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
68 FAITGUE CRACK GROWTH

Now, basically the Elber model says if one decreases U, daMN will decrease to
such a very low value that "retardation" will be observed. The average value of
this slow crack growth rate following the overload is specified by dividing the
minimum resolvable increment of crack extension by the number of delay
cycles.
Denote the conditions before overload, that is, the baseline growth rate, by
the subscript 1, and the average conditions after the overload by subscript 2.
Then
aa/aN2 = [ n (3)

Baseline growth rate data on these specimens give a value o f n = 3.76, U~ = 0.84
and da/d:Vl, = 3.8 x 10-Smm/cycle. The microscope and scale setup for
measuring crack length had a sensitivity of 0.051 mm. Using these values plus
the measured U values, one can predict the amount of delay for a given rest
time. The results of such calculations are given in Table 3. The U and N o values
in that table are an average of the U (IDG method) and N D (direct method)
measurements made on Specimens 1 through 6.
In examining the data of Table 3, it is evident that values o f N D predicted in
this manner are significantly less than actually observed. In this case the closure

TABLE 3 - Comparison of number of cycles of delay measured experimentally and

predicted by closure model.

NDa
Rest Condition U /VDa (closure model)

Baseline 0.84 . . . . . .
3-min rest 0.71 37 K 2.5 K
1-h rest 0.75 35 K 2.1 K
1-day rest 0.82 24 K 1.5 K

a In cycles.

predictions are in error by over an order of magnitude. Since it was assumed here
that U remained at its minimum value following the overload instead of
gradually increasing to U~, the predictions of Table 3 represent upper limits for
N D, Although the value of n used for these calculations was determined from
conventional da]dN-AK data rather than the more appropriate da]dN-Z~ef f
representation, the computations of Eq 3 are not very sensitive to n. Using the
prior approach, it would be necessary for the overload to cause U to drop from
0.84 to approximately 0.35 in order to achieve the number of delay cycles
actually observed. In addition, since there is still significant delay after a 24-h

rest period when U2 has nearly relaxed back to U1, it is evident that the closure
model does not completely account for retardation in this case.

Ultrasonic Measurements
Procedure

A through-transmission ultrasonic testing procedure was used to make

measurements of the fatigue crack response to loading variables. Two longitu-
dinal-wave piezoelectric ceramic transducers, 0.64 cm in diameter, were used for
the majority of the tests. The transducers were placed on opposite edges of the
specimen with the wave propagation direction normal to the crack plane (see
Fig. 4a for test conditions). Care was taken to place the transducers directly

~ TO
TRANSMITTER

RECEIVER

(o)

TO
i i TRANSMITTER

1o
TO
RECEIVER
(b)
FIG. 4-(a) Schematic illustration of ultrasonic transducer placement on the compact
specimen, fb) Approximate crack tip positions relative to ultrasonic transducer.

opposite each other along the centerline of the test specimen. Several schemes
were investigated to define a holding and coupling method which would give
reproducible results in the fatigue loading environment. The method finally
selected involved using a viscous liquid couplant (6-phenyl ether) and a rubber
band.metal clip arrangement around the transducers and specimen. The
transducer case was then visually aligned with a scribe mark on the specimen
surface, and the transducers were wrung against the specimen surface. An
uncracked specimen was repeatedly subjected to the fatigue overload test

spectrum, and it was shown that the vibrations involved had virtually no effect
on the ultrasonic signal.
To avoid introducing unwanted electronic artifacts in the ultrasonic results, a
very simple system was employed. The transmit transducer was excited at its
nominal 5 MHz center frequency by a l-ps high-voltage sinusoidal pulse burst
from a Matec Model 9000 Attenuation Comparator. The receive transducer was
connected directly to the high impedance input vertical amplifier of a wide
bandwidth Hewlett-Packard Model 180A oscilloscope. Initially, Polaroid photo-
graphs were taken of both the reflected signal as presented on the Matec cathode
ray tube and the transmitted signal on the H-P oscilloscope at each applied load
increment. The pulse-echo signal received by the transmit transducer is rectified
by the Matec circuitry and the video envelope is displayed. It was found that due
to stability problems the reflected signal was not sufficiently reliable and
sensitive enough for these experiments. It was therefore decided to photograph
and measure only the through transmitted r-f peak-to-peak signal amplitudes.
Figures 5 and 6 are representative plots of ultrasonic through-transmission
results versus applied load for 0.64 and 2.54-cm-thick specimens, respectively.
Each figure contains three sets of data which represent the three relative

600 POSITION 1
~ BEFORE
o~-~ OVERLOA0 "1-
~ ~ --EF' A F T E R &~t,45mV
550

> OPENINGLOAD
E
500 I I I I I

I,-:3
_1
250 POSITION 2
<[ BEFORE
O ~ OVERLOAD -I"
m v E ]AFTER
-(D- ""'-I~''-"r |
N 200
_u
z

o~ 150 I I ! I I

g ,
~,,,,~,~ POSITION 3 7-
-E>- BEFORE J
z OVERLOAD |
AFTER /
~ OVERLOAD ~V, 85mV
125

75 | t I I I
0 I0 20 30 40 50
APPLIED LOAD {% KMAx)

FIG. 5-Transmitted ultrasonic signal amplitude versus applied load, before application o f
overload and following a 1-h rest period. Specimen thickness is O.64 cm.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
SHARPE ET AL ON EFFECTSOF RESTTIME 71
450 POSITION1
~ . ~ BEFORE
OVERLOAD~iT
,~drTER
OVERLOAD
400I" ,,75mV

350 7 -

~ BEFORE
OVERLOAD
AFTER
OVERLOAD -7-
I00 ~V 9 3 8 mv

50 J O~EN,mLO? l. . .,

o
[
I00
+ BEFORE
POSITION 3

~ -.~ AFTER
OVERLOI~
5o[ "
0 I OPEN I J

0 It0 20 3O 40 5O
APPLIEDLOAD( %KMAx)
FIG, 6-Transmitted ultrasonic signal amplitude versus applied load, be[ore application of
overload and .following a 1-h rest period. Specimen thickness is 2.54 cm.

positions of the crack tip with respect to the ultrasonic field (refer to Fig. 4b).
Due to curvature of the crack tip through the specimen thickness there is a
certain degree of imprecision as to the exact position of the ultrasonic field with
respect to the crack tip. Roughly speaking, however, Position 1 corresponds to
the crack plane interrupting approximately 40 to 50 percent of the ultrasonic
beam, Position 2 corresponds to 60 to 80 percent, and Position 3 corresponds to
85 to 95 percent.
At each position the peak-to-peak r-f amplitude of the received signal is
plotted versus load before and after the application of the overload as shown in
Figs. 5 and 6. It should be noted that although these plots do not show the
ultrasonic signal amplitude at zero applied load, this value was periodically
checked and found to be proportionately greater than the 5 percent value. This
value was not always measured, however, because of the difficulty in
maintaining load control stability very near zero load. Also, at the other end of
the applied load range, that is, at the 225 percent overload value, the ultrasonic
signal was shown not to vary significantly from the 50 percent load condition
indicating the crack was fully open at this load. The crack growth due to the
overload was not sufficient to cause a detectable change in the ultrasonic signal
amplitude.

Results and Discussion

The following remarks are based on some general observations which were
commonly observed on several sets of similar experiments. In all cases the
ultrasonic signal amplitude at low values of applied load was less after the
overload than that observed prior to the overload, indicating that less energy is
being transmitted through the crack following the peak load. Since the signals
again match at higher loads when the crack faces are completely separated (see
Figs. 5 and 6), this result suggests that the effect of crack closure, as measured
by the ultrasonic technique, decreases with application of an overload. The
surface observations made by the laser interferometry method described
previously, however, indicated an increase in closure on the specimen surface
following the overload.
This discrepancy between the effects of closure observed on the surface and in
the specimen interior agrees with prior measurements of fatigue crack surface
displacements in a transparent polymer [10] and with electropotential measure-
ments in titanium [9]. In the latter reference, Shih and Wei suggested that due
to the differences in plane stress and plane strain plastic zone behavior, the
closure distribution through the thickness would be of the form shown in Fig. 7.

NOTCH~ - - - .
, / ~ F RONT
I PROBABLE I
l / CLOSURE-----.~ I

-~ ULTRASONIC J ~
CRACK.--z..___ ~ ~POS~TION 3 ~ TIP
TiP -~--~ POStTION Z " ~
~=~"'~ POSITION |

I-Os4-~ F 9 . Z5 4 ~
cm r
FIG. 7-Schematic illustration of probable closure distribution through specimen thickness
19/,

This is also in agreement with the interferometric measurements of crack

opening [10] which showed that the crack surfaces were closed around the
crack perimeter, but separated in the specimen interior.
The ultrasonic experiments reported here tend to support the distribution of
closure affected area shown in Fig. 7. Note in Figs. 5 and 6, for example, that
there are some basic differences indicated between the ultrasonic results as the
thickness of the sample increases. In the thinner specimen the magnitude of the
signal change between minimum and maximum applied load (AV) tended to
increase as the crack extended (going from Position 1 to Position 3). This is in
general agreement with 2-dimensional closure model predictions. However, it can

be seen in Fig. 6, .that this is not the case in the thicker material. A possible
explanation of this discrepancy can be seen schematically in Fig. 7. The relative
position of the transducers is indicated on the figure. If the closure affected area
is as assumed then the area sampled by the transducers would increase in the
thinner sample and stay constant or decrease in thicker specimen. Although
definite conclusions as to the exact shape of the closure affected zone cannot be
made on the basis of these experiments, it is clear that closure does not extend
uniformly back from the tip in the specimen interior.
Crack closure stresses were defined from the ultrasonic results before and after
the overload using the extrapolation approach as defined by Buck et al [11] and
are recorded in Table 2. It can be seen in Figs. 5 and 6 that the knee of the curve
(that point beyond which the ultrasonic signal does not change with increasing
load) either stays in the same place or decreases following the overload. This is
again in opposition to the surface measurements (and the closure model) which
indicated an increase in crack closure stress (that is, reduction in U) following
the overload.
The time dependency of the ultrasonic results following the overload were
somewhat ambiguous. In some cases there was a tendency for the after-overload
curve to shift toward the before condition with increasing time at zero toad. This
relaxation behavior was not clear cut and will require more detailed study. The
surface measurements on these samples indicated that this was not a very
pronounced effect in this alloy and it could be just outside the resolution
threshold of this ultrasonic technique.

Conclusions

1. Rest periods at zero load following a tensile peak overload had a slight
influence on subsequent fatigue crack retardation in 2024-T851 aluminum alloy.
For the conditions examined, the number of delay cycles (ND) decreased
slightly as the rest time was varied between 3 min and 24 h.
2. The zero load rest periods also had a slight influence on surface
measurements of the effective stress intensity range ratio U. Tensile peak
overloads initially decrease U below the steady-state value. As the rest time at
zero load was increased from 3 min to 24 h, U gradually returned to the original
value prior to the overload.
3. Varying specimen thickness from 0.64 to 2.54 cm had no appreciable
effect on N D or surface measured values of U.
4. Although the surface measured values of U decreased following an overload
as suggested by the closure model, the change was not of sufficient magnitude to
cause the actual amount of crack retardation. As U increased with rest time,
there was a corresponding decrease in N D. The fact that measurable fatigue
crack retardation remained after a 24-h rest period when U had returned to its
initial steady-state value further indicates that the closure model does not
completely account for surface measured crack delay.

5. Ultrasonic measurements of the internal crack behavior raise basic

questions regarding application o f the closure model to predict crack delay in
the specimen interior. Although these measurements are preliminary in nature,
they tend to support other published observations, Thus, the authors suggest
that future research be directed toward understanding the complex relation
between surface and internal measurements of fatigue crack retardation.

Acknowledgments
The authors appreciate the assistance of M. B. Strope and G. E. Mescher with
the experiments. W. N. Sharpe, Jr., very gratefully acknowledges the support of
the National Research Council and the Air Force Materials Laboratory while on
sabbatical from Michigan State University.

References
[1 ] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[2] Wheeler, O. E., Journal of Basic Engineering, Transactions, ASME, Vol. 94, Series
D, No. 1, March 1972, pp. 181-186.
I31 Willenborg, J., Engle, R. M., and Wood, H. A., "A Crack Growth Retardation Model
Using an Effective Stress Concept," AFFDL-TM-71-I-FBR, Air Force Flight
Dynamics Laboratory, Wright-Patterson AFB, Ohio, Jan. 1971.
[4] Gray, T. D. and Gallagher, J. P., "Predicting Fatigue Crack Retardation Following a
Single Overload Using a Modified Wheeler Model," presented at the Eighth National
Symposium on Fracture Mechanics, Brown University, Providence, R.I., 26-28
Aug. 1974.
[5] Christensen, R. H., Metal Fatigue, McGraw-Hill, New York, 1959, pp. 376--412.
[6] Wei, R. P., Shih, T. T., and Fitzgerald, J. H., "Load Interaction Effects on Fatigue
Crack Growth in Ti-6AI-4V Alloy," NASA CR-2239, National Aeronautics and
Space Administration, Washington, D.C., April 1973; also Wei, R. P. and Shih, T.
T., International Journal of Fracture Mechanics, Vol. 10, 1974, p. 77.
[ 7] Sharpe, W. N., Jr., and Grandt, A. F., Jr., "A Laser Interferometric Technique for
Crack Surface Displacement Measurements," Proceedings, 20th International
Instrumentation Symposium of the Instrument Society of America, Albuquerque,
N.M., May 1974; also published as Technical Report AFML-TR-74-75, Air Force
Materials Laboratory, Wright-Patterson AFB, Ohio.
[8] Sharpe, W. N., Jr., and Grandt, A. F., Jr., "A Preliminary Study of Fatigue Crack
Retardation Using Laser Interferometry to Measure Crack Surface Displacements,"
Mechanics of Crack Growth, ASTM STP 590, American Society for Testing and
Materials, to be published.
[9] Shih, T. T. and Wei, R. P., "A Study of Crack Closure in Fatigue," NASA CR-2319,
National Aeronautics and Space Administration, Washington, D.C., Oct. 1973; also
Shill, T. and Wei, R. P., Engineering Fracture Mechanics, Vol. 6, 1974, p. 19.
[10] Pitoniak, F. J., Grandt, A. F., Montulli, L. T., and Packman, P. F., Engineering
Fracture Mechanics, Vol. 6, No. 4, Dec. 1974, pp. 663-670.
[11] Buck, O., Ho, C. L., and Marcus, H. L., Engineering Fracture Mechanics, Vol. 5, No.
1, Feb. 1973, pp. 23-34.

DISCUSSION

R. 1. Stephens I (written discussion)-Your results indicate that hold or rest

time at zero load following a single tensile overload have little influence on crack
delay. You also indicate thicknesses from 6.4 to 25.4 mm have little influence
on delay and surface crack opening stress intensity. Are these results due to
plane strain conditions existing for all your thicknesses? What results would you
expect for thicknesses of say 1 to 2 mm? Can you perform these additional tests
for verification?
W. N. Sharpe, Jr., D. M. Corbly, and A. F. Grandt, Jr. (authors'closure)-As
Professor Stephens indicates, all the specimen thicknesses considered here were
sufficient for plane strain conditions based on the requirement B>2.5
(Kic/Oys) 2 . Using fracture toughness and yield strength values determined for this
plate of material in an earlier effort (Ref 8) where Kic = 23.2 MN/m 3/2 and
Oys = 448 MN/m 2 , the minimum thickness for plane strain fracture toughness is
6.7 ram. Since all testing was done at loads less than KIe it would be reasonable
to assume that plane strain conditions also prevailed for the 6.4-mm-thick
specimens tested here. We have not tested thinner specimens. The 1 to
2-ram-thick specimens suggested by Professor Stephens would be difficult to test
with the specimen geometry and ultrasonic techniques employed here. Although
we would not care to speculate on the results with thinner specimens, perhaps
the following discussion by Dr. Chanani will assist in understanding the closure
behavior for thinner specimens.
G. R. Chanani 2 (written discussion)-The crack closure results presented by
the authors, even though contrary to most of the reported results indicating
importance of crack closure in retardation, are in agreement with our results?
We used electrical potential and strain gage techniques for crack closure
measurements before and after an overload cycle during the delay period. The
strain gage technique met the "hysteresis loop" criteria for sensitivity as
described by Elber. For the specimen geometry (0.063-in. SEN), alloy
(7075-T6), and instrumentation used, we did not measure significant changes in
crack closure as the cycling progressed after the overload cycle (Fig. 8). The
small change was not enough to explain the delay. However, after relaxation for
16 h, as well as after a high overload ratio, significant crack closure was observed
by the potential method (Fig. 9). This crack closure disappeared within 200
cycles. Nevertheless, the overall conclusion from these preliminary observations
was that no substantial changes in crack closure were taking place to account for

1 MaterialsEngineeringDivision, The Universityof Iowa, Iowa City, Iowa 52242,

2 Northrop Corporation, Hawthorne,Calif. 90250.
3 Chanani, G. R. and Mays, B. J., "Observation of Crack-Closure Behavior After Single
Overload Cycles in 7075-T6 SEN Specimens," submitted for publication in Engineering
Fracture Mechanics.
75
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTMInternational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
76 FATIGUE CRACK GROWTH

NUMBEROFCYCLESAFTERTHEOVERLOADCYCLEp t ~UCO'RM~'ETR~:
~ DECCORVEcRKY'GOFOWTH
1
2 ! 10 1 ~ 1000 5000 10,000 15.000 20,000 2S.ObO 30,000 7S,O00

DISPLACEMENT 9

FIG. 8-Load versus displacement at intervals during fatigue testing," overload ratio of 2.

0.25-1N. 7075-T6 SEN SPECIMEN

10 CYCLES

"20 TO 200 CYCLESAFTERREI.AXATiONFOI~ 16 HOUI~S . . . . . . . .

~P-1 CYCLE'='=

POTENTIAL
DROP
7 ~O 11 CYCLESAFTER RELAXATION FOR 16 HOURS

PMAX = 2000 LBS, KMAx = 8 KSI

~ ' ~ P M I N = 200 LBS, KM/N = 0.8 KSI

2-6 CYCLESAFTER RELAXATION FOR tE HOURS

41 - 1ST CYCLE AFTER RELAXATION L

TIME

FIG. 9-Effect of relaxation of crack closure.

the observed number of delay cycles. These preliminary observations which

appear contrary to the results described in the past and at this symposium are in
agreement with your basic conclusion. However, we feel that these results are
applicable to the alloy, specimen geometry, thickness, and sensitivity of the
instrumentation used in this investigation. Further work is needed before
definite conclusions can be reached.
W. N. Sharpe, Jr., D. M. Corbly, and A. F. Grandt, Jr. {authors'closure)-The
authors find Dr. Chanani's results interesting and thank him for supplying a
preprint of the paper from which his comment is excerpted. Based on these
results, it is interesting to note that four different experimental approaches for
measuring closure lead to similar conclusions regarding the apparent inability of
the crack closure mechanism for completely explaining observed retardation.
The authors would agree with Dr. Chanani that much more work needs to be
done in this complex area.

Mechanisms of Overload Retardation

During Fatigue Crack Propagation

REFERENCE: Bernard, P. J., Lindley, T. C., and Richards, C. E., "Mechanisms of

Overload Retardation During Fatigue Crack Propagation," Fatigue Crack Growth
Under Spectrum Loads, ASTM STP 595, American Society for Testing and Materials,
1976, pp. 78--97.

ABSTRACT: The effect of single overloads on room-temperature fatigue crack

growth has been studied in two steels of markedly different yield stresses. The
observed retardation effects have been presented and the evidence suggests that
overload retardation is primarily due to residual compressive stresses generated in the
crack tip region, and associated with crack closure effects. The results have been
rationalized in terms of a fatigue crack growing through overload plastic zones of
different shapes and sizes associated with plane stress and plane strain deformation.

KEY WORDS: crack propagation, fatigue (materials), loads (forces), retarding,

plastic deformation

There are many reports in the literature that "overloading" a cracked

specimen reduces the crack growth rate or arrests the crack, for example, Jones
[ I ] , 3 Jonas and Wei,[2] etc. Many structures are overloaded for other reasons,
for example, during p r o o f tests on pressure vessels which may contain weld and
other defects, or overspeed testing of rotating machinery [3,4]. Overloading a
component or structure is therefore a possible method of providing protection
from fatigue failure. The retardation, or arrest o f fatigue cracks, is not cited as a
major reason for p r o o f testing, perhaps because of the quantitative, and even
qualitative uncertainties in overload effects reported in the literature and the
absence o f fundamental understanding of the mechanisms. The following models
have been proposed in an attempt to explain retardation following single
overloads or high-low amplitude block loading sequences:
1. residual compressive stresses at or near the crack tip [5-7] and associated
yield zone interaction effects [8],
2. crack closure [9-11],
3. crack tip blunting [12], and
4. crack tip strain hardening [1].
The purpose o f this investigation was to study the effect of a single overload in

i Now research metallurgist, Electricity Supply Board, Dublin.

2 Research officers, Central Electricity Research Laboratories, Leatherhead, United
Kingdom.
3 The italic numbers in brackets refer to the list of references appended to this paper.
78
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authoriz
BERNARD ET AL ON MECHANISMS OF OVERLOAD RETARDATION 79

a detailed and systematic manner to increase our understanding of the

mechanisms of overload retardation and thereby to relate laboratory data to
service conditions with greater confidence.
Fatigue crack propagation testing is usually carried out under either constant
load or constant stress intensity. In constant load tests, the maximum and
minimum loads in the fatigue cycle, Pmax and Pmin, are kept constant. In the
majority of fracture mechanics specimens, the maximum and minimum stress
intensities, Kmax and Kmin, increase with increasing crack length giving a
steadily increasing propagation rate and, hence, a curve, when crack length is
plotted against number of cycles (Fig. la(ii)).

Ill a ol) -I

D
CYCLES CYCLES

~fl b oN;

1/ CONSTANT,< LOAD
CYCLES CYCLES

/~OAO
-J ImJ C Nil

I ,~- CONSTANT~
CYCLES CYCLES

I/
/, ~

V CONSTANTK CONSTANTLOAD
CYCLES CYCLES

FIG. 1-The different types o f fatigue crack growth retardation behavior: (a] n~ f'fect, (b)
retardation, (c) delayed retardation, and (d] lost retardation.

During a constant K test the maximum and minimum loads are reduced in
steps that are sufficiently small (%2.5 percent of Kmax ) to avoid transient
effects. When crack length is plotted against number of cycles in this type of test

a straight line is obtained (Fig. la(i)). For given AK and Kmax conditions, the
fatigue crack growth rates obtained by the two methods will be the same.
In overload work, constant stress intensity testing has obvious advantages, in
that deviation from the straight line plot of crack length versus cycles (Fig. 1) is
more easily detected than deviation from a curve obtained under constant load
conditions. The present work involved constant stress intensity testing.

Types of Single Overload Effect

The types of single overload effects that have been observed are discussed in
this section.

No Effect
Small overloads generally have no detectable influence on fatigue crack growth
(see, for example, Jones [1], Jonas and Wei [2]). Curves of the type shown in
Fig. la are obtained. The threshold level, the lowest overload level at which
significant retardation occurs, is generally around 40 to 60 percent overload [1]
relative to the baseline Ks ,

%overload = Khmax-Ks xl0 01

Ks
but lower values have been reported [13]. There is apparently no explanation
for this variation in threshold level.

Retardation
Almost all investigations have revealed that retardation occurs above a
threshold overload value in all materials and specimen geometries so far tested.
Several empirical expressions have been proposed which quantify the retarda-
tions observed for particular specimen geometricw materials, and loading
conditions [1,8,9,14,15], but few attempts ha~,e been made to obtain a
generalized expression. Simple retardation is shown in Fig. lb.

Delayed Retardation
The phrase "delayed retardation" has apparently been used to describe two
different phenomena, Fig. le. It has been used to describe an increase in
propagation rate immediately following an overload application, followed by
retardation [13,16]. Delayed retardation has also been described simply as a
delay (Fig. lc) after an overload, prior to retardation [17]. It has been proposed
that a crack extends by brittle fracture in the center region of a specimen on
overloading [5] and this could account for an apparent increase in rate, thus
masking an immediate retardation.
Delayed retardation does not always occur, and some authors have provided
evidence of an immediate decrease in propagation rate [6,18].

Lost Retardation
Lost retardation is retardation followed by acceleration to a rate higher than
the baseline rate before returning to the original rate (Fig. ld). This effect has
only been observed by the present authors and will be described more fully later
in the text.
Most investigators have studied constant load specimens [1,5,17,18] and
under these circumstances, lost retardation would be difficult to detect, since
the crack accelerated continuously with increasing crack length.

Arrest
Complete arrest has been reported in the literature [1,13], but there is
evidence that the crack may propagate very slowly at a rate undetected by the
crack-following technique [18].
In general, the previous responses to overloads with increasing overload levels
are likely to occur in the following order: (a) no effect, (b) retardation, (c)
delayed retardation, (d) lost retardation, and (e) "arrest," but one or more may
be absent. All of these effects, with the possible exception of complete arrest,
have been observed during this investigation.

Experimental
Two steels with contrasting yielding and strain hardening characteristics were
examined, a low alloy pressure vessel steel, Ducol W30B, and a "corrosion
resistant" high strength steel, FV520B. The composition and mechanical
properties are summarized in Table 1.

TABLE 1 - Materialcomposition and mechanicalproperties at 20~C.

~ %
c Mn Ni Cr Mo V Cu Nb (MNm-2) (MNm-2) %Elongation

DucolW30B 0.12 1.17 0.8 0.6 0.26 0.08 0.13 ... 366 534 38
FV520B 0.04 0.92 5.6 15.1 1.35 ... 1.8 0.2 940 1110 20

The Ducol was normalized at 920~ and tempered at 640~ for 3 h. The
FV520B was normalized for 30 min at 1050~ air-cooled to 800~ and held at
800~ for 2 h before air-cooling to 20~ Finally, it was tempered at 450~ for 4
h.
Compact tension (CKS) test pieces were machined with all the dimensions as
for a 25-mm thick specimen, except for the thickness which was varied from 1
to 25 mm. The ratio H/W = 0.6 where 2H was the specimen height and W the
width measured from the pin-hole centers.

The crack length in side-grooved specimens [19] was measured by a d-c

potential drop method [20]. In specimens without side-grooves, crack lengths
were measured optically using 1-mm grid lines scribed on the .pre-polished
surfaces, several specimens being monitored by both methods. One 75-mm thick
CKS specimen of Ducol W30B with no side-grooving was also tested and was
monitored optically on both faces and at the center of the crack front by an
ultrasonic technique. When two methods of crack length measurements were
used on a single specimen, the surface readings were used for calculating loads
and stress intensities. Optical crack length measurements were estimated to
within 0.1 mm, and potential drop readings to • ram. The ultrasonic
technique could locate the crack tip position to within +-0.5 ram. Changes in
crack length of 0.2 mm could however be detected.
All fatigue cycling was performed on servohydraulic machines at a frequency
of 1 Hz for the 75-ram-thick CKS specimen, and at 3 Hz for all the other
specimens. Baseline cycling was carried out under "constant" 2ds and K max
conditions, the maximum stress intensity, Kmax , and minimum stress intensity,
gmi n being maintained to within +2.5 percent by reducing the loads every 0.5
,

mm of growth (discussed earlier). Cracks were initiated from spark machined

notches, and baseline
z3X = 30 MNm-3/2;R = Kg"rnin. =0.05
K~max

propagation rates were established, usually over a distance of 4 mm, before the
first overload. In many cases, several single overloads were applied to one
specimen provided that the baseline crack growth rate could be reestablished
following previous overloads. Overload levels were held for 30 s, and the load
returned to Kg.min before fatigue cycling was continued. Overloads were usually
applied in an increasing series, starting with an overload at 40 MNm -3/2 , and
subsequently raising this by increments of 10 MNm -3/2 .
In some cases a single specimen was overloaded to the same Khmax at
different crack lengths in order to establish whether the ratio, a/tf, of crack
length, a, to specimen width, W, influenced the degree of overload retardation.
Stress intensities were calculated from the tables of Walker and May, [21] and
load modifications for side-grooving as described by Freed and Krafft [19].

Results
Shapes of Curves
Figure 2 (Curve A) is representative of the results obtained on FV520B and
shows a sharp decrease in rate immediately following an overload. During the
retarded period, the crack grew at a very low, but measurable rate, for periods
which increased with increasing overload level. At the end of the retarded
growth rate region the rate abruptly reverted to the baseline rate, except at an
overload of 80 MNm-3/2 where the behavior was similar to that observed for
Ducol.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
BERNARD ET AL ON MECHANISMS OF OVERLOAD RETARDATION 83
CURVE B, 25rnm
CURVE A. 25 mm THICK FV520B THICK DUCOLW30B

3O
OVERLOAD TO
Khmax 80 ~

.~ 26

OVERLOAD TO j '
i 24 Khmox 7

2Z AD TO Khmax-- 60 ltN rn-3/2

2o ~ OvER,._O~TO J
./'- OV.LOAO TO ,.;. ,0
~s I I I ~-r I I "~/ I I I I
0 20 40 60 80 100 120 140 160 180 200 220
N x 1000CYCLES

FIG. 2-Typical crack growth retardation curves for 25-mm,thiek Ducol and FV520B
steels.

Figure 2 (Curve B) shows a curve representative of the softer Ducol W30B

steel. There are several differences in response compared to the high strength
FV520B.
1. The transitions from baseline to retarded rate, and retarded to baseline rate
were gradual in contrast to the FV520B.
2. The rates in the retarded position were higher than for the FV520B, even
though the baseline rates for the two steels were very similar.
3. There was a region following the application of the overload where the
crack continued to grow at approximately the baseline rate before slowing
down.
4. The rate of growth following the delay period was initially higher than the
baseline rate (as measured on the surface of the specimen) before returning
to the original rate (see earlier discussion).
The overload response due to changing specimen thickness and other variables
are treated in the appropriate sections.

Threshold Effects

In both materials there existed a threshold overload level below which

retardation was not detected. For a given specimen thickness, this threshold was
higher in the higher yield stress FV520B (Table 2). The threshold level in the
case of Ducol W30B decreased with decreasing thickness but was essentially
independent of thickness in FV520B over the range of thicknesses tested (Table
2).

Affected Calculated-Overload Calculated-Overload >

:-4
Khmax Retardation Crack Length Plane Stress Plastic Plane Strain Plastic
Specimen Thickness Overload Cycles (mm,) Zone Size (mm) Zone Size (mm) c
(mm) (MNm -3/2 ) (thousands) (Aa) m
No. (A,e*) (A~c *) ')
t--n

>
t')
DUCOL14 75 40 3 1.8 3.8 1.27
75 50 15 2.0 5.94 1.98
75 60 38 2.4 8.55 2.85 -n
0
75 70 130 4.5 11.64 3.85
-1-
DUCOL10 25 40 0 0 3.8 1.27
25 50 10 1.5 5.94 1~98
25 60 25 1.9 8.55 2.85
25 70 50 2.6 11.54 3.85

DUCOLI1 19 40 0 0 3.8 1.27

19 50 6 0.8 5.94 1.98
19 60 40 1.8 8.55 2.85
19 70 170 3.2 11.64 3.85

DUCOL7 19 60 33 1.8 8.55 2.85

19 60 29 2.2 8.55 2.85
19 60 38 2.0 8.55 2.85
19 60 29 1.7 8.55 2.85

DUCOL 2 19a 40 0 0 3.8 1.27

19a 50 0 0 5.94 1.98
19a 60 20 1.8 8.55 2.85

DUCOL4 19 40 0 0 3.8 1.27

19 50 6 1.0 5.94 1.98
19 60 12 1.8 8.55 2.85

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
~'~.
~" ~..
e- .
0" O', Ca 4~
O O O O O OO ~ O O O O O O O O ~ O ~ O O O O O O O
I=u
0
e~
O O O
K
OO k,O I ~
t.~ OO --~.1
s NOI/VOEIVJ_::IEI OVO'IEI=IAO dO s NO "IV 13 ONVNI=I:I8
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
86 FATIGUE CRACK GROWTH
Effect o f Overload Level
In both steels beyond the threshold, an increase in overload level was
accompanied by an increase in retardation. The relationship between the
percentage overload and retardation is not linear, as shown in Figs. 3 and 4, a
given increment being progressively more effective as the overload level
increases. However, limited results (Table 2) indicate that if the overload is
sufficiently large to cause general (nett section) yielding, smaller retardations are
obtained than for overloads within linear elastic conditions.

i80 -- 180

/
170
i 170
160 160
150 150
140 140
130 i t30
120 ~
o 120

iio IlO

I00

9o 9O ' / i eO sPILAINRooVED
O ~ SPECIMENS
8O 8O

70 70

60 I I I I I I 6o I I I I I
cq 10 20 30 40 50 60 h I0 2030 40 50 60
RETARDATIONXI000 RETARDATIONX l 0 0 0
FIG. 3-Effect of overload level on crack growth retardation in 5-ram (a) and 25-mm (b)
thick specimens of FV52OB.

Effect of Thickness
Steels of both strength levels behaved similarly in that, as specimen thickness
was increased, a given Khmax resulted in less retardation. The curves for Ducol
W30B are shown in Fig. 4.

Effect of Yield Stress

The higher yield stress material, FV520B, gave a smaller retardation than the
low yield stress material for the same overload level and specimen thickness (Fig.
2).

Effect o f Side Grooving

Table 3 gives details of the effect of side grooving on retardation. The lower
overload levels gave less retardation in the side-grooved specimens than in the

160

15O

140

t20

II0

~ 100 -

~ 90
N 80

7O
O --

50 SIDE GROOVED

4O
30 A'oo
0 10 20 30 40 50 60 70 80 90 I00 II0 120 130 140

RETARDATIONXl000

FIG. 4-Effect of overload level on crack growth retardation in Ducol W30B specimens of
various thicknesses.

plane-sided specimens, although the effect is relatively small. At higher overload

levels there was little difference.

TABLE 3 - Effect of side-grooving on crack growth retardation

(number of cycles lost) in Ducol W30B and FV520B.

Ducol W30B 19-ram Thick FV520B 25-mm Thick

% Overload Plain Specimen Side-Grooved Plain Specimen Side-Grooved

33 0a 0 0 0
66 6 000 a 0 1 000 0
100 34 000 a 20 000 6 500 a 4 000
133 170 000 ... 17 000 12 000
166 . . . . . . 50 000 50 000

a Average of two or more results.

Effect "of Crack Depth (a/W)

The retardations, in terms o f cycles lost, obtained in the two specimens loaded
to the same level at different crack lengths are shown in Table 4.
There was no systematic variation in retardation with crack length or a/W. The
scatter in results shows a variation o f 18 percent from the mean in the 19-mm

TABLE 4 - The effect of a/W on overload retardation.

B Overload
Specimen (mm) a/W (%) Retardation

Ducol 7 19 0.360 100 33 000

19 0.476 100 29 000
19 0.653 100 38 000
19 0.641 100 29 000

Ducol 9 5 0.380 66 28 000

5 0.476 66 27 000
5 0.548 66 44 000
5 0.624 66 15 000 (GY)

GY = general yield.

specimen, and 33 percent from the mean in the 5-mm specimen, excluding the
general yield overload.
For the largest overloads in Ducol specimens, plastic zone size corrections
became significant. For example, in specimen Ducol 7 (see Table 4 and Fig. 5)
overloaded to a nominal Khmax value of 60 MNm -3/2 at four different
values ofa/W, Khmax overload levels were estimated using surface measurements
of plastic zone sizes. Assuming plane stress conditions, the corrected values of
Khmax were approximately 66, 70, and 75 MNm -3/2 . Assuming the plastic
zone in the center is about one third of that on the surface, the corresponding
values were approximately 62, 63, and 64 MNm -3/2 . The results for the
overload at the largest a/W were just outside the ASTM calibration range and
were therefore probably invalid.
For overloads in the high yield strength FV520B and for lower overloads in
the softer Ducol material, the overload plastic zone size corrections were
insignificant.
There was an indication (Table 4) that general (nett section) yielding of the
specimen reduced the retardation, but further testing will be necessary.

75-mm Ducol W30B Specimen

The two crack growth curves shown in Fig. 6 are similar to the retardation
Curves A and B in Fig. 2, characteristic of FV520B and Ducol W30B behavior,
respectively. The center of the 75-mm Ducol W30B specimen behaved similarly
to the FV520B material and the edge regions reflected thin Ducol W30B
specimen behavior.
Under baseline AK conditions, the crack front in the 75-ram-thick Ducol
W30B achieved an equilibrium curvature whereby the crack surface lagged
behind the central regions by 6.8 mm (see Fig. 6). Following the application of a

FIG. 5-Surface o f specimen Ducol 7 overloaded at d~(ferent values o f a/W to same

overload Khmax o f 60 MNm - 3/2.

SURFACE, /
140 R E ~ / 74

120 "/2
CEI~ ~" f ULTRASONIREADI
C NGS , ~ f
9 ,~" ATCENTRE
OFCRACK~"
100 / n|~- FRONT
FRONT / - ~ / 70

i 80 68

60 / OVERLOAD
TONhmox70MNm"3'2 66

40 F i u ~ u ~ ~ ~ l / F 64
ACE
READINGS
20-- 62

0 I I I I I I I I J l l 60
20 40 60 80 iOO i20 140 160 180 200 22D
Nx1000.CYCLES
FIG. 6-Surface and midthickness crack growth retardation in 75-ram-thick Ducol W3OB.

large overload, the curvature of the crack front initially decreased since
retardation was immediate at the center but delayed at the surface. The
curvature was further reduced by the more significant growth through the
surface plane stress regions during retardation than through the central plane
strain regions. However, since retardation is completed after a smaller number of
cycles at the specimen center, the curvature then increased until retardation was
completed at the surfaces. Thereafter, the curvature decreased toward the
equilibrium curvature characteristic of the baseline fatigue conditions (lost
retardation). It is noted that the constant K conditions were calculated from
surface crack length measurements and that the shapes of the growth rate curves
in Fig. 6 would have been different if the crack length at the center of the
specimen had been used instead.

Macroscopic Fracture Appearance

The polished surface of a Ducol specimen overloaded at four different values

of a~ W to the same overload Khmax of 60 MNm- a/2 is shown in Fig. 5.
Examination of the fatigue fracture surfaces revealed a dark band associated
with the application of each overload. These bands were well defined at high
overload stress intensities but hardly discernible at low peak stress intensities.
The shear lip was small in the retardation region and then increased gradually to
approximately the original value found prior to the overload. This was reflected
in a change in direction of crack propagation at the specimen surface (Fig. 7).

FIG. 7-Effect o f overload on crack growth at surface of Ducol IV30B (x23}.

Discussion
Models f o r Crack Growth Retardation Following Overloads

Crack tip strain hardening has been discounted as a mechanism by Jones []]
following a study of overload effects on fatigue crack growth in pre-strained
material.
It is argued that the importance of crack tip blunting is different in plane
stress and plane strain situations, and it is convenient to consider separately the
evidence for either case.
A crack tip blunting mechanism would suggest that the crack length affected
by the overload, An*, should be restricted to the crack tip region which is clearly
not the case for plane stress surface measurements (Fig. 8a) where the affected
crack length spans several millimetres.

022 F ~ I 0 22 - m

020 F 020 -

018 L 018 -

016 F 016 -
014~ 014 _

0'2 -
012 F
0~0 -
010 F
<3 'd

i~
0 08 -

006 -

004I 0 04 -

~176 0 02

0
-
I I I I
2 3 4 5 6 h 04 08 I2
a
DISTANCEFROMOVERLOADPOINT ran' DISTANCEFROHOVERLOADPOINT H~m
FIG. 8-Growth rate through the overload retardation region for specimen Ducol 14
measurements at specimen surface (a) and center {b).

In the present investigation, the only case where complete crack arrest could
have occurred (and where crack tip blunting might be applicable) was at the
center of the 75-ram-thick Ducol 14 (see previous section) which is under
predominantly plane-strain conditions. The ultrasonic crack following technique
was incapable of discriminating between arrest and very slow growth over a very
short distance. The crack length affected by overloading (Fig. 8b) was much
more localized than for the plane stress case. However, clip (displacement) gage
measurements made immediately before and after a single overload on a Ducol
14 specimen indicate that an overload has no significant effect on edge opening
displacement for this predominantly plane-strain situation. Furthermore, the
shape of the retardation curve was similar to that for the high-strength FV520B
with side notching. In this case, which was also predominantly plane strain, the

crack-following technique (potential drop) was sufficiently sensitive to detect

very slow, but perceptible, crack growth over the small affected distance. These
observations are again inconsistent with a crack blunting model.
Preliminary results on overloads approaching general yield suggest a smaller
retardation effect than for those applied at small values ofa/W. Further work on
retardation approaching and beyond general yield might increase our under-
standing of the relevance of crack tip blunting which would be more severe
under these circumstances.
The results presented in Table 2, particularly those on the 75-mm-thick Ducol
specimen, may be rationalized by proposing that retardation effects differ
markedly between plane stress and plane strain conditions. It is proposed that
compressive residual stresses generated in the overload plastic zone, accompanied
by crack closure effects, are responsible for retardation and that the magnitude
and distribution of these stresses are different in plane stress and plane strain.
The variety of retardation effects described earlier are considered in terms of
these differences and to the variation in constraint that can occur from the
surface to the center of a test piece.
Following the arguments of Elber [9], it is probable that during growth
through the overload plastic zone, the surface (plane stress) regions of the crack
close above the minimum stress intensity Kmin of the fatigue cycle. This is
consistent with abrasion of the fracture surfaces previously observed where
specimens were tested at larger values of AK [11]. At present, it is unclear if
significant closure occurs in a predominantly plane-strain situation [22].

Residual CompressiveStress Mechanism for OverloadRetardation

The residual stress levels were estimated by measuring the crack propagation
rate at intervals during retarded growth after an overload (Fig. 8). The effective
AK responsible for crack growth, ~kKeff, is less than the applied AK, (AKap p ),
due to residual compressive stresses and consequent crack closure effects in the
affected crack length region. Values of/~Keff were obtained from the standard
(da/dN)/AK plots previously obtained for the two steels [22]. The magnitude of
the residual stresses are reflected in the quantity (AKap p - ~kKeff) which is
plotted in Fig. 9 as a function of distance from the point of overload. Two types
of curve were obtained, one characterizing plane stress deformation (Curve A in
Fig. 9) and the other plane strain deformation (Curve B in Fig. 9). The plane
strain plastic zone consists of two shear bands, inclined at 45 ~ or more to the
plane of the crack, with only a small region of plasticity ahead of the crack tip.
By contrast, the maximum dimension of the plane stress zone is in the plane of
the crack. The shapes of the plastic zones under monotonic loading have been
illustrated schematically by Hahn et al [23] and directly by etch-pitting a high
nitrogen steel (Griffiths and Richards [24], see Fig. 10). The residual stress
patterns reflected by ( ~ a p p -- ~ e f f ) in Fig. 9 are qualitatively consistent with
these plastic zones. Thus in predominantly plane-strain conditions, the retarda-
tion is immediate on application of the overload, and retarded growth occurs at
a very slow rate over a small distance (Curve A in Fig. 2 and Fig. 8b). For plane
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
BERNARD ET AL ON MECHANISMS OF OVERLOAD RETARDATION 93

22 22

20 20

IB 18

16 16

14 14

6
\ 1

4 4

a
2

0 I
I
~'~ 6
5
1
b
2

0
o
J
04
I
o.g
I
12

DISTANCE FROM OVERLOAD POINT. mrn DISTANCE FROM OVERLOAD POINT, mm

F I G . 9-z2~Kbaseline -- z2~Keffeetiv e as a function o] distance from point of overload for

specimen Dueol 14, (a) surfae~and [bJ center.

stress, the affected crack length after overload is more extensive and the rate of
growth through the overload zone (Curve B in Fig. 2 and Fig. 8b) diminishes
progressively to a minimum level, which is typically one quarter to one third of
the affected crack length, before eventually returning to the "baseline" rate.
This suggests that the maximum tensile strains and residual stresses after
overloading are not at the crack tip. The crack growth rate over much of the
affected crack length is significantly greater in plane stress (Fig. 8). This implies
that residual stresses will generally be higher under predominantly plane strain
conditions, and the present results are consistent with the view that residual
stresses will increase with increasing yield stress, specimen thickness, and
overload level (below general yield).
The monotonic plastic zone sizes, 2ry, due to overloading were calculated
using the relation

Aac* = 2ry = ~ Khmax

where d = 2 for plane stress and 6 for plane strain given in Table 2. Not
surprisingly, in view of the proposed importance of plastic zone shape, poor
correlation was generally obtained between Aa c* (either plane stress or plane
strain) and ,the measured affected crack length, An*, which was usually
appreciably less than Aac* (Table 2). In terms of this simple comparison, better
correlation would be expected for a plane stress situation than for plane strain.
In fact, best correlation was found for 5-mm-thick Ducol (appreciable plane
stress) and surprisingly for the 25-mm side-notched FV520B (predominantly
plane strain). In calculating plastic zone sizes, the use of ay s allows no
work-hardening and o u might be a more appropriate parameter, thereby
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
94 FATIGUE CRACK GROWTH

FIG. lO-The plastic zones at the midthicknesses o f (a) 24.ram and (b) 2-ram thick,
double edge notched specimens o f high nitrogen steel loaded to give the same nett-section
stress (~0.8 o f the general yieM stress} (x4].

reducing Aac* by 47 percent for Ducol and 72 percent for FV520B. In reality, a
flow stress intermediate between crys (giving an upper bound) and a~ (lower
bound) is probably relevant.
Further support for the residual stress mechanism was obtained by performing
stress relieving experiments. An initial stabilizing treatment of 40 h at 640~ was
performed to eliminate possible microstructural changes during stress relief of 25
and 5-ram Ducol specimens. In the 25-mm-thick specimen, a delay of 15 000
cycles was obtained following an overload of 50 MNm -3/2 . The same overload
followed by stress relief anneal (2 h at 640~ however, caused a delay of only
1000 cycles. Similarly, when a 5-mm-thick Ducol specimen undergoing what
would normally have been an extensive retarded growth region was subjected to
a stress relief anneal, continued cycling caused the growth rate to return
abruptly to the baseline rate.
There is a minimum cyclic stress intensity, AKo, below which fatigue cracks
do not grow significantly. If the effective stress intensity z~ef f does not exceed
AKo then the crack should arrest. These conditions were not achieved in this
investigation where the baseline AKs of 30 MNm -3/2 was relatively large and
further tests are planned at smaller values of baseline AKs
It was noted during this investigation that, with plane strain deformation,
there was a definite threshold level which must be exceeded during the overload
before significant retardation occurred. In both materials, this level was about
100 percent higher than K~.nax for the baseline fatigue cycling. In contrast, for
conditions approaching plane stress, there appeared to be no clearly-defined
threshold and small overloads ('~20 percent) resulted in detectable retardation
(Table 2). The prediction of fatigue life of a cracked component under
variable-amplitude loading using the Paris [25] or other crack growth laws
available in the literature will be more successful where the material is deforming
under predominantly plain-strain conditions, since transient effects due to
overloading will be absent (below the threshold) or minimal compared to
plane-stress situations.

Conclusions
The various retardation effects resulting from the application of a single
overload have been examined and discussed. The evidence suggests that overload
retardation is primarily due to residual compressive stresses, generated in the
crack tip region, and associated with crack closure effects. The results have been
rationalized in terms of a fatigue crack growing through overload plastic zones of
different shapes and sizes found for predominantly plane stress and plane strain
deformation.
A threshold level existed below which retardation did not occur, and this level
was higher for plane strain than plane stress. Above the threshold, the fatigue
crack growth rate through the overload region was higher in plane stress than
plane strain. The results are consistent with the view that residual stresses will
increase with increasing yield stress, specimen thickness, and overload level
(below general yield).
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
96 FATIGUE CRACK GROWTH

A cknowledgm en ts

The work was carried out at the Central Electricity Research Laboratories and
the paper is published by permission of the Central Electricity Generating Board.

References
[1 ] Jones, R. E., Engineering Fracture Mechanics, Vol. 5, 1973, pp. 585-604.
[2] Jona~, O. and Wei, R. P., International Journal of Fracture Mechanics, Vol. 7, 1971,
pp. 116-118.
[3] Nichols, R. W., Pressure Vessel Engineering Technology, Elsevier, 1971, p. 225.
[4] Nichols, R. W., British Welding Journal, 1968, Vol. 15, p. 21.
[5] Hudson, C. M. and Hardrath, M. F., NASA, TN D-1803, Aug. 1963.
[6] Hudson, C. M. and Raju, K.N., NASA, TN D-5702, March 1970.
[7] Schijve, J. and Broek, D., Aircraft Engineering, Vol. 34, 1962, p. 314.
[81 Wheeler, O. E., Journal of Basic Engineering, Vol. 4, 1972, pp. 181-186.
[9] Elber, W., Engineering Fracture Mechanics, Vol. 2, 1970, pp. 37-45.
[10] Elber, W., Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, p. 230.
[111 von Euw, E. F. J., Hertzberg, R. W., and Roberts, R., Stress Analysis and Growth of
Cracks, ASTM STP 513, American Society for Testing and Materials, 1972, p. 230.
[12] Christensen, R. H., MetalFatigue, McGraw Hill, New York, 1959.
[13] Corbly, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973,
pp. 479-497.
[14] Probst, E. P. and Hilberry, B. M., AIAA Paper No. 73-325, 1973.
[15] Schijve, J., Advances in Aeronautical Sciences, Pergamon Press, 1962, pp. 387-408.
[16] Vargas, L. G. and Stephens, R. I., Proceedings, Third International Conference on
Fracture, Munich, Germany, Vol. VI, April, 1973.
[17] McMillan, J. C. and Hertzberg, R. W. in Electron Fractography, ASTM STP 436,
American Society for Testing and Materials, 1968, pp. 89-123.
[18] McMillan, J. C. and Pelloux, R. M. N. in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, pp. 505--532.
[19] Freed, C. N. and Krafft, J. M., Journal of Materials, Vol. 1, 1966, p. 770.
[20] Ritchie, R. O., Garrett, G. G., and Knott, J. F., International Journal of Fracture
Mechanics, Vol. 7, 1971, pp. 462--467.
[21 ] Walker, E. F. and May, M. J., BISRA Report, MG/E/307167, 1967.
[221 Lindley, T. C. and Richards, C. E., Materials Science and Engineering, Vol. 10,
1972, p. 235.
[23] Hahn, G. T., Hoagland, R. G., and Rosenfield, A. R., Battelle Final Technical
Report, AR-016067, 1971.
[24] Griffiths, J. R. and Richards, C. E., CERL Report, RD/L/N 132/72, 1972.
[25] Paris, P. C. and Erdogan, F., Transactions, American Society of Mechanical
Engineering, Vol. 85, 1963, p. 528.

DISCUSSION

R. L Stephens ~ (written discussion)-Your results indicated the lower yield

strength steel had greater crack retardation than the higher strength steel for a
given overload and thickness with R~ = P~min/P~max > 0. Based upon greater
cyclic plasticity in the lower yield strength steel, I might think the opposite
effect could occur with R~ = - 1 . That is, residual stresses and crack closure can
1 Materials Engineering Division, The University of Iowa, lowa City, Iowa 52242.

be more readily removed in the lower yield strength steel-we are currently
investigating this behavior. Could the authors comment on this?
P. J. Bernard, T. C. Lindley, and C. E. Richards (authors' closure)-We have no
results concerning the effect of overloads on retardation for baseline tension-
compression cycling (Rs = -1). However, we suspect that residual stresses and
crack closure can indeed be more readily removed in a lower yield strength steel,
for a given overload and thickness. We look forward to the results of your
investigation.

Spike Overload and Humidity Effects

on Fatigue Crack Delay in
AI 7075-T651

REFERENCE: Buck, Otto, Frandsen, J. D., and Marcus, H. L., "Spike Overload and
Humidity Effects on Fatigue Crack Delay in AI 7075-T651," Fatigue Crack Growth
Under Spectrum Loads, ASTM STP 595, American Society for Testing and Materials,
1976, pp. 101-112.

ABSTRACT: This paper describes experiments on fatigue crack growth, Ao/~tN, in

part-through crack specimens of 7075-T651 aluminum alloy. The effects of dry and
humid environment in combination with single-spike overloads on ~2~a/Z~,the crack
growth delay, ND, and the "effective" stress intensity range, AKeff, are discussed.
Using an acoustic monitoring technique, it was observed that the crack closure not
only reflects the influence of the environment, but also of the spike overload on the
plastic deformation at the crack tip and therefore on Ao/AN. The relationship A,~/2t/V
= A (AKeff)n describes both environmental and spike overload effects, uniquely.
Delay is more pronounced in a dry environment and increases strongly with
increasing overload ratio.

KEY WORDS: crack propagation, fatigue (materials), aluminum alloys, humidity,

plastic deformation

Delay, or retardation, of the fatigue crack growth rate due to tensile overloads
is an important phenomenon in that the fatigue life of a structure can be
increased by the overloads. Only recently, however, have systematic studies been
aimed at the phenomenology o f the process [1-8] .3 Corbly and Packman [4]
have summarized many of the observations that are generally agreed upon in
various investigations and, therefore, a review will not be repeated here.
It has recently been indicated [3,5, 7] that crack closure [9] is a likely model
to quantitatively explain delay effects. Trebules et al [5] indeed have shown the
great potential of crack closure by using the Elber equation [9] to explain their
data. This paper describes results from closure measurements made directly
before, during, and after application of single overload spikes using an acoustic
technique [10] which was previously used to study the effects o f overload

1 Group leader and senior staff associate, respectively, Science Center, Rockwell
International, Thousand Oaks, Calif, 91360.
2 Formerly, group leader, Science Center, Rockwell International, Thousand Oaks, Calif.
91360; now, professor, Mechanical Engineering Department, University of Texas, Austin,
Tex. 78712.
a The italic numbers in brackets refer to the list of references appended to this paper.

101
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authoriz
102 FATIGUE CRACK GROWTH

blocks on fatigue crack delay. Also, if crack closure is important in explaining

delay, one might expect large effects of mildly aggressive environments on the
delay since an environment can change the crack closure stress and the crack
propagation rate severely [11]. Thus, it was decided to check these hypotheses
on the aluminum alloy 7075-T651 and apply single-spike overloads in dry and
humid gaseous environments. The results of this investigation are reported in this
paper.

Experimental Procedures
To study both load spike and environmental effects on fatigue crack growth,
the aluminum alloy 7075-T651 with a nominal yield stress ay s ~ 480 MN/m 2
was selected since its fatigue crack propagation rate is known to be very sensitive
to humidity [11]. Specimen geometry was of the part-through crack (PTC) type
(1.25 cm thick, 10 cm wide) with a semi-elliptical starter notch (1.3 mm deep
and 6 mm long). The starter notch was covered with a glass bell filled with
flowing dry nitrogen or moist air.
All experiments were performed in tension-tension loading with the load axis
perpendicular to the starter notch (Mode I). A sinusoidal gross section stress
with a maximum amplitude O~rna x = 0.225Oy s and a load ratio R =
(tT~min/O~max) = 0.1 was applied until a crack depth of about 0.2 to 0.4 cm was
obtained. At this time a single sinusoidal load spike with an overload ratio
Ohmax /(/~max of either 2.0 or 2.5 was applied using a paper-tape controlled
arbitrary function generator. Thereafter, the program reverted to the previous
stress conditions. During all measurements of crack depth and closure, the stress
was cycled at a rate of 0.1 s- 1 and in between measurements at 5 s- ~.
The acoustic device for monitoring crack depth and crack closure has been
described in several previous publications [10-13]. The device was positioned
across the starter notch and mounted within a glass bell [11]. Crack growth and
crack closure produce changes in the acoustic wave attenuation which result in a
direct measurement of true and "apparent" crack depth.
After fracture of each specimen, the fatigue crack growth region of the
fracture surfaces was examined by scanning electron microscopy (SEM) and by
replicas in the transmission electron microscope (TEM) to characterize the
fatigue fracture morphology.

Results
Crack depth measurements obtained from acoustic wave attenuation on five
specimens are shown in Fig. 1 as a function of the number of fatigue cycles
applied. In four of the cases the humidity was well controlled (0 to 80 percent
relative humidity (RH)); in the fifth case, the humidity was about 10 percent
RH at the beginning and dropped to 0 percent towards the end of the
experiment. Some of the effects observed on the fifth specimen were more
pronounced than they were in the other four samples, but this is more likely the

7 , , ~ . . . . . I .~ ' Q~ . . . . . .

6 0~RH Ohrnax/Olmax.Z 5

5 .~1 o -

4- ! %mJ~
,~ /Olmax 9 2. 0 80~ RH

3- ~ ~ % max/Olmax -

,~ j =2.00~RH

2--
~ 0 - 0-, ,,0 0 ',CO
ax = 2.5 0~RH

[ | l I I I l l I I = i = J f I i ;

I04 IOs 106

FATIGUECYCLES
FIG. 1-Crack depth versus number of fatigue cycles. Arrows indicate application o[ the
overload spike.

result of the slightly higher K levels involved rather than the variation within the
environment.
The four experiments with controlled relative humidity were performed as
outlined in the previous section. Fatigue cycling at O~max = 0.225ay s and R =
0.1 established an initial crack propagation rate, An/AN, which was appreciably
higher at 80 percent RH than at 0 percent RH (dry N2). After a number of
fatigue cycles in this initial condition, a single-spike overload of amplitude
Ohmax with (O'hrnax /O~rna x) equal to either 2.0 or 2.5 was applied (indicated by
arrows in Fig. 1) in both humidity conditions. The crack growth was delayed for
a certain number of fatigue cycles, No, before further growth occurred. The end
of the delay can easily be detected using the acoustic technique since the signal
strength starts to drop again. In one case, after application of the overload ratio
(Thmax /O~max ---- 2.5 at 0 percent RH, the specimen broke in the grip section
after fatigue cycling for 560 000 cycles without resumption of crack growth. All
external parameters as well as the results on these four experiments are listed in
Table 1. Also included are the stress intensity range, AK, just before the
overload and the corresponding effective stress intensity range, AKeff, as
deduced from crack closure data [10] on these specimens. AKeff has been
derived using the following extrapolation method [10]. Before and during
application of the overload AKeff is defined to be proportional to (O~max -
aop), with %p being the closure load (and equal to the opening load, since no
hysteresis was observed) given by the intersection of the tangents to the two
extreme segments of the closure curves. Although the AK levels prior to

'-rl
-..4
5
c
m
c~

TABLE 1 - Fatigue crack growth before application of various overloads, ~K, ZXKefj~ and o
delay data for four experiments with well controlled humidity {see Fig. 1).
o
9,X,,o/A, v
Relative ao(mm), (it/m/cycle), ~IK(MNm -3/2 ), z21Keff(MNm-3/2), "i"
Humidity Initial Crack Prior to Prior to Prior to
Ohmax/Oganax (%) Length Overload Overload Overload ND

2.0 0 2.35 0.012 7.60 3.88 24 000

2.0 80 3.50 0.20 8.59 6.18 5 000

2.5 0 2.50 0.019 7.71 3.31 > 5 6 0 000

2.5 80 4.50 0.25 9.14 6.03 93 000

overload were not quite the same for the four specimens, Table 1 demonstrates:
1, ~ a 0 / ~ and ZkKeff increase as relative humidity increases in agreement
with R e f t l ,
2. N D increases as O'hmax/O'~max increases in agreement with Ref 6
(correction to a constant zSX prior to overload is such that N D slightly
decreases with increasing AK), and
3. N o decreases as relative humidity increases.
These points will be discussed more fully in a later section.
The initial crack growth rate observed on the fifth specimen (at RH ~ l 0
percent) was intermediate to the ones observed in the 0 and 80 percent RH
experiments, in agreement with earlier observations [11]. During the spike
overload (ahmax /Os = 2.5) the acoustic signal indicated relatively large
crack growth (~300/.tin) as shown by Points 2 and 3 in Fig. 1. Point 4 shows the
crack depth as observed in the first load cycle after the overload: the crack does
not open up to the full depth it did during the overload cycle. At Point 6 the
crack depth again has reached the actual crack depth established during the
overload. The crack propagation rate thereafter is typical of the crack growth
rate observed in the dry environment (0 percent RH) experiments after delay.
Actual closure curves (crack depth versus applied stress) for the fifth specimen
tested in a mixed environment are shown in Fig. 2. The identification of each of
the closure curves corresponds to similar identifications in Fig. 1 for this
specimen. As has been pointed out before [10,11], the interpretation of these

I DENTIF ICATION
i
?
9
/-CRACK DEPTHDUETO

~
. . . . . . . . . . .

I I I
~ ~ ~
APPLIED STRESS
FIG. 2-Crack closure curves before, during, and after application of overload spike.
Identification as in Fig. 1.

closure curves is that the crack is fully open only at high load levels. The crack
stays open over a load range (defining z~tKeff and the true crack depth) as the
stress level is lowered and then the crack closes down on itself (a typical example
is shown in Curve 1). As may be seen, the acoustic measurements indicate that
the crack grew about 300 tan during the overload spike. After the overload the
crack does not fully open over the full crack depth established during the
overload for a large number of cycles, therefore Mfeff = 0. As an example, the
maximum crack depth indicated by Curve 5 is less than that indicated during the
overload cycle. Finally (Curve 6), the crack opens again over the full extent of
the crack depth established during the overload cycle. The load corresponding to
the crack opening decreases slowly with increasing number of fatigue cycles.
After the crack has started to move again, it is proposed that aop is given by the
intersection of the tangent at the upper inflection point of the closure curve
with the tangent at o9.max" The consequence of these definitions will be
discussed more fully in a later section. Curve 11 still shows an effect of the
overload on the low stress region of the closure curve as indicated by the two
inflection points on the curve. The crack has grown about 2 mm since
application of the overload as indicated by Curve 11. This distance is roughly
three times the plastic zone site generated during overload, in qualitative
agreement with Wei and Shih [6].
A typical example of the fracture surface features of the area, where the spike
overload has been applied, is shown in Fig. 3. A ledge, about 1 to 3/am high
(depending on the overload ratio) corresponds with the position of the crack at
the time the overload was applied. Striation spacing in front of the ledge is in
good agreement with the crack propagation rate observed acoustically. However,
the SEM pictures do not clearly reveal a crack extension of the magnitude
indicated by the acoustic measurements (Fig. 1). No striations have been found
for a zone greater than the measured crack growth; however, the lack of features
may be associated with the damage done by repeated loading during the delay.
Replicas taken in the vicinity of the ledges and examined by TEM have not
yielded conclusive results so far.

Discussion
The present results on constant-amplitude cycling are, in general, in very good
agreement with earlier observations [11,13] in that a gaseous, mildly aggressive
environment affects the crack propagation rate, Aa[2~V, the crack closure stress,
Oop , and, consequently, the effective stress intensity range, ZkKefr. Furthermore,
it was observed that both the size of spike overloads (as in Ref 6) and the
environment affect the subsequent delay in fatigue crack growth.
In particular, the present results seem to indicate that the acoustic technique to
observe crack propagation is also a valuable tool to study delay effects. As was
pointed out before, the closure results from the fifth sample at the higher stress
intensity level are amplified somewhat with respect to the other tests. Although
somewhat atypical in this respect, these results (Fig. 2) show the effects of the

0
Z
7~

C
m
C'~
-'n

7~
121
1"1"1
F-

p-
,,,d
FIG. 3-Scanning electron microscopy o f the fracture surface in the vicinity o f application o f an overload. The crack
propagation direction is f r o m b o t t o m to top with the application o f the load spike in the center of" the fractographs. The ledge
is indicated by an arrow. .-t

spike overload quite clearly: the crack tip stays dosed over the period of delay.
Once the crack starts to propagate again, AKeff increases slowly, which is
reflected in a much smaller ha/AN, even at the larger crack depth than before
application of the overload.
An attempt was made to quantize this observation by plotting Aa/AN versus
AK and AKeff . Both curves are shown in Fig. 4. The identifications given in this

L0 I I I I I I IIJ I I I I t I1

~Sm~ RH /

Y
IDENTIFICATION
"'~--2
",--I
A
t~

E
/ / "'--It

~10

""--9

K
0.0

-"--8

"-7

-4-6

0. 001 i I I I III I I I I I I I
1O I00
a K, ,~Keff (MN/m312)

FIG. 4-~a/ZkN versus ~ and AKeff, respectively. Identification as in Figs. 1 and 2. The
Z~KeffScatterband shows humidity experiments, obtained earlier [11].

figure again refer to the same identifications in Fig. 2. Before the overload is
applied both AK and the corresponding AKeff increase, with AKeff being
somewhat smaller than AK due to partial closure. After the overload Aa/AN
approaches zero, resulting in a vertical drop in the ha/AN versus AK curve. At
the same time AKefe goes to zero. As the crack starts to propagate again, AK
increases slowly above the value it had before the overload with a much reduced
An/AN. This results in a second branch of the ha/AN versus AK curve (6
through 11). At the same time ha/AN versus AKefe shows the same relationship
it had before the overload, resulting in a unique functional relation [9,11,13]
Aa/AN = A (AKeef)n for the crack propagation rate before and after the overload.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
SUCK ET AL ON FATIGUE CRACK DELAY IN AL 7075-T651 109
In other words, the effect of the overload is just to operate at a lower value of
~/f~ff after the overload but otherwise to follow the prior A,,/z~V versus AKeef
relation. Figure 4 also contains the Aa/AN versus AKeff scatterband on earlier
7075-T651 results [11] which were obtained by fatiguing the different gaseous
environments. As was shown, Aa/AN versus AKeff yields a unique functional
relation for three environments: dry N2, 15 percent RH, and 80 percent RH.
Although the present overload experiment is just at the lower boundary of the
humidity scatter band, both experiments are in good agreement on the exponent
n = 3.27 [11]. The values of AK and AKeff obtained during the overload have
not been plotted. The maximum K value during the overload approached the
expected fracture toughness of the material.
The Aao/AN versus AKeff data (Table 1) obtained from the four controlled
humidity experiments before application of the overload fell into the appro-
priate parts of the scatterband in Fig. 4, thus verifying the earlier results [11]. In
view of the limited data obtained so far, it is difficult to quantize the effects of
overload ratio and environment on the delay N o for this alloy. Indications from
the (tThmax /O'~max ) -----2.0 experiments are, however, that the delay N o is a
function of the form

N D ~xf [~/1____~ = f [(AKerf)-n] (1)

Assuming that this functional relation is merely

ND cx (AKeff)-n (2)
the relative change of the delay N D is in good agreement with the relative change
of AKerr (using n = 3.27 for (ahmax/a~a x ) = 2.0. Applying Eq 2 to the
(Ohmax /O]~ma x ) = 2.5 data one would estimate that the delay in the RH = 0
percent experiment should be about 660 000 cycles. Unfortunately, this
specimen broke at the grips after 560 000 cycles delay. Equations 1 or 2,
furthermore, would suggest that the delay decreases as the stress intensity
increases. Qualitatively, this is in agreement with observations on titanium alloys
[6], which show N D decreases as AK increases for a constant overload ratio.
Further indications are that for experiments at a constant Aa/AN (before
application of an overload), the delay N D is a strong function of the overload
ratio (O~rnax ]og_max ), again in qualitative agreement with observations on
titanium alloys [6]. Quantification of this effect has not as yet been
accomplished due to the limited data available.
The point most likely to be subject to criticism in the present paper is the way
the AKeff values have been derived. The authors do not deny the weaknesses in
their extrapolation technique used in the definition of the closure stress, Oop.
One has to keep in mind, however, that the present definitions yield information
on the relative changes in the closure stress, and this is the main objective of the
present investigations as well as of earlier ones [10,11,13]. Definition of aop by
the first deviation from the tangent to the closure curve at O~nax [9] would lead
to much smaller AKeff values than quoted in the present case. At present there

is no means to judge on the preference of one method over the other, however.
Future experiments have to decide on this question of the absolute value of
AKeff, as has been pointed out before [14].
One may speculate on the physical picture of the effects of single overload
spikes. A rough model based upon the residual strain produced by the
deformation in the plastic zone [9] (which does not intend to show actual
fracture features) is shown in Fig. 5. The identification numbers in Fig. 5 refer

FIG. 5-Model o f overload effects on the residual strain in the wake of the crack tip. (Not
intended to represent fracture features.} Identification as in Figs. 1, 2, and 4.

back to Fig. 2. At the top of Fig. 5 the crack is shown just before the overload
spike is applied. The shaded area symbolizes the residual strain causing closure.
The center part shows the large crack growth during the overload with the
increased residual strain due to an increase in the plastic zone size and the
amount of strain at the crack tip. This increased residual strain prevents the
crack from opening fully to the crack tip after the overload, thus causing the
crack growth delay. Furthermore, it seems likely that the ledges observed by
SEM (Fig. 3) are a result of increased crack tip opening displacement and
increased residual strain during the overload. Another effect predicted here
during unloading would be a large reduction of the area over which crack closure
occurs on the fracture surface, which indeed has been observed (compare Curve

2 and the unloading part of Curve 3 in Fig. 2). At the bottom of Fig. 5 a
situation is shown after the crack started to grow again. The residual strain is
reduced due to the smaller plastic zone size and reduced strain at the crack tip.
During unloading, the fracture surface will close down over the fracture surface
produced during the overload. As the crack continues to grow and move away
from the fracture surface produced during the overload, the effect of the
overload becomes less pronounced so that AKeff will increase with the
corresponding increase in crack growth rate.
It has to be mentioned that the closure phenomenon certainly is not only due
to the residual strain in the wake of the crack, but is also caused, to a certain
degree, by the residual stresses in the plastic zone. However, as has been
discussed before, in the present study the growth rate after an overload is
affected over distances many times the plastic zone diameter due to the
overload. Thus, it seems that the residual strain, as shown in Fig. 5, dominates
the delay effects after an overload.
Many more experiments are necessary to more fully quantize the delay due to
spike overloads in terms of AKeff. Furthermore, it will be necessary to include
block overloads into research of this kind, although some trends under such
conditions have been observed earlier [5,10]. In the aluminum alloy
2024-T851, it was noted [10] that crack growth is stabilized during the overload
block. Decreasing the maximum load by only 20 percent yielded a growth delay
with a large decrease in AKeff. Here, too, the AKeff increases slowly due to a
decrease in oop. The exact details of this delay had not been studied at that
time, due to experimental problems.

Conclusions
The following conclusions can be drawn from the present studies on the
aluminum alloy 7075-T651 :
1. The change in the crack growth rate due to a single overload can be
explained on the basis of crack closure and the corresponding AKeff.
2. Humidity strongly decreases the delay due to an overload which is also
explainable in terms of crack closure and AKeff.

Acknowledgment
The authors would like to acknowledge the experimental assistance of R. V.
Inman and discussions with M. W. Mahoney.

References
[1] Wheeler, O. E., "Crack Growth Under Spectrum Loading," General Dynamics
Report FZM 5602, 1970.
[2] Jonas,O. and Wei, R. P., InternationalJournal o/Fracture Mechanics, Vol. 7, 1971,
p. 116.
[3] Van Euw, E. F. J., Hertzberg, R. W., and Roberts, R. in Stress Analysis and Growth

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
112 FATIGUECRACKGROWTH
o f Cracks, ASTM STP 513, American Society for Testing and Materials, 1972, p.
230.
[4] Corbly, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973, p.
479.
[5] Trebules, V. W., Roberts, R., and Hertzberg, R. W. in Progress in Flaw Growth and
Fracture Toughness Testing, ASTM STP 536, American Society for Testing and
Materials, 1973, p. 115.
[6] Wei, R. P. and Shih, T. T., International Journal o f Fracture, Vol. 10, 1974, p. 77.
[7] Gardner, F. H. and Stephens, R. I., Fracture Toughness in Slow-Stable Cracking,
ASTM STP 559, American Society for Testing and Materials, 1974, p. 225.
[8] Adetifa, O. A. and Gowda, C. B., International Journal of Fracture, Vol. 10, 1974,
p. 614.
[9] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, p. 230.
[10] Buck, O., Ho, C. L., and Marcus, H. L., Engineering Fracture Mechanics, Vol. 5,
1973, p. 23.
[11 ] Buck, O., Frandsen, J. D., and Marcus, H. L., Engineering Fracture Mechanics, Vol.
7, 1975, p. 167.
[12] Buck, O., Ho, C. L., Marcus, H. L., and Thompson, R. B. in Stress Analysis and
Growth of Cracks, ASTM STP 513, American Society for Testing and Materials,
1972, p. 280.
[13] Buck, O., Frandsen, J. D., Ho, C. L., and Marcus, H. L. in Microstructure and
Design of Alloys, Proceedings, 3rd International Conference on Strength of Metals
and Alloys, Vol. 1, The Institute of Metals and The Iron and Steel Institute, 1973,
p. 462.
[14] Frandsen, J. D., Inman, R. V., and Buck, O., International Journal of Fracture, Vol.
11, 1975, p. 345.

Influences of Chemical and Thermal

Environments on Delay in a
Ti-6AI-4V Alloy

REFERENCE: Shih, T. T. and Wei, R. P., "Influences of Chemical and Thermal

Environments on Delay in a Ti-6AI-4V Alloy," Fatigue Crack Growth Under
Spectrum Loads, ASTM STP 595, American Society for Testing and Materials, 1976,
pp. 113-124.

ABSTRACT: The influences of chemical and thermal environments on delay in

fatigue crack growth under elemental load spectra were examined for a mill-annealed
Ti-6AI-4V alloy. The results indicate that delay is significantly affected by the hold
time at high load in 3.5 percent NaC1 solution at room temperature, and by both
temperature and thermal history. In 3.5 percent NaC1 solution, delay may be
increased by 40 times through increasing the hold time from zero to 900 s, the
specific amount depends on the overload ratio (Khmax/Ks and the maximum
value of the low-amplitude fatigue load (or K~.rnax). Delay is generally higher at the
higher test temperature (560~ or 293~ although the trend in behavior associated
with changes in loading variables is similar to that observed at room temperature.
Delay is significantly increased when the high load is applied at a high temperature,
with subsequent fatigue at room temperature. It is reduced by intermediate heating
at a high temperature following a high-load excursion at room temperature, and is
lowest when fatigue loads are applied at high temperature following a high-load
excursion at room temperature. The significance of these results in terms of modeling
and life prediction procedures is discussed.

KEY WORDS: crack propagation, fatigue (materials), delay, titanium alloys,

environmental tests, temperature, thermal environments

The importance of delay (retardation in the rate of fatigue crack growth),

produced by load,interaction in variable-amplitude loading, on the accurate
prediction of fatigue lives of engineering structures has been well recognized for
some time [1-11] .3 Delay properly refers to the period of abnormally low rate,
or approximately zero rate, of fatigue crack growth between a decrease in load
level and the establishment of a rate of growth commensurate with that for
constant-amplitude loading at the prevailing (lower) load; that is, between Points
a and c in a simplified schematic diagram shown in Fig. 1. It is usually measured

I Formerly, research assistant, Department of Mechanical Engineering and Mechanics,

Lehigh University, Bethlehem, Pa. 18015; now with Ebasco Services, Inc., New York, N. Y.
2 Professor of Mechanics, Department of Mechanical Engineering and Mechanics, Lehigh
University, Bethlehem, Pa. 18015.
3 The italic numbers in brackets refer to the list of references appended to this paper.

113
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
114 FATIGUE CRACK GROWTH

r
~t

NO. OF CYCLES

FIG. 1-Schematic illustration of delay in fatigue crack growth and definition of ND.

in terms of the number of elapsed load cycles. For experimental accuracy and
potential engineering utility, however, it is more convenient to define delay
(No) artificially as a period of zero crack growth, represented by the dashed line
segment (a-b) in Fig. 1, which is obtained by extrapolating the constant-
amplitude growth curve, Curve cd, to b. Following J o n ~ and Wei [5], this
definition of delay will be used here. The actual crack-growth response of
materials following a high load excursion is more complex, and is incorporated
indirectly in this definition for delay.
Systematic studies of delay on a mill-annealed Ti-6A1-4V alloy [5-8] showed
that the phenomenon is very complex and is affected by a broad range of
loading variables. The possible influence of chemical and thermal environments
on delay, however, has received little attention thus far. Although the influence
of environment on fatigue crack growth under constant-amplitude loading is
relatively well established [12-14], it is not clear that this information can be
applied directly to the case of fatigue crack growth under variable-amplitude
loads. Since engineering structures can be expected to encounter a wide range of
operating conditions, a systematic study of load and environment (chemical and
thermal) interactions on delay is needed. A brief examination of the effects of
chemical and thermal environments on delay was therefore carried out and is
reported here. The studies on the effect of chemical environment represent an
extension of the results reported previously [7], and are directed specifically at
the influence of hold time at high load on delay in a 3.5 percent NaCI (sodium
chloride) solution at room temperature. Studies of the effects of thermal
environment involve investigations of the influences of temperature and of
thermal history. Experiments were carried out within the framework of linear
fracture mechanics. The crack tip stress intensity factor K or AK was used to
characterize the mechanical crack driving force.

Material and Experimental Work

A 0.2-in.-thick (5.08-mm), mill-annealed, Ti-6AI4V alloy plate was used in
this investigation. The chemical composition, heat treatment, and longitudinal
and transverse tensile properties of this material are given in Refs 6 and 9.
Constant-load-amplitude fatigue crack growth data in several test environments
at room temperature, and in dehumidified argon at room temperature and at
about 580~ (305~ are also given in Refs 6 and 9.
Two-in.-wide (50.8-mm), wedge opening load (WOL) specimens and 3-in.-wide
(76.2-mm) center-cracked specimens, oriented in the long transverse (TL)
direction [15], were used [9]. The specimens were precracked in air at a stress
ratio, R, of 0.05 through a sequence of loads that reduced Kmax to a level that
is equal to or less than the selected starting Kmax level for the actual
experiments [16]. This precracking procedure provided fatigue cracks of at least
0.08 in. (2 mm) in length from the end(s) of the electro.discharge-machined
(EDM) starter notch, such that subsequent fatigue crack growth will be through
material that has not been altered by the notch preparation procedure and will
be unaffected by the starter notch geometry. The stress intensity factor, K, for
the WOL specimens was computed from Eq 1 [17,18]

K =Px/~"BW [ 3 0 . 9 6 - 1 9 5 . 8 ( ~ ) + 730.6(~) 2- 1186.3(_~)3 + 754.6(~)4] (1)

and from Eq 2, for the center-cracked specimens [19,20]

K= P ~/~'a s e c ~
w (2)
where P = applied load; a = crack length (or, half-crack length for the
center-cracked specimens); B = specimen thickness; and W = specimen width. A
continuous recording electrical potential system was used for monitoring crack
growth [9,21,22]. Resolution in the measurement of crack length, a, was
estimated to be about 0.001 in. (0.025 ram) for this material and these specimen
configurations [9]. The accuracy in crack length measurement was better than 1
percent.
The chemical environment used in these studies was 3.5 percent NaC1 solution.
The solution was made from reagent grade NaCI and triple-distilled water, and
was deaerated continuously in a reservoir and circulated through Plexiglas
environment chambers clamped onto the faces of the specimens. For conve-
nience, studies of the effects of temperature and thermal history were carried
out in air. Specimen heating was accomplished by the use of electrical resistance
heating tapes. The temperature was monitored continuously during the tests by
means of thermocouples spot welded to the test specimen. Temperature stability
was better than •176 (-+2.8~ during a delay test.
Delay experiments were carried out under tension-tension (sinusoidal) loading
in a closed-loop electrohydraulic testing machine operated in load control at 5

Hz. Deviations from this loading frequency or the sinusoidal waveform or both
were made only when a single or a small number of high load cycles were to be
applied (either manually or with the aid of the function generator). Load control
was estimated to be better than -+1 percent. Because previous experiments
indicated the existence of a significant effect on delay when a high-load
excursion occurs during delay produced by a previous high-load excursion [5,6],
and that this influence can be minimized or eliminated by allowing at least 0.08
in. (2 ram) of crack extension between the successive high-load excursions, such
a procedure was used in this investigation and permitted an average of six
individual delay experiments to be performed on each test specimen. (The actual
crack-growth increments between successive high-load excursions were selected
on the basis of crack growth response for the individual cases.) To minimize
error, delay (ND) data were determined directly from the electrical potential
(crack length) versus time (cycles) records.

Results and Discussions

Experimental results on the effects of chemical and thermal environments on
delay in fatigue crack growth will be reported and discussed separately.

Effect of Chemical Environment (3.5 percent NaCl Solution) on Delay at Room

Temperature
In a previous series of experiments [7], the effects of 3.5 percent NaCI
solution on delay in fatigue crack growth following single and multiple high-load
excursions were examined. The results for single high-load excursion tests (high
loads applied by manual control) did not conform to companion test results in
air and in dehumidified argon, and exhibited considerable scatter. The multiple
high-load cycle data indicated that delay may be dependent on the cycling
frequency, fh, for the high-amplitude loads. These results, along with prelimi-
nary test data, suggested that there may be a significant effect of hold time, t h,
at high load on delay for this alloy in 3.5 percent NaC1 solution. In this
investigation, a more detailed examination of the hold time effect was made.
Experimental work was limited to single high-load excursions at various K~rnax,
for prescribed values of Khmax/g~max, Rh, and R~. The test results are shown
in Figs. 2 through 4, and are summarized in the following.
1. For reference, single high-load excursion experiments with zero hold time,
using a triangular waveform at 0.5 Hz, were carried out. The results show that
delay, ND, decreases monotonically with increasing K~rnax at a constant ratio
ghmax /g~max , and increases with ghmax/g~max for fixed values of Ks max
(Fig. 2). They suggest that the previously observed anomalous behavior may
indeed reflect a hold time effect engendered by "manual" application of the
high loads [7].
2. Data on the effect of hold time (th), at Khmax/K~max = 2.0 and Khmin =
K~min = 0.05 K~max, indicate that delay, No, increases with hold time and

K lmox - MN - m'~
12 14 16 18 20 22 24 26
I I I I I I I I
Tria~ulor ~ v e , 0.5Hz

Khmox
-- Kfmox
- - 0 , 0 5 Kzmox

0 0~,. 0
' ~0 9~ x / K l r n ~ = 2"15

u
i

,,.I

I0 z
I I I 1 trio t
I0 12 14 16 18 20 22 24
Klmm, - ksi - in~Z

FIG. 2-Effect of Kg,nax and Khmax/Kp.ma x on delay in 3. 5 percent NaCI solution.

reaches a limit (Fig, 3). The effect of hold time on delay is dependent on
Ks For t h from zero4 to 900 s, N D can vary from about 103 to 4 x 10~
cycles (a 40-fold increase) at Ks -- 21 ksii ~ . (23.1 MN/m 3/2 ), and from
about 7 x 103 to 4 x 104 cycles (a six-fold increase) at Ks = 12 ksii,v~.
(13.2 MN/m 3/2 ). These results suggest that delay becomes much less sensitive
to K~max- at the longer hold times.
3. The same trend in the hold time effect was observed at Khmax /Ks ---
1.85, with K~max = 21 ksiv~. (23.2 MN/m 3/2 ) and ghmin = K~min = 0.05
Ks (Fig. 4). Here, delay varied from about 4 x 102 to 104 cycles with
increases in t h from zero to 900 s. Delay was shorter for the lower ratios of
&max /Ks
The experimental results suggest that delay in fatigue crack growth (No) for
this titanium alloy in 3.5 percent NaC1 solution, at a c o n s t a n t K h m a x / K I 2 m a •
may be represented by a surface in the N D - K ~ m a x - t h space as illustrated
schematically in Fig. 5. This surface (that is, N D) would be expected to increase
with increasing Khmax/Kgana x (Figs. 2 and 4). Delay produced by a single
high-load excursion with zero hold time (triangular waveform, Figs. 2 to 5)
decreased with increasing K~max in a weU-behaved manner, and is consistent

4 High Loads were applied using a triangular waveform at 0.5 Hz (see Fig. 2).

~, io4 o

i
~ (~.uMN-m%)
z

ac~ -- /~--Khmox = 2K lmax

9 / ~--K.max
-O.05Klr~x

, i, il ] , Iii I I III I i nnl u i in

9 I0 "1 I I0 10 2
t h - ~,,r

FIG. 3-Effect of hold time on delay with Khmax/K~nax = 2. 0 and two different values
o f Kp~nax in 3.5 percent NaCI solution.

-= Io* u
KhmOXlKimox = Z.O
i
--&

e...--e~ ,.ss--o/~-
'~ 9 . ~ &-'~" _

i O 0 5 Klmox

I0 "l I I0 I0 2
t h- seconds

FIG. 4-Effect o f hold time on delay with Kg~nax = 21 k s b ~ . and two different values of
K h m a x / K ~ a x in 3.5 percent NaCI solution.

~ ,o.!

ioz

FIG. 5-Three-dimensional representation o f effects of K~max and t h on delay in 3.5

percent NaCI solution.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
SHIH AND WEI ON CHEMICAL AND THERMAL ENVIRONMENTS 119

with results obtained in the other environments [5-7]. The small reduction in
delay in comparison with the other environments is consistent with the higher
rates of fatigue crack growth in 3.5 percent NaC1 solution for this alloy [7]. The
extensive increase in delay with increasing hold time is unique to the salt-water
environment. Although hold time effects have been observed previously on this
alloy in air [5], the increase in delay amounted to less than a factor of two for
hold times up to 15 h as compared to over one order of magnitude increase
observed here. The strong dependence of delay on hold time provides a
reasonable basis for explaining the anomalous behavior of the single high-load
excursion delay data reported previously [7]. A more extensive investigation of
hold time effects in 3.5 percent NaC1 solution is required to develop a clearer
understanding of this load/environment interaction effect in this titanium alloy.
Meanwhile, the significant influences of environment-induced frequency and
hold time effects on delay must be recognized and be taken into account in the
development of life prediction procedures. It is likely that each material-
environment system will need to be treated separately. Thus, generalization of
existing data should be avoided.
A plausible explanation for the increased delay at long hold times may be
advanced. As suggested by the ligament model [23], the material immediately
ahead of the crack tip has accumulated a substantial amount of fatigue damage
caused by the loading cycles prior to the high-load excursion. In the presence of
3.5 percent NaCI solution, stress corrosion cracking may occur during the
high-load excursion and cause the crack tip to penetrate through this zone of
damaged material. The extent of stress corrosion crack extension depends on
ghmax and t h. Thus, after the high-load excursion, the crack tip may now reside
in material that has experienced no prior fatigue damage. The residual
(compressive) stress introduced by the high-load excursion, however, is not
expected to be affected by stress corrosion cracking, since the zone of plastic
strain moves forward with the crack tip. It is suggested, therefore, that the
increase in delay is the consequence of a reduced (residual stress modified)
mechanical driving force encountering a more damage-resistant material. This
suggestion is consistent with the experimental observations that there was
significant stress corrosion crack growth during the high-load cycle [9], and that
the increase in N o with hold time, th, was substantially higher at the higher
Kg.max (Fig. 3) commensurate with greater amount of stress corrosion crack
extension. A more extensive study, however, will be needed to better quantify
this coupled load/environment interaction effect.

Effect of Thermal Environment on Delay (in Air)

In this part of the investigation, the influences of temperature and some
simple thermal history on delay under selected elemental load spectra were
examined. For simplicity, the experiments were carried out in air (with relative
humidity of 40 to 60 percent at room temperature). Experimental results are
shown in Figs. 6 to 9, and are summarized in this section. Wherever possible,
comparison data for the same load spectra at room temperature are also given.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
120 FATIGUE CRACK GROWTH

Klmox-MN-m"~'z
I0
~ ll
15
I
20
I
25I 30
I

l--Khmox

Id ~ ~'~ --K rain=O'C6Kzmox

i I i I i I ~ I ~ I
12 16 20 24 28
Klmox- ksiJ'~n.
FIG,6-Theinfluence of temperature on delay.

Klmox-MN-m'~t
18 22 26 30
' 1 ' I , I ' I

7 Khmox=28k~iJ'~'-j"-i

~5
\o~z4..~_~u
6 /~ KmiR=O

~J 560~
N3

R.T.
'

6 d., o'.2 o13 o'.. o'.5

STRESSRATIORs
FIG.7-Effectof Rj~on delay for fixed values of Khmax and Z~ 9 at two temperatures.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
SHIH AND WEI ON CHEMICAL A N D T H E R M A L ENVIRONMENTS 121

-=4 r 9 9 J,~
u

~3-

.J --Khmax = 2 Klmax
w 2--
--Klmox.,,142~-,/'~
(15.6 MN-mm'~)
I-- ~0

F---~ Hl~tino ot 560"F

t t I ~ I ,,.
I 2 3 16
INTERMEDIATE HEATING TIME" hr$
FIG. 8 - T h e effect o f intermediate heating at 560~ on delay at room temperature.

FIG. 9-Delay under changing temperature and loading conditions at Kp.max = 14.2
ksi~. and Khmax/Kgana x ----2.0.

1. For a single high-load excursion, delay at high temperature decreases with

increasing K~max at a given value of Khmax/K~max, and increases with
Khmax/Ks for fixed values of Ks (Fig. 6). These trends are identical to
those observed at room temperature [5,6]. Delay at 560~ (293~ however,
are some 2 to 4 times longer than at room temperature (Fig. 6), although

temperature had only a minor effect on the rate of fatigue crack growth [9,24].
2. For fixed values of AKh/AKs and Khmin , delay at high temperature is
strongly affected by Rs or by the ratio between Khmax and K~max (Fig. 7). If
Ks is equal to Khmax , no delay is experienced. These results are again
consistent with those obtained at room temperature [5,6].
3. The effect of intermediate heating at 560~ (293~ following a high-load
excursion, on delay was examined. Intermediate heating decreases delay at room
temperature by about 40 percent (Fig. 8). The reduction in delay appears to be
relatively unaffected by the intermediate heating time (Fig. 8).
4. Various simple combinations of changes in load and temperature were also
examined, and the results are summarized in Fig. 9 in order of increasing delay.
In Fig. 9, loadings c and d may be regarded as references for comparisons among
the different combinations. It is seen that multiple high-load excursions increase
delay at high temperature (Fig. 9e). This is consistent with previous results
obtained at room temperature. Comparisons between the various combinations
show that delay is longest when the high load is applied at a high temperature
followed by low-load cycling at room temperature (Fig. 9/) and is the shortest
when the high load is applied at room temperature and the low-load cycling at
high temperature.
These results show that the general behavior of delay with changing loading
conditions at an elevated temperature (560~ or 293~ is the same as that at
room temperature. Delay, however, is generally longer at the higher temperature,
Figs. 6 and 7. Since, for this alloy, relative humidity was found to have little
effect on delay [7] and the rate of fatigue crack growth is relatively insensitive
to temperature [9,24], the observed differences in delay are most likely related
to those differences pertaining to the high-load excursions. It is known that the
plastically deformed zone ahead of the crack tip is inversely proportional to the
yield strength, and that the yield strength of a material decreases with increasing
temperature. As such, the interaction zone (residual stress affected zone) would
tend to be larger at the higher temperature, and may account for the increased
delay. The increased plastic zone size also reduces constraint at the crack tip,
and would thereby increase delay [8]. The influence of yield strength on delay
was studied by using the same material heat treated to different strength levels
[25]. The results showed that delay was longer for the lower yield strength
material, and tend to support the foregoing rationale. The possible causes just
discussed are to be rationalized in terms of the influences of residual stresses
ahead of the crack tip. This residual stress concept, although difficult to
quantify, draws support from the observations that (a) intermediate heating
(some stress relieving) reduces delay (Fig. 8); (b) fatigue at elevated temperature,
following a high load excursion at room temperature, reduces delay (Fig. 9a);
and (c) high-load excursion at high temperature followed by fatigue at room
temperature tends to prolong delay (Fig. 9f). Even though residual stress may
not provide a complete explanation for delay, it certainly must be an important
part of any complete explanation.

Summary
The influences of chemical and thermal environments on delay in fatigue crack
growth under elemental load spectra were examined for a mill-annealed
Ti-6AI-4V alloy. Experiments carried out in 3.5 percent NaC1 solution showed a
significant effect of load/environment interactions. Specifically, delay was
strongly affected by the hold time at high load. With zero hold time, delay
decreased with Ks in a well-behaved manner and was slightly lower than that
observed in less aggressive environments (air, for example). By increasing hold
time from zero to 900 s, delay was increased by up to 40-fold depending on
K~max and Khmax/K~max. Tests at a high temperature (560~ or 293~
showed that delay was generally longer than that observed at room temperature
under identical loading conditions. Room temperature delay was reduced by
intermediate heating at 560~ (293~ Delay was found to be longest if the
high-load excursion occurred at high temperature and the subsequent fatigue was
at room temperature, and shortest if the temperature profile is reversed. These
results suggest that residual stresses ahead of the advancing crack play an
important role in delay, although this aspect of the problem is difficult to
quantify at this time. The significant effects of combined load and environment
(chemical and thermal) interactions must be recognized and be taken into
account in the development of life prediction procedures. It is likely that each
material-environment system will need to be treated separately, and that broad
generalization of existing data should be avoided.

Acknowledgment
Partial support of this work by the National Aeronautics and Space
Administration under Grant NGL 39-007-040, and by the Office of Naval
Research under Contract N00014-67-A-0370-0008, NR 036 097, is gratefully
acknowledged.
This paper is based on a part of the research performed in partial fulfillment of
requirements for a Ph.D. in Applied Mechanics, Lehigh University.

References
[1] Christensen, R. H. in Proceedings, Crack Propagation Symposium, Cranfield, The
College of Aeronautics, Vol. 2, 1961, p. 326.
[2] Schijve, J., Advances in Aeronautical Sciences, VoL 3-4, 1962, p. 387.
[3] Hudson, C. M. and Hardrath, H. F., "Effects of Changing Stress Amplitude on the
Rate of Fatigue-Crack Propagation in Two Aluminum Alloys," NASA TN D-960,
1961.
[4] Hudson, C. M. and Raju, K. N., "Investigation of Fatigue-Crack Growth under
Simple Variable-Amplitude Loading " NASA TN D-5702 1970.
[5] Jonas O. and Wel, R. P., Internauonal Journal of Fracture Mechamcs, Vol. 7, 1971,
p: 116.
[6] Wei, R. P. and Shih, T. T., International Journal of Fracture, Vol. 10, 1974, p. 77.
[7] Shih, T. T. and Wei, R. P. in Proceedings-The Prospects of Advanced Fracture
Mechanics, Noordhoff, 1975, p. 231.

[81 Shih, T. T. and Wei, R. P., Journal of Testing and Evaluation, Vol. 3, No. 1,, 1975,
p. 46-47.
I91 Shih, T. T., "Fatigue Crack Growth under Variable Amplitude Loading," Ph.D.
thesis, Lehigh University, Bethlehem, Pa., 1974.
[101 yon Euw, E. F. J., Hertzberg, R. H., and Roberts, R. in Stress Analysis and Growth
of Cracks, ASTM STP 513, American Society for Testing and Materials, 1972, p.
230.
Jill Trebules, V. W., Roberts, R., and Hertzberg, R. W. inProgress in Flaw Growth and
Fracture Toughness Testing, ASTM STP 536, American Society for Testing and
Materials, 1973, p. 115.
[12] Wei, R. P., International Journal of Fracture Mechanics, Vol. 4, 1968, p. 159.
[13] Wei, R. P., Engineering Fracture Mechanics, Vol. 1, 1970, p. 633.
[14] Hartman, A., International Journal of Fracture Mechanics, Vol. 1, 1965, p. 167.
[15] Goode, J.,MaterialsResearch and Standards, Vol. 1, 1961, p. 389.
[16] FitzGerald, J. and Wei, R. P., Journal of Testing and Evaluation, Vol. 2, No. 2,
1974, p. 67.
[17] Wilson, W. K., Analytical Determination of Stress Intensity Factors for the
Manjonie Brittle Fracture Test Specimen, Report No. WERL-0029-3, Westinghouse
Research Laboratories, 1965.
[18] Wilson, W. K., Optimization of WOL Brittle Fracture Test Specimen, Report No.
66-1B4BTLFR-R1, Westinghouse Research Laboratories, 1966.
[19] Brown, W. F., Jr., and Srawley, J. E., Plane Strain Crack Toughness Testing of
High-Strength Metallic Materials, ASTM STP 410, American Society for Testing and
Materials, 1966.
[20l lsida, M., "Crack Tip Stress Intensity Factors for the Tension of an Eccentrically
Cracked Strip," Lehigh University, Dept. of Mechanics Report, 1965.
[21 ] Johnson, H. H. Materials Research and Standards, Vol. 5, 1965, p. 442.
[22] Li, C. Y. and Wei, R. P., Materials Research and Standards, Vol. 6, 1966, p. 392.
[23] Majumdar, S. and Morrow, J. D., "Correlation Between Fatigue Crack Propagation
and Low Cycle Fatigue Properties," T & AM Report No. 364, University of Illinois,
Urbana, Ill., 1973.
[24] Wei, R. P. and Ritter, D. L., Journal of Materials, Vol. 7, No. 2, 1972, p. 240.
[25] Gallagher, J. P. and Hughes, T. F., "The Influence of Yield Strength on Overload
Affected Crack Growth Rate Behavior in 4340 Steel," AFFDL-TR-74-27, 1974.

DISCUSSION

R. L Stephens I (written discussion ] - I n Fig. 9, loadings c and d, your results

indicate delay was greater at 560"F than at room temperature with Rs =
Ks ~ 0. I would like to suggest the opposite might occur had you
tested with Rs equal to say - 1 . 0 . This idea is based upon greater cyclic plasticity
at elevated temperature and Rs = - 1 such that residual compressive stresses and
crack closure due to the overload are reduced. This in turn would decrease the
elevated temperature delay. Could the authors comment on this?
T. T. Shih and R. P. Wei (authors' closure)-It is the authors' experience that
the application of a compressive load following a high-load excursion reduces
delay. One would expect, therefore, fatigue at R~ < 0 would also reduce delay.
The authors can agree that the reversal in behavior suggested by Dr. Stephens is
within the realm of possibility (although very unlikely), but do not believe that
there is a valid basis or data to permit the type of extrapolation suggested by Dr.
Stephens at the present time.
1 Materials Engineering Division, The University of Iowa, Iowa City, Iowa 52242.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
S. 14/. Hopkins, 1 C. A. Rau, ~ G. R. Leverant, ~ and A. Yuen ~

Effect of Various Programmed

Overloads on the Threshold for
High-Frequency Fatigue Crack
Growth

REFERENCE: Hopkins, S. W., Rau, C. A., Leverant, G. R., and Yuen, A., "Effect of
Various Programmed Overloads on the Threshold for High-Frequency Fatigue Crack
Growth," Fatigue Crack Growth Under Spectrum Loads, ASTM STP 595, American
Society for Testing and Materials, 1976, pp. 125-141.

ABSTRACT: A threshold value of crack tip stress intensity range (AKtla) exists
below which Mode I cracks do not propagate in high-frequency fatigue. The response
of AKth, in a commercial nickel-base and titanium-base alloy, to single and multiple
cycle overloads (that is, the behavior of an overload modified threshold, AKtb*) has
been defined for a variety of fatigue conditions. Both alloys responded similarly and
showed an exponential increase of ~ t h * with increased magnitude of the overload.
The effects of overload rate, temperature, cycle shape and number, and the fatigue
conditions of mean stress and cyclic frequency have lesser, but significant, effects on
AKth*. Metallography and scanning electron fractography have been used to define
the changes in crack size and tip shape which contribute to the overload effects.

KEY WORDS: crack propagation, fatigue (materials), loads (forces), crack propaga-
tion, cyclic loads

The strong effects of spectrum loads on the rate of fatigue crack propagation
have been appreciated for some time [1] .2 The loading spectrums are often so
complex and involve such large numbers of small load excursions that there is a
need to define the limits below which load excursions do not contribute to the
crack growth rate (Aa/AN). Many loading spectrums, like those for rotating
equipment, consist of two major cyclic loads: low cycle fatigue (LCF) produced
by a relatively few number of large load (rotating speed or rpm) excursions, and
high cycle (vibratory) fatigue which may superimpose large numbers of
low-amplitude cycles on specific portions of the LCF cycle. For cases where
vibratory excitation occurs below the maximum LCF load, each LCF cycle acts
as an overload which may affect crack growth during the vibratory fatigue. The
present approach utilizes the observation that Mode I cracks do not extend in
fatigue for ranges of crack tip stress intensity factor below a threshold (AKth).
AKth can therefore be used directly to establish design allowables for some load
spectrums or as part of the lifetime prediction analysis of more complex load

1 Pratt Whitney Aircraft, Materials Engineering and Research Laboratory, E. Hartford,

Conn. 06108; Mr. Rau is now with Failure Analysis Associates, Palo Alto, Calif. 94304.
2 The italic numbers in brackets refer to the list of references appended to this paper.
125
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
126 FATIGUE CRACK GROWTH

spectrums. Because of experimental difficulties in valid precracking, much of the

early threshold data is suspect. Furthermore, the response of ,&Kth to various
operating conditions, including overload, has not previously been defined. The
present work determines AKth and AKth* (an overload modified threshold)for
two engineering alloy systems under several important special cases. Specifically,
AKth* is evaluated after single and multiple cycle overloads of various
magnitude under a range of fatigue (mean stress, frequency, temperature) and
overload (shape, temperature) conditions.

ExperimentalProcedure
Center-cracked sheet specimens, Fig. 1, were used to evaluate the threshold
and subsequent fatigue crack growth for two commercial alloys. The titanium-
base alloy (Ti-6A1-4V) specimens were machined from a pancake forging that

q 19MM

I F79MM

9s.2. t..,

~l.3mM
41.67_ ~(~t - ~
t +
12.7~m
VIEW A
STRESS-INTENSITY FACTOR

AK = AO"J-~ f(-~) - Ref."8

FIG. 1-Center notched crack growth and threshold specimen design.

had been forged at 1241 K (1775~ in the tx +/3 phase field and then annealed
at 1227 K (1750~ for 1 h, water quenched, and aged at 977 K (1300~ for 2
h. The chemical composition in weight percent was the following 0.02 to 0.033
carbon, O. 10 to 0.18 iron, 6.3 to 6.43 aluminum, 4.28 to 4.30 vanadium, 0.172
to 0.183 oxygen, 0.0050 to 0.0062 hydrogen, and 0.013 to 0.015 nitrogen. The
microstructure, Fig. 2a, consisted of 20 to 40 percent primary c~ and the

r"
o
z
o
<
m

0
z

FIG. 2 Photomicrographs showing the microstructure o f (a) Ti-6AI-4 F after being forged at 1241 K, annealed at 1227 K/1 h, water quenched, and m
aged at 977 K/2 h, and (bJ DS nickel-base alloy (MarM.200 plus hafnium} after being solutioned at 1477 K / 2 h, air cooled, aged at 1144 K / 3 2 h, and I
air cooled. 0
r

remainder Widmanstatten o~. The average a grain size, as determined by lineal

analysis, was 14/.an.
The nickel-base alloy (MarM-200 plus hafnium) specimens were machined
from two directionally-solidified (DS) castings so that loading was parallel to the
longitudinal growth direction. The DS alloy is composed of a 7 solid solution
matrix, precipitation hardened by ordered 7' particles and containing both bulk
and grain boundary carbides, Fig. 2b. The large grains are elongated in the
growth direction, have a <100> texture, and random transverse orientation. The
chemical composition was the following in weight percent: 4.8 to 5.0 aluminum,
0.009 to 0.018 boron, 0.11 to 0.16 carbon, 9.5 to 10.2 cobalt, 8.11 to 9.3
chromium, 0.03 to 0.08 iron, 1.9 to 2.25 hafnium, <0.001 manganese, 1.8 to
2.1 titanium, 0.04 to 0.06 silicon, 11.8 to 12.8 tungsten, and 1.0 (nominal)
columbium.
In past experience with the nickel-base alloy, it was found that ~Kth could
vary significantly if the final precracking was done at a temperature other than
the test temperature. For example, with a room temperature precrack, Z3dr~th =
7.6 MNm -3/2 , but with a 950 K (1250 ~ F) precrack, AKth = 4.4 MNm -3/2
when both specimens were tested at 950 K (1250~ and R (-~O'min/O'ma x ) =
0.695. Therefore, all specimens were precracked at 30 Hz and at the test
temperature. The fatigue precrack was extended a minimum of one sheet
thickness from the slot. The final step in precracking was done at a maximum
load less than or equal to the maximum load for the first step in the test
condition. Once the specimen was precracked, the AKth or AKth * value was
determined by cycling for 107 cycles and then examining either a cellulose
acetate replica or the specimen surface directly in an optical microscope at
xl000 magnification. If the crack had not propagated, then AK was incremented
up by 0.22 MNm -3/2 , keeping the ratio (R) constant, and cycled for an
additional 107 cycles. This procedure was continued until crack growth was
observed. The AKth or AKth * reported in this paper is the highest AK which did
not cause the crack to propagate in 107 cycles, that is, the load step just below
the growth step. Since crack extensions of 25 pm could be resolved, AKth or
$
AKth is equivalent to Aa/~uV < 2.5x10 -9 mm/cycle (10 -1 o in./cycle). Once
the AKth* was determined and crack growth occurred, the specimen was
overloaded again in various ways, and the resulting AKt~ * was evaluated in the
same manner. With this procedure, a number of ,Mrfth- values were obtained
from the same specimen. Care was taken to ensure that each successive
evaluation was not influenced by previous ones on the same specimen.
AKtt~* was determined after different magnitudes of the overload (ghmax)
ranging from no overload to Khmax approaching K c of the material. The
standard overload test sequence is shown in Fig. 3. This was done for five
different stress ratios R = (0.1, 0.3, 0.5, 0.7, and 0.9) in the titanium-base alloy
and three different R ratios (0.694, 0.785, and 0.887) in the nickel-base alloy.
Two different testing frequencies (1000 and 30 Hz) were examined and two
different overload rates were used (1 Hz and 1 min hold at maximum load) for
the titanium-base alloy. For the nickel-base alloy, three different overload

KkMAX

1
TIME
Precrack I Overload AKTN TEST
R: Constant
FIG. 3-Standard overload test sequence for determining AKth by uploading a non.
propagating crack.

temperatures (299, 866, and 1200 K) were also investigated, with AKth* always
being determined at 866 K (1100~

Experimental Results and Discussion

Basic Threshold and Slow Crack Growth

The basic AKth without overload is shown as a function of R in Fig. 4 for the
Ti-6AI-4V alloy at room temperature and for 30 and 1000-Hz testing
frequencies. The 1000-Hz frequency produces a slightly higher AKth than did
the 30 Hz testing frequency. The AKth at R = 0.1 is an extrapolated value
rather than a measured value because of the experimental problems involved in
obtaining a valid precrack without overloading. Irving and Beevers [2] have
stated that they observed a difference of 2.0 MNm-a/2 in the ~xgth at R = 0.35
for Ti-6A1-4V material cycled in high vacuum, depending on whether they
down-loaded with a growing crack or up-loaded a non-propagating crack to
obtain the AKth, with down-loading producing the lower results. 3 All the basic
AKth results presented here will necessarily be for up-loading to compare
directly with the overload results which can only be evaluated by up-loading
non-propagating cracks to keep the crack tip under the influence of the
overload's plastic zone. The basic Agth for the DS nickel alloy is shown in Fig. 4
as a function of R at 866 K and 1000 Hz frequency.
The basic Agth values for the Ti-6A1-4V at 294 K are about 50 percent of the
AKth for the nickel-base alloy at 866 K for the same R ratio, although both have
the same modulus of elasticity at their respective test conditions. The modulus

3 Down-load here means decreasing the load amplitude until a growing crack is brought to
"rest" (or stop), and up-load means increasing the load amplitude until a stationary crack
begins to grow.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
130 FATIGUECRACK GROWTH
%\
%%%x
4.5 .
9 x,,
%
A ~x
% -4.0
4.0 ~'- "q"-.. ",, "~'~
%x %%%%~% %\\ xx%
3.5 %'o,. "~ ~"x x ~
-3,0 [~

~2.5- D.S. Ni ALLOY "-1000Hz. "-% "*~ ~

z at T : 866 K ""',,,--~'%, "
-2.0
TI-6AL'4V 0-100011z. "". -o ",0
o-3OHz.
1.5. at T - 294 K 9 Extrapolated Tkrashold from
Overload Test.(3011z.)
1.0
~a o'.3 o~ o'.7 o'.9 1.0
II-ratio
FIG. 4-Basic threshold stress intensity factor as a function o f stress ratio for Ti.6AI-4 V at
294 K and 30 and 1000 Hz and for directionaIly solidified DS nickel alloy at 866 K and
1000 Hz.

of elasticity for Ti-6A1-4V at 294 K is 1.10 x 10 s MNm -2 (16 x 106 psi) and for
the nickel-base alloy at 866 K is 1.08 x 10 s MNm -2 (15.7 x 106 psi). Weiss and
Lal [3] have suggested that z~r('th is only a function of the material's modulus of
elasticity. The present results show that such models are at best crude
approximations of the basic threshold (L~tgth) that do not account for the
important effects of overloads.
The low crack growth rates (Aa/,SdV) as a function of stress intensity ranges
(AK) just larger than the basic ,~kgth were obtained and are shown in Fig. 5a for
the titanium-base alloy at 294 K for R = 0.5, 0.7, and 0.9 and in Fig. 5b for the
DS nickel alloy at 866 K for R = 0.785 and 0.887. It was noted that crack
growth rate just above ,:~tgth differed considerably between the two materials.
Specifically, in the titanium alloy, Fig. 5a, the crack growth rate jumps one
order of magnitude from below 2.5 x 10 -9 mm/cycle or no growth to 2.54 x
10-8 ram/cycle for R = 0.7, and only a factor of 5 to 1.3 x l0 -8 mm/cycle for
R = 0.9 when AK is increased 0.22 MNm -3/2 above z~/('th. With the DS nickel
alloy, Fig. 5b, the crack growth rate jumps three orders of magnitude to 2.0 x
10 -6 and 2.5 x 10 -6 ram/cycle f o r R = 0.785 and 0.887, respectively, when AK
is similarly incremented. Once this jump occurs, the crack growth rates are the
same for all three R-ratios in the titanium alloy and nearly the same in the
nickel-base alloy. In 2024-T3 aluminum, Schmidt and Paris [4] found the same
AK at a growth rate of 5 x 10 -7 mm/cycle from R = 0.5 to 0.8 at 300 Hz and
from R = 0.75 to 0.9 at 580 Hz. The observation from Fig. 5a is that AK to
produce A a / A N = 2.5 x l0 -7 mm/cycle is R-ratio independent; however, the
threshold Z~r(th at a growth rate of 2.5 x 10 -9 ram/cycle is strongly R-ratio

A K" KSI ~ 10 AK- K S I I ~

, , t J , J , 10
I i I I . . . .

D.S. Nickel Alloy

106- Temp. : 866 K /

Frequency " 1000 Hz. 9/

lOs_

qO s

167
~ 10 ~ D

E /
!
C-- 10
6 [
i Z
i
-o\ z

_10 9

Ti-6A -4V

ad s. ,, Temp. : 294 K IOs

J~
II
Frequency: 1000 Hz.
i
0 - R=0.9
" m "R=0.7
l o -R=0.5

~4 6 1610 /, 101o
r ~ , i T , , ,110
. . . . . . . 10 1.1
].1 AK-MNni~'2 AK-MNm3/2

FIG. 5-Fatigue crack growth rate immediately after the basic threshold as a function of
stress intensity factor for (a] Ti-6AI-4V at 1000 Hz and 294 K, and (b) DS nickel alloy at
1000 Hz and 866 K.

dependent. Great patience is obviously required to examine sufficiently slow

z~a/z~V where the large effects on z ~ t h are present. Schmidt and Paris [4] also
observed a lowering of the threshold with increasing cyclic frequency from 342
to 1000 Hz and suggested localized heating of the crack tip as a possible reason.
With the Ti-6AI-4V alloy, we do not observe a strong frequency effect.
Furthermore, the crack growth rate is insensitive to temperature up to 616 K
[5]. The slight increase in AKth with increased frequency is more likely the
result of less plasticity at the crack tip and less time for environmental
interaction.

ThreshoM After a Single Cycle Overload

The zSJfth is a strong function of the single cycle overload magnitude Khmax ,
as shown in Fig. 6 for the titanium alloy at 294 K and Fig. 7 for the DS nickel
alloy at 866 K. The open symbols represent the basic AKth * with its
corresponding Kmax and the filled-in symbols represent the AtKth after an

Kmax(Overload)- KSI
10
, 310 50
100 i 91
Ti-6 AL-4V
C>.R=.! at 30Hz.
'~.R=.3 at 30Hz.
~'R=.S at lO00Hz.
~.R=.7 at lO00Hz.
O.R=.9 at lO00Hz.
o . R at=30Hz. . 5
10-
.=
i=

...4---"

1.1 1
1'0 30 5'0
Kmax(Overload)-MNm89

FIG. 6-Overload modified threshoM stress intensity factor as a function of the maximum
single cycle overload stress intensity factor for Ti-6A l-4 V at 294 K and at both 1000 and 30
Hz. (Open symbols represent z~XKthwithout overload.)

K.~x (ovEItoAn),KSI,/IW'
0 10 20 ~ 40 so
10 I

I!
9'
sJ Ja sp~
8. /'e 9 f
R=.694--~ 82 // /'/~
7. ~ss s s. ~s eljs~
6.
/sJ 9 s ss s ~ 9 "5 e~
/ .+" qD/ / %~--R:.78S ..
~~"- " /o/ w"" I /s/
of
ss 94 <]
x4-

2.2 2
o ~o 2b ~'o 4'0 ~ 6o
K UAx(Overload),M N nl3/2
FIG. 7-Overload modified threshold stress intensity factor for the DS nickel alloy at 1000
tlz and 866 K after single cycle overloads to various Kma x. (Open symbols represent Z~th
without overload.)
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
HOPKINS ET AL ON OVERLOAD EFFECTS ON THRESHOLD 133

overload to the Khmax plotted. Figures 6 and 7 show that by overloading to

near the fracture toughness of the material, zlKth* at R = 0.1 can be increased
by 400 percent and ZkKth* at R = 0.9 can be increased by 50 percent. For
smaller overloads, the Z ~ t h * is increased by smaller amounts. For both alloys,
log ZkKth* increases nearly linearly with overload Khmax. Using the linear
relationship of log ~ffs versus Khmax along with the relationship ZkK = (1 -
R) Kmax, the basic zlKth at R = 0.1 in Fig. 4 was obtained by extrapolation.
In Fig. 4 it was shown that the basic threshold, zlKth, was slightly dependent
on fatigue frequency for the Ti-6A1-4V. However, Fig. 6 shows that the ~xKth*
after overload at R = 0.5 is not dependent on testing frequency between 30 and
1000 Hz. The specific shape and loading rate of the overload cycle affected the
resultant zlKth* in the Ti-6A1-4V as shown in Fig. 8. The standard overload

Kmax -(Overload)-KSIiv/~
0 110 3O 50
100 I I 91
Ti-6 AL "4V
R=0.5
Frequency:lO00 Hz.

Overload (IMin.) 7
Zero Load/
'E
Z
IE 10
-10 ""
i
z l
q
Z
cl

ii~ ~---Overload11 Hz.)

Zero Load

1.1
1'0 3'0 SO
K.Ax (Overload) - MNm-3"2

FIG. 8-Overload modified threshold stress intensity factor for Ti-6AL4 V after single cycle
overloads to Krnax at different overload rates at 294 K and 1000 Hz.

cycle was to load up to the maximum load (Khmax)and hold for 1 min, return
to zero load, and then load to mean load for the first attempt at determining
AKth* as shown in Fig. 3. The effect of the overload of magnitude Khmax was
increased (that is, AKth* increased) by not unloading to zero load and instead
returning directly to the mean toad for test after the overload as designated by
"Overload (1 min) Mean Load" in Fig. 8. Some of the apparent scatter in zlKth *
of the nickel alloy (Fig. 7) results from not unloading to zero after overload. By
not returning to zero load, the crack tip is apparently left more blunted and thus
produces a higher ~tKth*. Conversely, the effect of the overload on zlKth * is less
when applied at 1 Hz and eliminating the 1 min hold at maximum load as
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
134 FATIGUE CRACK GROWTH

designated by "Overload (1 Hz) Zero Load" in Fig. 8. This higher loading rate
apparently reduces the amount of crack tip deformation during overload and,
therefore, the effect of the overload.
Single cycle overloads were also applied to the nickel alloy at a lower (294 K)
and higher (1200 K) temperature, and the corresponding AKth* was obtained at
866 K after each overload, Fig. 9. The four different magnitude single cycle

~-Multi-LCF Overloads at 866 K.

9 -294K Overloads.
9 -1200 K Overloads.

KnAx (OVtJLOAe),KSIV ~
0 10 210 30 40 50
10
9. -8

.6
6-

'E 5-
z

~4-
i
"~
.3

2.21 , 2
0 1'0 2'0 3'0 4'0 5'0 60
K-Ax (Overload), MNm-3/2

FIG. 9-Overload modified threshold stress intensity factor o f DS nickel alloy at R =

O. 785, 866 K, and t 0 0 0 Hz as a function o f the overload magnitude Jbr multiple overload
cycles put on at 866 K and single cycle overloads put on at 294 and 1200 K compared with
the scatterband for single cycle overloads at 866 K.

overloads which were introduced at 294 K produced resultant AKth* at R =

0.78 and 866 K which were the same as the AKth* measured when the overload
was applied at 866 K. The two different magnitude single cycle overloads
introduced at 1200 K produced z~tgth* at R = 0.78 and 866 K which were on
the high boundary of the scatterband for the 866 K overloads. Because the yield
stresses for this alloy at 294,866, and 1200 K are 897, 890, and 573 MNm -2,
respectively, the size of the plastic zone in front of the crack tip due to K hmax
was different for each temperature. Even though the plastic zone size of
the overloads varied, the resultant AKth* was unchanged, which indicates that
the plastic zone size during overload cannot be the controlling parameter.
Multiple consecutive cycle overloads have also been investigated in a limited
number of tests. For the titanium-base alloy, 50 overload cycles were applied at

R = 0.1, and AKth* was then evaluated at both R = 0.5 and 0.9. With 50
consecutive overload cycles applied, the AKth* at R = 0.5 was only slightly (20
percent) above that after a single identical overload, while at R = 0.9, AKth*
after multiple overloads was indistinguishable from that after the single cycle
overload (Fig. 6). Similarly, for the DS nickel alloy, Fig. 9, A g t h * at R = 0.785,
was the same after 50 overload cycles as after a single overload of the same
magnitude. This limited multiple overload testing shows that multiple cycle
overloads do not produce markedly larger effects than a single overload.

Metallographyand Fractography
In order to better understand the mechanisms responsible for the large effects
of overloads on AKth*, extensive metallographic and scanning electron
fractography was performed. The crack length, path, and shape during and after
each overload was examined on the sheet surface. Since fractography revealed
some unusual crack front behavior near the sheet surface, for the DS nickel-base
alloy, some specimens were sectioned and examined along the mid-plane of the
sheet. Figure 10a shows the crack tip on the plate surface before an overload to

FIG. l O-Crack path for DS nickel alloy with a single cycle overload at R = O. 785 and 866
K. (a) Precrack before overload is applied (plate surface); (b) Crack growth due to a single
3/2
cycle overload o f 55 M N m - (plate surface); (c) Crack growth after single cycle overload
and threshold has been exceeded (plate surface); (d) Crack path at mid-thickness after single
cycle overload and threshold has been exceeded.

55 MNm-3/~ (almost Kc), and Fig. 10b is the crack tip after the overload. The
surface observations indicate that during the overload the crack grew approxi-
mately 0.25 mm along a plane other than the primary plane. Figure 10c shows
the same surface area after AKth had been exceeded, and the crack extended.
Note that on the sheet surface, the subsequent high cycle fatigue crack did not
propagate out of the overloaded crack tip. Figure 10d shows the same crack
location after polishing down to view the center thickness of the specimen. In
contrast to the surface behavior, the majority of the fatigue crack remained
planar through the overload zone. In this material, cracks on the surface often
deviated from planar and extended somewhat at ~K < AKth before arresting
permanently until AK was increased. Most of the crack away from the surface
did not extend until AKth was exceeded. Figure 1 la illustrates crack extension

FIG. l 1-Scanning electron fractographs of the DS nickel alloy after a single cycle
overload to 53 MNm -3/2 and threshold evaluation at 866 K and R = 0.694. (a) Low
magnification showing crack curvature and incremental surface growth; (bJ High magnifica-
tion showing the change in elevation at the center thickness due to the overload.

and arrest near the sheet surface at a typical fracture surface after a 53
MNm -3/2 overload. To ensure that these surface crack extensions were truly
arrested, further cycling at the same AK conditions was performed every time
surface extension occurred. No additional crack growth occurred in up to 5 x
107 additional cycles. The specimen shown in Fig. 1 la experienced five up-loads
before AKth*, and it can be seen that the surface crack length increased
incrementally and arrested while the center portion of the crack was stationary.
Figure 1 lb is a higher magnification of the center portion of the overload region
showing ductile crack advance during the overload and an elevation change of
the crack plane from before the overload to after the overload. The change in
elevation was measured from the fracture surface and found to be between 8 to

12/am. The value of the crack tip opening displacement (CTOD) was calculated
to be 7 #m using the approximate Eq 1 from Ref 6,

CTOD = 0.25 AK 2
(1)
gay s
where
E = Young's modulus,
AK = stress field intensity range, and
Oys = yield stress.
Scanning electron microscopy work on the DS nickel-base alloy specimens in
this investigation indicated, that during the larger overloads, the crack tip was
blunted in such a way that sharp corners were produced at the top and bottom
of the tip. After the overload, the high cycle fatigue crack propagated from
either the top or bottom corner of the overloaded crack tip, and in some cases
both. This crack tip bifurcation would certainly reduce the crack tip stress
intensity and thus contribute to the increased AKth that results from the
overload.
The fractographic observations of the Ti-6A1-4V are more difficult to interpret
due to the material's fine grain size which caused microscopic crack plane
irregularities. Replica fractography provided additional resolution and assisted in
identification of the following two regimes:
(a) Overload Khmax _< 33 MNm -3/2 -Macroscopically, the position of the
crack front at the overload (beach marks) is not well defined, Fig. 12a.
Microscopic observations in the overload areas reveal slight changes in the
fracture plane elevation and subtle differences in the crystallographic,
cleavage-like appearance [5].
(b) Overload Kmax > 33 MNm-3/2 -Macroscopically, distinct beach marks
on the fracture surface clearly delineate the crack front location at
overload application, Fig. 12b. Microscopically, each overload cycle is
accompanied by crack "advance by ductile rupture. The amount of crack
advance (width of the dimpled-rupture area) increases with increasing
Khmax of the overload.
In all cases, the fatigue growth just after Agth occursin the crystallographic
mode and proceeds from the overloaded crack tip.

General Discussion

The threshold (AKth*), below which cracks do not grow in Mode I fatigue,
has been shown to increase markedly with magnitude of a prior overload.
Figures 6 and 7 show that with overloads much larger absolute increases in
/~'th* can be obtained at low stress ratios than at higher stress ratios. The main
reason for this is that a much larger percentage overload can be applied at the
lower stress ratios without exceeding the fracture toughness of the material.
Figure 13 shows the same results normalized and plotted as relative threshold

FIG. t 2-Scanning electron fractographs of Ti-6AI-4V after single cycle overloads and the
threshoM evaluations at 1000 Hz and 294 K: (a} Overload Kmax = 2 7.5 MNm -a/2 showing
no apparent difference in fracture appearance due to the overload; (b) Overload K _ . ~ = 66
MNm ~ _ a / 2
showing crack advancement by ducttle rupture durmg the smgle cycle overload.
. . . r~Lu,.,.

Z~t~th*/Z~tgth , where AKth is the threshold after a given overload to K hmax

and AKth is the basic threshold at a given stress ratio and temperature, against
relative overload Khmax /Kmax , where Kmax is that reached in the basic zkKth
evaluation. Normalized in this way, data for all stress ratios investigated fall on
the same line indicating that the effect of relative overload on relative threshold
is independent of R. Both materials behave similarly, although the threshold of
the DS nickel superalloy is slightly more sensitive to overloads. Therefore, it may
be possible to write a relationship for ~sgth after overload which is
independent of R and only slightly dependent on microstructure. However, since
under the conditions investigated both materials had almost the same modulus
of elasticity and fracture toughness, more work with different materials and
overload conditions needs to be done to examine this and other possible
relationships.
The observation of crack closure at loads greater than zero has led to the
definition of an effective AK, less than the applied AK, which better predicts
many fatigue crack growth effects. The contribution of closure effects to the
overload effect on AKtn has been considered. For Ti-6A1-4V Shih and Wei [7]
did not observe crack closure above R --- 0.3. In the present work, we made some
limited closure measurements which agree with their results. In fact, for single
cycle overloads, we measured crack opening at K o = 0.25 Khmax . Effective
Z~r~th can be calculated by subtracting K o or gmi n , whichever is larger, from
Kmax of the fatigue cycle. Analysis of the effective ~ t h data indicated that

S- KOI . . . . . . . .

~e
,1=

e'~.
9
Ig
2- I9 9
/
0; Ti-Q AL'4V [,S, Nickel Alloy
11
9 ~ 294 K 866 K
9 - ~ r x'R=O.l A'R-----0.694

~J o ~ *.1:o.887
_ ~[ ~ R:0.7
y r 1=0.0

KoL/Ke
FIG. 13-The relative change in fatigue threshoM after single cycle overloads as a function
of the relative overload for both alloys and all stress ratios.

crack closure, in addition to residual compressive stresses at the crack tip, could
contribute to the increase in AKth* due to the overloads. However, closure is
not a factor at high values of R.
The effect of overloads on subsequent AKth have been shown to be quite large
and reproducible. The effect appears to be nearly independent of micro-
structure. However, the observations are such that no single mechanism appears
responsible. For instance, for overloads at higher temperature where the yield
stress is much lower, much more crack tip blunting and residual strains should
occur; but AKth was only marginally higher. Perhaps the lower flow stress at
high temperature reduces the magnitude of beneficial residual compressive
stresses. Crack tip blunting, bifurcation, closure, and residual stress fields may all
contribute to the overload effect, and additional work is necessary to
quantitatively define their individual and cumulative effects.

Conclusions

1. The basic fatigue threshold decreases nearly linearly with increasing stress
ratio (R) for both alloys and is frequency dependent for Ti-6A1-4V.

2. The overload modified fatigue threshold increases exponentially with

magnitude of the prior overload.
3. At low R, overloads produce much larger absolute magnitude increases in
the threshold than at high R; but the effect of relative overload on relative
threshold is independent of R, as shown in Fig. 13.
4. The number of overloads and their detailed shape and rate can affect
threshold but are much less important than overload magnitude.
5. Fatigue thresholds after overloads of various magnitudes can be extrap-
olated to obtain the basic threshold at low R ratios where valid precracking is
impractical.

Acknowledgments
The authors wish to acknowledge the assistance of K. V. Mattson and R.
Perkins for the care they took in the testing of these specimens.

References
[1] FractureMechanics of Aircraft Structures, Liebowitz, H., Ed., AGARD-AG-176,
Technical Editing and Reproduction Ltd., London, 1974.
[2] Irving, P. E. and Beevers, C. J., Metallurgical Transactions, Vol. 5, 1974, p. 391.
[3] Weiss,V. and Lal, D. N., Metallurgical Transactions, Vol. 5, 1974, p. 1946.
[d] Schmidt, R. A. and Paris, P. C. in Progress in Flaw Growth and Fracture Toughness
Testing, ASTM STP 536, American Society for Testing and Materials, 1973, p. 79.
[5] Yuen, A., Hopkins, S. W., Leverant, G. R., and Ran, C. A., Metallurgical
Transactions, Vol. 5, 1974, p. 1833.
[6] Hahn, G. T., Hoagland, R. G., and Rosenfield, A. R., Metallurgical Transactions,
Vol. 3, 1972, p. 1189.
[ 7] Shih, T. T. and Wei, R. P., Engineering Fracture Mechanics, Vol. 6, 1974, p. 19.
[8] Koiter, W. T., "Note on the Stress Intensity Factors for Sheet Strips with Cracks
under Tensile Loads," Technological University, Department of Mechanical
Engineering, Report No. 314, 1965.

DISCUSSION

Otto Buck ~ (written discuss&n)-You showed a figure that indicates a

dependence of Kth on R for Ti-6-4. How much of the effect is due to precracking
at low R? I also wanted to point out that one ought to be very careful with
threshold data since the crack closure load changes rapidly at low Aa/AN. The
effect is that threshold is much lower than you showed in your figure. 2
S. W. Hopkins, C. A. Rau, G. R. Leverant, and A. Yuen (authors' closure)-It
is agreed that one ought to be very careful with threshold data. Specifically at
low stress ratios, the ability to obtain a valid precrack without exceeding K max

1 Science Center, Rockwell International, Thousand Oaks, Calif. 91360.

2 Frandsen, J., Inman, R. V., and Buck, O., International Journal of Fracture Mechanics,
Vol. 11, 1975, p. 345.

of the threshold test is difficult. For all AKth data presented in this paper, low
stress ratio (R ~ 0) precracks were used at various final Kmax conditions which
never exceeded Kmax of the threshold test.
Additional work has since been done on this titanium alloy which shows an
effect of precracking condition on the low stress ratio threshold value and no
effect of precracking condition on the high stress ratio threshold value. Tubular
specimens were precracked at various negative stress ratios down to R = -9.0
prior to obtaining AKth at positive stress ratios. The largest reduction in AKth
was 1 MNm -3/2 for all the various negative stress ratio precracking conditions
investigated and this occurred at R = 0.1; the reduction in AKth value was
reduced with increasing stress ratio. At R = 0.7 and above no reduction in AKth
was obtained regardless of the precracking condition. For applications where
low-amplitude high cycle (vibratory) fatigue is superimposed on high-amplitude
low cycle fatigue, the high stress ratio results are of most technological
importance. Under these conditions the precracking effects do not exist. These
new results of negative mean stress precracking will be reported in the future.

A Model for Fatigue Crack Growth

Delay Under Two-Level Block Loads

REFERENCE: Adetifa, O. A., Gowda, C. V. B., and Topper, T. H., "A Model for
Fatigue Crack Growth Delay Under Two-Level Block Loads," Fatigue Crack Growth
Under Spectrum Loads, ASTM STP 595, American Society for Testing and Materials,
1976, pp. 142-156.

ABSTRACT: After reviewing previous attempts to quantify delay, a composite stress

intensity parameter, (Kp,ma /Khmax) 9 AK~, is proposed to correlate the number of
delay cycles in simple two-~evel vartable-load amphtude tests for various combina-
tions of stress levels. Its application is restricted in the present investigation to room
temperature tests in an air environment. Delay (cycles) is defined herein as the period
of apparent zero crack growth after the overload, and the stress intensity factors,
Kpmaax and Khmax refer to the stress intensities accompanying the lower and higher
load levels, respectively. It is hypothesized on the basis of experimental observations
that material under different two-level block loading sequences, but having equal
values of the parameter, will experience the same delay.
Good correlations for data on aluminum alloys, titanium alloys, and carbon steel
taken from the work of several investigators are obtained using the parameter. An
empirical model relating the delay to a power function of the parameter is suggested.
Limitations of the model are pointed out and discussed. The engineering significances
of the parameter with respect to the determination of crack propagation life and
crack arrest conditions are discussed.

KEY WORDS: fatigue (materials), crack propagation, stress ratio, aluminum alloys,
titanium alloys, carbon steels, delay cycles, loads (forces), cyclic loads

Qualitatively, the effects of tensile peak overloads on fatigue crack growth at a

lower level of cyclic load are now well known. A general conclusion reached by
most investigators [1-6] 3 is that high tensile overloads will cause crack growth
rate retardation (or delay), and sometimes even crack arrest, during subsequent
cycling at a constant lower load level. From the point of view of safely
predicting crack life endurance, this is beneficial because crack propagation life
predictions based on the usual crack propagation rate expression will be smaller
than the actual life. However, for better and less conservative predictions of
crack propagation lives for engineering structures subjected to variable amplitude
loading, delay effects should be considered.
Several explanations of the fatigue crack delay phenomenon have been

1 Research student and professor, respectively, Department of Civil Engineering,

University of Waterloo, Waterloo, Ontario N2L 3G 1, Canada.
2 Senior engineer, Advanced Reactors Division, Westinghouse Corp., Madison, Pa. 15663.
3 The italic numbers in brackets refer to the list of references appended to this paper.

142
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions author
ADETIFA ET AL ON DELAY UNDER TWO-LEVEL BLOCK LOADS 143

proposed [1,2, 7-9] but the most often discussed theories in this regard are: (a)
the residual compressive stresses theory [1,2], (b) the crack closure theory [7],
and (c) the crack tip blunting theory [8]. At present, no single mechanism
appears to be capable of satisfactorily explaining all the features of mixing high
and low load cycles on fatigue crack propagation. In reality, delay and other
load interaction effects are probably caused by a combination of these
mechanisms.
Few attempts [1,3,5,10-13] have been made to quantify the delaying effect.
An accurate quantitative description of this effect, which would allow the
prediction of the crack growth retardation behavior of several materials without
recourse to testing each individual material, would be invaluable to analysts and
designers. The objective of this paper is to present a simple method of predicting
delay cycle trends due to a two-level loading. Before the development of this
method, previous attempts to quantify delay will be briefly reviewed in order to
put the method in perspective.

Previous Quantitative Models

Hudson, Hardrath, and McEvily [1,3,5] made what are probably the first
serious attempts to correlate the number of delay cycles with the applied stress
levels. Graphs of the delay cycles versus the low stress for different peak stress
levels were presented for 2024-T3 and 7075-T6 aluminum alloys. Stulen et al
[10] and Wei et al [14] also presented data in graphical forms for titanium
alloys. Essentially, these graphical approaches are attempts to produce "delay
stress-life (S-N) curves" for the materials investigated.
Other methods used to quantify delay were based on data obtained from
constant stress intensity tests [11-13,15]. These methods assume that the
average growth rate during the retardation period can be determined using an
effective stress intensity in the basic growth rate equation [7]. For 2024-T3
aluminum subjected to a single peak overload, Von Euw [11] proposed the
parameter ro/bN o * as a measure of delay (ro is the overload plastic zone, b is
the steady growth rate after the overload, and N o is the number of delay cycles
corresponding to ro), while Probst and Hillberry [12] proposed the following
expression for the same material (R = 0.3)

ND = ro (1)
C [K~ max - EKhmax ] n
where K~max is the maximum lower load stress intensity factor, Khmax is the
maximum high load or overload stress intensity factor, E is a constant relating
the critical stress intensity to cause crack arrest and the overload stress intensity
(to be experimentally determined), C and n are constants from the constant-
amplitude growth rate expression (da/dN = CAKn). Equation 1 is applicable for
cases when Khmax > K~max and (K~max - EKhmax ) > 0.
Using an effective stress intensity concept, Himmelein and Hillberry [13] and

Gray and Gallagher [15] obtained N D by an integration of the constant

amplitude crack propagation equation within the overload plastic zone. The
overload residual stress intensity factor was experimentally determined by
Himmelein and Hillberry while Gray and Gallagher used a modified form of the
Wheeler retardation model [16] to account for load interaction effects. In all of
these methods, it was found that for a single peak overload, the correlation
between predictions and experimental data was within a factor of two.
In this paper, a simple phenomenological method of determining the number
of delay cycles for simple two-level loading is presented. A composite stress
intensity parameter is introduced to correlate the delay for single peak overloads
and for high-low block loadings for a variety of materials. When a consistent
definition of delay period is adopted, the proposed parameter provides a good
correlation of delay trends for two-level loadings.

Composite Stress Intensity Parameter

One of the difficulties encountered in using work from different investigations
is the lack of a consistent definition of delay. Consequently, delay will be
defined herein as the period of apparent zero macrocrack growth following a
peak overload or a block of overloads and is determined as illustrated in Fig. 1.
Experimental results [1,14,17-19] have shown delay to depend on a large
number of variables including AK~, AKh, R~, Rh, Nh, t h, and metallurgical,
environmental, and geometrical factors (temperature, thickness, yield stress
(Oys), strain hardening (n'), and stress intensity gradient (dK/da)). In functional
notation
N D = f[AKs ZXKh,Rs th, ay s, n', .......... ] (2)
where AK is the stress intensity factor range, R is the stress ratio (R --
Kmin/Kma x ), the subscript s is for the low load level, h is for the high load level,
N h is the number of applications of the high load, and t h is the hold time at the
high load level. The present work deals only with the influence of a subset of
these variables while others are held constant or treated separately. Specifically,
variations in delay with N h are bounded by dealing with the extreme cases in
which N h is one or very large. Environmental and hold time effects are
minimized by considering only results from room temperature tests in air
without deliberate hold periods. Metallurgical and thickness effects are dealt
with by considering materials and stress states (plane strain or plane stress)
separately. Since only step changes in load were introduced in the two-level tests
of this study, the effects of (dK/da) on delay is expected to be constant for all
cases considered. Within these restrictions, the following hypothesis relating the
delay cycles to various values of the variables in Eq 2 is now advanced.
Hypothesis-The composite stress intensity parameter (Ks max/Khrnax ) " zXK~
governs the number of delay cycles in variable-amplitude load fatigue crack
propagation, and, as a consequence, delay will be the same for a given Nh and
material in all two-level load sequences having the same value of the parameter.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorize
ADETIFA ET AL ON DELAY UNDER TWO-LEVEL BLOCK LOADS 145

N
(0) / d
I--
(.9
Z
14J
--I

~ - -- ~/bo
n,"
0
~ I- z"" ~1

No. OF CYCLES

//~- CONSTANT P~
(b) /' \ .-"

*6""

No. OF CYCLES
FIG. 1-Schematic illustration of delay in fatigue crack growth and definition of delay
cycles, ND [20].

The validity and usefulness of the composite stress intensity parameter will
now be examined for various metals.

Delay Data Correlation

The preceding hypothesis is tested for the materials listed in Table 1. The
results for API X-65 steel are from tests conducted at the University of Waterloo
while results for other materials are taken from the work of other investigators
(referenced in Table 1), and delay is determined as shown in Fig. 1. All of the
tests were load controlled, except those reported in Refs 11 thru 13 which
were constant-K tests. The tests were conducted under zero-tension or
tension-tension sinusoidal loading. Specimen types used by various investigators
included a compact tension specimen, a center-notched specimen, and single
edge notch specimen.
Figures 2 thru 8 show correlations between the proposed composite stress
intensity parameter and delay for the various materials. Different correlations
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
T A B L E 1 - Mechanical properties of metals investigated.

Yield Stress Ultimate Elongation "11

(0.2% Offset), Strength, in 2-in. Gage % Reduction
Material M N / m 2 (ksi) M N / m 2 (ksi) Length, % in Area Ref S p e c i m e n Details -t
c
m
Ti-6AI-4V 923.2 (133.9) 984.6 (142.8) 13.3 [20] center-notched specimen o
(titanium) t -----1.6 m m (0.06 in.)

975.6 (141.5) 1042.5 (151.2) 12.0 [141 center-notched specimen

t = 5.08 m m (0.20 in.) DO
0
2024-T3 358.9 (52.05) 497.4 (72.14) 21.0 [1] center-notched specimen
-1-
(aluminum) t = 2.29 m m (0.09 in.)

7075-T6 520.6 (75.50) 571.9 (82.94) 12.0 [1] center-notched specimen

(aluminum) t = 2.29 m m (0.09 in.)

SAE 1020 627.4 (91.0) 668.8 (97.0) . . . [211 acompaet tension specimen
cold rolled steel t = 6.35 m m (0.25 in.)

Auste~iticmanganese 420.6 (61.0) 1048.0 (152.0) 49.0(1") 34 I221 acompact tension specimen
steel t = 6.35 m m (0.25 in.)

A P I X - 6 5 steel b 449.0 (65.1) 688.3 (99.82) ... 51 University o f center-notched specimen

Waterloo t = 2.54 m m (0.10 in.)
modified compact specimen
, t = 6.35 m m (0.25 in.)

Ti-SAI-1Mo-IV 937.69 (136.0) 1051.45 (152.5) 12.5 I91 center-notched specimen

(titanium) t = 1.14 m m (0.045 in.)

a Flat fracture surface reported.

b
Cyclic strain hardening material.

(a) TI - 6 A f - 4 V (TITANIUM)
IOO SINGLE PEAK OVERLOAD I00
'E
50 5O
Z

0
o o

o ,,TONES (20)
-~ Rt VARIES ( R h , O ) /
~, R[ - R h 9 0 -~ (14)

, , I , , I , , I ,, , , ] , , I , J i , , I
I0 0 I01 tO2 I0 3 104 105 I0 6 i0 7
DELAY CYCLES, N D

FIG. 2-Relationship between (Kp.max/Khmax) " 12~K~and delay cycles for Ti-6AI-4Y
(titanium).

(b) 2 0 2 4 - T3 ALUMINUM
IOC -- | REE(I) CONSTANT LOAD'~L HI-LO TWO STEP I00
'E 0 REF,(II) CONSTANT-K J LOAD (R-O,O)
z 5O 50
2
v

; =;,c
o o ,_
9 [] I0 "~_==-
ir

:r "" 5 R L 0.~ 5
: REF" (12) R ^ _ t SINGLE OVERLOAD' ~ ,e,
REE(.~ ~'''ICONS-~N:r-~-'FES;~ ~ o.-
v REF.(II) RL=O ]
X

, I , ~ I , , 1 ~ , I I , I , = I , , I
i00 i01 102 103 104 iOs IOs IO
DELAY CYCLES, NO
FIG. 3-Relationship between (KpJnax/Khmax) 9 Z~j~ and delay cycles for 2024-7"3
(aluminum).

are shown for the single-cycle peak load and for high-low two-step tests. The
plots imply an empirical relationship for N o of the form

N D = A "Ks max. z~r~l?7 m (3)

Khmax
-j
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
148 FATIGUE CRACK GROWTH

,o o o IO "

5 5
o REF. ( I )

, * [ I I I , I I , , I , * [ , , 1 , L [
i0 o 101 102 103 104 105 iOs 107
DELAY CYCLES, ND

FIG. 4-Relationship between (Kp.max/Khmax) " ZkK~ and delay cycles [or 7075-T6
(aluminum).

(d) AUSTENITIC MANGANESE ,STEEL

SAE 1020 STEEL I00

"~ 10(2- (COLD ROLLED)

z 50 - - 50
=E

*1; ,o_
o o
-10
"N
E =
| REF. (17) SAE 1020 --5 v
__iv 5 A REF. (20) AUST,MANGANESE STEEL

, I l L I I , , J , : [ , J I , , I , , [
100 101 102 103 104 105 106 107
DELAY CYCLES, ND
FIG. 5-Relationship between (Kp~nax/Khrnax) " z~xK~and delay cycles for SAE 1020
steel and austenitic manganese steel.

where A and rn are constants for the data used in this work. Values of A and m
obtained from a least-square fit of the plots are given in Table 2. It should be
noted that Eq 3 is valid for only Khmax greater than Ks (for constant
amplitude loading Khmax = K~max, and Eq 3 would predict delay, which is not
correct).

Data plotted in Figs. 6 and 8 are obtained for different values of Rs and R h (R
_> 0). In Fig. 6, the data for R > 0 lie below the data for R = 0, indicating that
the delay for the former case is less than for the latter. There is no data available
for R = 0 for the titanium alloy in Fig. 8, however, it is evident from the graph

(e) AP~ x65 STEEL

HI-LO TWO STEP TEST
ioo - Ioo
'E
z 50 -50

;;,o
o o
--I0
~

v 5 o RI,Rh 9 0
D Rt, Rh VARY (0.17 to 0.60)

J , I = , I , , I , , I J , I , ~ I , , l
I0 0 I01 i0 2 I0 3 104 I0 5 106 I0 7
DELAY CYCLES, NO

FIG. 6-Relationship between (K~raax/Khmax) 9 z~iK~and delay cycles for API-X65 steel
(high-low),

N (f) API X65 STEEL

lOO - - I00
SINGLE PEAK OVERLOAD

z 50- -50
~E

O R! = Rh = 0.0 O~,,~,~ -IO 9

I00 10 1 102 103 iO4 I05 106 107

DELAY C Y C L E S , ND
FIG. 7-Relationship between (Kpanax/Khmax) 9 ~lKj~and delay cycles for API-X65 steel
(SO).

(g) Ti - 8A! - IMo - I V (TITANIUM) [

~E IO0-- HI- LO TWO STEP TEST -~100

o e h 9 o.z5 . [
e~ 9 (o.sr too.6r) ~ 4~ 9
,~ ~ I0-- rn Rh =(0.07 1o0'33' ~,~.~,_ l "~e
Rr = (0-30 ~o o sr) '~L,~ [ "--
v i:=: 5 -- a R h = (0-07 to 0.60) - ~ f15

i0 0 i01 102 103 104 10 5 i0 s 107

DELAY CYCLES, NO

FIG. 8-Relationship between (KpJnax/Khmax) 9 AK~ and delay cycles for Ti-8AI-IMo-1 Y
(titanium).

T A B L E 2 - Values o f A and m in Eq 3.

MATERIAL Aa m

8
Ti-6A1-4V 4.35 x 10 --5.5
(titanium)
11
2024-T3 b 1.21 x 10 ---5.69
(aluminum)
10
7075-T6 3.28 x 10 -5.86
(aluminum)
9
SAE 1020 steel 5.25 x 10 --4.22
12
Austenitic 2.43 x 10 -4.95
manganese steel
11
API-X65 steel 9.1 x 10 -5.72
(high-low)
9
(Single overload) 7.57 x 10 -4.49
6
Ti-8AI-IMo-IV 3.67 x 10 -3.01
(titanium)

1
a U n i t s in
(ksix/~ m
b High-low only.

that for a range of values of Rs and R h, the data fall essentially within a narrow
band. If it is assumed that data for R = 0 for the material will lie above those for
R > 0, as is the case in Fig. 6, conservative predictions for the former case will
be obtained from Fig. 8.
Figure 3 gives a plot of the data for single overload constant-K tests [11,12].
The parameter correlates relatively well with the data by Probst and Hillberry
[12], but the correlation is poor and there is considerable scatter in the data
taken from Von Euw [11].
It is interesting to note that specimen type and thickness do not appear to
greatly affect delay. In Figs. 6 and 7, data obtained from a 2.54-mm-thick
center-notched specimen cannot be distinguished from those obtained from a
6.35-mm-thick modified compact tension specimen. This finding is contrary to
that recently reported by Shih and Wei [19] who found delay to be a function
of specimen thickness for a 7075-T6 aluminum alloy. It may be significant that
the specimens for which data are given in Figs. 6 and 7 all showed the slant
fracture typical of plane stress. Thus, if a change in delay with specimen
thickness is due to differences between a plane strain mode and a plane stress
mode of cracking, the thickness effect is not expected in test results shown in
Figs. 6 and 7.

Crack Arrest Condition

The condition for crack arrest can be determined using the proposed
parameter if arrest is equated to some suitably long delay. If, for instance, crack
arrest is defined as a number of delay cycles in excess of 2 x 10 6 for the API
X-65 steel of Fig. 7, a value of the parameter of 7.91 is required for arrest. Then
if AKs were 21.98 MN/m 3/2 for a crack, the overload ratio for crack arrest
would be 0.34 for R = 0. A different value of overload ratio would be obtained
for a different AK~. From the foregoing, it is apparent that the overload ratio to
cause crack arrest depends on the stress intensity level range, AK~, and the
deduction is consistent with the experimental results of Corbly and Packman
[23]. In another example, Wei et al [14] found that for a low stress intensity
level of 10.99 MN/m 3/2 the crack did not propagate in 450 000 cycles. For this
condition, the value of the parameter is 3.92 which gives a predicted delay (Fig.
2) of 400 000 cycles-a conservative prediction.
Compared with the "zero-in" technique used by Probst and Hillberry [12] to
determine the arrest condition in Eq 1, this parameter offers a simpler procedure
and eliminates the trial-and-error approach inherent in the zero-in method.

Discussion

The results from the analysis of data on delay for a variety of materials
indicate that the parameter, (K~max/Khmax ) 9 AKs adequately correlates delay
for the data examined and Eq 3 can be used to determine delay cycles (defined
in Fig. 1). Effects of metallurgical and environmental variables (for example,
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
152 FATIGUE CRACK GROWTH

a~s, strain hardening (n'), temperature, etc.) can only be accounted for at
present through variations in the parameters A and m in Eq 3. Applicability of
the composite parameter so far has been shown only for first-order delay effects
where the influences of peak loads do not interact. Second-order effects due to
periodic overloads may cause an increase or a decrease in delay [17,18]
depending on the interval of repetition. Other load interaction effects [9] (for
example, delayed retardation) are not considered in this phenomenological
model. Furthermore, the effects of stress intensity gradient, dK/da, on delay
period pointed out by Schijve [9] and discussed by Bucci [24] is not
considered. In the context of the loading pattern considered here, the effect of
stress intensity gradient, dKfJda, should be constant since a step change in load
amplitude is applied in all cases. If this were not the case, the stress intensity
gradient effect would probably be important [24]. The proposed parameter
suggests that primary variables in Eq 2 governing the delay period are K~ max,
AK~, and Khmax.
It should be noted that only the two extreme cases of overloads, namely, the
single overload and two-step high-low block loading cases, were considered in the
preceding delay model. Therefore Eq 3 does not account for an increase in delay
cycles with an increase in the number of appfied overloads. However, it is
expected that the delaying effect of multiple overloads will be bounded by the
two cases considered here. In a situation where the overload is repeated often
enough that there is appreciable crack growth during its repeated application, a
reasonable estimate of the expected delay cycles should result from the high-low
loading case.
No similar parameter describing delay has been found in the literature, but it is
interesting to compare it with parameters used by other investigators to predict
crack initiation and crack propagation. Jack and Price [25] proposed a function
z~kg (]30/p')1/2 to predict crack initiation cycles in a mild steel under constant
amplitude loading. In the function, Po is the critical value of the crack tip radius
and p' is the effective crack tip radius. If we extend this to the case of variable
amplitude loading, (for example, a high-low two-step test), and assume that Po is
the crack root radius at the low load level and p' is the crack root radius at the
high load, we can show that the function reduces to the same form as the
parameter used herein. Let the crack root radius be represented by the crack
opening displacement (COD) at the respective load level. COD ~ ey (K2/Oys2).
By substitution, AK(po/p,)1/2 becomes 7 " (g~ma x/ghmax ) ~ ~ ' ~ where 7 is a
constant, which is, except for the constant, identical to the parameter proposed
in this paper.
During the delay period, crack growth can be visualized as continuing on a
relatively small scale (~10 -7 cm/cycle for a very long delay). In the high strain
region in the plastic zone ahead of the crack tip, many microcracks are
propagating and coalescing until a dominant crack is formed [26]. Under these
circumstances, initiation and propagation intermix and the growth process can
be described by a damage accumulation theory [25-28]. The equation for crack

propagation rate, based on a damage accumulation mechanism is given by [28]

d a = X A-eLo~"AK2 (4)
dN 4B~Oys
where A~p is the weighted average strain range close to the crack tip, B is a
measure of ductility, and X is a constant. It is evident from the expression that
the function, Aep.AK is the controlling variable for crack growth. For
variable-amplitude loading, Rice [29] has shown by analysis that the strain range
at a blunted crack tip depends on the ratio of the stress intensities, K~ and K h.
That is, Aep cc (K~/Kh). With this expression, the function can be written as A~p
AK o~ (K~/Kh) 9 AKs Assuming a proportionality factor of /3, it becomes
A~p.AK = j3(Ks 9 AKf~ which again is identical to the parameter proposed
except for the constant.
Finally, the engineering significance of the proposed parameter lies in its
provision of a simple method of estimating delay, thereby, making possible an
adjustment of crack propagation life obtained from a direct integration of a
constant-amplitude crack propagation "law," for example

~V-= faf da + ND
"ai f( O
Also, the delay model can be used to estimate crack growth behavior under
block programmed loading by accounting for the delay after a step down in
load. When using Eq 3 for multi-step increasing or decreasing block loading, it
must be borne in mind that it was developed for the simple cases of two-level
(high-low) block loading. It does not account for the initial acceleration of
growth rate after an increase in load level and the interaction of delay effects in
a multi-step decreasing load sequence. Many other variables will also have to be
included in the model to make it suitable for spectrum loading.

Summaryand Conclusions
This study has demonstrated that there is a relationship between the proposed
composite stress intensity parameter, (K~max/Khmax) 9 z2kKj~,and the number of
delay cycles due to simple two-step load blocks with different R h and R~ (R >
0) values for a variety of materials: aluminum alloys, titanium alloys, and carbon
steels. The functional form o f N D is

ND =A IKJ2mhT~,
axL '~max" AK~I m

where A and m are experimentally determined. In this form, the model is limited
to the estimation of delay for cases when Khmax is greater than K~max , and
AKs is due to constant-amplitude sinusoidal loading.
The engineering usefulness of this parameter at present lies in its provision of a

simple method o f estimating delay and, perhaps, in its utilization in determining

the combination of stress intensities that would produce crack arrest.

Acknowledgment

This work was supported by grants from the National Research Council under
Grant No. (A.1694), the Defense Research Board o f Canada under Grant No.
(9535-49), and the Office o f Scientific Research o f the United States Air Force
(AFOSR-71-2120). Mr. Adetifa would also like to acknowledge financial support
from the Canadian Commonwealth Scholarship Programme.

References
[1] Hudson, C. M. and Hardrath, H. F., "Effects of Changing Stress Amplitude on the
Rate of Fatigue-Crack Propagation in Two Aluminum Alloys," NASA, TN D-960,
Sept. 1961.
[2] Schijve, J. and Brock, D., Aircraft Engineering, Vol. 34, 1962, p. 314.
[3] Hudson, C. M. and Raju, K. N., "Investigation of Fatigue-Crack Growth Under
Simple Variable-Amplitude Loading," NASA, TN D-5702, March 1970.
[4] Trebules, V. W., Roberts, R., and Hertzberg, R. W. in Progress in Flaw Growth and
Fracture Toughness Testing, ASTM, STP 536, American Society for Testing and
Materials, 1973, p. 115.
[5] Hardrath, H. F. and McEvily, A. J. in Proceedings, Crack Propagation Symposium,
Vol. 1, Cranfield, England, Oct. 1961.
[6] Hudson, C. M. and Hardrath, H. F., "Investigation of the Effects of Variable-
Amplitude Loadings on Fatigue Crack Propagation Patterns," NASA, TN D-1803,
May 1963.
[7] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-251.
[8] Christensen, R. H. in Metal Fatigue, McGraw Hill, New York, 1959.
[9] Schijve, J., "The Accumulation of Fatigue Damage in Aircraft Materials and
Structures," AGARDograph No. 157, NATO, Jan. 1972.
[10] Stulen, F. B., Redfern, J. H., and Schuite, W. C., "An Approach to Metal Fatigue,"
NASA, CR-246, June 1965.
[11] Von Euw, E. F. J., "Effect of Overload Cycles on Subsequent Fatigue Crack
Propagation in 2024-T3 Aluminum Alloy." Ph.D. thesis, Lehigh University, March
1971.
[12] Probst, E. and Hillberry, B.,AIAA Journal, Vol. 12, No. 3, March 1974, p. 330.
[13] Himmelein, M. K. and Hillberry, B., "The Effect of Stress Ratio and Overload
Ratio on Fatigue Crack Delay and Arrest Behavior Due to Single Peak Overloads,"
presented at the Eighth National Symposium on Fracture Mechanics, Brown
University, Aug. 1974.
[14] Wei, R., Shih, T., and Fitzgerald, J., "Load Interaction Effects on Fatigue Crack
Growth in Ti-6AI-4V Alloy," NASA CR-2239, April 1973.
[15] Gray, T. D. and Gallagher, J. P., "Predicting Crack Retardation Following a Single
Overload Using a Modified Wheeler Model," presented at the Eighth National
Symposium on Fracture Mechanics, Brown University, Aug. 1974.
[16] Wheeler, O. E., Journal of Basic Engineering, Transactions, ASME, Series D, Vol.
94, March 1972, p. 181.
117] Petrak, J., Engineering Fracture Mechanics, Vol. 6, 1974, pp. 725-733.
[18] Jonas, O. and Wei, R. P., International Journal of Fracture Mechanics, Vol. 7, No.
1, 1971, pp. 116-118.
[19] Shih, T. T. and Wei, R. P., Journal of Testing and Evaluation, Vol. 3, No. 1, Jan.
1975, pp. 46--47.
[20] Jones, R. E., Engineering Fracture Mechanics, Vol. 5, 1972, pp. 585-604.

[21] Vargas, L. G. and Stephens, R. I., "Subcritical Crack Growth Under Intermittent
Overloading in Cold-Rolled Steel," Proceedings, Third International Conference on
Fracture, Munich, Germany, Vol. 6, April 1973.
[22] Rice, R. C. and Stephens, R. l. in Progress in Flaw Growth and Fracture Toughness
Testing, ASTM STP 536, American Society for Testing and Materials, 1973, pp.
95-114.
[23] Corbly, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973,
pp. 479--497.
[24] Bucci, R. J., Discussion on paper by J. Schijve, Engineering Fracture Mechanics,
Voi. 7, March 1975, pp. 181-185.
[25] Jack, A. R. and Price, A. T., International Journal of Fracture Mechanics, Vol. 6,
1970, pp. 401-409.
[26] Lehr, K. R. and Liu, H. W., International Journal of Fracture Mechanics, Vol. 5,
1969, p. 45.
[27] McClintock, F. A. in Fracture of Solids, Drucker and Gilman, Eds., Gordon and
Breach, New York, 1963, p. 65.
[28] Hahn, C. T., Rosenfield, A. R., and Sarrate, M., "Elastic-Plastic Fracture
Mechanics," AFML-TR-67-143, Part Ill, Air Force Materials Laboratory, Wright-
Patterson AFB, Ohio, Jan. 1970.
[29] Rice, J. R. in Fatigue Crack Propagation, ASTM STP 415, American Society for
Testing and Materials, 1967, pp. 247-311.

DISCUSSION

R. I. Stephens ~ {written discussion)-Could the authors indicate how this

composite stress intensity model might be extended to more complex
interaction spectra? How might this model be applied to design situations?
C V. B. Gowda {author's closure)-The proposed model estimates delay
caused by either a single overload or a block of overloads in a high-low load case.
However, delay cycles caused by most forms of variable-amplitude loads can be
estimated from the model in the following manner. It is known that delay
increases as the number of overloads increases. Therefore, the effects of multiple
overloads on delay should be bounded by the single overload and high-low step
load cases; with the extent of delay increasing toward the higher limit as the
number of overloads increases. If the overload is repeated until a quasi-constant
growth rate ensues before cycling at the low load level, delay can be reasonably
estimated from the high-low case data. Therefore, a combination of careful
analysis of load history, good engineering judgement during interpolation or
extrapolation, and enough test data to back the proposed model will enable one
to estimate delay due to most practical variable-amplitude load cases.
Once an estimate of delay is obtained, integration of crack growth equations
(for example, equation in text) or a cumulative damage evaluation procedure
will provide data necessary for design or integrity (serviceability) decisions.
For very unusual load histories like random load histories, it is not a simple
matter to adopt the proposed model. However, a thorough statistical analysis of
load histories may provide information suitable to estimate delay (or inter-

i Materials Engineering Division, The University of Iowa, Iowa City, Iowa 52242.

action) effects by procedures similar to the ones discussed in this paper. As a

conjecture (first attempt) one could start with

(g~max) rms
(ghmax)' rms ( ~ ) r m s

as the parameter which controls the extent of delay. In practical applications of

great importance and value, incorporation of simple models, like the one
proposed here, in computer piograms for crack growth integration would yield
quicker and physically interpretable results.

Mathematical Modeling of Crack

Growth Interaction Effects

REFERENCE: Bell, P. D. and Wolfman, A., "Mathematical Modeling of Crack

Growth Interaction Effects," Fatigue Crack Growth Under Spectrum Loads, ASTM
STP 595, American Society for Testing and Materials, 1976, pp. 157-171.
ABSTRACT: A combined analytical and experimental program was performed to
obtain additional insight into the growth of cracks subjected to simple variable-
amplitude loading. The experimental results were used to examine methods of
mathematically modeling crack propagation behavior to enhance predictive capability
for arbitrary spectrum loading.
The mathematical modeling consisted, primarily, of using the crack closure concept
to correlate predicted crack propagation behavior with observed behavior for various
simple loading spectra. Two analytical schemes were developed, one of which
calculated the crack opening load based on residual forces acting on the crack
surface, and another where empirical relationships between loading sequences and
crack opening behavior were utilized. Difficulties with the Residual Force Model in
correctly representing some of the experimentally observed crack opening and closing
effects led to emphasizing the development of the more empirical Crack Closure
Model.
The test program encompassed constant-amplitude tension loads, single and
multiple tensile overloads, compression spikes, and simplified variable-amplitude load
sequences applied to 2219-T851 aluminum and Ti-6AI-4V annealed titanium alloy
specimens. Detailed crack-growth and crack closure measurements were obtained.
These measurements were used to quantify crack closure behavior as a function of
stress ratio, including negative values, and various loading sequences.

KEY WORDS: crack propagation, fatigue (materials), retarding, stress ratio, tension,
aluminum alloys, titanium alloys, compressing

The introduction of fracture control criteria for new aerospace structures has
necessitated a mathematical model which accurately predicts crack growth
during spectrum loading. At the initiation of this development effort, existing
mathematical models attempted to predict crack-growth retardation resulting
from previous tensile overloads [1,2] .2 However, they did not account for a
number of factors which could critically affect crack growth. These included the
effect of the number of overloads on subsequent retarded crack-growth rates,
the effect of compression spikes, and the effect of stress ratio (ratio of minimum
stress to maximum stress in a load cycle).
This paper presents two new mathematical models for predicting crack growth

i Senior structural mechanics engineer and member, Systems Engineering Staff, respec-
tively, Grumman Aerospace Corporation, Bethpage, N. Y. 11714.
2 The italic numbers in brackets refer to the list of references appended to this paper.

157
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authori
158 FATIGUE CRACK GROWTH

under arbitrary spectrum loading. These models, which account for some of the
previously-mentioned deficiences, were empirically developed during the'course
of an analytical and experimental investigation of crack growth carried out at
Grumman and its subcontractors for the Air Force Flight Dynamics Laboratory
[3].

Results and Discussion

CrackClosureModel
The Crack Closure Model is an empirically based model which uses an effective
stress range concept. The closure variations were defined as functions of previous
load history. Since it is a strictly empirical model, certain specific test data, in
addition to constant-amplitude data, are required for each new material where
predictions are desired. Based on the results of the test programs used to verify
the model, it provides fairly good predictions for a variety of load histories.
Concepts-Elber [4] showed that cracks subjected to tension-tension loading
close when the remotely applied stress, S, is some value greater than zero.
Assuming that crack extension occurs only when the applied stress is greater
than the crack opening stress, the pertinent stresses involved in the crack-growth
process should be the crack opening and maximum stresses. It has been shown
[4-6] that the crack opening stress differs from the crack closure stress, but for
the purposes of this discussion, they will be considered to be equal and will be
referred to as the closure stress.
The effective stress range, ASeff, is then equal to the difference between the
maximum stress, Smax, and the closure stress, So:

z~Seff -----Smax - S c (1)

Defining the closure factor, Cf, as

Cf =So (la)
S max

Equation 1 becomes

~xSeff = Smax(l - Cf) (lb)

For the purpose of this model, a Paris type growth equation was modified and
written in terms of AKeff as

5a = A(z~Keff) n (2)
z2~N

Equation 2 can be written in terms of the effective stress range for a crack of
length 2a, in an infinite sheet as

~l =Z((_Smax_Sc) N/~ln (3)

The coefficient A and exponent n can be determined by fitting the crack growth
rate data plotted against the effective stress intensity range, say for R = 0.
Equation 3 is the basis for all crack-growth calculations performed by the
model. Elber [4] showed that S c was a function of Sma x and R during
steady-state (constant-amplitude) crack growth. If S c in Eq 3 is defined as a
function of R, then Eq 3 will predict the effect of stress ratio on crack-growth
rates. Further, if S c is defined as a function of the previous load history, then Eq
3 would predict crack-growth interaction effects. It remains then, to define S c as
a function of stress ratio and previous load history (and any other pertinent
parameters) so that Eq 3 will properly predict crack-growth rates for any
condition.
Application of Crack Closure to Mathematical Model-In the past, it has been
suggested that many parameters influence crack-growth interactions. The most
significant parameters were determined from the test data as far as possible. In
some areas, quantitative values were not divulged by the data and a certain
amount of intuition was employed. In all cases, the model was verified by
comparing predicted crack growth with the test data. The most significant
parameters found during this program were: effect of R, including compression,
on constant-amplitude growth; and the effects on subsequent crack growth of
overload stress, previous minimum stress relative to current minimum stress,
number of overload cycles, and compression.
Effect of Stress Ratio on Closure-Crack closure measurements were made for
both materials under constant-amplitude loading conditions. Some of the results
obtained for aluminum are shown in Fig. 1. The closure factor is plotted against
stress ratio, and the data show a definite trend towards increasing closure factor
with increasing stress ratio. The data scatter shown is typical of that obtained
throughout the program. Scatter was more extensive when measurements were
taken during and subsequent to transient loading sequences. It was determined

1.0
/
/

CLOSURE
FACTOR,
Cf

2! Eol,I l- " /
~-~. Kmin
- - =R
Kmax
0 I I I I I// I I I I I
-1.0 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8 1.0
STRESS RATIO, R

FIG. 1 -Comparison o f closure factor versus stress ratio for 2219-T851 aluminum.

that the measured closure values possessed too much scatter to be used directly
for modeling purposes. The absolute value of Cfo (at a stress ratio of zero) was
determined from multiple-overload (high-low) crack-growth data by utilizing the
crack-growth rates immediately after the load change. Elber [4] and the
qualitative trend of the measured closure factors indicated that Cf was stress
ratio dependent. As a result, Cf was determined indirectly from crack-growth
rate data as follows.
Equation 3 may be rewritten as
n
Aa ___A I(_~_._R (1-C f) (3a)
)
or more generally as
&7
--A ~ " (1-Cf (3b)
AN )
The parameter Cf is the closure factor at any stress ratio, R. It can be seen that
for a given value of AK, Cf can be related to Cfo, the closure factor at R = 0, by
the measured crack-growth rates. An analysis of the data revealed that Cf can be
expressed by the equation

Cf =Cf_ t + (Cfo -Cf_ 1)(1 + R ) p (4)

where Cf_ 1 is the closure factor at R = -1. The final values for the parameters in
Eq 4 are shown in Table 1.

TABLE 1 - Crack closure parameters.

2219-T851 Ti-6A1-4V
Aluminum Titanium

Cfo 0.4 0.4

Cf..l 0.347 0.332

p 3.93 3.33

These results were compared with the limited data available in the literature
(that is, with 2024-T3 from Ref 3). The behavior is similar for positive stress
ratios and agree within 20 percent with Elber's value at R = 0. No data were
available for negative stress ratios.
Equation 4 is presented in Fig. 1 for 2219-T851 aluminum. The most
significant aspect of these results is that the closure factor was determined
experimentally for stress ratios as low as -1, thus providing a convenient
technique for including negative stresses in crack-growth calculations.
Effect o f Overload Stress-It is well known that when many cyclic overloads

have been applied to a cracked specimen, the crack-growth rates during a

subsequent, lower cyclic loading are retarded. Investigators [3, 7,8] have shown
that after the crack propagates some distance, the degree of retardation
decreases and the rates return to constant-amplitude values. The Crack Closure
Model considered herein assumes that subsequent to such a load change the rate
is immediately retarded, and that the crack extension required for the rates to
return to constant amplitude is related to the plastic zone radius caused by the
overload.
Although delayed retardation could also have been included in the model, the
experimental data generated in this investigation did not provide sufficient
indications of this phenomenon. Further, it was concluded that delayed
retardation was a second-order effect during most variable-amplitude loading
applications.
Consider the loading shown in Fig. 2 for the case of many cycles of high stress,
S~, followed by a lower stress, $2. For convenience, the stress ratio for both

,c, AAAAAAA
V VVVVVV ..... JVVVVVV r
1~1 p l, I CYCLES, N,
== ~ta CRACK LENGTH, a

FIG. 2-Schematicof closurevariationthroughaffectedlengthcausedby overload.

stresses is taken as zero. The Crack Closure Model assumes that the closure
stress, So, varies as shown. Immediately prior to the change in stress, the closure
level is the stabilized level, Scl, associated with S~, and varies, as shown, through
some affected length, p, to the stabilized value, Sc2, associated with level $2.
The expression assumed to define the closure stress variation is given by

Se=Scl-(Scx-Sc2 ) (_.~) b for 0 < ~a < p (5)

An analysis of the aluminum and titanium data revealed that when P is taken
to be equal to the plane stress plastic zone radius and b is 1.0, they produced a
good fit to the data. Some typical results, obtained by numerically integrating
Eq 5 are shown in Fig. 3 for aluminum and in Fig. 4 for titanium.
Effect of Number of Overloads-It has been shown [7,9,10] that a single
overload produces a lesser degree of subsequent retardation than several overload
cycles. This effect was modeled by assuming that the increase in closure level is a
function of the number of overload cycles applied. An equation of the form

"~ -~" ~r + (1 -"~'1 ) \Nsat- 1 (6)

27
SPECIMEN:
NOL = 20 9 / AD-25-107
9 / 2219-T851 ALUMINUM
COMPACT TENSION
/ W = 63.5 ram, B = 6.35 mm
9 /
9 DATA
PREDICTION
USING CRACK
CLOSURE MODEL
26
Cfo = 40

NOL = 50 ./
/
9 /

25 0/
9

E
E [~-NoL

j-
,z, ~--~Z~N
_/

9 P = 2890N
< 24 . NOL = 100
g /" POL = 5200N
/. Pmin = 140N

o ~ ~

/
NOL = 500 /..
/o
/
/ o9
/
9 I
22 "e 9
J J I I I
4 8 12 16 20

AN s, CYCLES SINCE LOAD CHANGE, THOUSANDS

F I G . 3-Comparison o f predictions with data for high-low sequence, 2219-T851

aluminum.

was fitted to data where the number of overloads, No] , was varied. Here, 3" is the
ratio of the closure stress after No1 overloads to the stabilized overload closure
stress; 3'1 is the value of 7 for Nol = 1 ; and Nsa t is the number of overload cycles
required to achieve saturation (that is, beyond Nsat, the addition of overload
cycles produces no additional retardation). Values of 71 and Nsa t are shown in
Table 2 for both materials. These results indicate that only 13 overload cycles

SPECIMEN: P = 4000 N
TD-25-110 POL = 7200 N
36 TI-6AL-4V
Pmin = 200 N
COMPACT TENSION
W =55.9MM,B=6.38MM O/L = 1.8 5, /
._E 32
LOAD{N) I_ _l N 20 V /
,I-

c~ 2s
,2DOt- ", "i OL
40001" |III~--~MIIAI 20 " ~ / /
..1 .oo t. vvv v 6o
v 24 cYcLES ",oo L.I F
<
~ 20
"OL=5 EO,CT,ON
16 I I I I
1O0 150 200 250
CYCLES, THOUSANDS

FIG. 4-Predicted a versus N for Ti-6A I-4 V titanium.

are required to produce maximum retardation for aluminum, whereas 100

overload cycles are required for titanium, values which are indirectly supported
by Refs 10 and 9, respectively. Figure 4 also shows that the model provides a
good prediction for titanium where Nol < Nsa t . Similar results were obtained for
aluminum.

TABLE 2 - Increasing closure parameters.

2219-T851 Ti-6A1-4V
Aluminum Titanium

"/1 0.67 0.80

Nsa t 13 100

Effect o f Compression and Minimum Stress-The test program revealed that

whenever the minimum stress was reduced, the closure stress decreased to a new
value almost immediately. For example, given a constant maximum cyclic stress
with a positive stress ratio, when the minimum cyclic stress was reduced to zero,
usually one to three load cycles were required to produce a new stabilized
closure level. This effect was modeled by assuming that only one cycle of the
new lower minimum stress was required to produce a new stabilized closure
stress. The concept was then extended to more complex loading sequences (see
Fig. 5).
It was found that when the new minimum stress was compression, the same
approach held. The justification for extending the model to handle compression
is that, based on the data, the closure stress is always positive. Therefore, from a
closure standpoint, a compression load is the same as a reduction of the
minimum load. Figure 6 presents a prediction and data for an aluminum

STRESS
J

$1
S
c1

Smin 1

Stain2 [ _ _ . _

CYCLES, N

FIG. 5-Schematic of minimum stress adjustment.

SPECIMEN:
50 AG-25-gP
2219-T851 ALUMINUM
E CENTER CRACKED PANEL / PREDICTION
E /
~- W = 152.4 ram, B = 6.35 mm /
S = 68.9 MN/m2 /
Smin = 3,4 MN/m 2 , I / ~ ' 3 7~ . 9/ DATA
_J
25 68.9
(j 34.5 ~,j~..~,,,~ INDICATES ONE CYCLE
~ OF COMPRESSIONLOAD
rj APPLIED, VALUE SHOWN
is MAGNITUDEIN N
~0 ,~0 ~o ~o
CYCLES, THOUSANDS

FIG. 6-Predicted a versus N for occasional compression loads, 2219-T851 aluminum.

specimen subjected to several discrete compression spikes, and indicates that the
model properly accounts for compression spikes.
The model was verified by predicting the lives of specimens subjected to two,
three, and four-level block loading sequences. The stresses within the blocks
were ordered high-to-low, low-to-high, and randomized within each block.
Numerous comparisons, presented in Ref 3, show good correlation with the
data.

Residual Force Model

The Residual Force Model is essentially the same as the Crack Closure Model
except that it attempts to account for the actual closure forces acting on the
crack faces instead of basing closure on the externally applied loads. The
effective stress range is taken as the difference between the maximum remotely
applied stress, Smax,and the crack closure stress, S c. Elber [4] suggested that
crack closure resulted from plastic deformations left in the wake of a
propagating crack. The Residual Force Model assumes that when the crack is
closed, a compressive stress distribution develops along the crack surfaces. Figure
7 shows the development of the residual stresses. A crack of half-length ao, in an
infinite sheet, is subjected to the remote tensile stress, Smax,which produces a
plastic zone of extent py (Fig. 7a). When the remote stress is removed, a reverse

l l r~axT

t l TSmaxt t

(b)
~ ~Smaxl
FIG. 7-Residual stress distributions.

. . . . . . . . . . . . . SEQUENCE

RESIDUALSTRESSDISTRIBUTION

FIG. 8-Effect o f a single overload on residual stresses.

plastic zone of extent ry 3 remains defining an area of compressive residual

stresses. Figure 8b shows the crack after it has propagated to a new half-length,
a. The crack tip is part way through the envelope of compressive residual stresses
which were developed as the crack propagated to its new length. Figure 8c shows
an average residual stress distribution, Sres, which represents the integrated value
of the residual stresses from ao to a.
The ~following assumptions were made:

(a) ~kgeff = Kma x - gre s (7)

3 Here, ry is defined as the reversed plastic zone and differs from that defined in the
Glossary.

where Z~r(eff is the effective stress intensity range, K is the stress intensity
caused by Sma x , and Kre s is the stress intensity produced by the residual stresses
acting along the crack face only.
(b) Sres is proportional to Sma x.
(c) The extent of Sres due to the application of one cycle of tensile loading is
one quarter of the maximum plastic zone size and is given by

ry = ~ r : max 1 2 (8)
L2oy, j
where 13= 2LTfor plane stress and 13-~-~--
[ for plane strain.
Initially, it seemed that Sre s should be equal to the material compression yield
stress. However, a review of Eq 7 revealed that the crack could only be opened
by the application of S > oy s, the material compression yield stress. Experience
indicates that this is not the case and that Sres must therefore be something less
than ay s. It was therefore assumed that Sre s = o~Smax-
An interesting effect produced by this model is that delayed retardation is
predicted. It was found that the delay period depends on the number of
overloads applied and that as the number of overloads increases to the saturation
value, the delay period decreases to zero, and the minimum crack growth rate
occurs immediately after the load change. In addition, the model predicts
acceleration when changing from a low load to a high load. These results are
included in the following discussion of the development of the working model.
Development of WorkingModel-Figure 8 shows a crack which has propagated
to a half-length, ao, under the influence of remote stress $1, which has caused
the residual stress, Sresl , to be developed. At half-length, ao, a remote stress, $2
(where $2 > S1), has been applied and then removed. The plastic zone, py,
caused by $2 is shown along with the reverse plastic zone, ry, which exists when
$2 returns to zero. The crack then continues to propagate under the influence of
$1.
Immediately after the application of $2, the residual forces (stresses) acting
behind the crack are those caused only by the application of $1 and the residual
stress intensity is the same as for the constant-amplitude case for S~. However,
as the crack begins to propagate into the reverse plastic zone caused by $2, the
residual stresses, Sres2, begin to act behind the crack tip and an increase in K res
occurs. When the crack half-length is a + ry the influence of Sres2 is most
prominent and Kre s is a maximum. As the crack continues to propagate, S res2
becomes remote from the crack tip so that its effect on g r e s is reduced, and
when the crack half-length is large relative to a + ry, the effect of Sres2 on K res2
approaches zero.
If a crack propagation law of the Paris type is used such that

da=A(z~xgf)
n d N ef (9)

then Z~2eff is smallest when the crack half-length is a + ry and Aa/~dV is the

smallest at the same point. Therefore, the minimum crack growth rate occurs at
a point ry from where the overload was applied and delayed retardation has been
represented by the model.

[in
To test this approach, an equation of the form

was taken from Ref 11, and the equation was doubled to account for forces
acting on both surfaces of the crack. Dimensions b and c are shown in Fig. 7c.
At any crack length, a, it is a simple matter to sum the effects of the various
residual stress distributions over the crack half-length from zero to a, to obtain
the total Kre s.
Recent experimental investigations [6-8] indicate that the effect of the
overload should have dissipated by the time the crack has propagated through a
distance approximately equal to the plastic zone caused by the overload. It was
found, however, that when Eqs 7 and 9 were used, the effect of an overload on
subsequent crack growth persisted for crack extensions of several plastic zones.
It was therefore assumed that only those residual stresses which exist from one
plastic zone radius, py, behind the crack tip, to the crack tip, should be included
in the calculations. In this way, the effect of an overload on subsequent crack
growth is only included in the calculation while the crack tip is within the area
from ao to a0 + py in Fig. 8. Then, z~kgeff was normalized so that steady-state
crack-growth conditions would exist when appropriate.
The final crack-growth rate equation is given by

" t§
~__=a (1o)

Results-The model was tested using data from the literature [7] which
strongly exhibited delayed retardation. Figure 9 shows crack-growth rate data
versus crack-growth increment after a load change for the cases where the
number of overloads, Nol , were 0, 1,450, and 9000. The dashed lines represent
predictions for Nol = 1 and >_300. The values of a and /3 were selected to
provide a reasonable fit to the Nol = 1 data and were then used for the other
predictions. The basic trend is evident and shows that for the case ofNol = 1, a
minimum crack-growth rate occurs when Aa is about 0.2 mm. A comparison of
the predicted and test curves indicates that the curve shapes and magnitudes
agree fairly well. However, when the limiting case of 300 or more overloads is
applied, the residual force model predicts a crack-growth rate which is much
lower than the data suggests.
Figure l0 shows a case of acceleration where the crack-growth rate versus

10-3 -
/

"~" No -- 9

I~ TESTDATA FOR
~_ NO, =

450 9
cO 10.5 9000 9
L9 PREDICTIONWITH

~)
I! t -NoL o-
~=0.2
.E

10"6 N'~OL '

YI
>300 t
.2 .4 .6 .8
:: 1:0 112
~a AFTER LOAD CHANGE,turn

FIG. 9-Residual force predictions versus data.

,,z,' E
10 .2 - o
24,
lg!E , 'r
E
E
~~ . . T E S T DATA
"'~ RESIDUAL FORCEMODEL
w" / PREDICTION~ = 3.6

~ 10.3 - 1
CONSTANT AMPLITUDE

I I I I I I
.2 .4 .6 .8 1.0 1.2
Aa AFTER LOAD CHANGE, mm

FIG. l O-Residual force predictions versus data for low-high loading.

distance after load change is plotted for the case of a low-high loading. Here, the
predictions are based on the same values of a and 13 used for the delayed
retardation cases. Prediction and test agree fairly well.
Some predictions are compared in Fig. 11 to test data from more complex
loading cases [12]. Cases 1 and 2 were used to determine the Paris crack-growth
parameters while a and 13 were determined to be 2.6 and 0.I, respectively, from
Case 3. These values were then used to predict the lives for the remaining cases.
In general, the correlation between test and predicted lives is good. The
difference between the shapes of the predicted and test curves is attributed to
the block-wise integration scheme [13] used to calculate lives.

~ • a.

~~'~ oo
v

..i
w
r~
0

o
u.Z

C~
w
I- f,,r' o.

I
I
I

I I O
Q
&D o,I

' " " ' "e - H . t g N 3 1 ~ : ) V U : ) ~ ' I V H

Concluding Remarks
Both the Residual Force and Crack Closure Models offer new methods of
predicting crack growth under variable-amplitude loading. The Residual Force
Model is elementary and requires only basic input data for its use. However, its
predictive ability is currently limited to simple loading sequences. Conversely,
the Crack Closure Model is considerably more versatile but requires more
complex input data. The development o f both models and many comparisons of
predictions and data are presented in Ref 3. 4

Acknowledgments

The authors wish to acknowledge J. C. Newman, Jr., NASA Langley Research

Center, and R. Bucci, formerly of Del Research Corporation, for their
contributions in formulating the initial versions of the Crack Closure Model; and
P. C. Paris and M. Creagar of Del Research and Del West, respectively, for their
extensive efforts and advice in generating the experimental data and analytical
approaches reported in this paper. A grateful acknowledgment is also extended
to R. M. Engle, Air Force Flight Dynamics Laboratory, for his assistance in the
overall program.

References
[1 ] Wheeler, O. E., "Crack Growth under Spectrum Loading," Report No. FZM-5602,
General Dynamics Corporation, Fort Worth Division, June 1970.
[2] Willenborg,J., Engle, R., and Wood, H. A., "A Crack Growth Retardation Model
Using an Effective Stress Concept," Technical Memorandum 71-1FBR, Air Force
Flight Dynamics Laboratory, Wright-Patterson AFB, Jan. 1971.
[3] Bell, P. D. and Creager, M., "Crack Growth Analysis for Arbitrary Spectrum
Loading," AFFDL-TR-74-129, 1974.
[4] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[5] Newman, J. C., Jr., and Armen, H., Jr., "Elastic-Plastic Analysis of a Propagating
Crack," Paper 74-366, AIAA/ASME/SAE 15th Structures, Structural Dynamics and
Materials Conference, Las Vegas, Nov., April 1974.
[6] Newman, J. C., Jr., "Finite-Element Analysis of Fatigue Crack Propagation
Including the Effects of Crack Closure," Ph.D. thesis, Virginia Polytechnic
Institute, 1974.
[ 7] Trebules, V. W., Roberts, R., Jr., and Hertzberg, R. W. in Progress in Flaw Growth
and Fracture Toughness Testing, ASTM STP 536, American Society for Testing and
Materials, July 1973, pp. 115-146.
[8] yon Euw, E. F. J., Hertzberg, R. W., and Roberts, R. in Stress Analysis and Growth
of Cracks, ASTM STP 513, American Society for Testing and Materials, 1972, pp.
230-259.
[9] Wei, R. P. and Shih, T. T., "Delay in Fatigue Crack Growth," Lehigh University
Publications.
[10] Hudson, C. M. and Raju, K. N., "Investigation of Fatigue Crack Growth under
Simple Variable-Amplitude Loading," NASA TN D-5702, March 1970.
[11 ] Paris, P. C., and Sih, G. C. in Fracture Toughness Testing and lts Applications,
ASTM STP 381, American Society for Testing and Materials, 1965, pp. 30-8 3.

4 The Crack Closure Model has been programmed and incorporated into the Air Force
crack growth computer program, CRACKS II.

[12] Porter, T. R., "Method of Analysis and Prediction for Variable-Amplitude Fatigue
Crack Growth," EngineeringFracture Mechanics,Vol. 4, No. 4, 1972.
[13] Brussat, T. R. in Fracture Toughness and Slow-Stable Cracking, ASTM STP 559,
American Society for Testing and Materials, 1974, pp. 298-311.

DISCUSSION

Otto Buck I (written discussion )-Do you think it would be possible to include
stress relaxation and creep into your models?
P. D. Bell and A. Wolfman (authors' closure)-Yes, if creep is defined as either
stable crack growth at sustained load or as stable tear. Both models could be
modified to include relaxation of residual stresses and stable growth as functions
of both time and temperature if empirical data, similar to crack growth rate
data, are available. Stable tear has been incorporated in the Crack Closure Model
with some success.

t Science Center, Rockwell International, Thousand Oaks, Calif. 91360.

Experimental Results and a

Hypothesis for Fatigue Crack
Propagation Under
Variable-Amplitude Loading

REFERENCE: Jacoby, G. H., Nowack, H., and van Lipzig, H. T. M., "Experimental
Results and a Hypothesis for Fatigue Crack Propagation Under Variable-Amplitude
Loading," Fatigue Crack Growth Under Spectrum Loads, ASTM STP 595, American
Society for Testing and Materials, 1976, pp. 172-183.

ABSTRACT: Many as yet inadequately-understood phenomena contribute to the

sequence effects observed during fatigue crack propagation under variable-amplitude
loading. In order to get an insight in the sequence effects, an extensive test program
with 2024-T3 aluminum and Ti-6AI-4V sheet specimens with systematic variations in
the loading conditions was performed, where special emphasis was placed on the
accurate measurement of the crack propagation just after changes in the loading
conditions. A hypothesis is proposed which explains the observed behavior. The
hypothesis is based on deformation considerations within the plastic zones caused by
the propagating crack.

KEY WORDS: fatigue (materials), crack propagation, aluminum alloys, titanium

alloys, plastic deformation, fracture properties, damage, loads (forces)

Cumulative damage analysis of structures subjected to variable-amplitude

loading is complex because of sequence effects. Many studies have been
performed with respect to the accuracy of cumulative damage theories, program
and random tests [1,2] .3 There is still, however, a lack of information about the
crack propagation behavior resulting from changes in amplitude and mean stress.
This is mainly because of experimental difficulties in determining the exact
crack extension rates over a relatively short distance. With such crack
propagation measurements from tests where the loading conditions are systemat-
ically varied it should be possible to get further insight into the mechanism of
sequence effects [3,4]. Sequence effects are the consequence of the interaction
of several phenomena which have been outlined in many investigations with

a Director, Carl Schenck AG, Darmstadt, West Germany.

2 Department head and research fellow, respectively, Deutsche Forschungs- und Versuch-
sanstalt f~/r Luft- und Raumfahrt (DFVLR), lnstitut f{/r Werkstoff-Forschung, Porz: West
Germany.
3 The italic numbers in brackets refer to the list of references appended to this paper.

172
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
JACOBY ET AL ON VARIABLE-AMPLITUDELOADING 173
constant-amplitude loading (CAL). Some of these phenomena shall shortly be
considered.
Figure la shows schematically the stresses and strains along the x-axis and also
the plastic zones in front of the crack tip at the maximum and minimum load in

a) Stress
I , /loading Stress
+~,~~ m ~ ~ o a d l n g
loading o~ _

unloading~ x

Strain I
I
Strain %!!
l o a d m ~
~N. /loading
~zr / unloading unloading ~ .
EYS -
! I x
0 X
Plastic zones I
Plastic zones
I
~ I~oa~,ng yA I II~
I rKmax

unloading ~\, \ unlading

r~. Ir.K

FIG. 1-Stress and strain distributions along the crack line without and with considering
residual deformations behind the propagating crack and plastic zones.

a load cycle assuming an ideal elastic-plastic material behavior [5]. There exist
two areas with different plastic deformation characteristics. Within the area
denoted as rKmax the material is monotonically plasticly deformed. The size of
this area is controlled by the maximum stress in a cycle. Within the area denoted
as rAK, the material is subjected to reversed plastic deformations corresponding
to the stress variation in the cycle.
Actual materials usually exhibit monotonic and cyclic strain hardening or
softening properties. This leads to alterations in the plastic zone sizes and also to
the buildup of a cyclic strain hardening (or softening) profile along rAK [7,8].
During fatigue crack propagation residual deformations are built up in front of
the crack tip and are left behind the propagating crack. The presence of these
residual deformations is one of the main causes for the crack closure behavior as
observed in several investigations [9,10]. The distributions of the stresses and
strains along the x-axis, considering the residual deformations, are shown in Fig.
lb. The main consequence of the residual deformations is a reduction of the
crack tip stress intensity variation during a load cycle. The crack propagation

behavior is further influenced by the instantaneous stress state at the crack tip
[11] and by microstructural mechanisms. The latter may, however, become a
second order effect at higher loading levels [12].
In case of variations in the loading conditions, an interaction of the described
phenomena and mechanisms occurs. In the following sections, an attempt is
made to analyze these interactions on the basis of an extensive test program,
where the crack propagation behavior due to a variation in mean stress during
CAL, due to single and multiple high loads, and due to additional negative peak
loads is studied. The interpretation of the experimental crack propagation
behavior is further substantiated by the results from deformation measurements
on the specimen surface and by microscopic observations.
The investigations were performed with both cyclic strain hardening 2024-T3
aluminum and cyclic softening Ti-6A1-4V. Center-cracked 80-mm-wide speci-
mens were used. The thickness of the 2024-T3 aluminum specimens was 1 mm
and that of the Ti-6A1-4V specimens was 1.6 mm. The crack propagation
measurements were performed with two microscopes for both sides of the crack.
The magnification was x125. Special scales in the optical systems enabled the
observation of crack growth increments smaller than 0.01 mm.

Experimental Crack Propagation Behavior and Discussion

Constant-Amplitude Loading with an Increase in Mean Stress
Figure 2 shows the loading conditions; the observed crack length, a, as a
function of the cycle number, N; and the crack rates, &a/AN, as a function of
the crack length for 2024-T3 aluminum and Ti-6AI-4V. In both cases the crack
rates increase above the CAL-rates immediately after the change in mean stress.
The acceleration of the crack growth is, however, more pronounced for 2024-T3
aluminum resulting in a significant decrease in fatigue life in the crack stage. In
Fig. 3a the crack propagation behavior of 2024-T3 aluminum is schematically
replotted. Capital letters characterize the essential stages. In the figure the plastic
deformed regions before and after the change in mean stress are also shown. The
increase in the crack rates between @ and (~) can be explained as follows: At
the end of the low mean stress CAL-period, the plastic zones rKmax I and r A2c 1
exist. Residual deformations are present on the crack surfaces and along the
previous path of r K m a x I 9 Within rAK 1 a stable strain hardening profile
corresponding to the low mean stress loading level exists. During the first load
cycle after the increase in mean stress r K m a x 2 develops, the size of this zone is
mainly controlled by the monotonic strain hardening behavior of the material.
At the very tip of the crack the material is subjected to high tensile strains. Since
it is cyclic hardened from the preceding loading period, its deformation
capability is limited. This brittle characteristic of the material increases the crack
rates. Scanning microscopic observations revealed the formation of material
regions separated by deep grooves parallel to the crack front. During unloading,
the crack remains opened over a considerable range of the half cycle since the

a} 202Z,-T3 b) Ti6-4

2oot I
z 150 ..
I00 bml ' - 9
~5o~--I
I00 -,m,
/q~sm~
. . . . . .

01 u
Time

-5 ~<~Sm 2

--~ I I I I
0 500 I000 0 500 I000 1500
Cycle number, N

1.0-

~ 05-
o.,
_ L . _ ~ . . . . . . .- o.~

8 9 10 II I0 II 12 13
Crack length.a(mm)
FIG. 2 - C r a c k propagation behavior due to an increase in mean stress.

residual deformations along the fracture surfaces exhibit the low equilibrium
extent of the preceding loading period. Therefore, both the brittle crack tip
material and the high stress intensity variations, until crack closure occurs, are
the reasons for the high crack rates at (~) . Due to the fast crack extension, the
crack tip and rzkK2 rapidly enter material regions which are less cyclic strain
hardened than under CAL-conditions. From this and due to the fact that higher
tensile deformations can develop in the high mean stress period, which reduce
the stress intensity variation at the crack tip, Aa/&N decreases. The fact that the
crack rates decrease below the CAL-rates and that they then increase above the
CAL-rates indicates that the cyclic hardening of the crack tip regions plays an
essential part. Variations in the Aa/~uV behavior occur until new stable plastic
zones at the crack tip are built up at Q . The given explanations for the
&a/~uV-variations could further be substantiated by ey-strain measurements

FIG. 3-Aa//32~/and plastic z o n e behavior d u e to a variation in m e a n stress.

using a special grid technique. 4 The measurements showed that during the low
mean stress CAL-period up to ( ~ ) no high deformations occured in the vicinity
of the crack tip (ACYmax "~ 3 to 5 percent). Immediately after the change in
mean stress, however, extremely high strain variations (A6Ymax ~ 15 to 20
percent) were measured in the vicinity of the crack tip. These high strain
variations reduced when the crack propagated. A t ( D , where the Aa/AN-

4 A fine 0.l-ram spaced grid was engraved ahead of the crack tip with a modified
micro-hardness tester. Photographs were taken from the undeformed grids and from the
deformed grids at several instances during the loading histories. The grid spacings were then
measured with a special microscope and the ey-strains determined from the relative
displacements of the grid.

behavior has reached a stable condition, maximum strain variations of 5 to 7

percent were measured.
The considerations so far have referred to a strain hardening material. In the
case of Ti-6AI-4V, rAx~ at the end of the low load period exhibits a high
ductility. Consequently, the crack extension, due tothe increase in mean stress,
occurs in a ductile region accompanied by an extensive opening of the crack.
The cyclic softened material and the high stress intensity variations, due to the
comparatively small residual deformations left behind from the preceding
loading period, lead to crack propagation rates which are higher than under high
mean stress CAL. Along with the fast propagation of the crack, regions that have
experienced less cyclic softening as compared to the stabilized condition are
entered, and this increases the resistance against crack extension. In addition, the
formation of higher residual deformations decreases the crack rates significantly.
These low crack rates lead to a more extensive softening of the material within
rAr 2 resulting in an increase in crack rates. In conclusion, the crack propagation
behavior of cyclic softening materials and that of cyclic hardening materials is
similar although the mechanisms differ.

Constant-A mplitude Loading with a Decrease in Mean Stress

Figure 3b shows schematically the Aa/AN versus a behavior of 2024-T3
aluminum and the plastic deformed areas. At the end of the high loading period,
rKmax 2 and rAK2 are present. The residual deformations along the previous
path of rKmax 2 are in an equilibrium condition corresponding to the high
loading level. After the decrease in mean stress, the crack rates fall below those
under low mean stress CAL because the residual deformations generated in the
previous loading period reduce the crack tip stress intensity variation. A
satisfactory explanation for the further delayed decrease in Aa/AN could not be
found. An incompatibility of the crack front orientation from the high loading
period with the low loading conditions may have an influence. After passing
through the minimum at ~ , the crack rates grow up rapidly along with an
increase in rAx3- At ( ~ , the influences of rKmax 2 and the residual
deformations within this zone decay, and a new rKmax 3 develops. The described
plastic zone and residual deformation behavior became clearly visible in the
ey-distribution measurements using the previously mentioned grid technique.
Behind (~) , the crack rates are still lower than those under low mean stress
CAL, since the influence of the residual deformations from the preceding
loading period is still present. This influence gradually decays up to (~) .
Simultaneously, rKmax 3 and rAK 3 which coincide with those under CAL are
formed. The latter process can be recognized in the side view of the specimen
surface shown in Fig. 4. Further on the fracture surface shown in the figure,
dark areas in the range of the rKmax 2 path from the high mean stress loading
period can be seen. Scanning microscopic observations revealed that these are
material particles which have been abraded during the rubbing of the fracture
surfaces due to the action of the residual deformations.

FIG. 4-1nfluence of mean stress variations on the fracture surface and specimen surface
appearance.

Crack Propagation Under Single and Multiple High Loads

Under these loading conditions an interaction of the mechanisms described in

the two previous sections occurs. The Figs. 5 and 6a show the Aa/AN-behavior
for N h = 1, 40, and 100 high load cycles inserted in low load CAL for 2024-T3
aluminum, as well as for N h = 1 for Ti-6A1-4V. Before the details of the
Aa/AN-behavior in the subsequent low load period are discussed, the differences
in the plastic zone behavior and in the residual deformations as a function o f N h
will first be considered. The plastic zone behavior for a single and a small
number of high loads is shown in the lower part of Fig. 5. During the preceding
low load period, rKmax 1 and rAK 1 have been formed. Due to the increase in
load, rKmax2 is generated. Within rKmax2 high residual deformations are built
up. In the case of a low number of high loads, rKmax 2 propagates by a small
amount corresponding to the crack extension during the high loads. The area
where residual deformations are built up increases just a little. There is, however,
an essential difference compared to the behavior as described in the preceding
section. In that case the rKmax2 path and also the high residual deformations
within rKmax 2 exist along the whole previous crack path. Under a single or a

FIG. 5-Crack propagation behavior due to single and multiple high loads and plastic
zones.

small number of high loads, significant high residual deformations are only
formed ahead of the crack tip during the high loads.
The Aa/AN-behavior caused by the application of the high loads is the same as
described in the first section. Due to the aforementioned residual deformation
behavior within the rKmax2 path as a function of the number of high loads, the
minimum in the Aa/AN at 0~) is lower following multiple high loads and occurs
at a smaller crack length. During the subsequent loading period, the crack rates
increase again, whereby the crack extension mechanism is similar to that as
described in the preceding section.
From the experimental Aa/AN-behavior in Figs. 5 and 6a, it is seen that the
cyclic strain hardening 2024-T3 aluminum and the softening Ti-6A1-4V behave
in a similar manner. The experimental results further showed that the
retardation in crack growth caused by different numbers of peak loads N h is not
directly proportional to N h.

al Ti 6 - 4 b}

~ ~~176 ~ -~~176I-
-~ ,oo~~t~ ~ _:~,ool~
~o~
Izi
soO' '-'
9 '--'I ~ = ~
Time ~ -50 Time
Low load

~'~176 It ~:o.o~
~-o.o~- l
~176176 ,P 8 9
, 10
, ~176176 ,I
8 9 10
,
Crack length,a(mm) Crack length.a(mm)
FIG. 6-Crack propagation behavior due to a single high load (a) and clue to a single high
load followed by a compression load (b).

Crack Propagation Due to a Low Peak Load Following Single and Multiple High
Loads
These loading conditions represent a further approach to variable loading
conditions. Figures 7 and 6b show the Aa/AN-behavior for 2024-T3 aluminum
and for Ti-6A1-4V. The changes in the crack propagation mechanism due to a
negative peak load are schematically indicated in Fig. 7. Until the occurance of
the negative peak load the plastic zone and residual deformation behavior is the
same as already described in the preceding subsections. If the negative peak load
would not occur, rAK ~ would be generated upon unloading. When the negative
peak load is applied, however, a larger rAKe* is generated. This leads to a
reduction of the area of rK where residual deformations exist after
max2
unloading and also to a larger redeformation of the residual deformations.
Besides this, the crack tip is further resharpened. All these effects lead to the
formation of a larger r A ~ ~ than under a loading situation where no negative
peak loads occur, and the crack rates are higher. From these considerations it
immediately follows that the crack propagation is more influenced by negative
peak loads: the lower the peak loads are, the smaller the areas are where residual
deformations are formed, and the smaller the residual deformations are.

FIG. 7-Crack propagation behavior due to single and multiple high loads followed by a
compressive load and plastic zones.

This coincides with the experimental results in Fig. 7 and, also, with the
observations made by Schijve, where negative peak loads were applied during a
loading history with a comparatively high mean stress [11]. Since the effect of
the residual deformations along the crack surfaces is small, because of the high
loading level, a comparatively larger r a t 2 due to the negative peak load can
develop which also reduces the influence of the residual deformations ahead of
the crack tip. This together with a pronounced resharpening of the crack tip
leads to a nearly unretarded crack propagation.
The present test results further show that Ti-6A1-4V behaves in a similar
manner as 2024-T3 aluminum.

Conclusions

1. Experiments with systematic variations in the loading conditions, where

special emphasis was placed on an accurate measurement of the crack
propagation behavior after changes in the loading conditions, have shown that
significant variations in the crack rates occur until stabilization is reached.
2. A hypothesis is proposed which explains the observed behavior. This
hypothesis is based on the deformations caused by the propagating crack. The
deformations depend on:
(a) the maximum stress intensity and the variation in stress intensity
during a cycle and the stress intensity history, and
(b) the monotonic and cyclic deformation characteristics of the material.
3. Under constant-amplitude loading a monotonic deformed plastic zone and
a smaller cyclic deformed plastic zone exist directly in front of the crack tip.
From these plastic zones, residual deformations are also left behind the
propagating crack.
4. Sequence effects mainly occur due to the interaction of the plastic zones of
the respective loads in the subsequent load cycles. Of predominant importance
are the Kmax -controlled plastic zones. The residual deformations built up within
the area of the Kmax -controlled plastic zones hinder the formation of a stable
condition corresponding to constant-amplitude loading.
5. The ranges in crack length where sequence effects are observed are
considerably larger than calculated on the basis of the linear elastic fracture
mechanics equations for the plastic zone size.

A cknowledgm en t

The financial support for this work by the Deutsche Forschungsgemeinschaft

(DFG) is gratefully acknowledged.

References
[1] Schijve, J., "The Accumulation of Fatigue Damage in Aircraft Materials and
Structures," AGARDograph No. 157, NATO Advisory Group for Aerospace
Research and Development, 1972.
[2] Jacoby, G. in Effect of Environment and Complex Load History on Fatigue Life,
ASTM STP 462, American Society for Testing and Materials, 1970, pp. 184-202.
[3] Nowack, H. and van Lipzig, H. T. M., "ErmiJdungsrissausbreitung bei Belastungen
mit ver~/.nderlichen Mittelspannungen und Amplituden," DGM-Hauptversammlung,
Bonn, June 1974.
[4] Van Lipzig, H. T. M. and Nowack, H., "Rissfortschrittsverhalten und Restfestigkeit
yon Leichtbauwerkstoffen bei nicht-einstufigen Belastungen," Proceedings 6,
DVM-Arbeitskreissitzung "Bruchvorgange," Deutscher Verband f/Jr Material-
pr~/fung, 1974, pp. 129-141.
[5] Rice, J. R. in Fatigue Crack Propagation, ASTM STP 415, American Society for
Testing and Materials, 1967, pp. 247-309.
[6] Lehr, K. R. and Liu, H. W., International Journal of Fracture, Vol. 5, No. 1, March
1969, pp. 45---55.
[7] Bathias, C. and Pelloux, R. M., Metallurgical Transactions, Vol. 4, May 1973, pp.
1265-1273.
[81 Schwalbe, K. H., International Journal of Fracture, Vol. 9, No. 4, Dec. 1973, pp.
381-395.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
JACOBY ET AL ON VARIABLE-AMPLITUDE LOADING 183

[9] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[10] Nowack, H., "Ein Beitrag zur Untersuchung der Schadensakkumulation auf der
Grundlage biharmonischer Belastungsabl~/ufe," DLR Forschungsbericht 71-23,
Abteilung Wissenschaftliches Berichtswesen der DFVLR, Ed., Porz-Wahn
(Germany), 1971.
[11] Schijve, J., Engineering Fracture Mechanics, Vol. 6, No. 2, Sept. 1974, pp.
245-252.
[12] Freudenthal, A. M., Engineering Fracture Mechanics, Vol. 6, No. 4, Dec. 1974, pp.
775-793.

tn~

~ rind 9
m m
m m

0 0

Effect of Spectrum Type on Fatigue

Crack Growth Life

REFERENCE: Reiman, J. A., Landy, M. A., and Kaplan, M. P., "Effect of Spectrum
Type on Fatigue Crack Growth Life," Fatigue Crack Growth Under Spectrum Loads,
ASTM STP 595, American Society for Testing and Materials, 1976, pp. 187-202.
ABSTRACT: An ever-present problem in airplane design has been the accurate
prediction of fatigue crack growth rate in new (never flown) aircraft. This problem
has become even more critical in light of the new U. S. Air Force Damage Tolerance
Requirements (MIL-A-83444). Briefly, the purpose of this specification is to protect
flight structures from potentially deleterious defects which can be introduced into
the structure during processing, manufacturing, etc. One of the most important
things to consider when performing a damage tolerance analysis is the type of fatigue
loading spectrum to be used. The purpose of this paper is to present a method of
deriving a randomized flight-by-flight loading spectrum which can be used in
preliminary design, and to show the effect on crack growth life of varying certain
important parameters.
The random spectrum is derived using mission profiles, MIL-A-O08866Aor flight
recorder exceedance data, design loads, and stress data. The mission mix and actual
load sequence is determined randomly. The spectrum is then analyzed using a
conventional crack growth computer program which calculates crack growth (a versus
N) curves, and critical crack length.
The sensitivity of fatigue crack growth life to the variation in a number of
parameters was investigated. These parameters include different random sequences
and cycle counting methods. Random sequence effects include randomizing the loads
on a mission segment or on a flight-by-flight basis. The results of this study are
compared to the results obtained when the basic exceedance data was combined into
a conventional block loading sequence.

KEY WORDS: fatigue (materials), crack propagation, damage, randomization, flight

tests, counting, sampling, defects

The increased awareness of the effects of load interaction has underscored the
importance of defining a realistic loading spectrum for fatigue and crack growth
analysis and test. This has led to much effort being expended to quantitatively
determine exactly how the variation of key spectrum parameters affect crack
growth life [1-3] .~ We would like to know how closely a design spectrum must
simulate real life usage in order to accurately predict life.
The early fatigue spectra were all block-type. There were a number of reasons
for this: the test equipment available could not handle complex load time
histories, the contemporary analytical techniques such as Miner's Rule did not
1 Aerospace engineers, Fatigue and Fracture Branch, Structures Division, ASD/ENF,
Wright-Patterson AFB, Ohio, 45433.
2 The italic numbers in brackets refer to the list of references appended to this paper.

187
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions author
188 FATIGUE CRACK GROWTH

recognize sequence, etc. Unfortunately, it was learned from fleet experience that
these techniques of testing and analysis were not accurately predicting problems
which were occurring in service. Therefore, it became necessary to go to a more
complex test spectrum in order to improve the accuracy of test results by using a
load-time history which was a closer simulation of the actual spectrum
experienced by the fleet. Thus, the flight-by-flight spectrum was conceived, and
it has generally become the accepted standard. In recent years, the widespread
use of computers has allowed one to "ever increase the complexity of the
flight-by-flight spectra.
Although nearly everyone is in general agreement as to what kind of spectrum
should be used, the exact methodology to derive the spectrum is ill-defined. It is
the purpose of this paper to describe methods of deriving loading spectra which
can be used in the preliminary phases of airframe design. Furthermore, the effect
on crack growth life of varying certain important parameters will be demon-
strated. These parameters include various sequences and cycle counting methods.
Sequencing effects include block loading or randomizing the loads on a mission
segment or flight-by-flight basis.

Mission Profile Definition

In determining a fatigue loading spectrum, it is important to know how the
user plans to utilize the akcraft. To accomplish this, the user, together with the
agency which is responsible for design development, determines a set of "mission
profiles." These mission profiles characterize what type of missions will be
flown; for example, close air support, armed reconnaissance, training, etc. Each
mission profile outlines all the important parameters of the mission: the type of
mission, its payload, fuel load, mission duration, and total amount of time spent
flying this mission during the aircraft lifetime. Each mission profile is then
subdivided into a number of mission segments such as takeoff, ascent, cruise,
and combat. The important variables for each mission segment also are defined:
initial and final velocity, altitude, and gross weight; and time spent in the
segment. This is illustrated in Fig. 1.
With the mission profiles defined, a "load factor exceedance curve" is
constructed [4]. The load factor exceedance curve, Fig. 2, shows the number of
times a given load factor will be exceeded during the lifetime of the aircraft. As
the mission profiles tell how much time will be spent flying a given mission
segment, the total number of load factor exceedances for the aircraft in question
can be determined. MIL-A-008866A [4] is one of many sources of exceedance
data; there have been many flight load recorder programs which can provide
useful information, depending on the type of aircraft being analyzed.
After construction of the load factor exceedance curve, a bending moment
and then a stress-exceedance curve is developed. This is accomplished by
associating each load factor occurrence with a given weight, velocity, and
altitude. To determine how many occurrences one has of a certain load factor,
the exceedance curve is discretized. The bending moment is calculated using a

-!
iLL;
I--.

F-. 3
.J

LI.I
uJ~20 C

-- ]E
/;
LLJ
(n
-

--4
I1
1
== ~4,l

LI,( I
, li I I I
10 20
TIME (MINI)TES)

FIG. 1 -Mission profile." ground support.

known functional relationship between load factor, weight, velocity, and

altitude. Using another known relationship, the stress is calculated from the
bending moment. Curves for positive and negative maneuvers and their
corresponding 1 g stresses are developed.

Spectrum Development
For this study, two types of spectra for two different fighter aircraft were
developed. These spectra represent the most widely used means of deriving the
load-time (stress-time) history of an aircraft. The spectra developed were block
(low-high-low) and flight-by-flight (quasi-random). The flight-by-flight spectra
were developed by two different methods: one with the loads randomized on a
mission segment basis and one with the loads randomized on a total mission
basis. The load exceedance data used to generate these spectra were identical
[4,51.
The two aircraft for which these spectra were developed consisted of an
air-to-air fighter and an air-to-ground fighter. For the air-to-air fighter, both
flight.by-flight spectra, and a block spectrum were developed. The air-to-ground
aircraft did not have a flight-by-flight mission segmented spectrum.
A discussion of the methodology for developing each of these spectra follows.

t.d
t.t-

-.J

t.d
t.~

t.u

2 4 6 8 10
LOAD F,~.CTOR (~'s)
FIG. 2-Typical stress/exceedance curve.

Block
The block spectrum, the simplest to derive and test, consists of groups of
stress cycles, with each group having the same maximum and minimum stress.
The magnitude of the loads for each block are found by differentiating a load
(stress) exceedance curve at a given number of load (stress) levels and using the
resultant number of occurrences. The blocks so formed are ordered low-high-
low.
Block spectra were developed for both the air-to-air fighter and the
air-to-ground fighter. The air-to-air fighter's block spectrum consisted of ten
unique stress levels for each block while the air-to-ground fighter's spectrum
consisted of 14 unique stress levels for each block. To obtain one lifetime, each
block was repeated 80 times for the air-to-air aircraft and 325 times for the
air-to-ground aircraft. These spectra are illustrated in Fig. 3.

Flight-by-Flight
The importance of using a realistic flight-by-flight spectrum to predict life has
been frequently discussed. Derivation of these spectra has only recently been
accomplished. This is due to the increased use of high-speed computers,
especially when used in conjunction with fatigue test apparatus.

L 80 P,EPEATS-_~

a ~ "1 TIME =

I" 325 REPEATS=I LIFETIME "I

b
TIME " ~"

FIG. 3-Block spectra shapes for (aJ air-to-air and (b) air-to-ground fighter.

A program for developing such a flight-by-flight spectrum has been written by

the authors [6]. This program has the capability for pairing maximum and
minimum stresses on a random basis. The program is written such that any
number of missions and mission segments may be input. Ground loads also may
be input into the program. The resulting spectrum contains as many different
flights as desired. Each flight may be considered as a real-time history of the
maneuver loads. For this study, two different flight-by-flight spectra were
considered.
Mission Segment-by-Mission Segment-This type of spectrum is the most
realistic representation of an aircraft load-time history. The spectrum is first
broken into distinct flights, each flight being a certain mission-type such as a
special weapons mission, air-to-air combat mission, etc. Each mission type is
further divided into mission segments such as ascent, cruise, or descent. For the
air-to-air fighter, there are 3195 total flights comprised of five different missions.
Each mission contains up to seven segments. These 3195 flights represent 4000 h
of flight time. Figure 1 shows a typical mission profile. Each segment (ascent,
cruise, etc.) is comprised of a range of airspeeds, gross weights, and altitudes.
The mission ordering within the spectrum is random, but the segments retain
their order in each flight. Within each segment the stress history is random. The
stresses are divided among 20 unique stress levels, the levels being selected on a
statistical basis using a program developed by Lincoln [7]. The 20 stress levels
are unique for each segment of the mission, providing a total of 520 unique
stress levels in the spectrum.
Flight-by-Flight-This spectrum was constructed in a manner similar to the
mission segment-by-mission segment just described. The difference is that this
spectrum is broken down no finer than on a mission basis. This type of spectrum
was derived for both the air-to-air and the air-to-ground aircraft.
The air-to-air fighter's flight-by-flight spectrum was comprised of the same five
mission types as in the mission segment-by-mission segment spectrum. The

mission ordering was the same random ordering as in the mission segment-by-
mission segment spectrum. The stress history within each flight was random. As
in the mission segment-by-mission segment spectrum, there were 20 unique
stress levels in each mission; thus, there are 100 unique stress levels in all.
The air-to-ground fighter flight.by-flight spectrum was comprised of 2921
unique flights consisting of eleven different missions. As in the air-to-air fighter's
flight-by-flight spectrum, the mission ordering was random as was the ordering
of the stresses within each flight. There are also 20 unique stress levels in each
mission, thus giving 220 stress levels. The same method was used to generate this
aircraft usage spectrum [7] as was used with the air-to-air fighter.

Analytical Model Verification

The motivation behind development of a spectrum is to determine the life
characteristics, whether fatigue or fracture, of a component. Spectrum develop-
mentcannot, in itself, determine these life characteristics. To obtain them, it is
necessary to perform either a fatigue or fracture analysis or test of a particular
component with each of the spectra and compare their results. For the tasks
discussed within this report, a fracture mechanics analysis was performed.
It is believed by the authors that test results for each spectrum variation would
be beneficial, yet the expense in time, material, and test equipment would be
overwhelming. For this reason, the authors chose to utilize a crack growth
program, EFFGRO [8] with the Vroman retardation model [9] as the means of
comparing results. It is the purpose of this section to indicate that this analytical
procedure is capable of predicting crack growth behavior of aluminum alloys for
various aircraft spectra. The three spectra will be for a fighter, bomber, and
cargo aircraft.
The first example is that of the fighter spectrum. The material used was
2024-T3511 aluminum. The test coupon, shown in Fig. 4 was 101.6 mm (4 in.)

25 ,S1('TYP)"~
}*- _..L ANALYSI S " ~ S~
TEST
2D
t~
..d

t.9
15

I I I I I I I I I ,
29 40 60 80 1100 120 1~0 160 180
.BLOCKS

FIG. 4 - E x p e r i m e n t a l / a n a l y t i c a l correlation." fighter spectrum.

wide and 6.35 mm (0.25 in.) thick. A 6.35 mm (0.25 in.) hole was drilled into
the center of the sheet. The specimen was precracked according to ASTM Test
for Plane Strain Fracture Toughness of Metallic Materials (E 399-72). Actual
spectrum testing was initiated when two diametrically opposed precracks were
0.51 mm (0.02 in.) long. The test spectrum (maximum spectrum stress of 23
MN/m 2 (34 ksi)), shown in Table 1, was a modified low-high-low block
spectrum. Figure 4 shows the experimental test results and the analytical
correlation.
The second example, that of a bomber spectrum, Table 2, used 2024-T851
aluminum. The specimen geometry, shown in Fig. 5, is a "dog-bone" type
specimen, 25.4 mm (1.0 in.) thick. The initial flaw size was 1.78 mm (0.07 in.)
with a/2c = 0.5, and the critical crack size was 9.14 mm (0.36 in.). The
specimen was elox slotted, and the crack initiated using a bending fatigue
apparatus. Again, the specimen was precracked consistent with ASTM Method E
399-72. The test spectrum was a flight-by-flight type representative of a wing
lower surface. The maximum spectrum stress was 23 MN/m 2 (34 ksi). The test
result and its correlation with the analytical study are shown in Fig. 5.
The third example is indicative of a cargo spectrum. Its loading parameters are
shown in Table 3. The material is 7075-T6511 aluminum extrusion. The test
coupon used, shown in Fig. 6, was 203.2 mm (8 in.) wide and 4.877 mm (0.192
in.) thick. A hole 6.35 mm (0.25 in.) in diameter with a through crack 1.27 mm
(0.05 in.) in length was in the center of the specimen. Again, the crack was
initiated according to ASTM Method E 399-72. The test spectrum was a
low-high-low block spectrum with approximately 5600 cycles equivalent to
6000 flight hours. Figure 6 compares the experimental and analytical results.
The maximum spectrum stress was approximately 21.7 MN/m 2 (31.5 ksi).
Thus, it can be seen that for these types of aircraft (bomber, fighter, and
cargo) and for these types of spectra (flight-by-flight quasi-block, and block),
EFFGRO is able to predict the crack growth characteristics. It also should be
noted that the crack geometry for each of these specimens was different.

Results and Discussion

The analytical portion of this study was accomplished employing the same
crack growth program (EFFGRO) with the same retardation model (Vroman) as
used in the analytical model verification section. For this reason, justification
exists for accepting qualitatively the analysis performed herein for spectrum
studies. The studies performed for the air-to-air aircraft were as follows:
1. block spectrum versus flight-by-flight spectra,
2. segmented flight-by-flight spectrum versus unsegmented flight-by-flight
spectrum,
3. cycle counting techniques: rain flow and range pair, and
4. randomization effects for the unsegmented flight-by-flight spectrum.
The results of the first study, shown in Fig. 7, indicated that a flight-by-flight
spectrum was more damaging than a block spectrum: a result which agrees with

TABLE 1 - Fighter spectrum.

Max Load Min Load

(psi) (psi) Cyc~s

1 8370.00 3830.00 13.00

2 10870.00 5080.00 39.00
3 13620.00 4760.00 18.00
4 16010.00 5080.00 9.00
5 18320.00 5080.00 5.00
6 20530.00 990.00 1.00
7 22160.00 1320.00 1.00
8 24600.00 2010.00 1.00
9 20530.00 990.00 1.00
10 18320.00 5080.00 5.00
11 16010.00 5080.00 9.00
12 13620.00 4760.00 19.00
13 10870.00 5080.00 39.00
14 8370.00 3830.00 13.00
15 8370.00 3830.00 13.00
16 10870.00 5080.00 39.00
17 13620.00 4760.00 18.00
18 16010.00 5080.00 9.00
19 18320.00 5080.00 5.00
20 20530.00 990.00 1.00
21 22160.00 1320.00 1.00
22 20530.00 990.00 1.00
23 18320.00 5080.00 4.00
24 16010.00 5080.00 10.00
25 13620.00 4760.00 19.00
26 10870.00 5080.00 39.00
27 8370.00 3830.00 13.00
28 13620.00 4760.00 7.00
29 16010.00 5080.00 4.00
30 18320.00 5080.00 2.00
31 22160.00 1320.00 2.00
32 16010.00 5080.00 30.00
33 19250.00 6440.00 7.00
34 20530.00 990.00 5.00
35 27380.00 4260.00 5.00
36 29900.00 1670.00 1.00
37 22160.00 1320.00 1.00
38 20530.00 990.00 3.00
39 19250.00 6440.00 4.00
40 18320.00 5080.00 10.00
41 13620.00 4760.00 16.00
42 10870.00 5080.00 12.00
43 10870.00 5080.00 5.00
44 13490.00 5490.00 1.00
45 13620.00 4760.00 4.00
46 18320.00 5080.00 1.00
47 8370.00 3830.00 9.00
48 8370.00 3830.00 13.00
49 10870.00 5080.00 39.00
50 13620.00 4760.00 18.00
51 16010.00 5080.00 9.00
52 18320.00 5080.00 5.00

TABLE 1 Continued

Max Load MinLoad

(psi) (psi) Cycles

53 20530.00 990.00 1.00

54 20530.00 990.00 1.00
55 18320.00 5080.00 5.00
56 16010.00 5080.00 9.00
57 13620.00 4760.00 19.00
58 10870.00 5080.00 39.00
59 8370.00 3830.00 13.00
60 8370.00 3830.00 13.00
61 10870.00 5080.00 39.00
62 13620.00 4760.00 18.00
63 16010.00 5080.00 9.00
64 18320.00 5080.00 5.00
65 20530.00 990.00 1.00
66 22160.00 1320.00 1.00
67 24600.00 2010.00 1.00
68 20530.00 990.00 1.00
69 18320.00 5080.00 5.00
70 16010.00 5080.00 9.00
71 13620.00 4760.00 18.00
72 10870.00 5080.00 39.00
73 8370.00 3830.00 13.00
74 10870.00 5080.00 .50
75 16010.00 5080.00 .03
76 20530.00 9900.00 .17
77 29900.00 1670.00 .17
78 31063.00 0.00 .02
79 34260.00 1190.00 .01
80 34260.00 1190.00 .06
81 31480.00 2370.00 .06
82 24600.00 2010.00 .06
83 19250.00 6440.00 .56
84 13490.00 --5490.00 .04
85 8370.00 3830.00 .08

NOTE-Steps with fractional cycles are only applied at interval shown, for example, step
74 is applied every 2nd block, step 78 every 50th, etc.

many investigators [3,10,11]. This is due at least in part, to flight duration. It

has been shown [3] that the shorter the block the more closely the results agree
with a flight-by.flight spectrum. It cannot be stated categorically that a block
spectrum is nonrepresentative of actual flight conditions; however, it appears
that a flight.by-flight spectrum is more realistic.
The results of the second study, also shown in Fig. 7, reveal that no
discernable difference exists if a flight-by-flight spectrum is randomized on a
flight basis, or randomized on a mission segment basis. This result indicates that
it is not necessary to arrange a flight on a mission segment basis, which allows
much time savings both in computer hours and test equipment.

I "13

-I
0
c
i'rl

cJ
~'~
?
0
-I
I~~ -r

I
Z
0

o d~

~. ~ ~ ~ o " ~ ~ .

_~_3,56~I lJ- _~._1,

78 /

~-ANALYSIS

1 I I I I I I I I I I I I
0 tl 8 ]2 16 20 24 28 32 36 40
MISSIONS
FIG. 5-Experimental/analytical correlation." bomber spectrum.

The third study showed that no meaningful difference existed between rain
flow and range pair counting procedures. However, if the spectrum were not
subjected to a cycle counting procedure, a substantial difference in life exists.
Figure 8 shows this. The techniques of crack growth analysis demand that a
spectrum be input as a series of whole cycles where a cycle consists of a
maximum stress, a minimum stress, and a maximum stress. Thus', a whole cycle is
defined as a load excursion with constant mean stress across it. Figure 9
illustrates a cycle with and without constant mean stress. The block type
spectrum consists of, by definition, constant mean stress cycles and thus can be
used in the crack growth analysis directly. The random spectra, either
flight-by-flight or mission segment-by-mission segment, do not consist of whole

TABLE 3 - Cargo spectrum.

Max Stress Min Stress

(psi) (psi) Cycles

1 10103.00 505.00 1200.00

2 14064.00 5882.00 10823.00
3 17346.00 2540.00 94.00
4 17346.00 2540.00 94.00
5 14064.00 5882.00 10823.00
6 10103.00 505.00 1200.00

1.9 20
Z:
lad
....I

10
~ " ~ ANALYSIS
5

I I I I I I I I I I
6 12 18 24 3O 36 q2 48 54 60xioz
CYCLES

FIG. 6-Experimental/analytical correlation: cargo spectrum.

5 A B C

~ 4

~ 5

..-1

~ 2

'J 1

0 I I I I I I I I I I I

4 8 ]2 16 20 24 28 32 36 rio 44 x .~
FLIGHTS

F I G . 7-Air-to-air fighter: effect of spectrum type on crack growth life. (A) Random
flight-by-flight spectrum (stresses randomized on mission basis) rain flow counted; (B)
random flight-by-flight spectrum (stresses randomized on mission segment basis) rain flow
counted; and (C) block spectrum.

BA C

Lt.l
.-J

t_>
c::
t-}

I I I I I I I I
4 8 12 16 20 24 28 32xl/)2
FLIGHTS

FIG. 8-Air-to-air fighter." effect of cycle counting procedures on crack growth life. {A]
Random flight-by-flight spectrum rain flow counted; (B) same spectrum as (A) range pair
counted; and (C) same spectrum as (A} not counted.

0 NE EXAMB~ OF NONP
-~
,NDOMSPECTRUM
~( CYCLE :~ CONSS
I TN
I GOF~ CYC_IESWIIH
~ S'm~s ('rM)

O"M

B, I ~ON ~ ~ OF RAI~DOMSPECTRUMCONSISI'ING
I ~CYCLE(~) oF CYCI.ESWITH NON-C.CNST##ITIVEIIWSTRESS

'] I'-R 4 vv
N= WHOLE CYCLE (MAXIMUM-MINIMUM-
MAXIMUM)
R-FIRST HALF CYCLE OF N
S= SECOND HALF CYCLE OF N
~M
' R= MEAN STRESS OF R
~MS= MEAN STRESS OF s

FIG. 9-Definition of a cycle.

cycles but, rather, of half-cycles (maximum to minimum or minimum to

maximum). Thus, it is necessary to use a cycle counting technique that will
redefine the random spectrum in terms of whole cycles with constant mean
stress. The range pair method results in output consisting of whole cycles and
this output is used directly as input to a crack growth analysis (12). The rain
flow method results in output consisting of half-cycles (13), and different
techniques developed by the authors [6] were used to form the half-cycles into
whole cycles. These whole cycles are then used as input to a crack growth
analysis. Both methods preserve the sequencing effects of the random spectrum.
The final study, randomization effects, suggests that one random sequence
gives results similar to another, as shown in Fig. 10. This is not unexpected as

5.0

4.0

= 3.0
t.D

2,0

1,0'

I I I 1 I I I , I
g 8 12 16 20 24 28xkq 2
FLIGHTS
FIG. lO-Air-to-air fighter: effect of different spectrum randomizations on crack growth
life. ORandom flight-by-flight Spectrum A; [:]random flight-by-flight Spectrum B;
and Q random flight-by-flight Spectrum C.

the randomization was determined by the computer according to a random

number scheme, and the laws of probability indicate that this will be the result.
The two studies performed on the air-to-ground aircraft were: a block
spectrum versus a flight-by-flight spectrum, and an evaluation of cycle counting
procedures. It should be noted that each block of the block spectrum contained
five low-high-low excursions as is shown in Fig. 3.
The first study indicated (Fig. 11) that a flight-by-flight spectrum results in
shorter life than does a block spectrum. The second study, the evaluation of
cycle counting procedure again showed, Fig. 12, that the difference between the
range pair and rain flow techniques is negligible.

Conclusions
1. Flight-by-flight spectra appear to result in a more conservative life estimate
than block spectra if proper cycle counting procedures are used.

25 A B

~ ~11-.51(TYP) i
"='
.~ lo ~ FI I:1 (B,35 /

I i i I i I I I
4 8 12 16 20 24 28 32 x 102
FLIGHT HOURS
FIG. 11-A ir-to-ground fighter: effect o f spectrum type on crack growth life. {A} Random
flight-by-flight spectrum rain flow counted," and (13) block spectrum.

25 A B

"~ ~ 51 (~P)

i-~ I i I I I i i i
0 4 8 12 16 20 24 28 32x102
FLIGHT HOURS
FIG. 1 2 - A i r - t o - g r o u n d fighter." eflect of qvcle counting procedures on crack growth li~.
{A) Rain flow counted; and (B) range pair counted.

2. Negligible differences in analytical life exist between mission segmented

and unsegmented flight-by-flight spectra.
3. In this study, a randomized flight-by-flight spectrum gives repeatable
results independent o f the exact random sequence.

Acknowledgments

The authors would like to thank the various System Program Offices from
Wright-Patterson AFB which allowed us to use their test data for the Analytical
Model Verification portion of this paper.

References
[1] Schijve, J., "Load Sequences for Fatigue Testing of Components and Fullscale
Aircraft Structures," Paper No. 70-37, ICAF Congress, Rome, 1970.
[2] Barrois, W., "A Short Survey on Possibilities of Fatigue Life Assessment of Aircraft
Structures Based on Random or Programmed Fatigue Tests," Fracture Mechanics of
Aircraft Structures, AGARD-AG-176, NATO, Jan. 1974.
[3] Schijve, J., "The Effect of Load Sequence on Fatigue Crack Propagation Under
Random Loading and Program Loading," NLR TR 71014, Amsterdam, 197t.
[4] Department of the Air Force, "Airplane Strength and Rigidity Reliability
Requirements, Repeated Loads, and Fatigue," MIL-A-008866A(USAF), 31 March
1971.
[5] Clay, L., Nash, J., Rockafeilow, R., and Shope, C., "Structural Flight Loads Data
from F/RF-4 and Thunderbird F-4E Aircraft, January, 1972 through June, 1973,"
ASD-TR-74-2, Jan. 1974.
[6] Landy, M., Kaplan, M., and Reiman, J., "Derivation and Analysis of Loading
Spectra for USAF Aircraft," ASD-TR-TX-XXXX, to be published.
[7] Lincoln, J. W., "Development of an Aircraft Maneuver Load Spectrum Based on
VGH Data," ASD-TR-7X-XX, to be published.
[8] Szamossi, M., "Crack Propagation Analysis by G. Vroman's Model," NA-72-94,
Rockwell International, 1 Feb. 1972.
[9] Vroman, G. A., "Analytical Prediction of Crack Growth Retardation Using a
Residual Stress Concept," B-1 Division, Rockwell International, May 1971.
[10] Swanson, S. R., Cicci, F., and Hoppe, W. inFatigue Crack Propagation, ASTMSTP
415, American Society for Testing and Materials, 1966, pp. 312-362.
[11 ] Breyan, W. in Effects of Environment and Complex Load History on Fatigue Life,
ASTM STP 462, American Society for Testing and Materials, 1970, pp. 127-166.
[12] Tischler, V. A., "A Computer Program for Counting Load Spectrum Cycles Based
on the Range-Pair Cycle Counting Method," AFFDL-TM-72-4-FBR, Nov. 1972.
[I3] Dowling, N. E., Journal of Materials, Vol. 7, No. 1, March 1972, pp. 71-87.

Stress Spectrums for Short-Span

Steel Bridges*

REFERENCE: Klippstein, K. H. and Schilling, C. G., "Stress Spectrums for

Short-Span Steel Bridges," Fatigue Crack Growth Under Spectrum Loads, ASTM STP
595, American Society for Testing and Materials, 1976, pp. 203-216.
ABSTRACT: U. S. Steel's Research Laboratory is currently completing research on
Project 12-12, sponsored by the National Cooperative Highway Research Program of
the National Academy of Sciences, and entitled "Welded Steel Bridge Members
Under Variable-Cycle Fatigue Loadings." The main objectives of the program are to
develop fatigue data on welded bridge members under variable-amplitude random-
sequence stress cycles representative of stress spectrums such as occur in actual
bridges and to develop an analytical method of predicting the fatigue behavior under
variable-amplitude stress spectrums from constant-amplitude fatigue data.
This paper summarizes available field measurements of stresses in short-span bridges
under traffic, and describes the stress spectrums that were developed from these
measurements for use in the testing program. The available field data show that the
passage of a vehicle over a bridge produces a single major stress cycle and
superimposed vibrational stress cycles. In most types of bridges, the vibrational
stresses are small enough to be neglected; however, the described method of deriving
a simple equation for the stress spectrum is applicable to field data with or without
vibrational stresses. The frequency of occurrence of stress cycles can be defined by a
family of skewed Rayleigh probability-density curves; a particular curve from the
family is defined by the modal value, Srm, and the dispersion, Srd. The major stress
cycles (or stress ranges) are added to the dead-load stress, Stain, which remains
essentially constant during the life of the bridge. In general, the stress cycles are
arranged in random sequence. The stress spectrums used in the main test program,
therefore, are defined by three parameters: Srm, Srd , and S_. n. The fatigue-test
l Ill . 9

control tapes corresponding to these spectrums record 500 individual loads that
satisfy one of the family of Rayleigh probability-density curves and are arranged in a
random sequence. The tape is continuously cycled throughout a fatigue test.

KEY WORDS: crack propagation, fatigue (materials), steels, stresses, loads (forces),
stress cycle

U. S. Steel's Research Laboratory is currently completing research on Project

12-12, sponsored by the National Cooperative Highway Research Program o f the
National A c a d e m y o f Sciences, and entitled "Welded Steel Bridge Members

* The opinions and findings expressed or implied in this paper are those of the authors.
They are not necessarily those of the Highway Research Board, the National Academy of
Sciences, the Federal Highway Administration, the American Association of State Highway
Officials, nor of the individual states participating in the National Cooperative Highway
Research Program.
1 Senior research engineer and section supervisor, respectively, Application Research
Division, Research Laboratory, United States Steel Corporation, Monroeville, Pa. 15146.

203
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM Intemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
204 FATIGUE CRACK GROWTH

Under Variable-Cycle Fatigue Loadings." The main objectives of the program are
to develop fatigue data on welded bridge members under variable-amplitude
random-sequence stress cycles representative of stress spectrums such as occur in
actual bridges and to develop an analytical method of predicting the fatigue
behavior under variable-amplitude stress spectrums from constant-amplitude
fatigue data.
To accomplish these objectives, field measurements of stresses in bridges under
traffic were analyzed, and various hypotheses of predicting the fatigue behavior
of small specimens and beams tested during the program were used and
compared. In the experimental part of this study, variable- and constant-
amplitude fatigue tests of wedge-opening-loading specimens and plate specimens
with a simulated cover-plate-end detail were conducted to obtain crack-growth
data and to determine the effects of various stress-spectrum parameters on
fatigue life. Also, more than 200 variable- and constant-amplitude fatigue tests
of relatively large beams simulating bridge members were performed. Welded
beams with and without cover plates and fabricated from either A36 or A514
steel were used to obtain the approximate upper and lower bounds of the
variable-amplitude fatigue strength of fabricated bridge members. Various stress
spectrums, defined in terms of the mode and dispersion of the frequency of
occurrence of stress ranges, were used in the tests.
Currently, the last set of beams is being tested at stress levels representative of
actual stresses observed in field measurements. The final report, which
summarizes the results of this project, is being drafted and should soon be
available through the sponsoring agencies. This paper describes the method
employed to represent the available field measurements of stresses in short-span
bridges under traffic by simple equations for stress spectrums that were
subsequently used in the testing program of the research project.

Field Measurements
Individual Vehicle Passages
The stress or strain response of a bridge to the passage of a vehicle depends on
the type of bridge, the weight, speed, and dynamic characteristics of the vehicle,
the roughness of the pavement preceding the bridge, and many other factors
[1] .2 Therefore, the exact shapes of stress-time curves from available field
measurements [2,3,4] vary considerably, as shown (in Fig. 1) for five different
bridges. However, most of the available curves can be characterized as illustrated
by an idealized stress-time relationship (Fig. 2a).
Without a vehicle on the bridge, the weight of the bridge produces a static or
dead-load stress that is not recorded during field measurements, because it
existed when the strain gages were installed. The passage of a vehicle produces a
single major cycle of additional stress that is controlled primarily by the weight
of the vehicle. Smaller vibrational stress cycles, which depend on the dynamic

2 The italic numbers in brackets refer to the list of referencesappended to this paper.

)4 See
,,o .... -~ /

'=~
rx L~.
MID-POIN'T MIO-POINT MID- POINT
BRIDGES
BRIDGE 2
o.4,
iZO ~p~, --t 0,4 Ser
i

,J\ j ~ 9 zO/ ~ " 4

-- A/L.A. 4,A tUtA,%D AA,-

I I ~v ,
ENOOF COVERPLATE ENO OF COVER PLATE

FIG. 1 - E x p e r i m e n t a l load traces for passage of a single vehicle (traces reproduced from
Ref2).

STRESS
j SUPERIMPOSEO VIBRATIONAL STRESS CYCLES

~
---..---~0
MAJOR L I V E - L O A D STRESS CYCLE

nn~
NiHliHiINNnnnnnnr~n ........ "t

L
- gUUUUUUUUU<' . . . . . . . " .....

J ~DEAD-LOAD STRESS
a i IL TIME

STRESS

Smln !
Sr; I Smax

b
I= TIME

STRESS
Kt = SVI/~ r

KZ = In(Svz / S v l )
Sr I

s,,I "-"

Ii- TIME

FIG. 2-Idealized load traces for passage o f a single vehicle.

characteristics of the bridge and vehicle, are superimposed on this major stress
cycle. Vibrational stress cycles also occur after the major stress cycle is complete
and the vehicle has left the bridge. These vibrational stress cycles occur at the
natural frequency of the bridge and usually decrease logarithmically due to
viscous damping. 3
In most of the available stress-time curves, the vibrational stress cycles are
small compared with the major stress cycle. As confirmed by tests described in
the final report, the small vibrational stress cycles superimposed on the major
stress cycle can be neglected. Also, the vibrational stress cycles occurring after
the major cycle can be included in the stress-cycle histogram. Thus, the stress
caused by the passage of a vehicle can be approximated by a single stress-time
cycle (Fig. 2b). The cycle is defined by any two of the three parameters: (1) the
maximum stress, S m a x , ( 2 ) the minimum stress, S m i n , and (3) the stress range,
Sr-
In a few stress-time curye6, particularly curves for a girder bridge with a
suspended span, large vibrational stress cycles occur after the major cycle. This
type of curve can be approximated by logarithmically descending multiple
stress-time cycles (Fig. 2c).

Frequency of Occurrence
The stress spectrum, or stress history, for a particular location in a bridge can
be conveniently defined in terms of the frequency of occurrence of the
maximum (peak) stress in each stress cycle. Usually, frequency-of-occurrence
field data are presented as a histogram showing the percentage of recorded
maximum live-load stresses that fall within a certain stress interval. Figure 3a
illustrates a similar histogram for the total (dead plus live) stresses. For example,
the figure shows that 20.2 percent of the recorded peak stresses were between
7.5 and 8.5 ksi (52 and 59 MPa). The frequency of occurrence of stress ranges
can be represented by a plot with the vertical (stress) scale changed according to
the relationship among Smax , S m i n , and S r. Since stress range is the most
important stress parameter controlling the fatigue strength of bridge members
[5], stress range is used to define the major stress cycle in the present program.
The frequency-of-occurrence data can be presented in a more general form by
dividing the percentage of occurrence for each interval (Fig. 3a) by the interval
width to obtain a probability-density curve4 (Fig. 3b). The probability density is
independent of the interval used in classifying the data and is a continuous
curve. Thus, data from sources that use different stress-range intervals can be
compared by using the probability-density curve. The area under the curve
between any two stress-range values represents the proportion (or percentage) of
stress ranges that fall within this interval. The probability-density curve can be
defined mathematically by an appropriate equation; a nondimensional equation

3 Viscous damping, in which a force proportional to the velocity opposes motion, causes a
progressive decrease in the peak amplitudes, such that the logarithm of the ratio of any two
consecutive peaks (the smaller divided by the larger) is a constant.
4 The points representing each interval are connected by a smooth curve.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
KLIPPSTEIN AND SCHILLING ON SHORT-SPAN STEEL BRIDGES 207

FIG. 3-Frequency of occurrence data.

that can be used to represent a family of different probability-density curves is

most convenient.

Curve Fitting
Two mathematical expressions illustrated in Fig. 4 were considered for use in
representing the frequency of occurrence of stress rangesS: (1)a two-parameter
Rayleigh probability-density function and (2) a three-parameter Erlang proba-
bility-density function. The Rayleigh function is defined for x' > 0 by

p'=x'e -(1/2) (x')2 (1)

In this equation, p' is the nondimensional probability density, and

S r - Srmin
x' - (2)
Srd
s Two additional expressions, normal and log-normal distributions, were considered during
initial evaluations but were discarded.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
208 FATIGUE CRACK GROWTH

RAYLE|GH ERLAN6

I , 2 k k-I
p'= --~-(k) (x') e -kx'
p ' , x , e - "Z( x )

WHERE
x'- Sr- $r~t-o-
Srd

r(k,. f,zk-,,
NON01MENSIONAL
PROEIABILI~'Y DENSITY, p'

Srd P.0 ~ l k=t

- \
0.6
Srmm= 0 O6 =3

0,4 0.4

0.2 0.2

0,0 o.D I
I 2 3
NONDIMENSIONAL STRESS RANGE j X'

PROB~ILITY DENSITY , p

MODAL MEAN
VALU~ VALUE
Srmm S~d
/

3 4 I 2 3 4

STRESS RANGE,S r

FIG. 4-Rayleigh and Erlang probability density.

in which S r is the independent variable (stress range), and Srmin and Srd are the
two parameters (constants) that define any particular probability-density curve6
from the family represented by Eq 1. The Erlang function is defined for x' > 0
by
kk
(x')k-le -Kx (3)
P' - r (k)
in which x' is defined by Eq 2, and k is the third parameter necessary to define a

6 In both the Rayleigh and Erlang funtions, Srmin represents the distance from the origin
to the starting point of the function. Srd represents the distance from the starting point to
the modal value in the Rayleigh function and from the starting point to the mean value in
the Erlang function. The Rayleigh function is a special case of the Erlang function.

particular probability-density curve, k is a nondimensional parameter greater

than 0, and r(k) is the gamma function defined by

r(k) = zk-le-Zaz (4)

The Rayleigh curve always starts with zero probability density at the lowest x'.
In contrast, the Erlang curve starts with the highest probability density at x' = 0
and steadily decreases if the paramenter, k, is equal to 1.0. If the parameter, k, is
greater than 1, the Erlang curve has a shape similar to that of the Rayleigh curve;
if k < 1.0, the curve is asymptotic to the vertical axis at x' = 0.
The two mathematical expressions, Eqs 1 and 3, were fitted to frequency-of-
occurrence field data by using a curve-fitting computer program that is based on
the least-square criterion. The NLWOOD program was selected from a group of
available programs. Specifically, the program determines optimum values for the
two or three parameters of the Rayleigh or Erlang functions, respectively, and
defines an individual probability-density curve of each type. Using trial values of
these parameters, the computer program calculates the theoretical frequency of
occurrences (in percent) of stress ranges within each experimental interval. The
theoretical frequency of occurrence is obtained by integrating the probability-
density equations and by subsequently evaluating the integral over the limits
defined by the stress-range interval of the experimental data. For the Rayleigh
function, integrating Eq 1 for the interval defined at its left side by SrL and at
its right side by SrR results in

SrR _ S r m i n X 2

PR = e -e (5)

The Erlang function, Eq 3, cannot be readily integrated. Consequently, the

numerical integral utilizing Simpson's rule with 16 divisions for the stress-range
interval was used to obtain the frequency of occurrence, PE, for the Erlang
function.
The algebraic difference between the experimental frequency of occurrence
and the corresponding theoretical value is the residua! (Fig. 5). The computer
program automatically changes the parameters and recalculates the residuals
until a minimum value of the sum of the squares of the residuals is obtained.
When the difference in the sum of the squares between two successive iterations
is less than 0.01 percent, it is assumed that the minimum value has been reached.
Thus, the selected parameters give the best possible fit (according to the
least-squares criterion) over the range of experimental S r values.
The results of the curve fitting are summarized in Table 1. Fifty-one data sets
collected from six sources [2,4,6-9] 7 were used. For convenience in comparing

7 This includes all data available during the early part (1971) of the project. Sources that
did not contain suitable data because measurements were made under specific truck loadings
rather than under normal traffic conditions were excluded.

F--I I--] L__ RES~OUAL

r-'-I

-- "7

O I. 0 2.0 5 .O 4.0
STRESS R A N G E , ksl

FIG. 5-.4 ctual versus computed frequency o f occurrence.

the Rayleigh and Erlang curves for a given set of data, the mean, Srmean , and
the minimum, Srmin , instead of Srd and Srmin , are given to define a particular
probability-density curve. The mean is equal to Srmin + 1.23Srd for the
Rayleigh curve and Snniu + Srd for the Erlang curve. The parameter k is also
given to complete the definition of the Erlang curve. The sum of the squares of
the residuals, which is a measure of the closeness of fit, is listed.
As expected, the sum of the squares for the three-parameter Erlang curve was
less than the corresponding sum for the two-parameter Rayleigh curve for most
of the 51 sets of data. The Erlang curve provides a much closer fit than the
Rayleigh curve for data (classified as "descending" in Table 1) that starts with
the highest probability density at or near S r = 0 and steadily decreases; such a
probability-density curve usually results if the small vibrational stress cycles are
included in the data. The closer fit provided by the Erlang curves, of course,
results primarily from the use of the third parameter, k, which varied from 1.0
to 8.2 for the 51 sets of data.
The Rayleigh curve was chosen for use in the test program because two
parameters were found to be sufficient to represent a wide variation of skewed
data and because many more tests would be required to establish the fatigue
strength in terms of three parameters than in terms of two parameters. Also, the
asymptotic tail of the theoretical Rayleigh probability-density curve was
truncated for the testing program. Specifically, a standard Rayleigh curve s that
has a width of 3Srd was used as shown in the upper part of Fig. 6. This width
was chosen to permit a reasonable factorial experiment within the limitation
that the peak loads must not exceed the yield load. The available field data were
considered insufficient to provide an accurate representation of the probability
8 "Standard Rayleigh curve" is used herein to refer to the family of truncated Rayleigh
curves that are used as the standard probability-density curves for the testing program.

of extreme values of Sr (3Srd and above) in actual bridges. The probability of Sr

values above 3Sra is only 1.1 percent and suggests that the cutoff at 3Srd is of
little practical significance. Furthermore, the value of Srrms, which is a very
meaningful parameter to represent variable-amplitude fatigue data, is shifted
only 2.6 percent by truncating a Rayleigh curve at 3Sra and would cause an
estimated shift in the fatigue life of less than 10 percent.
The constant 1.011 has been inserted into the mathematical expression
defining the standard Rayleigh curve to make the area under the truncated curve
equal to 1.000, so that the curve represents 100 percent of the occurrences.
Thus
P'= 1.Ollx'e -0/2) (x')2 (6)
The modal, median, mean, and root-mean-square values of x' for the standard
truncated curve are shown in the figure. These values are slightly different from
the corresponding x' values for a full curve, which are equal to 1.000, 1.177,
1.253, and 1.414, respectively. The root-mean-square (rms) value is equal to the
square root of the mean of the squares of the individual values.
Based on the definition of x' (Eq 2), the standard Rayleigh curve represents a
family of probability curves for stress ranges that may be shifted from the origin
of the coordinate system by an amount equal to Srmin and changed in width by
varying Srd as shown in the center of Fig. 6. Each particular probability-density
curve from the family can be defined by any two of the following three
parameters: Srm , Srd , and Srmin. In the testing program, the curves are defined
in terms of Srm and Srd/Srm. Four values of Srd/Srm are used in the testing
program: 0.00, 0.25, 0.50, and 1.00. Curves for these values are shown at the
bottom of Fig. 6. While the root mean square of the x' value (X'rms) is a
constant for all curves from the Rayleigh family, the root mean square of the Sr
values (Srrms) varies slightly with the ratio Srmin/Srd. However, the maximum
difference between the two root mean square values, which occurs when
Srmin/Srd is between 1 and 2, is less than 3 percent.

Control Tapes for Fatigue Tests

Several methods were considered to generate and control the variable-
amplitude random-sequence loads used in the experimental portion of this
project such that these loads would follow a specific Rayleigh probability
density function. A filtered one-sided signal of a random-noise generator was
considered but rejected because this method would have caused extensive
operating problems and the load sequence generated in a trial setup was not
random. Also, a permanent computer hookup with an available random-
sequence length of up to 2 million cycles or a three-track magnetic tape were
considered. These possibilities were rejected because their reliability over a
three-year period was considered questionable. Instead, it was decided to use a
digital-tape control with looped tapes that contain 500 load cycles per tape in a
random sequence. The chosen number of load cycles (500) approximately
represents the observed average number of truck passages per day [2,4,6-9], and

TABLE 1 - Comparison of Rayleigh and Erlang curves.

Type Rayleigh Curve Erlang Curve

Data of Srmin ' Fit . Fit
Set Curve ksi S r ~ an' Parameter Sr~sln' Sr~ an' k Parameter

1 D 0.52 0.79 12.1 0.56 0.81 1.87 2.8

2 P 0.23 2.24 29.1 0.22 2.35 3.30 18.7
3 D 0.46 0.80 59.6 0.60 0.90 1.10 0.6
4 P 0.20 1.94 18.0 0.00 1.99 4.06 30.2
5 P 0.00 1.36 42.0 0.25 1.50 1.84 2.2
6 P 0.14 1.52 80.9 0.45 1.71 1.61 8.3

7 D 0.42 0.97 13.7 0.48 1.00 2.91 2.3

8 P 0.47 2.04 157.0 0.00 2.10 5.85 180.8
9 D 0.52 0.81 28.7 0.60 0.84 1.05 1.0
10 P 0.32 1.28 20.0 0.34 1.34 3.08 5.8
11 D 0.43 0.89 34.3 0.55 0.96 2.03 2.8
12 P 0.21 2.37 17.2 0.25 2.49 2.96 12.4

13 P 0.07 1.54 3.5 0.00 1.60 3.55 4.0

14 P 0.11 1.47 29.3 0.03 1.52 3.54 28.8
15 P 0.00 1.01 55.2 0.35 1.15 1.11 0.6
16 P 0.05 1.33 68.2 0.38 1.49 1.48 2.6
17 P 0.52 1.84 208.5 0.00 1.86 6.86 225.9
18 P 0.30 0.95 14.1 0.41 0.82 2.02 2.1

19 P 0.09 1.37 20.2 0.08 2.02 3.22 12.4

20 P 0.17 0.95 80.2 0.07 1.00 4.11 81.6
21 P 0.12 0.35 0.1 0.00 0.34 8.18 0.2
22 P 0.22 0.74 19.2 0.16 0.78 3.99 10.6
23 P 0.15 1.05 59.8 0.30 1.18 1.97 31.6
24 P 0.23 1.00 56.0 0.26 1.03 2.93 49.5

25 P 0.14 0.68 30.8 0.17 0.71 2.86 25.5

26 P 0.20 1.02 35.3 0.22 1.07 2.86 28.7
27 P 0.23 1.91 6.8 0.00 1.97 4.11 9.2
28 P 0.20 1.56 37.2 0.00 1.59 4.28 35.4
29 P 0.01 1.02 48.1 0.16 1.16 2.04 19.0
30 P 0.23 1.61 48.8 0.38 1.71 2.33 36.6

31 P 0.08 0.88 118.4 0.40 1.00 1.05 43.7

32 P 0.00 1.26 72.8 0.35 1.47 1.42 31.6
33 D 0.50 0.96 9.8 0.48 0.95 4.19 9.0
34 P 0.31 1.15 4.3 0.21 1.16 4.29 4.6
35 P 0.40 1.14 0.0 0.00 1.15 7.82 2.1
36 D 0.31 0.88 6.1 0.33 0.92 3.61 1.8

37 D 0.37 0.79 2.8 0.36 0.81 3.80 0.6

38 P 0.28 1.08 4.9 0.25 1.12 3.76 6.1
39 D 1.33 1.69 3.0 1.33 1.70 3.51 0.4
40 P 0.82 2.37 5.7 0.44 2.39 5.15 9.6
41 D 0.80 1.79 123.3 0.92 1.86 2.70 79.2
42 P 0.46 1.26 18.2 0.51 1.28 2.95 6.0

43 D 0.00 1.07 151.4 0.43 1.34 0.98 5.9

44 D 0.05 1.01 42.4 0.57 1.28 1.08 8.9

TABLE 1 Continued.

Rayleigh Curve Erlang Curve

Type
Dataset Curve~ Sr~in, Sr~ian, Fit
Parameter Sr~s~n, Sr~ian, k Fit
Parameter

45 D 0.26 0.78 22.9 0.33 0.80 2.19 9.6

46 P 0.21 1.20 4.6 0.40 1.29 2.46 0.0
47 D 0.16 0.81 110.6 0.35 0.88 1.08 36.7
48 D 0.00 1.19 70.3 0.40 1.38 1.63 31.6

49 P 0.27 1.17 3.9 0.15 1.21 4.52 7.1

50 P 0.32 1.20 3.6 0.30 1.23 3.78 0.6
51 D 0.31 0.75 15.8 0.43 0.76 1.11 13.8

NOTES-1 ksi = 6.895 MN/m2.

Curves that start at a peak and steadily decrease are referred to as descending curves
and axe identified by a D; curves that increase to a peak and then decrease are called
peak curves and are identified by a P.

Srmin is the minimum stress range in the spectrum; Srmean is the mean stress range
for the spectrum; k is the nondimensional shape parameter; and the fit parameter is the
sum of the squares of the residuals, lower values of the fit parameter indicate a closer fit.

the random sequence o f the load cycles within each tape resembles the observed
randomness of recorded stress-range histories [8].
The punched tapes that are used to control the fatigue tests were generated by
a computer program [10]. Two methods were considered to determine the
individual loads necessary to generate stress-range cycles that correspond to a
given probability-density curve: (1) subdivide the width of the curve (3Srd) into
a certain number of stress-range intervals of equal widths to produce a
frequency-of-occurrence graph similar to that shown in Fig. 5, and (2) subdivide
the area below the probability-density curve into a certain number of intervals
with equal areas but different widths. For the first method, each interval would
contain a different percentage of the total area and thus would represent a
different percentage of the total number of cycles on each test control tape.
Each percentage would be multiplied by the total number of cycles on the tape
to determine the number of cycles at the stress range corresponding to the
midpoint of that interval. Since the number of cycles corresponding to each
stress-range interval would not be an integer, rounding off of fractional cycles
would be required. Therefore, the method was rejected and the computer
program was written to make use of the second method.
The program calculates 500 individual loads that satisfy the standard Rayleigh
probability-density curve, arranges them in a random sequence, and punches a
control tape defining these loads in ASC II code. A separate tape is required for
each different value of Srd/Srm , but different levels of Srm and Smin are set

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
214 FATIGUE CRACK GROWTH
NONDIMENSIONAL PROBABILITY DENSITY , Dr
I ,.3rB _,tx'R.~)_
E 1.230 (X'mean)
[ I.,6B /i (X'm,(li,n)

0 . 4 0 " 6 ~ ' : 1.011x'll -I/2(x')2

(St- Srrnin ) /$rO
O.Z
0.0 2 3

NONDIMENSIONAL STRESS RANGE,x'

PROCaABILtTY DENSITY iPl I/ks;

p : p'/Srd

Srmax
L~
TRESS RANGE, Sr

PROBABILITY DENSITY , p , I/ksi

Er,,Sro"
~1~./ Srd/Srm = O'SO

Srm
STRESS RANGE

FIG. 6-Characteristics of Rayleigh probability curves.

manually on the testing-machine controls. The program is written in FORTRAN

IV.
The 500 individual loads are calculated by dividing the area below the
nondimensional probability-density curve, Eq 1, into 500 vertical segments
(bars) of equal area. The width of the bars varies to provide equal areas. The
midwidth (or more precisely, the value o f x ' that bisects the bar into segments of
equal area) of each of these bars corresponds to a load with a frequency of
occurrence of 1/500. The value x'n corresponding to the midwidth of the nth bar
is calculated by integrating the nondimensional probability-density curve from 0

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
KLIPPSTEIN AND SCHILLING ON SHORT-SPAN STEEL BRIDGES 215
I
to x n and equating the result to the desired area, (n - 1/2)/500. The result of
the integration is 1.011 times (1 - e- 0 / 2 ) (x~)2). Thus

x 'n = x/-2 l n i l -0.00198 (n-i3.5)] (7)

r
The values of x n vary from 0 to 3. In generating the tapes for the main testing
program, the values o f x ' n corresponding to the right side of the bars, rather than
to midwidths, were used for convenience. This procedure is equivalent to
omitting the 0.5 in Eq 7 and is permissible because the x' interval is very small.
The resulting increase in Srrms does not exceed 0.3 percent.
For convenience in operating the fatigue-testing equipment, the corresponding
stress ranges, Srn , on the control tape are expressed as a three-digit percentage of
the maximum stress range, Srmax ; thus the control-tape values vary from a
minimum of 000 to a maximum of 999 representing 0.0 and 99.9 percent,
respectively. A different tape is required for each of the four different values of
S r d / S r m in the program: 0.0 (constant amplitude), 0.25, 0.50, and 1.00.
The computer program arranges the calculated Srn/Srmax values in a random
sequence by generating 500 random-sequence numbers, consecutively assigning
these sequence numbers to the 500 Srn/Srmax values, and finally arranging the
control-tape values according to the assigned sequence numbers. The random
numbers are generated by an available CDC (Control Data Corporation)
computer subroutine.
The control tape defines both the peak and valley of each load cycle; the
valley is equal to 0, and the peak equals the calculated control-tape value. In the
actual test, the level of maximum cyclic stress corresponding to a tape
command of 999 is set by the testing-system controls, and a constant minimum
stress may be superimposed on the cyclic stress. The ends of the tape are
connected to form a loop which is continuously cycled through the test. Thus,
within the loop of 500 cycles, the sequence of the load levels is fully random;
however, since the loop is repeated many times before the specimens fail, the
overall sequence is pseudo-random.
The control tape just described was used for the major test program. To assure
that the chosen number of load levels within the spectrum and the number of
load cycles and their relative positions within a sequence (or loop) would not
affect the test results, 24 tests were conducted. In these tests, 100 and 500 load
levels were used to subdivide the load spectrum, and sequence lengths of 100,
500, and 5000 cycles were used within a loop. The results of the additional tests
were not significantly different from the tests conducted with the tape used for
the major test program.

Conclusions

The investigated field data of steel bridges indicate that the passage of a
vehicle over a bridge produces a single major stress cycle, St, with superimposed

vibrational stress cycles that may continue after the vehicle has left the bridge.
For most types o f short-span bridges, the vibrational stress cycles are small
enough to be neglected. The major stress cycle is added to the existing dead-load
stress. Therefore, the stress spectrum for bridges can be def'med in terms of the
minimum (dead load) stress, Stain, which remains essentially constant during the
life of the bridge, and the frequency of occurrence o f major stress cycles, S r. The
chosen probability.density function, along with the available controls of the test
equipment and the command signals generated by continuous control tapes,
adequately permits variable-amplitude random-sequence fatigue testing repre-
sentative of the actual stress spectrums for short-span steel bridges.

References
[1 ] Veletsos, A. S. and Hnang, T., Journal of the Structural Division, American Society
of Civil Engineers, VoL 96, No. EM5, Oct. 1970, pp. 617-619.
[2] Cudney, G. R., "The Effects of Loadings on Bridge Life," Research Report No.
R-638, Michigan Dept. of State Highways, Jan. 1968.
[3] Cudney, G. R., Journal of the Structural Division, American Society of Civil
Engineers, Vol. 94, No. ST12, Dec. 1968, pp. 2725-2737.
[4] Heins, C. P. and Sartwell, A. D., "Tabulation of Dynamic Strain Data on
Three-Span Continuous Bridge Structures," Civil Engineering Research Report No.
33, University of Maryland, Nov. 1969.
[5] Fisher, J. W., Frank, K. H., Hirt, M. A., and McNamee, B. M., "Effect of Weldments
on Fatigue Strength of Steel Beams," NCHRP Report 102, 1970.
[6] Galambos, C. F. and Armstrong, W. L., "Loading History of Highway Bridges,"
Highway Research Record 295, Highway Research Board, 1969.
[7] Heins,C. P. and Sartwell, A. D., "Tabulation of 24-Hour Dynamic Strain Data on
Four Simple-Span Girder-Slab Bridge Structures," Civil Engineering Research
Report No. 29, University of Maryland, June 1969.
[8] Heins, C. P. and Sartwell, A. D., "Tabulation of Dynamic Strain Data on a
Girder-Slab Bridge Structure During Seven Continuous Days," Civil Engineering
Research Report No. 31, University of Maryland, Sept. 1969.
[9] Christiano, P., Goodman, L. E., and Sun, C. N., "Bridge Stress Range History and
Diaphragm Stiffening Investigation," report by the Department of Civil Engineering
and Hydraulics, University of Minnesota, June 1970.
[10] NCHRP Project 12-12, Interim Report, Research Laboratory, United States Steel
Corporation, Project No. 76.019-001, 1 Oct. 1972.

Fatigue Crack Growth Under

Variable-Amplitude Loading in
Various Bridge Steels

REFERENCE: Barsom, J. M., "Fatigue Crack Growth Under Variable-Amplitude

Loading in Various Bridge Steels," Fatigue Crack Growth Under Spectrum Loads,
ASTM STP 595, American Society for Testing and Materials, 1976, pp. 217-235.

ABSTRACT: Well-conceived procedures used to study the safety and reliability of

structures recognize that the performance of a structure or a structural component is
governed not only by material properties but also by the design~ fabrication,
inspection, erection, and use of the structure. These parameters govern the initiation
of subcritical cracks and their propagation to critical dimensions, and therefore,
determine the useful fatigue life of structural components subjected to load
fluctuations.
This paper presents the results of the first phase of an investigation sponsored by
the National Cooperative Highway Research Program, Project 12-14, to study
"Subcritical Crack Growth in Steel Bridge Members."
The paper describes the fatigue-crack-growth behavior of various bridge steels (A36,
A588-A, A588-B, A514-B, A514-E, and A514-F) under variable-amplitude random-
sequence stress spectra such as occur in actual bridges. The fatigue crack growth-rate
data were obtained by using wedge-opening-loading specimens tested under variable-
amplitude random-sequence load spectra that are represented by a Rayleigh
distribution function.
The data obtained for these steels showed that the average fatigue crack growth
rates, da/dN, under variable-amplitude random-sequence load fluctuation and under
constant-amplitude load fluctuation agreed closely when da/dN was plotted as a
function of the root-mean-square stress intensity factor range, AKrm s. Thus, within
the limits of the present investigation, the average fatigue crack-growth rates, da/dN,
of various bridge steels subjected to variable-amplitude random-sequence load
fluctuations, such as occur in actual bridges and to constant-amplitude load
fluctuations, can be represented by the equation
da =A (AKrms)n
dN
where AKrms is the root-mean-square stress intensity factor fluctuation, and A and n
are material constants.

KEY WORDS: crack propagation, fatigue (materials), inspection, steels, loads

(forces)

Steel bridges have an excellent service record extending over millions of

operational years. However, the collapse o f the Point Pleasant Bridge [1] 2 has

1 Section supervisor, Mechanical Behavior, Research Laboratory, United States Steel

Corporation, Monroeville, Pa. 15146.
2 The italic numbers in brackets refer to the list of references appended to this paper.

217
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM International
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
218 FATIGUE CRACK GROWTH

led to an increasing concern about the possibility of catastrophic fractures in

steel bridges.
The safety and reliability of steel bridges are governed by many interrelated
factors. Fracture of a bridge detail can occur as a result of improper material
properties, design, fabrication, inspection, erection, or operating conditions.
Consequently, a need exists for a systematic evaluation of the procedures used
to select steels for bridges, of the philosophy and trends in bridge design and
analysis, and of the procedures used to fabricate and to inspect bridge details. A
thorough understanding of the several technologies used in bridges is the basis
for the development of a systematic procedure to evaluate the safety and
reliability of steel bridges. A description of these technologies and the
development of fracture-control plans for steel bridges are beyond the scope of
this paper.
The rate of fatigue crack growth has been investigated in many materials and
has been found to depend on the magnitude of the stress range, AO, the crack
length, a, and the material properties. The stress range and crack length can be
incorporated in a single-term parameter, zkKi, which represents the fluctuation
of the stress intensity in the vicinity of the crack tip. Consequently, the rate of
fatigue crack growth, da/dN, has been related to AK I by the empirical
relationship
= A(AK~)n (I)
dN
where
a = crack length,
N = number of cycles, and
AKI = fluctuation of the stress-intensity factor.
A and n are constants that reflect effects of material properties and
environments. Sufficient data obtained under constant-amplitude cyclic-load
fluctuations are available in support of this observation [2,3].
Several investigators [4-8] have noted that changes in cyclic-load magnitude
and the order (sequence) of the variation of cyclic-load magnitude can lead to
accelerated or retarded rates of fatigue crack growth. Consequently, fatigue
crack-growth rate models that account for these interaction effects must be
developed to predict the crack growth of structural components under
variable-amplitude random-sequence load fluctuations. Accordingly, the present
investigation was undertaken to study fatigue crack growth in various bridge
steels under constant-amplitude, and random variable-amplitude cyclic-load
fluctuations such as occur in actual bridges. The data were analyzed by using
linear-elastic fracture-mechanics methods.

Materials and Experimental Work

Materials
The steels studied in the present investigation are A36, A588-A, A588-B,
A514-E, and A514-F. One-in.-thick (25.4-mm) plates of each steel were used in

this study. The chemical composition and mechanical properties of the steels are
given in Tables 1 and 2, respectively. This paper includes some data for A514-B
that were published earlier [9,10].

Experimental Procedure
The details for the experimental procedure used in the present investigation
have been published elsewhere [9,10]. A brief description of the procedure is
presented in this section. The fatigue crack propagation data were obtained by
using wedge-opening-loading (WOL) specimens. The data were obtained at a
cyclic-load frequency equal to 300 cpm. The tests were conducted in a
room-temperature air environment by using a 50-kip (222-kN) high-strain-rate
Materials Testing System (MTS) machine. Alignment was obtained by carefully
machining specimens and other auxiliary parts and by using universal joints to
toad the specimens.
The specimens were tested under constant-amplitude and under variable-
amplitude cyclic-load fluctuations in the form of a sine wave. In each test the
fatigue crack was initiated and propagated in tension-to-tension at a constant
minimum load of 200 lb (890 N) and at a constant- or variable-amplitude
maximum load, which were controlled within -+1.0 percent. The crack was
initiated and propagated from the notch root so that at the time crack-length
measurements were begun, the total crack length, a, was equal to 1.0 -+0.001 in.
(25.4 -+ 0.0254 mm).
Fatigue-crack-growth rates were measured optically with a Type M-101
Gaertner microscope mounted in a micrometer slide. To improve the accuracy of
measuring the rate of crack extension, a series of hardness indentations was
made on the surface (with a Vickers pyramid hardness testing machine) along a
line parallel to the plane of the initial crack and in the direction of expected
crack extension.
The variable-amplitude random tests were conducted in a 50-kip high-strain-
rate MTS machine, used in conjunction with a tape-controlled digital program-
mer [11]. The programmer accepts manually entered commands or programmed
digital commands from punched tape, and produces an electronic control signal
that varies with time as specified by the input instructions. Continuous cycling
of tapes can be achieved by splicing the tape to form a closed loop.
Stress spectra, typical of bridge loadings, have been thoroughly investigated
[9,12]. In the crack-growth tests, all loadings followed a Rayleigh distribution
curve, with the ratio of the load-range deviation to the modal (peak) load
(Pra[Prm) equal to either 0 or 1.0. A block of 500 individual (usually different)
loads satisfying one of these distribution curves was repeated throughout each
test. Within the block, the loads were arranged in random sequence.
Although the cyclic-loading spectrum was not changed during a test, the stress
intensity factor range, AK, for successive blocks increases as the crack length
increases. Thus, a single test gives crack-growth rates for a range of AK values.
The value of AK corresponding to a given crack length and loading was
calculated from an available theoretical analysis [13].
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
bo
r,o
o
-aq
-4
c
m
TABLE 1 - Chemical composition o f steels tested-percent {check analysis}. t~
2o
c)
Sol Insol Total 7~
Steel Heat No. Plate No. C Mn P S Si Cu Ni Cr Mo V Ti A1 AI AI Na Ob B 63
20
O
-4
A36 74C515 195264 0.26 1.14 0.009 0.023 0.038 0.031 0.042 0.05 0.016 ~-O.005 ~--,O.005 0.001 0.001 0.002 0.004 90 ND c "r

A588-A 67C611 193804 0.13 1.14 0,011 0.026 0.22 0.31 0,022 0.57 0,020 0.033 0,005 0,040 0.001 0.040 0.006 15 ND

A588-B 662J487 551528 0.11 1.16 0,011 0.018 0.27 0.26 0.35 0.63 0.016 0.052 ~,0.005 0,021 0.001 0.022 0.009 27 ND

A514-E 50343 P70074 0.16 0,61 0.008 0,018 0.28 0.21 ND 1.84 0.53 ND 0,063 0.033 0.001 0,034 0.010 41 0.0020

A514-F 70C125 79873A1 0.17 0.60 0,014 0,013 0.22 0.28 0.78 0.56 0.45 0,050 0,005 0.024 0.001 0,025 0.005 26 0.0041

a Kjeldahl determination.
b Parts per million.
c ND = Not determined.

Yield S t r e n g t h Tensile Elongation Reduction Charpy V-Notch

(0.2% Offset), Strength, in 2 In., o f Area, Energy Absorption,
Steel Heat No. Plate No. ksi ksi % % at 72~ ft " lb

Longitudinal
A36 74C515 195264 43.6 78.2 28.2 62.4 28
A588-A 67C611 193804 54.9 81.7 28.2 68.3 69
A588-B 662J487 551528 55.6 82.1 27.8 76.2 66
A514-E 50343 P70074 107.9 122.6 19.0 65.1 68
A514-F 70C125 79873A1 126.0 134.0 18.3 57.8 45
co
Transverse
A36 74C515 195264 43.9 78.6 25.3 57.4 26 0
A588-A 67C611 193804 54.6 81.3 24.2 54.0 31 o
A588-B 662J487 551528 55.8 82.7 24.2 60.5 29 z
A514-E 50343 P70074 106.1 123.1 17.3 55.5 43 <
A514-F 70C125 79873A1 126.0 134.0 18.3 58.0 32
0
C
a Tension and impact s p e c i m e n s were taken in t h e l o n g i t u d i n a l o r i e n t a t i o n f r o m the q u a r t e r - t h i c k n e s s p o i n t o f the plates, w h i c h were all 1 in. co
thick. Tension-test results are the average o f three 0.505-in.-diameter t e n s i o n specimens, a n d i m p a c t results are t h e average o f two C h a r p y V - n o t c h
specimens. t~
Ill
r
N O T E - C o n v e r s i o n factors: .q
1 i n c h = 25.4 m m m
m
1 ksi = 6.895 M N / m 2 r-
1 ft 9 lb = 1.36 J
C = 5/9 ( F - - 3 2 ) bO
ro

Results and Discussion

Fatigue Crack-GrowthBehavior

Most fatigue crack-growth tests are conducted by subjecting a fatigue cracked

specimen to constant-amplitude cyclic-load fluctuations. Incremental increase of
crack length is measured and the corresponding number of elapsed load cycles is
recorded. Various a-versus-N curves can be generated by varying the magnitude
of the cyclic-load fluctuation or the size of the initial crack or both. These
curves reduce to a single curve when the data are represented in terms of
crack-growth rate per cycle of loading, da/dN, and the fluctuation of the
stress-intensity factor, AKI, because ' ~ I is a single-term parameter that
incorporates the effect of changing crack-length and cyclic-load magnitude.
Incremental increase of crack length and the corresponding number of elasped
load cycles can be measured under variable-amplitude random-sequence load
spectra. However, unlike constant-amplitude cyclic-load data, the magnitude of
AK I changes for each cycle. Reduction of data in terms of fracture-mechanics
concepts requires the establishment of a correlation parameter that incorporates
the effects of crack length, cyclic-load amplitude, and cyclic-load sequence.
It is desirable to determine the magnitude of constant-amplitude cyclic-load
fluctuation that results in the same a-versus-N curve obtained under variable-
amplitude cyclic-load fluctuation when both spectra are applied to identical
specimens (including initial crack length). In other words, the objective is to find
a single stress intensity parameter, such as mean, modal, or root mean square,
that can be used to define the crack-growth rate under both constant and
variable-amplitude loadings. The selected parameter must characterize the
distribution curve.
Fatigue crack-propagation behavior under variable-amplitude random-sequence
stress spectra such as occur in actual bridges has been investigated as part of
NCHRP Project 12-12 [9-13]. The tests were conducted on A514-B steel under
variable-amplitude random-sequence load spectra having Prd/Prm values of 0.5
and 1.0; a typical portion of the 500-cycle loading block for each is shown in
Fig. 1. The data showed that the average fatigue crack-propagation rates under
variable-amplitude random-sequence and ordered-sequence load fluctuation, Fig.
2, are approximately equal to the rate of fatigue-crack propagation under
constant-amplitude cyclic-load fluctuation equal to the root-mean-square (rms)
value of the variable-amplitude function. The root mean square is the square
root of the mean of the squares of the individual load cycles in a spectrum; it is
related to the modal value of the spectrum, as indicated in Fig. 3. The average
fatigue crack-propagation rates, da/dN, under variable-amplitude random-load
fluctuation, ordered-load fluctuation, and constant-amplitude load fluctuation
were found to agree closely when da/dN was plotted as a function of the
root-mean-square stress intensity factor range, Agrm s , Fig. 4. Thus, within the
limits of the available experimental data, the average fatigue crack-growth rates

FIG. 1-Two variable-amplitude random-sequence load fluctuations studied in this

investigation." (a) Prd/P rm = 1.0 and (b) Prd/Prm = 0.5.

per cycle, da/dN, under variable-amplitude stress spectra such as occur in actual
bridges can be represented by the equation

da = A ( ~ r m s ) n (2)
dN

where A and n are material constants.

The root-mean-square value of the stress-intensity factor under constant-
amplitude cyclic-load fluctuation is equal to the stress intensity factor
fluctuation. Consequently, the average fatigue crack-growth rate can be
predicted from constant-amplitude data by using Eq 2.

FIG. 2- Various random-sequence and ordered-sequence load fluctuations studied in this

investigation [18,19]: (a] random sequence, (b) ascending sequence, (c] descending
sequence, (d) ascending-descending sequence, and (el constant amplitude.

Fatigue Crack Growth in Various Bridge Steels

The preceding results were obtained by testing A514-B steel. Because several
investigators [4-8] have noted that changes in cyclic-load magnitude can lead to
accelerated or retarded rates of fatigue crack growth, the applicability of the
root-mean-square model for correlating crack-growth rates under random
loading, such as occurs in actual bridges, must be established for bridge steels of
various yield strengths. Consequently, the fatigue crack-growth rates under
constant-amplitude and variable-amplitude random-sequence load fluctuations

J
O'rd/O'rm 9 0.5
O'r rain" 0~50"rm
O'rd/O'rm" 0
O'r max 9 20"rm = 40'rd
O'rm t 9 1.207 O'rm
b=

i
o-too x = O, mllx

.E ~, b o'r
b
. ~

9 VARIES FOR
b[ 1 EACH CYCLE
F - I00 50 O t-
IF I00 50 0 t"
O"
o" O-rd/O.rm, 1.0
~t O'r min = O
~r max ~ 3~rm " 3 ~
J T Crrd/O'rm-O
~rm$ " 1.4i4 ~ r m

i~ x9Om
'xo

t
!,
F'IO0 50 0

FREQUENCY TIME
tr F" IOOSO 0
FREQUENCY TIME
t-

OF OCCURRENCE OF OCCURRENCE
OF O"r VALUES, % OF o,r VALUES, %

FIG. 3-Stress spectra, constant firms.

were investigated in A36, A588-A, A588-B, A514-E, and A514-F steels. All
loadings followed a Rayleigh distribution curve, with the ratio of the load-range
deviation to the modal (peak) load (Prd/Prm) equal to either 0 or 1.0. Data of
crack length and the corresponding number of elapsed load cycles obtained by
subjecting identical specimens of A588-A steel to the variable-amplitude
random-sequence load spectrum are presented in Fig. 5. Similar data were
obtained for the other steels. The figure also includes data obtained under
constant-amplitude cyclic-load fluctuation (Prd/Prm = 0). The toad range, ~d~,
for every cycle in the constant-amplitude tests was equal to APrm. The data
show that the fatigue life under constant-amplitude cyclic-load fluctuations was
longer than the life obtained under random-sequence load spectra having the
same value of APrm. The data for each steel are represented in Figs. 6 through
10 in terms of crack-growth rate, da/dN, and the root-mean-square stress
intensity factor range, AKrms .

~ I 'I I ~/I ' I ' I '_

o
8
#
r

d
~- 2

o
? i6 6 _
~ -
U ONSTANT AMPLITUDE
< 8
o ASCENDING ORDER
0 o DESCENDING ORDER
i
W 6 /~ v ASCENDING-DESCENDING
ORDER
P z= RANDOM-SEQUENCE
< 4 I ksi ~ 9 1.0998 MN/m 3/2
l i n c h , 25.4 mm
I i l i I i l i
6 8 I0 20 40 60 80 I00

Z~Krms, ksl

FIG. 4-Summary of crack growth rate data under random-sequence and ordered-sequence
load fluctuations for A 514-B steel [18,19].

The data presented in Figures 6 through 10 show that, within the limits of the
present experimental work, the average fatigue crack-growth rates, da/dN, in
various bridge steels subjected to variable-amplitude load spectra as occur in
actual bridges can be represented by Eq 2. The root-mean-square value of the
stress intensity factor range under constant-amplitude cyclic-load fluctuation is
equal to the stress intensity factor range. Consequently, the average fatigue
crack-growth rate under variable-amplitude load spectra as occur in actual
bridges can be predicted from constant-amplitude data by using Eq 2.

Summary
The results of this phase of an investigation to study the fatigue crack-propaga-
tion behavior of various bridge steels under variable-amplitude stress spectra such
as occur in actual bridges may be summarized as follows:
1. The average fatigue crack-growth rate, da/dN, under variable-amplitude
random-sequence load spectra that can be represented by Rayleigh distribu-
tion functions can be related to various stress intensity factor ranges, ,~urf,
that are characteristics of the distribution function (for example, mean,
root-mean, or root-mean-square AK).
2. The average fatigue crack-growth rates, da/dN, under variable-amplitude

1T WOL SPECIMEN 1 LB = 4.448 N

1 INCH = 2.5.4 mm
Pmin = 200 LB A K IN KSI ~ / I N C H
1 KSI ~ / I N C H = 1.099 MN~m 3/2
2.0 Ap ~ CONSTANT AMPLITUDE ,
&Prm --~ VARIABLE A M P L I T U D E ~Prm[Pr~ = 1.0)

1.9 AP = 3000 LB AP = 2500 LB

O
13 O
1.8 O O
[]
APrm = 2000 LB O
[] O
1.7 [] 9 O
[] 9 O
tu O
..u APrm = 3000 LB
[] 9 O O0
1.~3 I 9 O
=< E []# t o r
~? n 6O-,I~,~K = 17.6 0 0
1.5 E 0
m [] 9 0
i [] 9
(3 o
k- | 9 0 Z
1.4 0
u. 0 9 0
3~
1,3 9 o
9 o 9 ~K= 17.3
O o
o 9 0 c
1.2 0
0
n...q,~
9 A K= 1 7 . ~ eO.q.,~ K = 17.g 0 DO
1.1 O A 5 8 8 G R A D E A STEEL
iO ~ 0: O O 111
1,0 I I I I .... I I I I, I
50 100 150 200 250 300 350 400 450
m
ELAPSED CYCLES X103 m
f-
u~
FIG. 5-Fatigue crack growth under constant-amplitude and variable-amplitude load fluctuation/or A588-A steel.

,,,J

./:I
o
2

0j
w
10_5
A36 STEEL

Q
r 6

gi 4

u. 9 RANDOM
A RANDOM
O CONSTANT
9 CONSTANT

1 KSI ,/INCH = 1.099 MN/m 3/2

1 INCH = 2,64 mm
10--G

6 I I I I I I I I
t0 2O 30 40 50 60 80 100
~.Krms, ksi i ~ ' h "

FIG. 6-Crack growth rate as a function of the root-mean-square stress intensity factor for
A 36 steel.

random-sequence load spectra such as occur in actual bridges and under

constant-amplitude load fluctuations can be represented by the equation

d.~a = A(~Krms )n
dN
where z21Krms is the root-mean-square stress intensity factor range, and A and
n are constants for a given material.
3. The preceding relationship was found applicable to the following steels: A36,
A588-A, A588-B, A514-B, A514-E, and A514-F.
4. The average fatigue crack-growth rates, da/dN, of the various steels studied
under variable-amplitude random-sequence load fluctuations such as occur in
actual bridges are equal to da/dN obtained under constant-amplitude load
fluctuation when the stress intensity factor range, z2~K, under constant-

z
-~ 1o-S
d
8

A588 GRADE A
0 STEEL

9 RANDOM
u. 2
& RANDOM
RANDOM
O CONSTANT
9 CONSTANT
10-6
1 KSI ,,/INCH = 1.099 MN/m 3/2
8 1 INCH = 25.4 mm

6 f,
I I I I I I i I I I
10 20 30 40 50 60 80 100
AKrms, ksi~/i-nch

FIG. 7-Crack growth rate as a function of the root-mean-square stress intensity factor for
A588-A steel.

amplitude load fluctuation is equal in magnitude to the AKrms of the

variable-amplitude spectra.

Acknowledgment

The author appreciates the assistance of E. J. Imhof, Jr., in developing the

experimental procedure, conducting the experimental work, and analyzing and
interpreting the data.
This study was conducted under National Cooperative Highway Research
Program 12-14. The opinions and findings expressed or implied in this report are
those of the author. They are not necessarily those of the Highway Research
Board, the National Academy of Sciences, the Federal Highway Administration,

10-4

"G
4

z
~o

2
I-
<
gE

O
cE
10-~
v
A588 GRADE 13
< 8 STEEL

~ 6

<
u. /
6•h 9 RANDOM
9~, RANDOM

O CONSTANT
9 CONSTANT

I KSI ~ / I N C H = 1 099 MN/m 3/2

1 INCH = 25.4 mm

10 -6 I I I I | I I I
10 20 30 41) 50 60 80 100
AKrms, k s i ~

FIG. 8-Crack growth rate as a function of the root-mean-square stress intensity factor for
A588-B steel.

the American Association of State Highway Officials, nor of the individual states
participating in the National Cooperative Highway Research Program.

References
[1 ] "Collapse of U. S. 35 Highway Bridge, Point Pleasant, West Virginia, December 15,
1967," Point Pleasant Bridge, National Transportation Safety Board, Report No.
NTSB-HAR-71-1, 1971.
[2] Barsom, J. M., Journal of Engineering for Industry, Transactions, ASME Series B,
Vol. 93, No. 4, Nov. 1971, p. 1190.
{3] Paris, P. C. and Erdogan, F., Journal of Basic Engineering, Transactions, ASME,
Series D, Vol. 85, No. 3, 1963, p. 528.
[4] Hardrath, H. F. and McEvily, A. T., "Engineering Aspects of Fatigue-Crack

I I I I/I 7

-,/

10-5

w
<

A514 GRADE E
STEEL
=o
o,

1 KSI ~ / I N C H = 1.099 M N l r n 3/7

C3 1 INCH = 25,4 m m
I-<
u-

9 RANDOM
RANDOM
O CONSTANT
9 CONSTANT

II I I I I I I I I I
10 20 30 40 50 60 80 100
A K r m s, ksi ~/inch

FIG. 9-Crack growth rate as a junction o f the root-mean-square stress intensity factor for
A514-E steel.

Propagation," Proceedings, Crack Propagation Symposium, Vol. 1, Cranfield,

England, Oct. 1961.
[5] Schijve, J., Jacobs, F. A., and Tromp, P. J., "Crack Propagation in Clad 2024-T3A1
Under Flight Simulation Loading. Effect of Truncating High Gust Loads," National
Aerospace Laboratory, NLR, The Netherlands, NLR TR-69050-U, June 1969.
[6] Hudson, C. M. and Hardrath, H. F., "Effects of Changing Stress Amplitude on the
Rate of Fatigue-Crack Propagation of Two Aluminum Alloys," NASA Technical
Note D-960, Sept. 1961.
[ 7] McMillan,J. C. and Pelloux, R. M. N. in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, p. 505.
[8] Wheeler,O. E., "Spectrum Loading and Crack Growth," General Dynamics Report
FZM-5602, 30 June 1970.
[9] "NCHRP PROJECT 12-12: Interim Report," U. S. Steel Research Laboratory
Report 76.019-001, 1 Oct. 1972.

I10] Barsom, J. M. in Progress in Flaw Growth and Fracture Toughness Testing, A S T M

STP 536, American Society for Testing and Materials, 1973, pp. 147-167.
[11] Blake, G. T., Klippstein, K. H., and Schilling, C. G., "NCHRP PROJECT 12-12:
Test Setup and Procedures for Beam and Plate-Specimen Tests," Research
Laboratory Report 76.019-001(2), 30 June 1972.
[12] Klippstein, K. H. and Schilling, C. G., in this symposium.
[13] Wilson, W. K., "Review of Analysis and Development of WOL Specimen,"
67-7D7-BTLPV-R1, Westinghouse Research Laboratory, 8 March 1967.

10-4 i I J/l Ii

"/
i

ca
2

Z
13

u~
10-5
o:
-c
8
I-

o= A514 GRADE F
6 STEEL
L~
<
4
uJ
1 KSI ,,/INCH = 1.099 M N / m 3/2

C- 1 INCH = 25.4 mm
<
u.

9 RANDOM
9 RANDOM
O CONSTANT
10-6
I8 9 CONSTANT

6
10 20 30 40 50 60 80 100
AKrm s, ksi ~/inch

FIG. l O-Crack growth rate as a function o f the root-mean-square stress intensity factor
for A514-F steel.

DISCUSSION
R. P. Wei1 {written discussion)-Dr. Barsom is to be complimented on his
comprehensive and careful experimental work on fatigue crack growth under
variable-amplitude loading in bridge steels. For the benefit of the unwary, I feel
that one needs to emphasize the fact that the loading sequence used by Dr.
Barsom is not truly random in the strictest sense. Although the selected load
levels obtained from a Rayleigh distribution had been randomized within a given
load block, this same load block was repeated ad infinitum in the actual
experiments. As such, the actual load sequence may be viewed as being periodic
in nature; the sequencing of loads within each period (block) being identical but
complex. A plausible interpretation to be applied to the load sequence used by
Dr. Barsom is that it represents, on average, the loading experienced by short
span bridges, with each block representing the traffic load on a day-to-day or
week-to-week basis.
A. J. McEvily 2 (written discussion)-The following expression based upon
crack-opening displacement considerations has been developed to account for
the contribution of fatigue and static modes to the amount of crack growth per
cycle, Aa/ AN 3
/2u/ = A (z~tK2_z~Lgth2)(l+ zxg ) (3)
AtaV O'ysE Kc - Kmax
(Since Kmax = (AK/1 - R), the effects of mean stress are incorporated in the
expression.) Equation 3 can be modified to incorporate the effects of variable
amplitude loading. Barsom [10] has reported on crack growth tests of an A514-B
steel (yield strength 128 ksi) subjected to random loading. He found that the
rate of crack growth could be expressed as
An -- t n
5jv-A (AKrms) (4)

where A' and n are material constants, and Z ~ r m s is the root-mean-square

stress-intensity-factor fluctuation. It is of interest to compare predictions based
upon Eq 3 with this result. Since Barsom observed that crack growth in this steel
was independent of mean stress for crack growth less than 10-4 in. per cycle,
the contribution due to static modes in Eq 3, that is, the zXK/(Kc-Kmax) term,
can be neglected. An average rate of crack growth under variable amplitude

I Department of Mechanical Engineering and Mechanics, Lehigh University, Bethlehem,

Pa. 18015.
2 Department of Metallurgy, School of Engineering, The University of Connecticut,
Storrs, Conn. 06268.
3 McEvily, A. J., "Phenomenological and Microstructural Aspects of Fatigue," presented
at the Third International Conference on the Strength of Metals and Alloys, 1973,
Cambridge, England; published by The Institute of Metals and The Iron and Steel Institutes,
Publication No. 36, Vol. 2, 1974, p. 204.
233
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 bybyASTM International
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
234 FATIGUE CRACK GROWTH

loading, ( A a ] / ~ a v g , can be taken to be equal to the average contribution from

each of the cycles over the increment of crack growth considered, that is

i+N*~r

(5)
i+N

Note that the quantity XzkK2/&N is the square of/~grms, so that Eq 5 can be
written

= (&Krm s - ~K~h) (6)

vg eys E
Fig. 11 shows a comparison between Eq 4 and Eq 6 and the data obtained by

io-s
II
.Iu

J=
u
l=
'" 4

0
1

~ 2

o
? IO 6
x
U 9 CONSTANT A M P L I T U D E
< 8 P o ASCENDING ORDER -
u o DESCENDING ORDER
,.', 6 ASCENDING-DESCENDING
ORDER
", R A N D O M - S E O U E N C E
1-
< 4 I kl~ ~ 9 1 . 0 9 9 8 M N / m 3/z
I inch 9 2 5 . 4 mm /
I 1 I 1 9 I , I i J
6 8 10 20 40 60 80 I00

AKrms, ks,

FIG. 1 1 - D e p e n d e n c e o f the rate o f crack growth for variable~mplitude loading on

Z~rmr: Eq 4 represented by the solid line, and Eq 6 represented by the dashed line (data
fro~"lJarsom [101).
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
DISCUSSION ON VARIOUS BRIDGE STEELS 235

Barsom [10]. (In evaluating Eq 6, A was taken to be 0.023 and Z~r(th to be 8

ksiV~.) Agreement of both equations is quite good, however, for other more
extreme types of loading the agreement may not be as good. For example, a
single overload can greatly retard the rate of crack growth rate at a lower
amplitude, and the preceeding equations would over-estimate the rate of crack
growth. Nevertheless, for a large number of random cycles, the amplitudes of
which do not fluctuate too widely, the agreement shown here indicates that
reasonable predictions can be made.
J. M. Barsom (author's closure)-The author thanks Dr. McEvily for his
comments concerning the applicability of the AKrms parameter to analyze
fatigue crack-growth rates under random loading and the fatigue crack growth
expression that he developed using crack-opening displacement considerations.
Similar expressions that are based on energy release rate or plastic zone size can
also be developed. All these expressions result in a relationship between da/dN
and AK2. However, expressions based on energy release rate, crack opening
displacement, and plastic zone size suggest that da/dN is not related to the yield
strength, Oys, for the material, to Oys- ~ and Oys-2, respectively. The available
fatigue crack-growth-rate data do not show any consistent dependence of growth
rate on oy s. Consequently, further work is necessary to establish the correct
fatigue crack-growth-rate model.

Equivalent Constant-Amplitude
Concept for Crack Growth Under
Spectrum Loading

REFERENCE: Elber, Wolf, "Equivalent Constant-Amplitude Concept for Crack

Growth Under Spectrum Loading," Fatigue Crack Growth Under Spectrum Loads,
ASTM STP 595, American Society for Testing and Materials, 1976, pp. 236-250.

ABSTRACT: A concept based on the crack-closure phenomenon has been developed

to replace random-load spectra with constant-amplitude loading in both analysis and
tests. The maximum load and the crack-opening load in the constant-amplitude
loading are chosen to be equal to those for the spectrum, so that both crack-growth
mode and the crack length at failure are equivalent to those under the random-load
spectra. The number of cycles of constant-amplitude loading is chosen so that the
amount of crack growth is equal to that due to a given sequence or block of the
random spectrum loading. The concept was tested experimentally after predicting the
equivalent number of constant-amplitude cycles for six different random-load
sequences. The agreement between predictions and test results was good.

KEY WORDS: crack propagation, fatigue (materials), analyzing, loads (forces)

To ensure the safety of aerospace structures, designers calculate the growth of

possible cracks for the expected service loading conditions. The calculated crack
growth is useful in determining inspection intervals, as well as in selecting
materials and in establishing nominal operating stresses.
Currently available models of the crack-growth process require repetitive
analyses, either cycle-by-cycle or block-by-block, of the service loading spectrum
from the initial to the critical crack length. Analyses o f the crack-growth
process, such as those o f Wheeler [1] 2 and Willenborg et al [2], are usually based
on plastic zone sizes which change with crack length. Even when combined with
fast calculation methods such as that by Brussat [3], these crack-growth
analyses are cumbersome.
The equivalent constant-amplitude concept developed here is based on the
crack-closure phenomenon [4] and on results of pilot tests that showed that the
crack-opening load remained essentially constant while cracks grew under
repeating random-load sequences containing several thousand load peaks. When
the crack-opening load is essentially constant and known, the equivalent

1 Research engineer, NASA Langley Research Center, Hampton, Va. 23665.

2 The italic numbers in brackets refer to the list of references appended to this paper.
236
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions author
ELBER ON AN EQUIVALENT CONSTANT-AMPLITUDE CONCEPT 237

constant-amplitude concept replaces the random loads with constant-amplitude

loads in crack-growth calculations.
The concept hinges on the determination of the crack-opening load. To obtain
a design method based on this concept will require empirical or analytical
methods of predicting the crack-opening load for a particular load spectrum.
Newman [5] has developed a numerical analysis to calculate the crack-opening
load; however, at the moment such an analysis is more complex than desired for
design use. Also, empirical rules for determining the crack-opening loads for an
arbitrary spectrum do not exist.
The search for an equivalent constant-amplitude test to replace the random-
load test is not new. Barsom [6] showed that for some random-load
distributions the rate of crack growth was generally equivalent to the rate of
crack growth under a constant-amplitude test with the same minimum load and
an amplitude representing the root-mean-square amplitude of the random test.
However, Barsom's approach did not attempt to obtain equivalent failure crack
lengths or equivalent crack-growth modes.
In this report an equation of crack-growth equivalence was developed, and the
validity of the concept was tested experimentally on six different random-load
sequences. In the tests the crack-opening loads were measured in crack-growth
tests run with both the random sequences and their predicted equivalent
constant-amplitude sequences. To check the predicted number of equivalent
constant-amplitude cycles, the number of random-loading sequences required to
cause failure was compared to the number of equivalent constant-amplitude
cycles required to cause failure.

Analysis
The Crack-Growth Law
The crack-growth law proposed by Elber [4], and experimentally verified
under constant amplitude and some two-level variable amplitude loadings, states
that the crack-growth rate is a power function of the effective stress intensity
range only, that is,

da = a (Ageff) n (1)
dN
where the effective stress intensity range,/2kgeff, is measured relative to the load
at which the crack fully opens.

The Equivalent Constant-Amplitude Concept

The equivalent constant-amplitude concept was developed to replace a
repeating random-load sequence containing several thousand load excursions by
a shorter constant-amplitude sequence. Random-load sequence, in this context,
represents a fixed number of load excursions whose distribution is known. The
constant-amplitude sequence which replaces this random-load sequence is

selected so that the total crack growth, the crack-growth mode, and the critical
crack length are equivalent for the two loading sequences. To achieve this, the
maximum gross section stress for the equivalent constant-amplitude sequence
was chosen to be the same as the largest maximum stress in the random
sequence. Therefore, the crack length at failure under the constant-amplitude
loading represents the shortest possible failure crack length under the random-
load sequence. Also, the minimum stress for the equivalent constant-amplitude
sequence was chosen so that the crack-opening stress for that sequence is the
same as the crack-opening stress in the random sequence. This produces
equivalent maximum effective stress intensity ranges and, hence, the plastic zone
envelopes are essentially equal for the two loading conditions. Also, this choice
simplifies the equation of equivalence developed later. Last, the number of
cycles in the equivalent constant-amplitude sequence was chosen so that the
crack growth caused by those cycles is the same as the crack growth caused by
the random-load sequence. This equivalent number of cycles, Neq , was
determined as follows.
If S i is the maximum and S'i is the minimum stress in the ith excursion of the
random sequence, Sop is the crack-opening stress, ~i is the effective minimum
stress, and a is the crack length. Then from Eq 1, the growth increment due to
the ith load excursion is

~ a = A(Si - ~i )~ ( ~ ' ~ ) n (2)

where

'Sop,S i ~ S o

(3)

"Si,"Si > Soy

If Smax is the highest maximum stress in the random sequence, then this
growth increment, ~a, can also be expressed as a fraction Xi of the growth caused
by one cycle of the equivalent constant-amplitude loading
~ia = Xi A(Sma x - Sop) n ( x / ~ ~)" (4)
Equating Eqs 2 and 4, yields the equation of equivalence

a ( s i _ ~i) n (N]~ ~))n : ~i a(Smax _ S o p ) . (%/r~ t~)n (5)

which, when solved for h i and then summed over all excursions in the random
load sequence, simplifies to

Neq =~]Xi =~-" (Si - ~i)" (6)

z-~'(Sma x - Sop) n

With the crack-opening ratio, ~, defined by

S Op - SB
a= (7)
Smax - S B
Soo was expressed in terms of the highest maximum stress Smax and the lowest
minimum stress S B as

Sop = S B + o~(Srnax - SB)

Then the final form of the equation of equivalence becomes
(s,.
xT' (8)
-

=z., 0 _ (Smax - n

Equation 8 can be evaluated from the distribution of loads in the random

sequence. The resulting relationship between Neq , a, and n is a unique
relationship for the particular spectrum. When the concept is applied, and a
particular value of Neq is obtained, that value depends on the crack-opening
stress and the materials' crack-growth exponent, n. Differences in configuration
and environment affect the crack-opening stress and, therefore, through ~, will
affect Neq.
The relationship among Area, ~, and n was obtained for the two spectrum
shapes in this test series. The necessary steps are explained in the next section.

Spectrum Analysis

A pseudorandom noise generator was used to produce a continuous analog

signal which is identically repeated after a given sequence length. Changes in the
shape of this spectrum were made using a variable nonlinear amplifier in the
output stage of the noise generator. Two spectra were selected. The main stress
parameters are defined in Fig. 1. To calculate the equivalent number of cycles
for these spectra from Eq 8, the distribution of stress excursions ( S i - "ffi) and
the corresponding relative maxima (S i - SB) was evaluated for the load
spectrum. For simplicity of analysis and data presentation, the spectrum load

SMAX SMAX

SOP
STRESS

SMIN

TIME TIME
a b
FIG. 1-Defi'nition of stress parameters." (a) spectrum load sequence, and (b) equivalent
constant amplitude.

excursions were sorted into a two-dimensional matrix of normalized stress

excursions (S i - Si)/(Smax - SB) and normalized relative maxima (SI
- S B ) / ( S m a x - SB). The spectrum range (Smax - SB) was subdivded into
approximately 20 intervals. The number of occurrences in each interval was then
tabulated in the matrix.
The data from Spectra I and II are tabulated in Figs. 2 and 3, respectively.
Using the values from these tables, Eq 8 was evaluated for each spectrum for a
range of c~, (0 < a < 1), and for a representative range of n, (2.5 < n < 4). The
resulting relationships between the equivalent number of cycles, Neq, and the
parameters n and a are plotted for Spectra I and II in Figs. 4 and 5, respectively.

Testing
S p e c i m e n s a n d Material

Sheet specimens (100-mm wide, 3.29-mm thick) with 2.5-mmqong central

notches were tested. The test section configuration is shown in Fig. 6. The
specimens were made of 7075-T6 aluminum alloy, having a nominal tensile
strength of 595 MPa, and a 0.2-percent offset yield strength of 540 MPa.

Testing E q u i p m e n t
Both spectrum load tests and constant-amplitude tests were conducted in a
100-KN servo-hydraulic testing machine. The mean cyclic frequency for the

O.00 - 0 . 0 5 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.O5 - 0 . 0 9 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 . 0 9 - O, 14 1 5 5 8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.14 - 0.18 5 8 5 0 2 2 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0,18 - 0.23 158 168 126 68 15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.25 - 0.27 150228242201 99 8 0 0 0 0 0 0 0 0 0 0 O 0 0 0 0 0
0.2"/ - 0,32 44105158197167 74 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.32 - 0.36 2 11 25 36 37 40 15 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.36 - 0.41 0 4 2 1 8 1 9 2 8 3 0 7 2 0 0 0 0 0 0 0 0 0 0 0 0 0
0.41 - 0,45 0 0 1 4 1 2 1 8 3 2 1 9 5 1 0 0 0 0 0 0 0 0 0 0 0 0
0.45 - 0.90 0 0 0 0 4 4 1 3 1 3 1 2 4 0 0 0 0 0 0 0 0 0 0 0 0
0.30 - 0.5~ 0 0 0 0 0 1 1 3 1 7 3 1 0 0 0 0 0 0 0 0 0 0
O.b~ - 0 . 5 9 0 0 0 0 0 0 2 0 4 4 4 1 0 0 0 0 0 0 0 0 0 0
0.59 -0.64 0 0 0 0 0 0 0 0 0 1 4 4 2 0 0 0 0 0 0 0 0 0
0.64 - 0.68 0 0 0 0 0 0 0 0 0 0 3 4 2 0 0 0 0 0 0 0 0 0
0.~,8 - 0 . 7 3 0 0 0 0 0 0 0 0 0 0 1 1 1 2 2 0 0 0 0 0 0 0
0.73 0.77 0 0 0 0 0 0 0 0 0 0 0 1 3 0 1 0 0 0 0 0 0 0
0.77 - 0.82 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0
0.82 - 0.86 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 1 0 0 0 0 0
0.86 - 0.91 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0
0.91 - 0.95 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.95- 1.00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

d G d d S d o d 6 d 5 d d o 5 d d o o o

FIG, 2-Data from Spectrum I.

0.00 - 0.05 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.05 - 0.11 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.11 - 0.16 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.16 - 0.21 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.21 - 0.26 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.26 - 0,52 0 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.32 - 0,37 3 0 4 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.37 - 0.42 2 6 1 1 6 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0,42 - 0.47 8 4 ? 6 3 9 2 8 8 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0.47 - 0.55 120 163 151 99 52 11 0 0 0 0 0 0 0 0 0 0 0 0 0
0.~3 - 0.58 92 147 182 186 115 69 10 0 0 0 0 0 0 0 0 C 0 0 0
0.56 - 0.63 24 69 98 131 177 119 47 3 0 0 0 0 0 0 0 0 0 0 0
0 . 6 5 - 0.68 4 14 41 52 60 85 67 19 4 0 0 0 0 0 0 0 0 0 0

0.68 - 0.74 0 3 6 8 15 25 27 22 9 I 0 0 0 0 0 0 0 0 0
0.74 - 0~.79 0 0 0 0 0 4 4 4 3 0 0 0 0 0 0 0 0 0 0
0.79 - 0.84 0 0 0 0 0 1 4 0 1 4 0 0 0 0 0 0 0 0 0
0.84 - 0.89 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0

0.89 - 0.95 0 0 0 0 0 1 0 0 0 2 2 0 0 0 0 0 0 0 0
0.95 - 1.00 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0

$i - Smln 9

q ~. ~ ~. ~ ~. ~ X ~ ~ ~ '~ "~ L ~- % co ~ c

, p i i i i i i i i ~ i i i i , i , p

o o o o o o o o g g g J d o g o g ~ g

FIG. 3-Data from Spectrum II.

I000
- (I) n-Z.5
(~) n-&O
| n-3.5

tOO (~ n-4.0
NUMBEOFR
EQUIVALENT
CYCLES,neg
10~" ! ~

1 , I, I I I I 1 [ I
0 .2 .4 .6 .8 1.0
CRACKOPENINGRATIO,a
FIG. 4-Equivalent cycles function for Spectrum L
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
242 FATIGUECRACKGROWTH

10001 |
(~ n.2.5
| n - 3.0
'| | n-3.5
looi_ | n-4.o
n=4.5
NUMBEROF
EQUIVALENT
CYCLES,neg
10-

1 i I i I i I i 1 i I
0 .2 .4 .6 o8 1.0
CRACK OPENING RATIO. a
FIG. 5-Equivalent cycles function for Spectrum II.

m ~ m ~ ~ - w ~

I
I
!
0-251
!

I DETAILA

I
L 100 -i
FIG. 6-Specimen test section configuration.

spectrum load tests was 5 Hz. The cyclic frequency for the constant-amplitude
load tests was 1 Hz. Load tracking accuracy at those frequencies was within 1
percent.

Crack-Closure Measurements
The crack closure and opening behavior of all specimens was continuously
measured with a crack-opening displacement (COD) gage [4]. The COD-gage
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
ELBER ON AN EQUIVALENT CONSTANT-AMPLITUDE CONCEPT 243

output and the testing-machine load-cell output were analyzed to determine the
stresses at which the cracks opened fully. The compliance method of Ref4 was
used for that analysis.

Spectrum Loading
Load Spectra I and II were applied as tensile loads to three specimens each at
selected values of minimum stress level SB, and spectrum range Smax - S B. The
matrix of test parameters is shown in Table 1.

TABLE 1 - Test matrix for spectrum load tests.

Specimen SBn, Smax - SB, Smax,

Number Spectrum MPa MPa MPa

1 I 20 180 200
2 I 10 90 100
3 I 50 100 150
4 II 20 180 200
5 II 10 90 100
6 II 50 100 150

Constant-Amplitude Loading
The equivalent constant-amplitude stresses were determined analytically from
the measured values of the stabilized average crack-opening stresses in each of
the six spectrum tests. Because of lack of data for 7075-T6, the crack-closure
behavior was taken from the published results for 2024-T3 [4], where

U = Smax - Sop = 0 . 5 + 0 A R f o r R > 0

Smax - S m i n
from which Stain, the minimum constant-amplitude stress, was determined as

Smin = l'25 {N/l'6 Smax Sop-O'79 Sm2ax-O'l Smax} (9)

The resulting test matrix for the constant-amplitude tests is tabulated in Table 2,
where Specimen 7 is the constant-amplitude specimen corresponding to
Specimen l, as indicated in the first column.

Results and Discussion

Spectrum Load Tests
The equivalent constant-amplitude concept was based on the assumption that
crack-opening stresses remain essentially constant during short (several thousand
load peaks) random-load sequences. Figure 7 shows the relationship between
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authoriz
244 FATIGUE CRACK GROWTH

TABLE 2 - Test matrix for constan t-amplitude tests.

Specimen Smax, Sop ,a

Number MPa MPa

7, 1 200 104 0.13

8, 2 100 53 0.18
9, 3 150 87 0.34
10, 4 200 102 0.07
11, 5 100 56 0.28
12, 6 150 89 0.37

a Measured values from spectrum tests.

00 -
/ - INITIAL CRACKLENGTH

1~-

CRACK 120- ~0 0 ^~ _/A _^/--~

OPENING
STRESS, 1- 0 0 -0 0 0
SOp. MNIm2 o o
80-
~---O SPECIMEN 6

40-

I I I I I I
0 2 4 6 8 I0 12
HALF-CRACKLENGTH,ram
FIG. 7-Crack-opening stress as a function o f crack length for Specimens 1 and 6.

crack length and crack-opening stress for two typical specimens from the test
series. Specimen 1 was tested under Spectrum I with a minimum stress o f
approximately zero. Specimen 6 was tested under Spectrum II with a minimum
stress of one third o f the maximum. For b o t h specimens, the crack-opening
stresses were above the stabilized average just after initiation, and then stabilized
to a constant value for the remainder o f the test. This initiation effect has also
been observed in surface crack growth in titanium alloy. The stabilized average
crack-opening stresses and the number of sequences to failure are given in Table
3.

TABLE 3 - Crack-opening stresses and number of sequences to failure

Ns for spectrum tests.

Specimen Sop,
Number MPa Ns

1 104 372
2 53 3400
3 87 930
4 102 26
5 56 640
6 89 68

Constant-Amplitude Tests

The stresses for the equivalent constant-amplitude tests were calculated from
the measured crack-opening stresses and Eq 9. The cyclic stress ratios and
crack-opening loads are tabulated in Table 2. The values of crack-opening ratio ct
are obtained from Eq 7. Values for the equivalent number of cycles, Neq, were
obtained from Figs. 4 and 5 for these values of a, and the material's
crack-growth exponent n = 2 9 7 [7]. These values are tabulated in Table 4.

TABLE 4 - Equivalent constant-amplitude parameters.

Desked
Specimen Smax, Sop,
Number MPa MPa R O~ Neq

1, 7 200 104 0.13 0.465 8.1

2, 8 100 53 0.18 0.48 7.8
3, 9 150 87 0.34 0.37 12.9
4, 10 200 102 0.07 0.45 126
5, 11 100 56 0.28 0.51 83
6, 12 150 89 0.37 0.39 191

Results from the constant-amplitude tests consist of the stabilized average

crack-opening stresses and the number of cycles to failure, Nca. The values are
given in Table 5. The measured crack-opening stresses were in good agreement
with the opening stresses in Table 4, based on the 2024-T3 aluminum data.
Failure crack lengths and failure modes of corresponding specimens were
compared. Failure crack lengths were generally equivalent, except for the
short-lived Specimens 10 and 4. The change of crack-growth mode from normal

TABLE 5 - Measured results from constant-amplitude tests.

Measured
Specimen Sop,
Number MPa Nca

7 105 3 240
8 56 32 200
9 84 13 200
10 104 2 800
11 53 41 300
12 86 14 300

to slant mode generally occurred at equal crack lengths. In all cases, the fracture
surfaces from the spectrum tests showed more discoloration due to corrosion or
fretting than the corresponding constant-amplitude specimens.

Comparison o f Results
The number of constant-amplitude cycles to failure, Nea, was divided by the
equivalent number of cycles, Neq, for each test and compared with the number
of spectrum sequences to failure, N s. The results are shown in Table 6.

TABLE 6 - Comparison of results.

Neq ' Nca Nca

Specimen Nca' cycles/ N s,
Numbers cycles sequence Neq sequences Neq x N s

1, 7 3 240 8.1 400 372 1.08

2, 8 32 200 7.8 4100 3400 1.21
3, 9 13 200 12.9 1020 930 1.10
4, 10 2 800 126 22 26 0.85
5, I1 41 300 83 500 640 0.78
6, 12 14 300 191 75 68 1.10

The ratio Nca/(Neq x Ns) ranges from 0 9 78 to 1 9 21, a range that is no

greater than scatter that might be expected in fatigue crack-growth data. The
results show no systematic differences between the test results and the predicted
number of equivalent cycles, and that the crack-closure-based crack-growth law
and the equation of crack-growth-equivalence gave valid predictions for these
tests.

Concluding Remarks
To simplify crack-growth calculations, a concept has been developed for
replacing relatively complex random-load tests and analyses with simpler
constant-amplitude tests and analyses. An equation of crack-growth-equivalence
resulting from derivations based on the crack-closure crack-growth law was
obtained and was used to determine a relationship between an equivalent
number of cycles of constant-amplitude loading (producing the same amount of
crack growth as a fixed sequence of the random-load spectrum) and the
distribution of the random loads, the exponent in the crack-growth law, and the
ratio of the crack-opening load to the maximum load. That relationship, which is
independent of crack length and stress level, is unique for a given spectrum.
The concept was tested experimentally for six different spectrum loadings and
the six corresponding equivalent constant-amplitude loadings. Good agreement
was obtained between the experimental results and the predictions.

References
[1 ] Wheeler,O. E., Journal of Basic Engineering, Vol. 94, March 1972, pp. 181-186.
[2] Willenborg, T., Engle, R. M., and Wood, H. A., "A Crack Growth Retardation
Model Using an Effective Stress Concept," AFFDL Tech. Memo 71-1-FBR, Air
Force Flight Dynamics Laboratory, Wright-Patterson AFB, Ohio, Jan. 1971.
[3] Brussat, T. R. in Fracture Toughness and Slow-Stable Cracking, ASTM STP 559,
American Society for Testing and Materials, 1974, pp. 298-311.
[4] Elber, Wolf in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[5] Newman,J. C., "A Finite-Element Analysis of Fatigue Crack Closure," presented at
the Eighth National Symposium on Fracture Mechanics, Providence, R. I., 26-28
Aug. 1974; also NASA TM X-72005.
[6] Barsom,J. M. in Progress in Flaw Growth and Fracture Toughness Testing, ASTM
STP 536, American Society for Testing and Materials, 1973, pp. 147-167.
[7] Hudson, C. M., and Newman, J. C., Jr., "Effect of Specimen Thickness on
Fatigue-Crack-Growth Behavior and Fracture Toughness of 7075-T6 and 7178-T6
Aluminum Alloys," NASA TN D-7173, National Aeronautics and Space Adminis-
tration, 1973.

DISCUSSION

C E. Richards I (written discussion)-I would like to ask where the compliance

measurements were taken and what differences, if any, may have been
encountered if compliance measurements are taken from different faces or
regions of test pieces.
In our experience, 2 we would expect, for example, compliance measurements
obtained from the sides of single edge notched specimens (that is, the two

I Materials Division, Central Electricity Research Labs., Leatherhead, Surrey, England.

2 Lindley, T. C. and Richards, C. E., 3rd International Conference on Fracture, Munich,
1973.

parallel surfaces intersecting the fatigue crack and perpendicular to the crack
front) to be different from those obtained from, say, the front face (that is, the
surface from which the starter notches are machined). I would appreciate any
information you have on this.
Wolf Elber (author's closure)-In our tests on center-cracked thin sheet
specimens, the compliance measurements are taken on the surface just ahead of
the crack tip. The nonlinearity of the compliance curve is, indeed, a function of
gage location. Our gage location maximizes the curvature just as the crack
becomes fully open.
J. Schijve 3 (written discussion)-The original papers of Dr. Elber should be
considered as a mile-stone in the development of the theory of fatigue crack
propagation. His crack-closure model has significantly contributed to the
understanding of variable-amplitude loading. The present paper is also promising
for predicting crack growth under spectrum loading. The measurements of the
crack opening stress during spectrum tests are of great interest. During the
meeting he indicated that spectra should be classified into either "short" spectra
or "long" spectra. It would be helpful if the meaning of these concepts could be
defined in some more detail. In this respect it would also be appreciated if the
number of cycles of the two spectra adopted in his tests could be indicated (that
is, the recurrence period). Were both spectra symmetric with respect to positive
and negative stress ranges?
The equivalent N approach apparently avoids a cycle-by-cycle calculation.
However, a prediction on the same basic assumptions on crack closure could also
be made by employing
=

Would the result have been essentially the same?

Wolf Elber (author's closureJ-The terms "short" spectra and "long" spectra
can only be given relative and approximate definitions. I consider a spectrum
"short," if the crack growth during one repeat interval is less than the plastic
zone size created by the highest load in the spectrum. In this case, the
fluctuations in crack-opening load are usually very small and the spectrum shows
very little sequence effect. By corollary, a "long" spectrum is associated with
crack growth larger than the maximum plastic zone size and can show significant
sequence effects.
The spectra in my tests contained about 5000 cycles. They were assymetrical.
The precise form can be derived from the tables of load maxima and associated
load ranges.
Your summation equation indeed is the prime condition for my derivation, so
the result would be identical.
R. L Stephens a (written discussion)-The previous paper by Barsom and your
3 Department of Aeronautical Engineering, Delft University of Technology, Delft,
Netherlands.
4 MaterialsEngineering Div., The University of Iowa, Iowa City, Iowa 52242.

own paper indicated that constant-amplitude fatigue crack growth behavior can
predict crack growth rules and life in some complex loading histories. This
implies sequence effects are not of primary concern in these spectra, Would the
author include his presented explanation of this situation in the discussion?
Wolf Elber (author's closure)-There may indeed be a large group of spectra
that encompass the majority of industrial applications, which could be
considered "short" spectra. Sequence effects for those spectra would then be at
worst secondary effects.
T. R. Brussat s (written diseussion)-Progress in understanding and predicting
crack growth under spectrum loading has been impeded by a general hesitation
to identify and separate the important and less important features of a loading
history and to appropriately simplify the history to a manageable form. This
paper has succeeded in demonstrating how three primary features of a random
spectrum loading history can be preserved in the equivalent constant-amplitude
loading history.
The author correctly limits his approach to "short" spectrum loading histories,
wherein the crack growth between any two consecutive repetitions of the
maximum spectrum load is a very small fraction of the overload-effected zone,
so that the opening stress, Sop, remains constant. It is well-known that service
loading histories for aircraft structure often include very high tensile or
compressive loads that occur only a few times during the crack growth life. The
temporary changes in Sop caused by these overloads (and underloads) introduce
major discontinuities in the slope of the crack length-time (a versus N) curve.
This differs from the smooth a-N curve that is characteristic of constant-ampli-
tude loading. Thus, it is clear that the author's approach does not apply to
"long" spectrum loading histories.
The following is suggested to extend the equivalent constant-amplitude
approach to long spectrum loading histories. A short spectrum loading history is
created by removing all infrequent overloads (and underloads) from the long
history. This short history is then replaced by the equivalent constant amplitude
loading history using the author's approach. Finally, the infrequent overloads
and underloads are reinserted into the loading sequence.
This simplified loading history is expected to display the correct failure mode
and critical crack size. In addition, Sop will vary in the appropriate manner, and
the a-N curve will have the appropriate shape. In general, the extent of
retardation due to the infrequent overloads will be somewhat overestimated, but
this source of error would be expected to be small for spectrum loading histories
of practical interest.
Using this extension of the author's approach it would be possible to estimate
fatigue crack growth for even the most complex "long" spectrum loading history
using retardation modeling that has up to now been applicable only to simple
sequences of constant-amplitude loading with intermittent overloads and
underloads.

s Stress Department, Lockheed-CaliforniaCompany,Burbank, Calif. 91520.

For aircraft structure, variations in service usage lead to significant differences

in the number of infrequent overloads that occur among various fleet members.
This, in turn, may lead to large deviations from the crack growth obtained from
one test or analysis conducted using a single arbitrarily selected overload
sequence. Multiple spectrum testing to cover this source of variation would
usually be unfeasible. It would now appear to be feasible, however, to conduct
multiple analyses. These analyses would operate upon the equivalent constant-
amplitude loading history with various numbers of overloads, inserted at various
times, covering the range of possible service histories. These analyses, supplemen-
tary to a single spectrum test result, could form a basis for improved assessment
of structural reliability.
Wolf Elber (author's closure)-I agree with Mr. Brussat that his proposed
modifications would simplify his particular problems. It was Mr. Brussat's
integrating calculation scheme which first stimulated my search for a simple
calculation scheme. This chain may develop a few more links.

Crack Growth in Ti-8AI-1Mo-IV

with Real-Time and Accelerated
Flight-by-Flight Loading

REFERENCE: Imig, L. A., "Crack Growth in Ti.8AI-IMo-IV with Real-Time and

Accelerated Flight-by-Flight Loading," Fatigue Crack Growth Under Spectrum
Loads, ASTM STP 595, American Society for Testing and Materials, 1976, pp.
251--264.

ABSTRACT: Crack growth in Ti-8AI-IMo-IV was measured and calculated for

real-time and accelerated simulations of supersonic airplane loading and heating.
Crack-growth rates calculated on the assumption that an entire flight could be
represented by a single cycle predicted the experimental rates poorly. Calculated
crack-growth rates were slower than the experimental rates for all tests with
flight-by-flight loading. For room-temperature accelerated tests, the calculated rates
agreed well with the experimental rates; but the calculations became progressively
less accurate for progressively more complex test conditions (tests that included
elevated temperature). Calculations of crack growth using the crack-closure concept
can probably be improved through study of crack-opening stresses using finite-ele-
ment models that account for variable-amplitude loading, residual stresses, and
temperature effects. The calculations of crack growth could also be improved
through detailed studies of material properties and interactions among stress,
temperature, and time as appropriate for the real-time operating conditions of a
supersonic transport airplane.

KEY WORDS: fatigue (materials), stress cycle, high temperature tests, supersonic
transports, titanium alloys, flight simulation, axial stress, accelerated tests, crack
initiation, crack propagation

Practical aspects of fatigue-crack growth are a concern shared by both

researchers and designers. Evidences of that concern are the many conferences
held by the American Society for Testing and Materials (ASTM) to discuss and
interchange ideas about crack growth. Crack growth related to airplane materials
and structures has been studied predominantly for subsonic airplane flight
conditions [1,212 for recent examples; but crack growth for materials and
conditions representing supersonic flight also requires study to stay abreast of
advancing flight capabilities. (Such advancement has led to the Anglo-French
Concorde and the Russian Tu-144 supersonic airplanes.) The objectives of the
present paper were to determine the effects on crack growth of the real-time and
temperature environment of simulated Mach 3 flight, and to assess the

1 Research engineer, NASA Langley Research Center, Hampton, Va. 23665.

2 The italic numbers in brackets refer to the list of references appended to this paper.

251
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authoriz
252 F A T I G U E C R A C K GROWTH

applicability of the crack-closure and linear summation methods to crack-growth

calculations for these flight-simulation tests.

Experimental Considerations
Materials and S p e c i m e n s

The crack-growth tests for this investigation were conducted on duplex-

annealed Ti.8AI-IMo-IV, titanium alloy sheet which was 1.27-mm thick. The
tensile properties of the sheet agreed closely with those from Ref 3 for this
alloy.
The fatigue specimens made from the sheet titanium alloy are shown in Fig. 1.
The central notch produced a stress-concentration factor of 4.1. Parallel

~--- ll.7 ----1

--] 1--572
o
Enlarged view
of notch

N~h

- 25(
o 1
a b
FIG. 1-Configurations of fatigue specimens. Dimensions are in millimeters. Sheet
thickness is 1.27 ram." (a) accelerated tests and 0o) real-time tests.

reference lines spaced at 1.27 mm were photographically placed on accelerated

test specimens for use in recording crack lengths.

Stress Sequences
The present fatigue tests used programed variable-amplitude stress sequences
(flights) representing the stresses expected in the lower wing skin of a
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
IMIG ON CRACK GROWTH IN TI-8AL-1MO-1V 253

commerical supersonic transport during its operation. The flights were derived
[4] considering atmospheric turbulence, flight maneuvers, and landing as sources
of wing load. Reference 4 describes the formulation of variable-amplitude flights
whose largest amplitude was calculated to occur, on the average, once per flight,
once in five flights, once in a hundred flights, and so forth. The stresses within
the flights were determined separately for the climb, cruise, and descent
segments of flight as shown in Fig. 2 for the flight whose largest amplitude
occurred once per flight. The design mean stress referred to in Fig. 2 represents
the stress in a lower wing skin during level unaccelerated flight at maximum
airplane mass.
The flights developed in Ref 4 are called Type A flights. They were simplified
to produce an additional sequence of variable-amplitude flights called Type G
flights, see Fig. 2b. Type G flights produced variable-amplitude tests. The flights
consisted of a single cycle having the same maximum and minimum stresses as
each of the flights from which they were taken.

Fatigue Tests
Two kinds of fatigue tests, called "real-time" and "accelerated," were
conducted to compare crack growth for the realistic time and temperature
environment of a supersonic airplane relative to the crack growth from fatigue
tests which neglect time and temperature effects. Both the real-time and
accelerated tests employed identical sequences of stress amplitudes within the
flights. In accelerated Type A tests, both the stress sequence within each flight
and the order in which the flights were applied were scheduled automatically by
a computer; the flights of Fig. 2a were applied at a rate of about 2 s each, and
the test was conducted either at room temperature or constant elevated tempera-
ture. In real-time tests, the flights were inscribed on charts for a chart-following
device; each flight, Fig. 2c, lasted about 96 rain and included a 90-min-long
elevated temperature cycle.
All fatigue tests employed hydraulically-actuated, closed-loop, servocontrolled
fatigue testing machines [5] which operated 24 h a day. The specimens were
loaded axially. Stresses for accelerated tests were based on the initial net area at
the test section of each single notch specimen; for real-time tests, stresses were
based on the average net area of each six-notch specimen. In a given six-notch
specimen, the individual net areas of the notches were within 1 percent of the
average area.
Crack-Growth Recording-Specimens for real-time tests were visually in-
spected approximately monthly for crack initiation. After cracks were discov-
ered, their growth was logged weekly by an optical inspection with a lO-power
microscope. In accelerated tests, crack growth was recorded by a 70-ram camera
which simultaneously photographed the specimen and a flight counter. A special
command included in the sequence of flight loads activated the camera and
advanced the film by one frame. The image on the film was approximately full
size and crack lengths were determined from the spacing of the reference lines
on the specimen.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
2.5
tj R
20 i j~, -Climb
"13
1.5 - Cruise
Stress 1.0 ~~~/~-o.c.o,
Design mean slress ~.,.,
[,.,.,
f'''r"
IT'''
"['''''I['''
'~ c
m
.5 c)
"n
~>
0 c~
Time l C)
-.5
0
a --t
b "1"

2.5

2.0 / Ctimb Descent----~

1.5 / /--Temper o1 u re ~176176

Stress ,.o ',',~I
'l,',I',',
! ',
',','#~ ..---? ,;~,
Design meor, slress " v~liJ~ J 5 0 0
,5 ,I~,,rq/ Temperot ure,
I! K
I
0 I 't1400
l

-,5 ~ L ~ 300
90 rain
,- 96 rain
C
FIG. 2-Flights for accelerated and real-time tests: (a) a Type A flight for accelerated tests, (b) a Type G flight for accelerated tests, and
(cJ a Type A flight with cyclic temperature for real-time tests.

Test Conditions-The Type A flights were used in tests with design mean
stresses of 172,207, and 241 MPa, as shown in Table 1. One accelerated test was

TABLE 1 - Test program and results.

Duplex-i~nnealed Ti-8Al-lMo-1 V, 1.27 mm thick.

Crack Initiationa Crack Growthb

Design
Mean Fatigue Percentage Percentage
Stress Stress, Life, Period, of Total Period, of Total
Sequence MPa flights flights Life, flights flights Life, flights

Accelerated Tests at Room Temperaturec

172 137 158 127 000 93 10 160 7

A 207 18 540 13 300 72 5 240 28
241 7 472 5 300 71 2 170 29

172 105 988 78 000 74 27 990 26

G 207 57 290 40 300 70 16 990 30
241 22 290 12 000 54 10 290 46

Accel~a~d Te~s at 560K c

172 36 243 25 500 70 10 740 30

A 207 12 498
241 4 580 2 850 62 1 730 38

ReabTime Te~sd

172 19 014 12 000 63 7 010 37

A 207 10 420 8 100 78 2 320 22
241 5 093

a For cracks extending 1 mm from the notch.

b For cracks from 1 mm long until failure.
c One test at each design stress.
d Median value from test.

conducted at room temperature (300 K) and one at 560 K for each design stress.
In real-time tests, the temperature was cycled from 300 to 560 K in each flight
(see Fig. 2c).
Type G flights were used on one set of room-temperature accelerated tests (see
Table 1). Tests with Type G stress sequences were conducted as a potential aid
in analyzing the crack growth in tests with Type A stress sequences.

Crack-Growth Analysis

Crack growth from both constant-amplitude tests [6] and the present
variable-amplitude tests was analyzed in terms of growth rate and stress
intensity. The analyses are discussed in the two following sections.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authori
256 FATIGUECRACK GROWTH
Constant-Amplitude Tests
The crack-growth data for constant-amplitude tests of Ti-8AI-IMo-IV [6]
were analyzed using the crack-closure concept of Ref 7 to define an effective
stress-intensity factor. According to that concept, the crack is closed during part
of each load cycle and therefore cannot grow during that time; thus, the applied
stress range AS is reduced to an effective stress range of UAxS.The factor Uwas
determined [ 7] for crack growth at positive R values in an aluminum alloy. For
expediency, the same factor U was used in this paper to analyze the
constant-amplitude data from Ref 6. Effective ranges of stress intensity were
calculated from Z2~Keff = UAxSx/~aa, where the values of a are average crack
lengths for the increments of crack growth [6], and U = 0.5 + 0.4R (R > 0)
[7]. A least-squares technique using the relation Zka/~dV-=A [Z~geff ] n produced
a good correlation between crack-growth rate and 2tKeff as shown in Fig. 3 for
tests at room temperature and 560 K, respectively. The best-fit constants
determined for the two test temperatures are given in the following tabulation.
Test Temperature A n
Room Temperature 5.5 x 10-8 3.15
560 K 1.1 x 10.7 2.72

These values were used to make the following calculations of crack growth for
the present variable-amplitude tests.

Variable-Amplitude Tests
Crack-growth rates for the present variable-amplitude tests were calculated in
three ways:
1. by assuming one cycle per flight, and using effective stress intensity,
2. by considering only the stresses exceeding the crack-opening level and using
effective stress intensity (the "crack-closure method"), and
3. by considering all stresses in the flight, and using the conventional stress
intensity, zXK (the "linear-summation method").
Factors to account for the narrow specimen width and the eccentricity of the
individual cracks at the two ends of the notch (see Fig. 4) were used to calculate
the stress intensities. The width of the specimen was accounted for by the
factorx/sec (Tra/gO [8], and eccentricity was accounted for by factors L
interpolated from the tabular values in Ref 9. Thus, the expression for stress
intensity was U~SL x/~ra sec (ira/W) , where a is the average crack length for the
current growth increment, L is the factor for either the long or the short crack,
as appropriate, and U = 1 to calculate conventional stress intensity. The stress
concentration due to the notch in the specimen was neglected because it
influenced the stress intensity by less than about 3 percent for the present
calculations [10].
For all three methods of calculating crack growth, the stresses selected were
from flights of the severity shown in Fig. 2, because they represented about 80

,~ 10-3
E
E e=

z
o = 5,5Ox10 e(AKeff)s 15 1.09 x IO'r(AKeff) 2"r2
<1
~ IO -4
o
O
f f AKef f = U ASv'~-d z
r
U=OS+04R (ref 7)
c)
7~
-~ IO-S 6~
. ? ' R=O | 50
o O
,3 I- 25
,~" .43 -
67
~ / 85 9
~o
10-6
I0 0 101 I02 10 3 IOn ,r"
I0 ~ I0 j 102
b
AKeff, M Po,/'~ AKeff, M Po/"m- ,<

F I G . 3-Correlation between crack-growth rate and Z ~ e f f for constant-amplitude fatigue tests o f duplex-annealed t,a
Ti-8Al-lMo-1 I1, 1.27 m m thick. Data f r o m R e f 6. (a) tests at rob~ha temperature and (b) tests at 560 K. oa

FIG. 4-Crack and geometric nomenclatures for calculating crack eccentricities.

percent of the flights applied. All three calculations also assumed that cracks
would grow at the rates determined in constant-amplitude tests for the same
values of stress intensity.
One Cycle per Flight-This calculation assumed that a single cycle between the
minimum and maximum stresses in the flight (the ground-air-ground cycle)
produced all the crack growth. Both Type A and Type G tests were analyzed in
this way using effective stress intensity. The R value for the ground-air-ground
cycle is --0.37. The effective stress range for constant-amplitude tests, with
negative R values, was determined by elastoplastic, finite-element analysis of a
cracked sheet in Ref 11. Those calculations produce U = 0.54 for the present
tests (R = -0.37). For this first method, crack-growth rates (growth increments
per flight) were calculated from z3a = A (ZkKeff) n using A and n from the
previous section, and using the average crack length from each increment of the
experimental crack growth, in 2tKeff. Experimental rates were assumed to be
constant during each increment of crack growth.
Crack Closure-For this method, effective stress intensities were calculated for
all stress cycles fully exceeding the crack-opening stress. Only the stress cycles in
climb fully exceed the crack-opening stress when U = 0.54. The cycles in cruise
are only partially above (and were neglected), and the cycles for descent are
below the crack-opening stress. Therefore, the effective stress range selected was
from the maximum stress of each cycle to, arbitrarily, the minimum stress for
the smallest amplitude of the climb segment. As Fig. 2a shows, the climb
segment of Type A flights had 30 cycles of small amplitude, 6 cycles of an
intermediate amplitude, and 1 cycle of large amplitude. Thus, for this method,
crack growth per flight was calculated from
Atl = A [30(,dkgeff)l n + 6(z~kgeff)2 n -4-(z~kgeff)3 n]

where the subscripts correspond to the three ranges of the effective stress, and
the other nomenclature is the same as previously described.
The values of A and n used in the calculations depended on the kind of test.
For accelerated tests, the calculations used the values of A and n corresponding
to the constant test temperature. For real-time tests, each flight included a
temperature cycle, but the temperature for the climb segment of each flight was
near room temperature; therefore, values of A and n for room temperature were
used.
Linear Summation-This calculation summed the crack growth for the
complete range of each cycle in the flight, including the ground-air-ground cycle.
Crack growth was calculated using ZkK. To calculate crack growth using ZkK, a
different value of.4 was required for each stress range in the flight. The A's were
obtained by factoring the term U from the expression for Z~eff, and combining
U with the value of A determined earlier in this paper for constant-amplitude
data. Thus, 2xa/AxN= A U n (zXK)n = A(zkK)n . The values of.4 obtained using U
= 0.5 + 0.4R [7] and the constant-amplitude data from Ref 6 are shown in Fig.
5. A value of.4 was determined for each stress range in the flight. The previously

8x10"8
= AU n
6
J

A
4

"~176 .... 4

-.4 -.2 0 .2 .4 .6 .8 1.0

R
FIG. 5-Relations between ~ and R for constant-amplitude fatigue tests of duplex-
annealed Ti-SAl.lMo-1 V, 1.27 mm thick. Data from Ref 6.

determined values of the exponent n were retained. The crack-growth rates per
flight were calculated from
= 30~, (zu<,)" + 6~T~ (a/<~)" + . . . .

where the subscripts on .A and ZkK correspond to the various stress amplitudes,
the number of terms in the equation corresponds to the number of different
amplitudes, and the coefficient of each term is the number of cycles for each
amplitude.

Results and Discussion

The test results, presented in Table 1, are discussed in terms of both crack
initiation and crack growth.

Crack-Initiation Periods
The smallest cracks detectable from the film records of room-temperature
accelerated tests, or from observations of the real-time tests, were usually less
than I mm long; but the cracks were about 2 mm long before becoming
detectable on film records of accelerated tests at 560 K.
In most tests, cracks at the two ends of the notch initiated at different times.
Therefore, in general, an eccentricity factor was required to calculate stress
intensity for these cracks, as described earlier. The values given in Table 1 for
accelerated tests are for the first crack to initiate. For real-time tests, the values
are for the first crack at the notch having the median life.
To establish a consistent base of comparison, the number of flights to produce
a crack 1 mm long (see Table 1) was determined from plotted curves of crack
length versus number of flights for each test. For hot accelerated tests, the
curves were extrapolated. Table 1 indicates that most of the cracks reached a
length of 1 mm (initiated) at between 60 and 80 percent of the total life. In all
of the accelerated tests, the crack initiation periods were consistently larger
fractions of total lives for lower design mean stresses. The data for real-time tests
show the opposite trend, but its significance cannot be adequately determined
because data are available for only two real-time tests.

Crack-Growth Periods
The periods of crack growth in Table 1 are differences between the fatigue
lives and the initiation periods. The crack-growth periods for Type A accelerated
tests at the two temperatures were about the same, consistent with the
crack-growth periods for constant-amplitude tests of this alloy [6] for the same
temperatures. For a given design stress, the growth periods for real-time tests
were somewhat shorter than for the accelerated tests.
The growth periods for Type A tests were much shorter (growth rates were
faster) than for Type G tests. The faster rates for Type A tests indicate that the
small stress cycles contributed significantly to the crack growth.

Crack-Growth Rates
The calculated crack-growth rates are plotted against the experimental rates
from the persent tests in Figs. 6 and 7. The points in each figure represent the
rates for both the long and short cracks and for all design stresses. The solid line
in each figure is a least-squares linear fit of all the points. The dashed lines
indicate where the points would lie if the calculated rates equaled the
experimental rates.
Type G Tests-Figure 6 shows the crack-growth rates for Type G tests where
both the tests and calculations simulated the flights by a single cycle. As shown
by the slope of the solid line, the calculated rates were faster than the
experimental rates by an average of about 70 percent. That result should be
expected qualitatively, because the higher stresses in fifth flights, 100th flights,

.00~

iu
~- /Lt68 "
../3
;.7: -/1 / 1
..
u E ~ ~Cruck-opening

~ ,, // .
u
d
I I I ,0~5

Experimental crack-growth rale,

mm/fllghl

FIG. 6-Experimental and calculated crack-growth rates for Type G fatigue tests of
duplex-annealed Ti-8AI-IMo-1 V, 1.27 mm thick.

and so forth, probably retarded the crack growth, and these higher stresses were
not accounted for in the calculations.
Type A T e s t s - T h e experimental crack-growth rates for Type A tests and the
rates calculated in the three ways discussed earlier are shown in Fig. 7. Parts a, b,
and c of Fig. 7 contain the rates for accelerated tests at room temperature,
accelerated tests at 560 K, and real-time tests, respectively. The top parts of
Figs. 7a, 7b, and 7c compare the experimental rates to those calculated,
assuming the flight could be represented analytically as a single cycle; the center
parts compare the experimental rates to those calculated, accounting for only
the cycles exceeding the assumed crack-opening level (all the cycles in the climb
segment of each flight); the bottom parts compare the experimental rates to
those calculated, using linear summation of the crack-growth contributions of all
cycles in the flight.
Figure 7 shows better agreement between calculated and experimental rates
for room-temperature accelerated tests (Fig. 7a) than for elevated temperature
or real-time tests (Figs. 7b and c). It shows that the flights were poorly
represented in the calculations by a single cycle (top row of figures). It shows
that the rates calculated by the crack-closure and the linear summation methods
were about equal for each test condition (center and bottom figures).
The present calculations using linear summation and other calculations of
linear fatigue damage both predicted longer fatigue lives (slower damage
accumulation) than the experiments produced [5]. In contrast, Refs 1 and 2
reported that linear summation calculations for crack growth indicate higher
crack-growth rates than their experiments on 2024-T3 and 7075-T6 aluminum
alloys. These different observations probably result from the interaction effects
of the spectrums on the materials, the relative severities of the spectrums, or the
combined effects. As shown in Fig. 8, the spectrum from Ref 1 contains

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
,0~ ,/ O)
i/ ///~ /'-- Crack-apllninll
j/ I /. ," / I 9 / ""*
// / I I r'-l--t~
J //" P I I J---II-#~'F-'~
// .--t
j' . . .
/s// / -- 1 -7
B
/ i ! c
,,," . 2_~....-., ~ pt _1 I- m
i///// . . ~P -. I
llllkl

~
e G')
9 .02[ +,r+ f*" -11
/1/ [ // ~ I r--- Crack........ 0
///. /// levul .-I
m_~ 8 ii II
-'r
"~ '/ , ~I o

,J c,~ ........... ,

.02 9 .
Is
/ i,/*
i I
i/p I ill s
iI
i I
i I

;,'.J. , 9 ,,f 9 9
. ./ Linear-summation concept
i 1 | .I)2 ' " .Z'ju,I

a C
Experimental craek*growthrate,
mm/|llllkt
FIG. 7-Experimental and calculated crack-growth rates for Type A fatigue tests o f duplex-annealed Ti-8Al-lMo.1 V,
1.27 mm thick." (a) accelerated tests at room temperature, (b] accelerated tests at 560 K, and (c] real-time tests.

.4
I~f. 1,Aluminumalloy
1 ~Ref. S, Titaniumalloy
,~ '~ .2

10 -4 10 -2 10 0 10 2

Cumulative frequency per flight

FIG. 8-Gust spectrums.

significantly higher stresses than the present spectrum, relative to the yield
strengths of the two materials. Conceivably, the higher relative stresses in the
infrequent flights of Ref 1 produced larger delays in crack growth than in the
present tests.
In contrast to the good agreement between calculated and experimental rates
for the room-temperature tests, the calculated crack-growth rates for accelerated
tests at 560 K and for real-time tests were much slower than the experimental
rates. This discrepancy is difficult to explain for the hot accelerated tests
because the calculation method was identical to that for room-temperature tests,
and the supporting constant-amplitude data for both temperatures seemed
equally well correlated. Possibly, the high loads in the infrequent flights
produced less retardation at 560 K than at room temperature. For the real-time
tests, Fig. 7c, insufficient information is available about local high-stress creep,
stress-strain relations, crack-opening stresses, and their interactions during cycles
of temperature, to properly predict crack growth. A first attempt to calculate
local stresses at the notch for real-time flights [12] indicated only slight
differences between the stresses for real-time flights and those for room-tempera-
ture flights. Thus, much basic study will probably be required before crack
growth can be calculated accurately for this type of loading.
Overall, the present calculations indicate that the linear summation and
crack-closure concepts produced the best calculations of crack growth. The
calculations by both methods could probably be improved by considering a
crack-growth relation which allows the rate to accelerate at high values of stress
intensity [13], and by considering the plasticity and stress distribution effects
induced by the notch [14]. In addition, calculations with the crack-closure
concept could probably be improved through further study of the crack-opening
stress levels for variable temperature and loading.

Concluding Remarks
Crack growth in Ti-8AI-IMo-IV was measured and calculated for real-time and
accelerated simulations of supersonic airplane loading and heating. Crack-growth
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
264 FATIGUE CRACK GROWTH

rates, calculated on the assumption that an entire flight could be represented by

a single cycle, predicted the experimental rates poorly. Calculated crack-growth
rates were slower than the experimental rates for all tests with flight-by-flight
loading. For room-temperature accelerated tests, the calculated rates agreed well
with the experimental rates; but the calculations became progressively less
accurate for progressively more complex test conditions (test that included
elevated temperature). Calculations of crack growth using the crack-closure
concept can probably be improved through study o f crack-opening stresses using
finite-element models that account for variable-amplitude loading, residual
stresses, and temperature effects. The calculations o f crack growth could also be
improved through detailed studies o f material properties and interactions among
stress, temperature, and time as appropriate for the real-time operating
conditions of a supersonic transport airplane.

References
[1] Schijve, J., Jacobs, F. A., and Tromp, P. J., "Fatigue Crack Growth in Aluminum
Alloy Sheet Material Under Flight Simulation Loading. Effects of Design Stress
Level and Loading Frequency," NLR-TR-72018U, National Aerospace Laboratory,
Amsterdam, Feb. 1972.
[2] Brussat, T. R. in Damage Tolerance in Aircraft Structures, ASTM STP 486,
American Society for Testing and Materials, 1971, pp. 122-143.
[3] "Metallic Materials and Elements for Aerospace Vehicle Structures," Department of
Defense, MIL-HDBK-5A, 1966.
[4] Imig, L. A. and Illg, Walter, "Fatigue of Notched Ti-8AI-IMo-IV Titanium Alloy at
Room Temperature and 550 F (560 K) With Flight-by-FlightLoading Representa-
tive of a Supersonic Transport," NASA TN D-5294, National Aeronautics and
Space Administration, 1969.
[5] lmig, L. A. and Garrett, L. E., "'Fatigue-Test Acceleration With Flight-by-Flight
Loading and Heating to Simulate Supersonic-Transport Operation," NASA TN
D-7380, National Aeronautics and Space Administration, 1973.
[6] Hudson, C. M., "Studies of Fatigue Crack Growth in Alloys Suitable for Elevated
Temperature Applications," NASA TN D-2743, National Aeronautics and Space
Administration, 1965.
[ 7] Elber, Wolf in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[8] Brown, W. F. and Srawley, J. E., Plane Strain Crack Toughness Testing of High
Strength Metallic Materials, ASTM STP 410, American Society for Testing and
Materials, 1966, p. 78.
[9] Isida, M., "Crack Tip Stress Intensity Factors for the Tension of an Eccentrically
Cracked Strip," Dept. of Mechanics, Lehigh University, Bethlehem, Pa., 1965.
[10] Newman, J. C., Jr., "An Improved Method of Collocation for the Stress Analysis of
Cracked Plates With Various Shaped Boundaries," NASA TN D-6376, National
Aeronautics and Space Administration, 1971.
[11] Newman, J. C., Jr., "A Finite-Element Analysis of Fatigue Crack Closure,"
presented at the Eighth ASTM National Symposium on Fracture Mechanics,
Providence, R. I., Aug. 1974.
[12] "Cyclic-Stress Analysis of Notches for Supersonic Transport Conditions," Boeing
Aerospace Co., NASA CR-132387, National Aeronautics and Space Administration,
1974.
[13] Forman, R. G., Kearney, V. E., and Engle, R. M. in Transactions, American Society
of Mechanical Engineers, Series D, Journal of Basic Engineering, Vol. 89, No. 3,
Sept. 1967, pp. 459--464.
[14] Crews, J. H., Jr., and White, N. H., "Fatigue Crack Growth From a Circular Hole
With and Without High Prior Loading," NASA TN D-6899, National Aeronautics
and Space Administration, 1972.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
Life Prediction and Applications

Prediction of Fatigue Crack Growth

Under Irregular Loading

REFERENCE: Nelson, D. V. and Fuchs, H. O., "Prediction of Fatigue Crack

Growth Under Irregular Loading," Fatigue Crack Growth Under Spectrum Loads,
ASTMSTP 595, American Society for Testing and Materials, 1976, pp. 267-291.
ABSTRACT: Fatigue crack growth under irregular loading is calculated from
constant-amplitude test data. The results of such calculations are compared with
measurements of crack growth in modified compact tension specimens made of two
commonly used structural steels and subjected to three different irregular load
histories, each applied at several load levels.
Two crack growth rate relations are utilized: I, the Forman relation, and I1, a
relation based on an effective stress intensity range concept. The Forman relation is
used with two different methods of "counting" load ranges, that is, crack growth is
calculated: (a) for all rising tensile load ranges as they occur in the load history and
(b) for the load history condensed by a method which forms overall ranges and
includes only the top decile of all rising tensile ranges. Crack growth rate Relation II
is used only with counting Method a. The three resulting prediction methods are
designated Ia, Ib, and II. Methods Ia and Ib disregard sequence effects, while Method
II includes them.
Predictions of crack growth using Methods Ia and lb agree with average test lives to
within a factor of two in nearly all cases, with the exception of those instances where
large compressive loads cause gross yielding and accelerated growth. Predictions based
on Method II are superior to those for Ia and lb when such gross yielding occurs and
are comparably good for the other cases investigated here.

KEY WORDS: crack propagation, predictions, loads (forces), stresses, crack initia-
tion, stress ratio, retarding, residual stress, fatigue (materials), fracture properties,
crack closure, compression

Data obtained from constant-amplitude tests permit the calculation of crack

growth under irregular loading. Virtually all predictions of fatigue crack
propagation to date have been concerned with load histories which can be
characterized by cycles. For simulated aircraft loadings, crack growth has been
calculated for varying layers of cycles, with attempts to account for crack
growth retardation due to periodic tensile overloads [1-4]. 3 For random
loadings, particularly those with load spectra describable b y the Rayleigh
distribution function, crack growth has been characterized in terms of equivalent

i Research assistant, Mechanical Engineering Dept., Stanford University, Stanford, Calif.

94305 ; now with General Electric Co., Sunnyvale, Calif. 94086.
2 Professor, Mechanical Engineering Dept., Stanford University, Stanford, Calif. 94305.
3 The italic numbers in brackets refer to the list of references appended to this paper.

267
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authori
268 FATIGUE CRACK GROWTH

root-mean-square cycles [5-7]. However, little consideration has previously

been given to crack propagation for those irregular loadings where the definition
of a cycle is not straightforward. A similar problem exists in the prediction of
crack initiation, where various "cycle counting" methods [8-12] have been
developed for selecting damaging "events" (for example, large overall ranges
interrupted by smaller ranges) in irregular load histories. Constant-amplitude
fatigue data are then used to predict the crack initiation damage of such events.
A similar type of approach will be applied here to the prediction of crack
propagation for the same histories for which initiation was calculated in Ref 12.
Previous crack growth predictions have also neglected the possible effects of
compressive loadings. A number of experimental investigations [13-15] have
shown that compressive overloads can increase crack growth rate, while other
studies [14-17] have found that if a tensile overload is followed immediately by
a compressive overload, the effects of crack growth retardation are greatly
diminished. Furthermore, Stephens [15] has recently found that retardation can
be reduced by application of cycles with compressive loading following single
tensile overloads. A new prediction method with the ability to account for
important load sequence effects due to both tensile and compressive overloads
will be presented. The effect of compressive loadings which cause gross yielding
will be given special consideration.
All predictions are made using a linear elastic fracture mechanics approach.
The following items are needed for this purpose: (a) a forecast of loading,
including not only the magnitude but sequence of loads as well; (b) a relation
between stress intensity, K, and loading for a given component geometry and
anticipated type and size of crack, which will be referred to as a "K-calibration"
for the sake of brevity; (c) constant-amplitude crack growth rate data (Aa/AN
versus AK) for a given metal, including knowledge of fracture toughness, Kc,
threshold stress intensity, AKth, and stress ratio, R, effects on growth; and (d) a
way of using these items. Item d is the main concern of this paper.

Input Data for Predictions

Load Histories
Three distinctly different irregular load histories were used in this study and
are depicted in Fig. 1. These histories have recently been used in an extensive
crack initiation and propagation test program of the SAE Fatigue Design and
Evaluation Committee [18]. Each history was applied repeatedly in testing, so
that test data were reported in terms of blocks (that is, repetitions of the
history). Due to their contrasting nature, these load blocks are especially useful
for evaluation of prediction methods. The suspension history, with 2506
reversals, has a strong compressive bias. The transmission (or axle) history, with
1710 reversals, has a tensile bias with frequent reversals into compression. The
bracket history, with 5936 reversals, is a narrow-band, random loading with little
or no mean bias.

FIG. 1-Three irregular load blocks representative of ground vehicle service (courtesy of
the Cumulative Damage Division of the SAE Fatigue Design and Evaluation Committee).

Specimen and Materials

The specimen used in the SAE test program was a modified compact tension
type with a large keyhole notch [18]. Two commonly used structural steels,
U. S. Steel Man-Ten and Bethlehem RQC-100, were tested. Mechanical
properties and estimated fracture toughness values for these steels are given in
Table 1.

TABLE 1 - Mechanical properties and fracture toughness of U.S. Steel

Man-Ten and Bethlehem R QC-IO0.

Man-Ten RQC-100

Monotonic yield strength (0.2%), MPa (ksi) 324 (47) 815 (118)
f
Cyclic yield strength (0.2%), Oys, MPa (ks1) 324 (47) 62O (90)
Ultimate tensile strength, MPa (ksi) 565 (82) 863 (125)
Reduction in area, % 67 44
Fracture toughness, a MPa X/~ (ksi X~.) 121 (110) 154 (140)

a Estimated from knowledge of critical crack lengths and corresponding maximum

tensile loads for tests reported in Ref 18.

K-Calibration
The K-calibration was derived from experimental measurements [19] of
specimen compliance (C) at various normalized crack lengths (a/W) ranging from
0.35 to 0.72. First, these data were divided into two intervals: (a/W) from 0.35 to
0.53 and from 0.53 to 0.72. Data in each interval were fitted by a cubic function
in a least-squares sense and subject to the constraints that at (a/W) = 0.53, the
two adjoining functions have equal values, slopes, and curvatures. This produced
a single continuous curve (with continuous first and second derivatives) which
represented the overall trend of the scattered data without introducing
inflections in the fitted curve. A K-calibration in the form (K/P) = f(a/W) was
then derived using Eqs 1 and 2.

G =, ..... (1)
2B
K=~ (plane stress) (2)

where
G = elastic strain energy release rate,
E = modulus of elasticity,
B = specimen thickness, and
P = applied loading.
The experimentally derived K-calibration compares favorably with an analyt-
ical (boundray collocation) calibration [20] for a compact tension specimen of
similar geometry. In particular, the experimental calibration is 5 percent lower
than the analytical calibration at (all49 = 0.35. The difference between the two
K-calibrations diminishes as (a/W) increases, and they converge at (a/W) ~ 0.6.
For (a/W) greater than 0.72, the analytical K-calibration was used for predictions
since no experimental data were available for larger (a/W) values.

Constant-Amplitude Aa/AN versus AK Data

Measurements of crack length versus cycles [21] under constant amplitude, R
= 0, loading for both Man-Ten and RQC-100 were made using compact tension
specimens (ASTM Standard Method of Test for Plane-Strain Fracture Toughness
of Metallic Materials (E 399-72)) machined from the same heats of steel as the
modified compact tension specimen. The "raw" crack growth data were
converted to Aa/AN versus AK form using an incremental-polynomial data-
fitting technique [22].
Crack growth rate data were available only in the range of &a/AN from about
5 x 10 -7 to 3 x 10 -s cm/cycle for both metals. The threshold stress intensity
and fracture toughness values and possible stress ratio effects were not
determined. 4 This lack of data might be a serious hindrance to predicting crack

4 Further tests are planned by the Fracture Mechanics Division of the SAE Fatigue Design
and Evaluation Committee to determine these data.

growth. Yet, it provided an opportunity to examine the capability of the

prediction methods to perform with limited input data, a situation typical of
engineering practice.
For both Man-Ten and RQC-100, Aa/AN at AK = 10 MPax/~'is roughly 10-7
cm/cycle. In addition, threshold stress intensities for a number of other steels
fall in the range of 5 to 15 M P a x / ~ [ 2 3 ] . Thus, two values of &Kth, 5.5 and
11.0 MPax/~, were used in this work to examine the effect of ZkKth on
predictions.

Crack Growth Rate Relations

An important consideration in any crack growth prediction is the selection of

a growth rate relation. For irregular loadings, the ability of such relations to
account for stress ratio effects on crack growth rate is of particular concern.
Two relations were used here.
The Forman relation [24], that is,

Aa/AN = A(AK)n (3)

(1-R )K c - AK

where R = gmi n / g m a x , was selected because of its ability to predict possible

stress ratio effects using only R = 0 data, all that was available for Man-Ten and
RQC-100. Whether it adequately predicts such effects for these metals remains
to be investigated experimentally. Constants A and n in Eq 3 were determined
by a linear least-squares fit of log [(Aa/z2CO(Kc - AK)] versus log (z2uZ)using R
= 0 data for Man-Ten and RQC-100.
A crack growth rate relation based on an effective stress intensity range
concept was also used, that is,

Aa/~h~r A(~r~eff )n (4)

K c - K max
where AKeff is an effective stress intensity range equal to (gma x - gre s) or
(gma x - gmi n ) if gmi n is larger than Kre s. The Kres term is a "residual stress
intensity" which corresponds at least conceptually to the crack opening stress
described by Elber [25]. In this work, Kre s was taken as q(Kmax - zMr~th),
where q is a fraction. Threshold stress intensity range was included in the (K max
- &Kth ) term to make Kres vanish as Kmax approaches &Kth; however, this
condition is not an essential element of the growth rate relation and, with
hindsight, could probably be discarded.
In general, the q value is likely to vary from metal-to-metal, for different
specimen geometries and thicknesses, and will even increase with increasing
Kmax, as suggested by the studies of Newman [26] and Schijve [27]. As a first
approximation, q was considered a constant.
The &Keff term in Eq 4 will be used to account for both stress ratio effects
and load sequence effects such as crack retardation. Keeping this in mind, values
of q suitable for exploratory purposes were estimated as follows. The application

of a single tensile overload of sufficient magnitude, Kol , in constant-amplitude

(R = 0) loading, Kca, will usually cause crack arrest. The ratio (Kol/Kca) varies
between two and three for a number of metals [13,17, 28-33]. If AKeff is to
account for crack arrest or retardation, then q should be taken as the reciprocal
of (Kol/Kca), thus varying between 0.35 and 0.50. Constants A and n in Eq 4
were fitted for both "extreme" values of q by the same procedure used to
calibrate Eq 3.
The use of Eq 4 to account for load sequence effects will be described in detail
later. First, its ability to predict stress ratio effects on crack growth will be
considered.
Equation 4 was calibrated with R = 0 growth rate data, K c and z~r(th values
for aluminum alloys 2024-T3 and 7075-T6 [34] and titanium alloy Ti-6A1-4V
[35]. In all cases, a value of q = 0.35 was used for exploratory purposes. Stress
ratio effects were then predicted and compared with test data. Figure 2 shows
such a comparison for 2024-T3 and is typical of the agreement between
predictions and test results for all three metals. Unfortunately, no comparable
stress ratio data exist for Man-Ten or RQC-100 to allow Eq 4 to be checked for
these metals.

Crack Growth Predictionsfor Constant-Amplitude Loading

Before considering the prediction of crack growth for the three irregular
loadings, predictions based on Eqs 3 and 4 for constant-amplitude loading

MPa ~/~
IO_Z 0 IO 20 50 40
I i I
10-2

9 0.50
- 10-5 o 0.70 ,o~..
10-3
l PREDICTION~

tO~

A~ 9 D
i0- 5 ix o

10-5

i .I o~
10-6
M ~ u/ K~ = STksiJT~ 10-6
~L ~ AK th = 2 k $ i ~ ' ~ -

,o-'" v&,d , , , (~i~M~~ ~-I ,

0 5 ~0 15 20 Z5 30 55
MAXIMUM STRESS INTENSITY (Km0x), ksi ilTn"

FIG. 2-Comparison o f predictions with test data for stress ratio effects on fatigue crack
growth rate in 2024-7"3 [34].

applied to the modified compact tension specimen will be compared with

available test data [18]. All predictions were based on growth from an initial
crack length of 33 mm, as measured from the specimen load line (or 2.5 mm
from the specimen notch tip).
Predictions were first made for fully-reversed loadings. Only the tensile
half-cycle was considered in computing AK. For RQC-100, predictions were
within experimental scatter for tests with a 17.8 kN load amplitude and within
20 percent of total crack propagation life for single tests at 13.4 and 15.6 kN
load amplitudes. For these load amplitudes, predictions were the same for AKth
= 5.5 and 11.0 MPax/~. For Man-Ten, predictions were within 10 percent of
total life for a single test with 8.9 kN load amplitude and were again insensitive
to ~ t h -
Although only limited constant-amplitude test data were available for
comparison, these predictions were reassuring. They suggested that the limited
Aa/AN versus AK data generated for Man-Ten and RQC-100 and the
experimentally derived specimen K-calibration were probably adequate for
making reasonable predictions for the irregular loadings.
Predictions were also made for two single constant-amplitude tests conducted
at stress ratios of R = 0.17 and 0.20 for Man-Ten, at maximum load levels of
26.7 and 22.3 kN, respectively. In both cases, predictions based on Eqs 3 and 4
were about a factor of 2.5 too conservative, indicating that perhaps stress ratio
effects were being predicted which do not actually occur in Man-Ten. To gain
some insight into this possibility, predictions were also made using the Paris
relation, Aa/AN = A (AK)n, which does not account for stress ratio effects.
Predictions based on this relation agreed very closely with those based on Eqs 3
and 4 for the fully-reversed tests. As expected, however, the Paris relation
predicted less growth for the two stress ratio tests with Man-Ten. Yet, it still
predicted too much growth, by a factor of about 1.5. It is conceivable that the
high tensile loadings in these two tests created compressive residual stresses in
the vicinity of the specimen notch, the influence of which may have retarded
initial crack growth beyond 33 mm. In any case, it is difficult to draw
conclusions based on the results of only two tests.

Prediction of Crack Growth for Irregular Loadings

All prediction methods were based on the following two key assumptions: (a)
that compressive loadings cause no growth and (b) that only rising tensile load
ranges cause growth. The first assumption seems physically plausible, since under
compressive loading, a crack tip should be shut and therefore unable to produce
growth. The second assumption is supported by fractographic studies of
McMillan and Pelloux [36], which showed that crack growth apparently occurs
only during the rising portion of load cycles. Based on the first assumption, the
three irregular load blocks described previously were converted to blocks
containing only tensile loadings, as illustrated in Fig. 3.
Each prediction method also utilized the following procedure to compute and
sum crack growth. First, an initial crack length, ao, was selected. The first rising
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
274 FATIGUE CRACK GROWTH

Q -~ Q
~IB ID L ~
" E- ~'V .... ~ M E T,~
I

FIG. 3-Conversion ofan irregular load sequence to only its tensile loadings.

load range in a given block was converted to a corresponding stress intensity

range. Crack growth for the first range, (Aa/AN)I , was computed, giving a new
crack length a~ = ao + (Aa/AN)I. Next, the second rising load range was
converted to a stress intensity range using a~. Growth due to this range,
(Aa/AN)2, was computed giving a2 = a~ + (Aa/AN)2. This process was
continued throughout the entire load block. The sum of growth for a block gave
a growth rate per block associated with the initial crack length, ao, that is,
(Aa/Ablock) at ao. This procedure was repeated for a number of selected crack
lengths, ai, which, in turn, allowed construction of a curve of crack length versus
blocks by a simple numerical integration of

af_
blocks = . / ' (Aa/Ablock)- 1 Aa (5)
-a0

Prediction M e t h o d Ia

Equation 3 was used to calculate growth for each rising load range in a given
block. No growth was calculated for ranges with AK less than AKth. As
mentioned before, two values of AKth (5.5 and 11.0 MPax/~) were used, which
provided information on the sensitivity of predictions to Agth at various
extreme load levels. This method, of course, neglects possible load sequence
effects.

Prediction Method lb

This prediction method is the same as la except that crack growth was
calculated for load blocks which were first condensed to 10 percent of their
original number of reversals by the Ordered Overall Range Method [11,12]
before compressive loads were removed. This method of condensing irregular
load blocks selects overall ranges and screens out smaller ranges which interrupt
the overall ranges, as illustrated in Fig. 4. It is qualitatively similar to "rainflow"
cycle counting [8], except that it condenses a loading while preserving its
sequence. The size of ranges which are screened out is taken as a fraction of the
largest overall range in a given load sequence. Fractions are selected depending
on the extent to which one wishes to condense a sequence. The method was
originally developed for use in making crack initiation predictions, where
long-life damage, D, varies with stress range, AS, to a high power, for example,

5~- ~B D

,-, 5 I / ~ \ / ~ F :-

_5L3 ,~. TIME

BE = LARGEST OVERALL RANGE

CD= OVERALL RANGE WHICH INTERRUPTS BE
AB, BC, DE, EF = OTHER OVERALL RANGES
b
FIG. 4-Selection of overall ranges in an irregular load sequence (a), screening out smaller
interrupting ranges which are less than 40 percent of the largest overall range; (b)
corresponding condensed sequence.

D cc (AS)I o, and thus the damage of smaller ranges relative to larger overall ranges
could be neglected in most cases without undue loss of accuracy. For crack
propagation, damage usually varies with stress intensity range to a lower power,
for example, D = Aa/AN = (AK) 3 , and thus smaller ranges are relatively more
damaging than their counterparts for long-life crack initiation. However, the use
of condensed load blocks may still offer substantial savings of test and
computing time for crack propagation studies without an unreasonable sacrifice
of accuracy, especially in view of such uncertainties as scatter in experimental
data, approximate kno.wledge of stress intensity factors for complex geometries
and real flaws, etc.
A second purpose of using condensed load blocks was to investigate how crack
growth, calculated on the basis of overall ranges, would compare to that
calculated range-by-range (for every range), as done with Method Ia (see Fig. 5).

K B

-~-~"-A7"-v ~(AK)2 I ~ (&K)0vERALL

A TIME A TIME
a b
FIG. 5-Range-by-range (a} and overall range (b) methods o f calculating crack growth,
which poses the question of how crack growth due to (AK]I, (AK}2 , and (AK)3 compares
to that for (AK)overa u.

Prediction Method H
This prediction method uses the Afffeff term in Eq 4 to account for the
following possible load sequence effects: (a) crack retardation, (b) the decrease
of retardation by sufficiently large compressive overloads, and (c) the accelera-
tion of crack growth rate by gross yielding in compression (but not in tension).
The general philosophy of using zSXeff to account for such effects is
illustrated in Fig. 6. The Z2Urfefffor the cycles from Points A to B is that which

KI ~-- TENSILEPEAK, K'mox

l/ //
COMPRESSIVE LOAD WHICH
CAUSES GROSS YIELDING

FIG. 6-1llustration o f the use o f an effective stress intensity range concept to account for
various sequence effects in a simple variable-amplitude loading.

would normally be active in the absence of large tensile or compressive

overloads. At Point B, a large tensile peak occurs, causing the /(res level to
t
increase to q(Kma x - ZXKth). As a simplification, the effects of "delay and
decay" of retardation [13,28,29] are not shown. The AKeff for the cycles from
Points B to C is reduced, producing retarded crack growth for these cycles.
At Point C, a large compressive load is applied. Based on test results [14-17]
mentioned earlier which show that compressive overloads tend to eliminate
crack retardation, the Kre s level is returned to the normal level associated with
the next loading, which in this case happens to be the same as that at Point A.
Retarded crack growth ceases, and the growth rate is returned to normal values.
In this work, any compressive load greater than or equal to a previous tensile
overload was assumed to destroy the retarding effect of the tensile overload.
At Point D, a compressive overload large enough to cause gross yielding in
compression occurs. To try to account for the increase in growth rate which is
likely to result from such yielding, the/(re s level is reduced to zero, increasing
~ e f f (as long as a crack tip is within the compressively yielded zone). Recent
X-ray diffraction measurements [37] have shown that the usual compressive
residual stresses in front of a crack tip are eliminated in a specimen subjected to
gross yielding in compression. Without such residual stresses to exert a clamping
influence on a growing crack, growth rate should be accelerated in accordance
with crack closure concepts. Note that throughout all of this, the assumption
that compressive loadings do not cause growth directly is maintained. It is
proposed, however, that large compressive loadings may greatly influence the

ability of tensile loadings to cause growth through the mechanism of residual

stresses.
The ZM~eff term could also be used to try to account for the acceleration of
crack growth due solely to the application of compressive overloads even in the
absence of gross yielding. However, this was not done here because other effects
seemed to be much stronger in the given load histories.
Since tensile peaks are so closely spaced in the load blocks considered here and
since the blocks are applied repeatedly in testing, it was assumed, as a further
p
simplification, that the highest tensile peak in each bIock would establish a Kre s
t i f
= q(Kma x - ZkKth), where Krnax and Kre s are the applied and "residual" stress
intensities associated with the highest peak. As noted previously, two values of
q, 0.35 and 0.50, were used to see the effect on predictions. As another
simplification, since large compressive loads occur frequently in the load blocks,
r
it was assumed that Kre s due to the highest tensile peak would be eliminated
t
if [Kmintwas greater than tKmaxt' where K'rain is a "stress intensity"
associated with the largest compressive load in a block.
The effect of gross yielding in compression was taken into account as follows.
The elastic compressive stress distribution in the modified compact tension
specimen, based on simple beam theory and as determined by finite elements, is
shown in Fig. 7 for the case of -40.0 kN loading with Man-Ten. Under

mrn
0 10 20 30 40 50 60
I , , , , , I-I000
I /---- FINITE ELEMENT I
";, h / ELASTIC STRESSES I
~-iO0 ~'V r " - ~ ' S = O-y' =4'Tksi (324 MPo) | ~;
L\ I ~PROeAeLE STRESSES -~-500
OOE TO,,ELO,, / "
l ",,~.:.~-... ,--S:ELAST,C STRESSES

0 0.5 1.0 2.0

DISTANCE FROM NOTCH, inches

FIG. 7-Finite element and simple beam theory estimates o f elastic stresses in the
modified compact tension specimen plus an estimate o f the yield zone size in compression
for Man-Ten at --9.0 kips (-40.0 kN) load level.

compression, it was assumed that a crack closes completely and that the
specimen behaves as if no crack were present. It was further assumed that if for
any crack length, the nominal, elastic compressive stress at the crack tip, S
(based on simple beam theory), exceeded the cyclic yield strength, then Kre s
would drop to zero, producing accelerated growth. This assumption gives only a
rough approximation to the size of the zone of gross yielding in compression for
this specimen, but it is suitable for exploratory purposes. When the crack tip
grows out of the influence of this zone, Kre s is returned to the normal level
associated with the next loading.

No crack growth was calculated, of course, for load ranges with Z~eff less
than AKth. Predictions were made only for full load blocks, as with Method Ia.

Comparison of Predictions with Test Results for Irregular Loadings

Predictions were made only for those tests in which gross yielding in tension
did not occur so as not to violate the usual limits of applicability of linear elastic
fracture mechanics upon which the prediction methods are based. All predic.
tions were based on growth from an initial crack length of 33 mm.
Predictions at various extreme load levels are compared with average test life
to fracture in Table 2. To save space, only plots of crack growth versus blocks
for selected cases are shown in Figs. 8 through 12.

Transmission (Axle) Load History

Predictions were made at extreme load levels of +15.6 and +35.6 kN for
Man-Ten and at +35.6 kN for RQC-100.
Method Ia predictions are overly conservative for the Man-Ten tests for both
AKth values but are well within a factor of two of average test life. Figure 8
illustrates this for the +15.6 kN extreme load level. On the other hand, Method
Ia predictions are exceptionally good for the RQC-100 test.
Method Ib predictions are in excellent agreement with test data for the
Man-Ten tests. As shown in Fig. 8, predictions are within experimental scatter
and are also relatively insensitive to AKtu at the +15.6 kN extreme load level
(which is not the case for Method Ia). For RQC-100, Method Ib predictions fail
to account for enough growth yet are still within a factor of two of average test
life.
Method II predictions with q = 0.35 "bracket" experimental scatter for the
two Agth values, as shown in Fig. 9 for Man-Ten at +15.6 kN extreme load
level. The prediction with q = 0.50 and AKth = 5.5 MPax/'~ is within
experimental scatter while that with AKth = 11.0 MPax/~ is nonconservative by
about a factor of two. For Man-Ten with +35.6 kN extreme load level, Method
II predictions with q = 0.35 are overly conservative for both AKth values, but
are still within a factor of two of average test life. Predictions with q = 0.50 are
within experimental scatter for both AKth values. For RQC-100, Method II
predictions with q = 0.35 are exceptionally good, while with q = 0.50, they fail
to account for enough growth but are still within a factor of 1.4 of average test
life.

Bracket Load History

Predictions were made for Man-Ten at extreme load levels o f - 1 3 . 4 and -15.6
kN, and for RQC-IO0 at -15.6 and -35.6 kN.
Method Ia predictions are overly conservative for the Man-Ten tests for both
AKth values but are within a factor of two of average test life. Predictions based
on this method are quite good for RQC-IO0 at -35.6 kN extreme load level, as

Load History Transmission Bracket Suspension

Prediction la lb II II Ia Ib I1 II la Ib II II
Method q = 0.35 q ----0.50 q = 0.35 q = 0.50 q = 0.35 q=0.50
AKth, ksiN/~.

aN/~) 5 10 5 10 5 10 5 10 5 10 5 10 5 10 5 10 5 10 5 10 5 10 5 10
Load, ~. (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11) (5.5) (11)
kips ~k~ Z
Ill
r-
eD
i 0
'3.0(13.4) No Tests 0.45 0.80 0.90 1.10 0.40 0.90 0.40 0.90 No Crack Initiation Z
i
13.5(15.6) 0.65 0.90 1.05 1.05 0.80 1.40 1.20 2.10 0.50 0.70 0.95 1.05 0.45 0.65 0.45 0.65 No Tests Z

16.0(26.7) No Tests No Tests 1.55 2.00 2.35 3.00 1.40 2.00 1.40 2.00 C
I
8.0(35.6) 0.60 0.60 0.90 0.90 0.60 0.60 0.85 0.85 No Predictions No Tests
0
9.0(40.0) No Tests No Tests 8.5 9.4 13.2 14.6 4.6 5.7 3.4 4.2 Z

3.5(15.6) No Crack Initiation 0.15 0.20 0.25 0.30 0.15 0.20 0.15 0.20 No Tests -n
m
6.0(26.7) No Tests No Tests 0.45 0.60 0.65 0.900.40 0.60 0.40 0.60 C
r-
i7.0(31.2) No Tests No Tests 0.60 0.70 0.90 1.05 0.60 0.65 0.60 0.65
r-
6 8.0(35.6) 0.95 0.95 1.55 1.55 1.00 1.00 1.40 1.40 0.90 0.90 1.70 1.70 0.85 0.85 0.85 0.85 No Tests 0

9.0(40.0) No Tests No Tests 0.60 0.70 0.90 1.05 0.55 0.65 0.55 0.65 Z
G3
16.0(71.2) No Predictions No Predictions 12.4 12.5 19.1 19.3 5.8 5.8 3.9 3.9
bO
tD

2.7 , t ] I r I ~ i
2.6 TRANSMISSION TEST D A T A 9 9 9
MAN-TEN 65
2.5 + 3.5 KLBS. (+15.6KN) EXTREME LOAD =')~.li )1(
/,i
2.4 60
~; 2.3 PREDICTION METHOD
Ta, A K t h = 5 KspJ~
2.2 55
"I- (5 5 MPa ~/~)
. . . . Ta, A K t h - 10
~-
(3 2.1 -~ Ib, AKth - 5
Z
UJ 2.0 ...... rb,~.Kth = 10 50 ~
g 9
~, 1.9 If.
1.8 45
1.7 j.
1.6 40
BLOCKS TO CRACK / _ _ /
1.5 INmATION (AVE.)= 4.6 x 103 / _ : - /
1.4 35

1.3q01 10 2 10 3
BLOCKS
FIG. 8-Comparison of Methods la and lb predictions with test data for the transmission
history with Man-Ten at +3.5 kips (+15. 6 kN) extreme load level.

2.7 , ~ I ~ I I
2.6 TRANSMISSION TEST D A T A 9 9 9
65

2.5 + 3.5 KLBS. 1+15.6KN) EXTREME LOAD

2.4 60

__Z 2 . 3 PRED,CTIONMETHOD i I
:Z:" 2.2 .., ]z, q=.35, ~ K t h : 5 ks, ~ Ill | 55
I~ (5 5 MPa ~ ) II I
~.~ 2 . 1 . . . . ]I.q=,35, AKth = 1{)
Z ]1", q=.50,~.Kth = 5 l I
uJ 2.0 50
,J ...... ~'. q=.50, ~.Kth = 10 |9
~, 1.9 ;9

~C 1.8
r~
~
1.7
/i;: 45

40
1.5 BLOCKS TO CRACK
1.4! I N I T I A T I O N (AVE.) = 4.6 x 10 3 ~'~/~'~'i"',,,/ " I "
35
1.31 i = , r'-~ i I I
101 102 103
BLOCKS
FIG. 9-Comparison o[ Method l/predictions with test data for the transmisston history at
+3.5 kips (+15.6 kN) extreme load level.

2.7 , ~ I f ~ J ' '

BRACKET liP. q
2.6 ROC- OO II !',
2.5 --8.0 KLBS. (35.6KN) ._ In II
2.4 EXTREME LOAD II I 60
:~ 2.3 PREDICTIONMETHOD li 9 'I
- ~ 9 ,
2.2 " Z.,~K,h : ~ K . , ~ li [' 55
(fi.5 MPa vrm) Ii *'1
2.1 .... Z,.aKth:, o li " ~ It
~-~ Zb.AKth S 9 II 'l
2.0 ...... Ib, AKth 10 II " /] 50
1,9 hi :,
NOTE ME THOCJ~ PREI~ICTIONS 9 # f
~1.8
ARE-"~THE SAME FOR q 35 and 50
AHOA.E VE.uCLOSE,O T.OSE Llun /
i 45
o 1.7 ~o. METHODI,, FOR BOTH 9 LO /
~.Ktt VALUES FEB I
1.6 ' 9 2/~= / 40
1.5 BLOCKS TO CRACK ~ /
INITIATION (AVE.) = 9 ~ - ' ~ ' / ' " TEST DATA 9 9
1.4 8.3 x 101 ~.,~=.~.~.~_~.7.--'- . . . . 9 35
1.3 J "~ 9 ~ I I I i ,

1 10 102
BLOCKS
FIG. lO-Comparison of Method la and Ib predictions with test data for the bracket
history with RQC-IO0 at -8. 0 kips (-35.6 kN) extreme load level,

2.7 I I I F I I i i

2.6 SUSPENSION
RQC-100 65
2.5 - 1 6 . 0 KLBS. (71.2KN) EXTREME LOAD
2.4
60
2.3 PREDICTION METHOD
m
Ia. AKth 5 KS, ~/~
2.2
l- (5 5 MPa vr~) 55
ED 2.1 -------- Ia. AKth : 10
Z ~ - - ~ [b, AKth 5
UJ
_J
2.0 ...... [b, AKth : 10 50 ~
v 1.9
1,8 45
BLOCKS TO CRACK 9149
rj 1,7 INITIATION (AVE.) = 9 TEST DATA 9 9 9
1,6 3.6 x 101 4O
1,5 &O
1.4 O O 9 9 35
1.3 I I I
1 10 102
BLOCKS
FIG. 11-Comparison of Method la and Po predictions with test data for the suspension
history with RQC-IO0 at -16.0 kips (-71.2 kN) extreme load level.

27 r t l I 1 [ I l
2.6 SUSPENSION TEST DATA 9 9 9
RQC-100 i r 65
2.5 --16.0 KLBS. (-71.2KN) EXTREME LOAD I
9 '=
2,4
9
9 Ii
tJ
i
i
r 60
;~ 2,3 PREDICTION METHOD 99 ]i i
i
]I, q = 3 5 , A K t h = 5 ksl v/~ 9 II
i
. 22
"1- ( 5 5 MPa ~ ) 9 [! L 55
. . . . ] I , q = 3 5 , ~ K t h = 10
i
(,9 2.1 ~ - ~ ]I, q=.50,,~Kth = 5
i
I~
Z 2,0
.J
...... ] I , q= 50, A K t h = 10 9 9 so :
1.9
< 1.8 BLOCKS TO CRACK / ]l 45
INITIATION (AVE.) = 9149 ,/ n/1
o 1.7
3 6 x 101 9 9 ~ . J j
1.6 9 ,~" dw, 40
1.5
1.4 I I 9 9 /./ / 35
1.3
10 102
BLOC KS
FIG. 12-Comparison of Method H predictions with test data for the suspension history
with R QC-I O0 at -16.0 kips [-71.2 kN) extreme load level.

shown in Fig. 10; however, five to six times too much growth is calculated at the
-15.6 kN extreme load level, depending on the &Kth value used.
Method Ib predictions are in excellent agreement with the Man-Ten test
results; however, for RQC-100 at the -35.6 kN extreme load level, they fail to
account for enough growth, as shown in Fig. 10, but are still within a factor of
two of average test life. Method Ib predictions are somewhat better than those
for Ia at the -15.6 kN extreme load level with RQC-100 but are still three to
four times too conservative.
Method II predictions are the same for q = 0.35 and 0.50 in all cases and are
very close to those for Method Ia for both ~ t h values.

Suspension Load History

Predictions were made for Man-Ten at extreme load levels of -26.7 and -40.0
kN, and for RQC-100 at extreme load levels of -26.7, -31.2, -40.0, and -71.2
kN.
Method Ia predictions fail to account for enough growth for the Man-Ten
tests. For the -26.7 kN extreme load level, predictions are nonconservative by
about a factor of two (in terms of total life). At the -40.0 kN extreme load
level, predictions are nearly an order of magnitude nonconservative. In both
cases, initial crack growth was more rapid than later growth. For the -26.7 kN
extreme load level, growth from 33 to 36 mm took only about 400 blocks while
that from 38 to 41 mm took about 3000 blocks. For the -40.0 kN extreme load

level, growth from 33 to 38 mm took about 25 blocks, while that from 41 to 46

mm took about 200 blocks. The influence of compressive yielding at the notch
may well be causing this initial accelerated growth. In contrast to the results for
Man-Ten, Method Ia predictions for RQC-100 tend to be overly conservative but
are generally within a factor of two of average test life for the -26.7,-31.2, and
-40.0 kN extreme load level tests. At -71.2 kN extreme load level, predictions
fail to account for enough growth by over an order of magnitude, as shown in
Fig. 11. In this case, as with the -40.0 kN test for Man-Ten, gross yielding in
compression but not in tension occurred. Again, initial crack growth was
accelerated.
Method Ib predictions are even more nonconservative than those for Method
Ia for the Man-Ten tests and for the -71.2 extreme load level tests with
RQC-100, as shown in Fig. 11. On the other hand, Method Ib predictions agree
better with test results for RQC-100 at -26.7, -31.2, and -40.0 kN extreme load
levels.
Method II predictions are the same for q = 0.35 and 0.50 and agree very
closely with Method Ia predictions for all tests except those at extreme load
levels of -40.0 kN for Man-Ten and -71.2 kN for RQC-100. In both cases,
predictions based on this method try to account for the acceleration of crack
growth due to gross yielding in compression. Method II predictions are clearly
better than those for Method I, as shown in Fig. 12. However, they still fail to
account for enough growth. The discontinuity in the crack growth prediction
curves indicates the point where the effects of compressive yielding are predicted
to cease. Note that if these effects were allowed to continue for another 3 mm
or so beyond this point, which they actually may, then the predictions would
agree quite well with total crack propagation life. However, they would still fail
to account for enough of the initial accelerated growth.

Discussion
Predictions based on Methods Ia or Ib were almost always within a factor of
two of average test life with the exceptions of the suspension load history when
gross yielding in compression occured and the bracket history with RQC-100 at
-15.6 kN extreme load level. Method Ib predictions always accounted for less
growth than those based on Method Ia, but never by more than a factor of two.
It is interesting to note that in many cases, predictions based on Method Ib were
in better agreement with test data than those based on Method Ia, particularly
for the transmission and bracket load histories with Man-Ten.
In general, reasonably good predictions could be obtained from a limited
amount of R = 0, constant-amplitude Aa/AN versus AK data, and by estimating
stress ratio effects and using assumed values of AKth. This is significant for a
realistic design situation where the use of incomplete data is often necessary.
Use of condensed histories offers several advantages: (a) substantial savings of
test and computing time (in this case, a factor of ten reduction) may be achieved
without undue loss of accuracy in view of experimental scatter, and (b) the

effect of AKth on predictions is minimized. Furthermore, condensed load blocks

(by neglecting smaller load variations) are composed of overall ranges virtually
all of which have R -~ 0. Thus, stress ratio data need not be known precisely to
predict growth using condensed blocks. This point and Item b are significant
because to determine AKth and stress ratio effects for a particular metal involves
considerable testing and expense.
The fact that Methods Ia or Ib produced fair predictions although they neglect
load sequence effects suggests, of course, that the load histories considered here
do not produce significant sequence effects. In addition, the assumption that
only tensile loadings cause growth appears reasonable in all cases studied here
with the notable exception of the suspension history tests when gross yielding in
compression occurred.
The success of Methods Ia and Ib, which neglect the effects of load sequence,
should be viewed with some caution. For load histories with occasional targe
overloads, tensile or compressive, they may produce overly conservative or
dangerously nonconservative predictions.
Method II predictions are comparable to those for Method I for the
transmission and bracket histories and are clearly superior for the suspension
history when gross yielding in compression occurs.
The use of an effective stress intensity range concept to account for the
acceleration of crack growth by sufficiently large compressive loading, especially
that which causes gross yielding, has a worthwhile potential. It is consistent with
crack closure concepts and experimental measurements of the destruction of
compressive residual stresses at a crack tip by high compressive loadings,
especially in lower yield strength metals.
Finally, the calculation of crack growth range-by-range compared to using
overall ranges warrants further investigation. For example, tests should be
conducted to see how crack growth for a load sequence such as that shown in
Fig. 13a compares to that shown in Fig. 13b. Such test results would provide a
good check on the suitability of crack propagation prediction methods for
irregular loadings.

TIME TIME
a b
FIG. 1 3 - T w o load sequences for which comparative crack growth tests should be
conducted.

Conclusions
1. For the load histories studied here, predictions based on Method Ia
(calculating growth range-by-range with full histories and using the Forman

relation) were within a factor of two of average crack propagation test life in
nearly all cases and were always on the conservative side, with the notable
exception of the suspension load history when gross yielding in compression
occurred.
2. The use of condensed load histories with Method Ib (calculating growth
with overall ranges and neglecting smaller load variations) also produced
comparably good predictions.
3. Predictions based on Method II, which tries to account for possible load
sequence and stress ratio effects, were essentially the same as those for Method
Ia for the bracket and transmission histories but were considerably better for the
suspension history when gross yielding in compression occurred.

Acknowledgments

Financial support from the Association of American Railroads, Deere &

Company, and the National Science Foundation through the Stanford Center
for Materials Research and the cooperation, encouragement, and advice of the
members of the Fracture Mechanics Division of the SAE Fatigue Design and
Evaluation Committee and of the reviewers of this paper are gratefully
acknowledged.

References
[1 ] Katcher, M., Engineering Fracture Mechanics, Vol. 5, 1973, p. 793.
[2] Wheeler,O. E., Journal of Basic Engineering, Transactions, ASME, Vol, 94, Series
D, No. 1, March 1972, p. 181.
[3] Willenborg,J., Engle, R. M., and Wood, H. A., "A Crack Growth Model Using an
Effective Stress Intensity Concept," Technical Memorandum AFFDL-TM-73-
137-FBR, Air Force Flight Dynamics Lab, Dayton, Ohio, 1971.
[4] Gallagher, J. P. and Stalnaker, H. D., "Methods for Analyzing Fatigue Crack
Growth Rate Behavior Associated with Flight-by-Flight Loading," paper presented
at the AIAA/ASME/SAE 15th Structures, Structural Dynamics and Materials
Conference, Las Vegas, Nev., April 1974.
[5] Barsom, J. M. in Progress in Flaw Growth and Fracture Toughness Testing, ASTM
STP 536, American Society for Testing and Materials, 1973, p. 147.
[6] Smith, S. H. in Structural Fatigue in Aircraft, ASTM STP 404, American Society
for Testing and Materials, 1966, p. 74.
[ 7] Swanson,S. R., Cicci, F., and Hoppe, W. in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, p. 312.
[8] Endo, T., Mitsunaga, K., Takahashi, K., Kobayashi, K., and Matsuishi, M., "Damage
Evaluation of Metals for Random or Varying Load," paper presented at the 1974
Symposium on Mechanical Behavior, Kyoto, Japan.
[9] Dowling, N. E., "Fatigue Failure Predictions for Complicated Stress-Strain
Histories," Report No. 337, Department of Theoretical and Applied Mechanics,
University of Illinois, Urbana, I11.,Jan. 1971.
[10] Schijve, J., "The Analysis of Random Load-Time Histories with Relation to Fatigue
Tests and Life Calculations," Fatigue of Aircraft Structures, MacMillan Co., New
York, 1963.
[11] Fuchs, H. O., Nelson, D. V., Burke, M. A., and Toomay, T. L., "Shortcuts in
Cumulative Damage Analysis," Paper No. 730565, presented at the SAE Auto-
mobile Engineering Meeting, Detroit, Mich., May 1973.
[12] Nelson, D. V. and Fuchs, H. O., "Predictions of Cumulative Damage Using
Condensed Load Histories," Paper No. 750045, presented at the SAE Automotive
Engineering Congress and Exposition, Feb. 1975.

[13] Rice, R. C. and Stephens, R, I. in Progress in Flaw Growth and Fracture Toughness
Testing, ASTM STP 536, American Society for Testing and Materials, 1973, p. 95.
[14] Hsu, T. M. and Lassiter, L. W. "Effects of Compressive Overloads on Fatigue Crack
Growth," paper presented at the AIAA/ASME/SAE 15th Structures, Structural
Dynamics and Materials Conference, Las Vegas, Nev., April 1974.
[15] Private communication, R. I. Stephens, Materials Engineering Division, University
of Iowa, Iowa City, Iowa.
[16] Schijve, J., "Fatigue Crack Propagation in Light Alloy Sheet Material and
Structures," Technical Report No. MP-195, National Aerospace Laboratory, The
Netherlands, Aug. 1960.
[17] Wei, R. P., Shih, T. T., and Fitzgerald, J. H., "Load Interaction Effects on Fatigue
Crack Growth in Ti-6A1-4V Alloy," NASA CR-2239, National Aeronautics and
Space Administration, April 1973.
[18] Bussa, S. L. and Tucker, L. E., "The SAE Cumulative Fatigue Damage Test
Program," Paper No. 750038, presented at the SAE Automotive Engineering
Congress and Exposition, Detroit, Mich., Feb. 1975.
[19] Private communication, R. C. Rice, Battelle Memorial Institute, Columbus, Ohio.
[20] Srawley,J. E. and Gross, B., Engineering Fracture Mechanics, Vol. 4, 1972, p. 587.
[21] Private communication, D. R. Galliart, Materials Research, Deere & Company,
Moline, I11.
[22] Private communication, N. E. Dowling, Westinghouse Research Labs, Pittsburgh,
Pa.
[23] Donahue, J., Clark, H. McI., Atanmo, P., Kumble, R., and McEvily, A.,
International Journal of Fracture Mechanics, Vol. 8, No. 2, June 1972, p. 209.
[24] Forman, R. G., Kearney, V. E., and Eagle, R. M., Journal of Basic Engineering,
Transactions, ASME, Vol. 89, Series D, No. 3, 1967, p. 459.
[25] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, p. 230.
[26] Newman, J. C., Jr., and Armen, H., Jr., "Elastic-Plastic Analysis of a Propagating
Crack under Cyclic Loading," AIAA Paper No. 74-366, presented at the
AIAA/ASME/SAE 15th Structures, Structural Dynamics and Materials Conference,
Las Vegas, Nev., April 1974.
[27] Schijve, J., The Accumulation of Fatigue Damage in Aircraft Materials and
Structures, AGARD-AG-157, NATO Advisory Group for Aerospace Research and
Development, 1972, p. 75.
[28] Von Euw, E. F. J., Hertzberg, R. W., and Roberts, R. in Stress Analysis and Growth
of Cracks, ASTM STP 513, American Society for Testing and Materials, 1973, p.
113.
[29] Gallagher, J. P. and Hughes, T. F., "Influence of Yield Strength on Overload
Affected Fatigue Crack Growth Behavior in 4340 Steel," Technical Report
AFFDL-TR-74-27, Air Force Flight Dynamics Lab, Dayton, Ohio, 1974.
[30] Corbly, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973, p.
479.
[31] Trebules, V. W., Jr., Roberts, R., and Hertzberg, R. W. inProgress in Flaw Growth
and Fracture Toughness Testing, ASTM STP 536, American Society for Testing and
Materials, 1973, p. 113.
[32] Hudson, C. M. and Raju, K. N., "Investigation of Fatigue Crack Growth under
Simple Variable Amplitude Loading," NASA TN D-5702, National Aeronautics and
Space Administration, 1970.
[33] Jones, R. E., Engineering Fracture Mechanics, Vol. 5, 1973, p. 585.
[34] Hudson, C. M., "Effect of Stress Ratio on Fatigue Crack Growth in 7075-T6 and
2024-T3 Aluminum Alloy Specimens," NASA TN D-5390, National Aeronautics
and Space Administration, 1969.
[35] Fitzgerald, J. H. and Wei, R. P., Journal of Testing and Evaluation, Vol. 2, No. 2,
March 1974, p. 67.
[36] McMillan, J. C. and Pelloux, R. M. N. in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, p. 505.
[37] X-Ray Diffraction Division, SAE Fatigue Design and Evaluation Committee,
unpublished test results, April 1974.

DISCUSSION

C E. Richards 1 (written discussion)-I have three questions or discussion

points concerning this paper.
1. Following our work which is presented in this symposium (P. J. Bernard, T.
C. Lindley, and C. E. Richards), I would suggest that we would require at least
two empirical laws for irregular loading-one for plane stress and one for plane
strain conditions. Conditions between these extremes would require even more
complex combinations of these expressions.
2. Concerning the assumption in Nelson and Fuchs' analysis that only the
rising load is the only important part of the fatigue cycle, Laird has presented a
model where both the rising and falling parts of the fatigue cycle are important
to fatigue crack growth.
3. I do not see how one can argue that crack closure can completely explain
the large effect of R ratio (Ormin/Oma x ) on threshold values of AK for fatigue
crack growth, AK0, when it has been shown that environment can have such a
large influence on AKo. For instance, by testing in vacuo, it has been shown that
under these circumstances R ratio has little or no effect on AKo (C. J. Beeves et
al at Birmingham University, Dept. of Physical Metallurgy and Science of
Materials, England).
D. V. Nelson and H. O. Fuchs (authors' closure)-In this paper, we have
presented several methods of crack growth prediction for review and improve-
ment by others. We appreciate the comments of the discussers. We are aiming at
a method which can be used in design and which must therefore neglect some
effects in order to remain reasonably simple.
Regarding Dr. Richards' first point of discussion, we agree that crack growth
and load sequence effects such as retardation will vary under different types of
multi-axial stress or strain. Plane stress and plane strain are, or course, merely
special cases of multi-axial stress or strain which happen to be more convenient
for analysis. In our paper we avoided much of this problem by making
predictions based on constant-amplitude test data from specimens with the same
degree and pattern of multi-axiality. The prediction of crack growth for
specimens of varying thickness or subject to more general multi-axial loading
conditions is a topic for future studies.
Regarding the second point of discussion, we note that in Laird's model, 2
both the rising and falling parts of a cycle are important, but in very different
ways. New crack surfaces are created during the rising part; crack tip

1 Central Electricity Research Laboratories,Leatherhead, Surrey, England.

2 Laird, C. in Fatigue Crack Propagation, ASTM STP 415, American Society for Testing
and Materials, 1967, p. 131.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright
9 1976 bybyASTMIntemational
Downloaded/printed www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
288 F A T I G U E C R A C K GROWTH

re-sharpening occurs during the falling part. We believe that increments of crack
extension are best related to the AK values associated with rising load ranges.
Re-sharpening may well influence subsequent growth; the effect of the falling
range is implied in the use of AK instead ofKma x
In response to the third point, we note that the crack closure concept has been
used previously as a possible explanation for the effect of R-ratio on threshold
stress intensity range for tests conducted in air. ~ We appreciate Dr. Richards'
reminder that environment often has an important effect on crack growth
behavior. We believe that crack closure is a very useful concept but do not want
to imply that it can explain all effects observed under different environmental
conditions.
C. C. Osgood 4 {written discussion)-These methods appear to predict life with
reasonable conservatism for the cases considered, but with one notable
exception: that wherein the mean stress was at a significant level in compression.
This condition, together with the lack of recognition of compressive applied
stress by the methods would seemingly lead to over-prediction, the severity of
which may be related to the ratio of the total compression to the residual
compression left after the previous tension.
The difference in responses of Man-Ten and of RQC-100 may be partially
explained by Figs. 14 and 15 which show the difference in initiation time (or
blocks) for the two steels. The rather high ordinate location of most of the

I I I ] I ~ X
~J

~.9
,.>,.8 x
x x x-f'~"~
RQ l,-"~lxa" ' -

z
_0.5

re"

% I
5
I
10 15
I
20
I I
25 30
LOAD RANGE x 103 lb.

FIG. 14-Ratio o f initiation to total reversals versus load range at constant amplitude.

3 Schmidt, R. A. and Paris, P. C. in Progress in Flaw Growth and Fracture Toughness

Testing, A S T M STP 536, American Society for Testing and Materials, 1973, p. 79.
4 Forrestal Labs, Princeton University, Princeton, N. J. 08540.

I.O-- I I I ~ I

~--.7
0

Z
0
~-.5 .~. 0 ~
F-
z4
o
~'-.3
o3

0 I L 1 I I
0 5 I0 15 20 25 30
LOAD RANGE xl0 3 lb.

FIG. 15-Ratio of initiation to total blocks versus load range at variable amplitude.

curves indicates that the crack length of 0.1 in. (originally defined as
"initiation") is itself too long with respect to the crack length at failure.
D. K Nelson and H. O. Fuchs {authors' closureJ-Dr. Osgood's plots of crack
initiation blocks divided by total blocks to fracture shown in Figs. 14 and 15 are
most interesting. They put into perspective the relative amount of life spent in
crack propagation versus initiation (for a "transition" crack length of 0.25 mm
(0.1 in.)), which is significant for design purposes.
The question: "When has a crack been initiated and starts to propagate?" has
often been asked but has not been discussed in the literature. We propose the
following operational definition: "A crack has completed its initiation and
begun its propagation when the damage per cycle or per block calculated from
the local shear stress ranges at a notch (based on the stress distribution in the
uncracked part) becomes less than the damage calculated from stress intensity
ranges (based on crack length and nominal stresses)." The influence of residual
stresses must be included in both sets of computations.
Figures 16 and 17 illustrate this definition and show that the "transition"
crack length will depend on load level, stress ratio, notch stress gradient,
material, and size of the part. In these figures, damage per cycle or block is taken
as (1/NI) for the crack initiation 0) curves, and as (~a/W Z~dV) for the
propagation (P) curves, where N[ is the life to failure at the calculated local
stress range, a is crack length, lq is the width of the part, and N is the number of
cycles. In Fig. 16, the indices 1, 2, and 3 indicate three levels of loading, with 1
being the lowest and 3 the highest. For crack propagation, damage increases with
load and crack length; for initiation, damage is proportional to another power of
load (a higher power at long life, a lower power at short life) and usually
decreases with crack length because of the notch stress gradient. For the curves
assumed in Fig. 16, the "transition" crack length is greater for higher loads.

P3 P2 P1

I[[i I 'I,
(a0l 1 (a0l3 CRACK LENGTH (a)

FIG. 16-Schematic of the variation o f crack initiation (1) damage and propagation (P)
damage with crack length and load level for a notched part.

P(R =-1) P(R =0)

uJ
r3
,r
:Z
,<
O

(a0lR = 0 (a0lR =-1 CRACK LENGTH (a)

FIG. 17-Schematic of the variation o f crack initiation (I) damage and propagation (P)
damage with crack length and stress ratio at a given maximum load level for a notched part.

Figure 17 shows how the "transition" length would vary for a given maximum
load at stress ratios of R = -1 (fully-reversed loading) and R = 0 (fluctuating
tension). The propagation damage is only slightly larger at R = -1 than at R = 0.
In this case, the "transition" length would be larger for R = -1 than R = 0
loading.
If a complete, correct theory of fatigue were available, we would not be
concerned with a "transition" from crack initiation to propagation. Infinitesimal
cracks on smooth parts and the local progress of fatigue damage at the tip of
larger cracks would be considered in one complete formulation. Due to our
current ignorance and because of greater convenience, we use two distinct
theories. In the early stages of fatigue, we consider primarily the effect of local
shear stresses at a notch root and compute damage from smooth specimen test
data. In the later stages, we consider mainly the tensile stress field in the vicinity
of a crack tip and compute damage from crack growth rate test data. This
approach has worked quite well here for the bracket and transmission load
histories, using a somewhat arbitrary "transition" crack length of 2.5 mm (0.1
in.). We doubt that calculations with another "transition" length would improve
the results noticeably.
For the suspension history, we agree that improvements in our prediction

methods are desirable. We did recognize the effect of gross yielding in

compression in Method II and thus arrived at better predictions, but additional
work is needed to achieve more reliable predictions when the influence of
compressive loading is significant.

Structural Reliability Prediction

Method Considering Crack Growth
and Residual Strength

REFERENCE: Varanasi, S. R. and Whittaker, I. C., "Structural Reliability Predic-

tion Method Considering Crack Growth and Residual Strength," Fatigue Crack
Growth Under Spectrum Loads, ASTM STP 595, American Society for Testing and
Materials, 1976, pp. 292-305.

ABSTRACT: An analysis method to estimate structural reliability based on crack

growth and residual strength of aircraft structures is presented. The method is based
on linear elastic fracture mechanics theory and allows for the variability of crack
initiation and growth found in the experimental data of various metals. At a
reference stress intensity factor, the central tendency and the variance values of
material crack-growth parameters are determined. Combinations of these parameters
are selected by Monte Carlo simulation techniques, and are used to describe the
characteristically stochastic behavior of crack growth in a material. This description
of material crack-growth behavior is then applied to the typical case of the built-up
skin-stringer configuration of fail-safe type airplane structures to predict the number
and size of cracks in a fleet at any time during its life. Thus, inspection routines may
be established based on realistic fleet performance, to provide suitable levels of
structural reliability for a fleet of airplanes during its operational lifetime.

KEY WORDS: crack propagation, fatigue (materials), predictions, analyzing, residual

stress, aircraft, structural analysis, fleet reliability

The integrity of aircraft structures had been considered, originally, to be a

function only of the demonstrated static strength. Recently, the emphasis has
changed to include fatigue evaluation techniques and damage tolerance methods
as part of the structural integrity task.
Application of these differing technologies often results in inconsistent
requirements which are resolved by arbitrary procedures. To rationalize the
influence of these requirements on design, a study [1,2] 2 was accomplished on
the interaction of static strength, fatigue damage initiation and crack growth,
residual strength, and the environmental load exposure. This paper extends the
analysis model [1,2] by utilizing a realistic airplane structural configuration for
crack growth and residual strength calculations. The resultant analysis scheme
provides a methodology to evaluate or weigh the design and operational

1 Specialist engineers, Stress and Fatigue Research Group, Boeing Commercial Airplane
Company, Seattle, Wash. 98124.
The italic numbers in brackets refer to the list of references appended to this paper.

292
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by International www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions autho
VARANASl AND WHITTAKER ON STRUCTURAL RELIABILITY 293

parameters of a modern aircraft structure. This analysis involves the modeling of

the material and structural characteristics in the presence of a fatigue crack, and
the modeling of procedures for inspecting the structure for possible damage.
Thus, this modeling of the complex interaction of material characteristics,
structural design configuration, operational load exposure, and operational
inspection provides a measure of the reliability of a structure at any time.
The results of some exploratory parametric studies using this reliability
analysis system are presented to illustrate some of these interactions.

Analytical Development
This analysis model presumes that a fatigue crack will initiate in some critical
piece of structure due to a spectrum of loads through service usage. The crack
will propagate at a rate dependent on the material, structural geometry, and
applied loads, until it is either detected during an inspection, or is arrested by
some design feature, or the structure fails.
Thus, the essential elements of a structural reliability analysis system are:
1. A crack initiation and growth model to calculate the time to crack
initiation and subsequent growth to a critical size for a realistic structural
configuration.
2. A residual strength model to calculate the strength of the structure in the
presence of a crack of a given length.
3. A structural inspection and crack detection model which calculates the
probability of detecting a crack and repairing it during a scheduled
inspection.
4. A loading model which calculates the probability of a load, which is greater
than or equal to the strength of the structure, occurring within some
particular time.
The residual strength of structure diminishes as the crack grows, but the
chance of detection improves with increase in crack size. This dependency of
residual strength and inspection functions on crack size places considerable
emphasis on the utilization of a crack-growth model in the reliability analysis.

Crack Growth Model

A model of crack development must consider crack initiation and growth to
critical size. The time to initiation will be different for each structure, even
though nominally identical structures of the same material are subjected to the
same applied loads. Despite the use of very sensitive detection methods, the
fatigue crack is known to have proceeded through an incubation phase and a
physically ascertainable growth even before it reaches detectable size. Thus, a
practical assumption of initial crack size includes the incubation and stress
intensity threshold crack-growth phases. In this paper a crack is said to be
initiated whenever it reaches a size which can be detected.
Following the lead of Refs 3 and 4, it is assumed the crack initiation life to

detectable size is a random variable No with an extreme value distribution,

namely, a two-parameter Weibull distribution with given characteristic life 13and
shape parameter t~
No ~ w(a,~)
Linear fracture mechanics theory is applied in the crack-growth phase
following initiation. For the sake of simplicity, a constant-amplitude crack-
growth model relating the crack-growth rate, ~a/~uV, to the stress intensity
factor, K, by a power law is assumed
Aa/~V= A K n (1)
In Eq 1, the proportionality constant,A, and the exponent, n, are determined
from crack-growth tests of simple specimens of a material. Test data indicates
that the values of A and n exhibit variability between identical specimens of the
same material. For convenience, the crack-growth Eq 1 is expressed as
log Aa/&N = log A + n log K (2)
Literature [5,6] provides information on typical observed values for the
material constants in Eq 2. Figure 1 illustrates the variability of observed values

10 6 - -

SHEET THICKNESS - 0 - 0 . 3 2 Cm
GRAIN DIRECTION L
STRESS R A T I O , R 0.01 - 0.10
10 5 - C Y C L I N G SPEED 10 - 30 Hz
ENVIRONMENT ROOM A I R

9(J 10 4

d 10 $

10 2

1 -

I I I
S 10 $0

MAXIMUM STRESS INTENSITY FACTOR, K m x , ( l U C / m 3 / 2 )

FIG. 1-Fatigue crack growth behavior of 2024-T3 bare aluminum alloy sheet.

in similar specimens for the two parameters-the slope, n, and the intercept, log
A. In order to determine empirically the stochastic relationship between these
two parameters, it may be observed that Fig. 1 shows the representative data in
the form
y = d + nx (3)
where
y = log &a/AN, x = log K, and d = log A
If K R denotes a material reference stress intensity factor, and g = log K R, it
may be seen from Fig. 1 that, about this reference value, there is a variation in
the slope, that is, the exponent, n, and a small varaiation in the growth rate. This
is illustrated schematically in Fig. 2. This variation in slope and growth rate may

Y
~1 = MEAN REFERENCE C R A C K i* " /
" GRQWTNRATE ~///
"2 = ~EA. SLOPE /.~I
El/jilt/!
K =-O,AR, Tm, OF REFERENCE /."//
STRESs INTENSITY FACTOR /,#1~/ /

Ii I
!

I D-X
g

FIG. 2-Schematic of crack-growth simulation.

be expressed, using capital letters to denote random variables, as

y = D + A(x - g ) (4)
where D and A are jointly determined once measurements are made on a test
specimen. Over the population of such specimens, these values would have a
distribution which is presently unknown. Presuming this stochastic variability to

be the result of multitudinal influences makes the assumption of normal

distribution plausible. These are specified as
D ~ N (//1, al 2)
A "" N (//2,022)
Here the parameters governing the distributions of D and A, namely the mean,
~, and the variance, e z, are chosen to simulate the crack-growth behavior
observed experimentally. Use of Eqs 1 and 4 results in

A = 10~ (5)
rR A
n=A
which define the material crack-growth behavior completely.
This description of material crack-growth behavior is then applied to the
typical case of the built-up skin stringer configuration of a fail-safe type airplane
structure. This is done by considering a stiffened sheet with riveted and
uniformly-spaced stringers, containing a symmetrical crack centered at a stringer.
The stiffened sheet is subjected to a remote applied stress, S, normal to the
length of the crack (Fig. 3). The stress factor for the stiffened sheet is

tl t J-

J
S

t F---2~ =,
FIG. 3 - S t i f f e n e d sheet with uniformly-spaced stringers and a stringer-centered crack.

calculated by superimposing stress intensity factors for the fastener forces and
the remotely applied stress [7,8]. The effects of parameters such as stiffener
spacing and stiffening amount, fastener spacing, size and flexibility, and crack
length on the stress intensity factor of a stiffened sheet is given conveniently by
a stress intensity correction factor % This factor is defined to be the ratio of
stress intensity factors of the cracked sheet with and without the stringers.
Thus, with the generation of three random variables No, D, and A from their
respective distributions and the calculation of the structural stress intensity
factor, it is possible to compute the crack growth simulating one structure's life.

Residual Strength Model

It is presumed that the strength of a structure remains constant until the
initiation of a fatigue crack. As the crack propagates, the residual strength of the
cracked component diminishes until the structure fails under the applied load.
Current structural design philosophy emphasizes the fail-safe approach, where
materials with slow crack-growth rates and good fracture properties are often
used in conjunction with positive crack stoppers. Crack stoppers provide a means
of arresting the crack at some predetermined fail-safe length, say a,. Thus, the
fail-safe length defines the lower limit of the residual strength (fail-safe residual
strength). This fail-safe residual strength is a design constraint; it is usually
substantiated by test or analysis or both; and it is available, therefore, as an
input parameter for this analysis procedure.
There is a critical value of the stress intensity factor, usually labeled K c for
plane stress conditions, which defines the critical crack length, ac, for a specified
loading condition. This length is essentially a material parameter and for a
single-element, monolithic type structure, it defines the fall-safe length. In the
case of built-up, skin-stiffener type airplane structure, the critical crack length is
generally considered to be only a function of the skin material and load.
However, the fail-safe length is also a function of the structural geometry, the
location of tear-straps, or other similar crack stopping devices. Using considera-
tions such as these, the following residual strength model is proposed.
Let L(a) be the residual strength of some structure containing a fatigue crack
of length a. The original strength of the structure is assumed to remain constant
until the initiation of a crack of length ao.
Therefore let

L(a) = 6 u for a _<ao (6)

L(a) = 6 for a = min (a c, al)
where 6 u is the ultimate strength parameter, 6 is the fail-safe residual strength
parameter, a c is material critical crack length, and a 1 is fail-safe length. These
strength design parameters are usually related to the design limit strength of the
structure, 5s as follows
0<6<~s u

Furthermore, on current fail-safe airplane structures, typically: ~u = 1.5 ~s

Taking into account the foregoing definitions, the residual strength of a
structure containing a fatigue crack of length a, is given by

Structural Refiability Model

Typical airplane operational procedures require the periodic examination of
the structure as part of its structural integrity program. In the event that any
damage is detected, the structure is repaired. Thus, the reliability of an airplane
fleet, as a function of time, can be maintained at a high level. These operational
variables of structural inspection and damage detection are considered, together
with the structural performance variables of life to damage initiation, crack-
growth, and corresponding residual strength degradation, in a developed
reliability analysis system [1,2]. For the sake of conciseness, in this paper only
the essential elements of this reliability analysis system are outlined as follows.
Reliability of a structure is defined to be one minus the probability of failure
of a structure subjected to the imposed conditions of fatigue crack initiation and
growth to critical size under a cumulative loading environment, periodic
inspections, and repairs. Fleet reliability is defined to be the reliability of the
weakest structure in the fleet and is given by the product of the individual
reliabilities of structures in the fleet.
The probability of failure of a structure can be simply described as follows:

[Probability of I [Probability of
Probability oft lsurvival of a ~ Idetection and repair
failure of a = 1 - lstructure when ! - / of a structure before
structure |a crack is [ |failure
~undetected ] /
In other words, at anytime during a service interval between inspection periods,
a structure has either failed or survived. Structural survival is due to one or the
other of the following separate situations.
1. The structure survives because the strength of the structure in the presence
of an undetected crack exceeds the imposed loads. The probability of
structural survival in this situation is determined from load occurrence
probability and residual strength of a structure.
2. The structure survives because the crack is detected and repaired before
failure. The probability of structural survival in this situation is determined
from a crack detection model which takes into account the variables such
as crack length and quality of inspection.

Results
The results of some exploratory applications of the reliability analysis system
are discussed in this section. Constant-amplitude, crack-growth test data are

obtained for 2024-T3 bare aluminum alloy sheet specimens. For each specimen,
a least square fit of a straight line relationship in log-log scale between the stress
intensity factor and the crack-growth rate is obtained from the experimental
crack size versus cyclic life data (Fig. 1). These fitted data are used to calculate
the central tendency and variance values for the slope and intercept parameters
of the crack-growth model described earlier. Monte Carlo simulation techniques
have been utilized to select random combinations of these two parameters. Each
combination of the parameters defines a simulated crack-growth relation for the
material. Figure 4 shows a single set of simulated data and the bounds from

106 CRACK GROWTHPARAMETERS:

"~
01 = 0.03
////
105
0 2 = 0.6 //SIMULATION

Io
104
KR =17,6 MN/m3/2

//
N R,b~DO~SIIIULATIONS
~ 103

i 102
RANDOM SIMULATIONS

/
10

I I I
S 10 50

MAXIMUMSTRESS INTENSITY FACTOR,Kmux, ( JJN/m3/2)

FIG. 4-Simulation o[2024-T3 bare sheet crack-growth rate behavior.

2000 such simulations, along with the central tendency and variance values for
the two crack-growth parameters used in this simulation. A comparison of the
results of Fig. 4 with those of Fig. 1 demonstrates quite clearly the realism of
the crack-growth model.
The structural representation used in this analysis for a typical case of the
built-up skin-stringer configuration of a fail-safe type airplane is that of a
stiffened sheet with uniformly-spaced stringers, containing a crack centered at a
stringer (Fig. 3). Typical values of the structural parameters such as stiffener
size, spacing and stiffening amount, fastener spacing, and size are used in this

example to calculate the stress intensity factor of this stiffened sheet. The
fasteners attaching stringers to the sheet are assumed to be rigid for this analysis.
However, provision is made in the analysis to include the effect of fastener
flexibility on the stress intensity factor of a stiffened sheet. The results are
shown in Fig. 5 where the stress intensity correction factor, 7, is plotted against

--t,,1"
: 34.9 (PERCENT STIFFEHIHG)
I.O -.P-: 0.18

d~.= 4.0
o.e

0.6

Vl~
0,2

I I I I
0.6 1.0 1.5 2.0
3EMI 9 CRACK LENGTH 0
STRIHGER SPACIHG W

FIG. 5-Stress intensity correction factor for a stiffened sheet.

nondimensionalized crack length, for a symmetrical crack centered at a stringer.

It can be seen from this figure that the effect of stringers is to lower the stress
intensity factor for a stiffened sheet compared to an unstiffened sheet. The
fail-safe crack length assumed here is 35.5 cm (a crack length spanning
two-bays).
The interaction between the time to initiation of a detectable crack, the
crack-growth, and residual strength models is illustrated in Fig. 6. The results
shown here have been based on the assumptions that the characteristic life to a
fatigue crack (0.25 cm in length) in a critical detail in the structure is 60 000
ground-air-ground cycles; the threshold size for crack detection is 0.08 cm; the
maximum distance between positive crack stoppers is 35.5 cm; and the fail-safe
residual strength of the structure containing a crack of this magnitude is 80
percent of the limit strength. It is further presumed that the structural material
is 2024-T3 bare aluminum alloy with crack-growth parameters from Fig. 4 and a
critical stress intensity factor, Kc, of 77 MN/m 3/2 . Finally, the applied loads
data are based on recorded velocity-acceleration-altitude (VGH) data from large

I zSo ~.
|

\,,,/
125 ~-
J

I00
sT. o..

75 ~
I ~ I / " I
15 30 45 60
GROUND-AIR- GROUHDCYCLES ( xl03)

FIG. 6-A simulation of structural crack growth and residual strength.

jet transport airplanes [2]. Power spectral density techniques are used with these
data to develop structural loading histories, both operational and extreme. The
operational loads are used in the determination of the characteristic life to
initiation, /~, as well as the stress intensity factor for crack-growth. For the
examples presented, the operational stress variation of 0-137.9 MN/m 2 is used.
The extreme loads are used to determine the probability of an overload
occurrence in the calculation of the probability of structural survival. Figure 6
shows an example of the crack propagation and residual strength values given by
a single random simulation based on the hypothetical parameters just defined.
This particular example was one of the more extreme cases, that is, early crack
initiation life, selected from the very many simulations performed at each
computation. It demonstrates how the life to detectable crack can be low
despite the initial assumption of the 60 000 cycles characteristic life. This figure
also helps to illustrate the cycle-dependent nature of both crack length and
residual strength, with the former an increasing function and the latter a
decreasing one.
Several parametric studies based on the previously discussed models in
conjunction with the reliability analysis system [1,2] have been conducted, and
some of the results are now presented to illustrate the scope of application. The
baseline example case assumes a fleet of 300 structures of which one-quarter,
namely, 75 structures, are examined at each of the inspection periods, which are
scheduled at 7500 ground-air-ground cycle intervals, and repaired when found to
be cracked.
The impact of characteristic initiation life on fleet reliability is shown in Fig. 7
where fleet reliability is plotted as a function of ground-air-ground cycles for
three assumed characteristic initiation lives (fl = 60 000, 30 000, and 1000) for

a fail-safe type aircraft with a fail-safe residual strength equal to 80 percent of

the limit strength. As would be expected, Fig. 7 shows that the fleet reliability
decreases as the characteristic initiation life decreases. It may also be noted that

O, 999i

O. 995

o
fl
U4
N 0,995

r
O. 99
CHAR CTERIST
,.I
\ CymEs Y0 A C,A
SPECIFIC DETECTASkE $ 1 l E z
O, \\
6QOa n CYCLES -~ \
O.98 9 30000 CYCLES
I000 CYCLES

,\ /
0.95

0.90 I I i -"~~ i
1S 30 45 60
GROUND-AIR - GROUND CYCLES ( ,clO J)

FIG. 7-Effect o f initiation life on fleet reliability.

the fleet reliability does not drop to zero even with a low characteristic initiation
life of 1000 cycles. This is due to the fact that although a majority of the
airplanes in the fleet now develop cracks, the probability of detection and
strength renewal at scheduled inspections is also higher. Furthermore, the
residual strength of the cracked structure remains high due to fail-safe design.
The effect of fail-safe residual strength constraint on fleet reliability is shown
in Figs. 8 and 9. Fail-safe residual strength is the strength of the structure when
it contains a crack of fail-safe size (see prior discussion). Single load path,
monolithic structures, without a similar provision for arresting a growing crack,
are termed safe-life structures. These safe-life structures are not considered to
possess any significant residual strength when fatigue crack reaches critical
proportions. In this analysis, fail-safe type structures possess residual strength
which is 80 percent of its limit strength. This represents a residual strength level
which is common in current fail-safe airplane structural design. In contrast, the
safe-life structure has been assumed to possess a residual strength which is 10
percent of its limit strength. The reliability of a safe-life structure designed to

0.999

0.998 ~ FAIL- SAFE TYPE

~.,~ STRUCTURE
\ ~, ('c~,,cx iw#rlArmN,r
\ "~,.. 6o, ooo G-A-C cYCI.ESJ-'7
\ \'--.. /
0.995 SAFE-LIFE TYPE ~ '~." ~ . %% /

fcxAcx imrlArlo~ Ar ~ \ ~ ' ~ /

iN,it,rE, o-A-~c,sj~ "l ~ /
--z-- ', ,. ',-"-<
O. 99

IIO, OOO | I l O , IO0 ~ 210, OlO

0.90
9 I t l

I I l
I I
I I
I I
F,.L-S*FE TYPE ~. I
STRUC'ruRE ~,~1 : I
0.95 cxacx ~mr#Ar~oH,r ~. ' ~ , i I
'~176~176 I I
I I ~ I
I I " ~ I
O. 90 I I I i I "'~ I
15 30 45 60
GROUND- AIR - GROUND CYCLES (x 103)

FIG. 8-Effect of residual strength on fleet reliability.

the same characteristic life to initiation as the fail-safe structure (60 000
ground-air-ground cycles) is presented in Fig. 8. To facilitate the comparison,
some of the results for the fail-safe structure shown in Fig. 7 are repeated. It is
clear from Fig. 8 that the reliability of the safe-life structure is much less than
the equivalent fail-safe structure. However, safe-life reliability can be improved
by increasing the characteristic life to initiation as illustrated.
The importance of the crack detection and residual strength parameters is
clearly illustrated in Fig. 9. Here, all the cases illustrated are identical except for
the particular variable noted against each case. A comparison has also been made
between assumption of a stochastic behavior for crack-growth and the
application of a discrete (average) crack-growth rate. It can be seen that in this
example the fleet reliability based on the deterministic crack-growth behavior is
optimistic. However, it is also clear that for the particular set of parameters
assumed for these examples, the difference between the results of stochastic and
discrete crack-growth rate assumptions is not large. It must be realized that as
the input parameters vary, the influence of the stochastic crack-growth behavior
on structural reliability can increase or diminish. For example, an assumed large
value of characteristic life to initiation tends to reduce the importance of the
crack-growth phase. As a result, the effect of crack-growth variability on
structural reliability will be minimal.

ALL CASES: CRACK INITIATION CHARACTERISTIC

Cl FE = 60,000 G-A-G CYCLES
0.999

0.998
,\ ,.SPE D .A.REO
\ " % ST
.A,L SAFECTUR
Y
.PE
\ ".~. DISCRETECRACK

0.995 \ .~ ---~./. oe CRACK~RO~. -

/
0.99

,, \
0.98
INSPECTED AND REPAIRED ~ ~
;AFE-LIFE TYPE STRUCTURE~

I k UNINSPECTED
\ FAIL- SAFE TYPE

\/
0.95 STRUCTURE

,,
!

0.90 I II I ~ I
1S 30 45 60
GROUND-AIR- GROUNDCYCLES ( 9 103)

FIG. 9-Effect of major parameters on fleet reliability.

These few examples have been presented to illustrate how the crack initiation,
growth, detection, and residual strength models, discussed earlier, can be applied
in an analysis to determine the structural reliability of a fleet of airplanes.
Although these examples have been presented against a quantitative background
for clarity, their main intent is to demonstrate qualitatively the impact and
interaction of the various elements on structural reliability.

Conclusions
The development of an interdisciplinary analysis system for calculating the
structural reliability of a fleet of airplanes during its operational life has been
discussed. This system has been applied to hypothetical, but plausible, examples
to illustrate the impact of a few of the major elements on structural reliability.
With the assumption of a reliability criterion, this system will allow an optimum
selection of parameters which define structural material and configuration, crack
initiation, crack-growth, and inspection frequency. Although the initial applica-
tion of this system has been to airplane structures, it is anticipated that it can be
extended with minor modifications to other structures which have similar design
and operational characteristics.

References
[1 ] Whittaker, I. C. and Sannders, S. C., "Application of Reliability Analysis to Aircraft
Structures Subject to Fatigue Crack Growth and Periodic Structural Inspection,"
AFML-TR-73-92, Air Force Materials Laboratory, Wright-Patterson AFB, Dayton,
Ohio, 1973.
[2] Whittaker, I. C., and Saunders, S. C., "Exploratory Development on Applicationof
Reliability Analysis to Aircraft Structures Considering Interaction of Cumulative
Fatigue Damage and Ultimate Strength," AFML-TR-72-283, Air Force Materials
Laboratory, Wright-Patterson AFB, Dayton, Ohio, 1973.
[3] Freudenthal, A. M., "Reliability Analysis Based on Time to the First Failure," 5th
ICAF Symposium, Aircraft Fatigue Design and Operational Aspects, Melbourne,
Australia, May 1967.
[4] Whittaker, I. C. and Besuner, P. M., "A Reliability Analysis Approach to Fatigue
Life Variability of Aircraft Structures," AFML-TR-69-65, Air Force Materials
Laboratory, Wright-Patterson AFB, Dayton, Ohio, 1969.
[5] Donaldson, D. R. and Anderson, W. E., "Crack Propagation Behavior of Some
Airframe Materials," Proceedings, Crack Propagation Symposium, Cranfield,
England, Sept. 1961.
[6] McEvily,A. J., Jr., and Illg, W, "The Rate of Crack Propagation in Two Aluminum
Alloys," NACA Technical Note 4394, Sept. 1958.
[7] Poe, C. C., Jr., "Stress Intensity Factor for a Cracked Sheet with Riveted and
Uniformly Spaced Stringers," NASA TR R-358, National Aeronautics and Space
Administration, Washington, D. C., 1971.
18] Varanasi,S. R. and Carlson, C. M., "CRISP-A Computer Program for the Analysis
of Cracks in Stiffened Panels," The Boeing Commercial Airplane Company
Document D6-41314, Feb. 1974.

Spectrum Crack Growth Prediction

Method Based on Crack Surface
Displacement and Contact Analyses

REFERENCE: Dill, H. D. and Saff, C. R., "Spectrum Crack Growth Prediction

Method Based on Crack Surface Displacement and Contact Analyses," Fatigue Crack
Growth Under Spectrum Loads, ASTM STP 595, American Society for Testing and
Materials, 1976, pp. 306-319.

ABSTRACT: A method for prediction of crack growth behavior has been developed,
based on evaluations of stress intensity caused by crack surface contact. The
potential interference of the crack surfaces is determined from analyses of elastic
displacements during loading and unloading and of the permanent deformation left in
the wake of a growing crack. The potential interference is treated as a wedge acting
behind the crack tip and the contact stresses created by this wedge are computed
through an elastic-plastic analysis. The effective stress intensity range used for crack
growth prediction is found by subtracting the stress intensity caused by these contact
stresses from the applied stress intensity range. Comparisons of crack growth
behavior predicted by this method and that measured in constant amplitude tests,
with and without high loads, and in block spectrum tests have shown that the
method accounts for load interaction effects in these cases. These effects include
delayed retardation following high loads, crack growth acceleration during high loads,
and dependence of growth rates on number of high loads.

KEY WORDS: crack propagation, fatigue (materials), stress analysis, predictions,

loads (forces)

A method for prediction of crack growth behavior has been developed, based
on analysis o f crack surface contact stresses. Several experimental investigations
[1-8] 2 have indicated the existence o f crack surface contact prior to complete
unloading o f the cracked material. Crack surface contact increases the minimum
stress intensity range effective for propagating the crack. The concept o f an
effective stress intensity range influenced b y crack surface contact is attractive as
a basis for spectrum crack growth prediction since it qualitatively explains many
observed load interaction effects. These effects include delayed retardation
following high loads, crack growth acceleration during high loads, and
dependence o f growth rates on number of high loads.
This method o f crack growth analysis is based on evaluations o f stress
intensity caused by crack surface contact. An analysis of crack surface

1 Project strength engineer and strength engineer, respectively, McDonnell Aircraft

Company, St. Louis, Mo. 63166.
2 The italic numbers in brackets refer to the list of references appended to this paper.
306
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorize
DILL AND SAFF ON SPECTRUMCRACK GROWTH PREDICTION 307

displacements during loading and unloading is used to determine the permanent

plastic deformation left in the wake of the growing crack. The potential
interference of the crack surfaces is determined from an analysis of the
interference free displacement remaining at minimum load, in conjunction with
the analysis of the permanent deformations. The contact stresses behind the
crack tip are found by treating the potential interference as a wedge between the
surfaces and performing an elastic-plastic analysis of the stresses caused by this
wedge. The effective stress intensity range used for crack growth prediction is
found by subtracting the stress intensity caused by these contact stresses from
the applied stress intensity range. Comparisons of crack growth behavior
predicted by this method and that measured in constant amplitude tests, with
and without high loads, and in block spectrum tests have shown that the method
accounts for load interaction effects in these cases.

Crack Surface Displacement Analyses

An analysis method has been developed to determine displacements along the
crack surfaces caused by loading and unloading during a single load cycle. The
displacement analysis is based upon the Dugdale model [9] of the plastic zone,
treating the plastic zone as an extension of the elastic crack surface over which a
constant yield stress acts. The plastic zone size is determined to be that length of
constant yield stress such that the stress intensity caused by the yield stress
equals the applied stress intensity; hence there is no singularity at the extended
tip. The crack surface displacement at Kmax is found from the superposition of
the Westergaard solution [10] for elastic displacement near the crack tip and the
displacement due to the constant yield stress of the plastic zone [11,12]. The
Westergaard solution for displacements under remote loading is shown in the
upper portion of Fig. 1. The constant yield stress of the plastic zone tends to
reduce the Westergaard displacement by the amount shown in the middle
portion of Fig. 1. The resultant displacement is shown in the lower portion.
The displacements during unloading are similarly shown in Fig. 2. During
unloading, the reversed plastic zone stress is increased to twice the yield stress
used during loading, accounting for an elastic stress range equal to the difference
of tensile and compressive yield stresses.
The crack surface displacements at the minimum applied load, shown in Fig.
3, are found by subtracting the displacements occurring during unloading from
those at maximum load. Even when the minimum load is zero, the crack surface
remains wedged open by the stress distributions within the plastic zone, when
permanent plastic deformations caused by prior crack growth are not
considered.
To qualitatively understand the residual plastic deformation left in the wake
of a growing crack, consider a ligament of material ahead of the crack tip being
traversed by a crack as in Fig. 4. The ligament accrues plastic deformation as it is
encompassed by the plastic zone of the crack. As the crack tip passes through
the ligament, the plastic deformation remains as a displacement of the crack

Y
Plastic
Zor~
Elastic DisplacementsDueto
PlasticZone YieldStress.~f o - ' ~
'

'-max Yield 5tress,t O

bKmax = ~rernote+~fo

MaximumCrackOpening; ~
Displacement'CO~ ~ j

FIG. 1-Elastic crack surface displacements at maximum load.

surface due to permanent extension of the ligament. This is referred to as the

plastic crack surface in Fig. 4.
The bottom portion of Fig. 4 shows that the elastic surface is propped apart
by the plastic deformation ahead of the crack tip. However, the extended plastic
crack surfaces will interfere upon unloading. The interference is visualized as
though the surfaces were allowed to pass through each other. The potential
interference of either surface is the displacement of that surface past the
centerline of the crack.
In modeling the residual deformation and surface interference, the plastic
deformation at the crack tip at minimum load is considered to be equal to the
crack opening displacement (COD) at that load [I1,13]. This crack opening
displacement, as obtained from Rice [11], is approximated as

CODKmin = a (K2max _ 12 A K 2 )/2Efo (1)

where a = 1 for plane stress, a = (1 - v~)/2 for plane strain, and fo = ays =
yield stress.
This deformation exists just behind the crack tip as well since material

Elastic DisplacementsDueto
PlasticZ o n e r [ [[
Plastic Zone Yield Strels, 6[o-~ ] I
\ I 1

/--Net DisplacementDuring
Unloading 6A K

~ K = ~'ramote+ 6'fo

ChangeinCrackOpening
l
Displacement, ACODll ",,j
iI

FIG. 2-Elastic crack surface displacements during unloading.

adjacent to the crack tip, both ahead and behind, is compressed to this value on
unloading. Correlation with constant-amplitude closure stress intensity data
from Elber [1] shows that a closer approximation to the permanent plastic
deformation is
2
residual = O~ (gma x - 0.4 AK 2)/2Efo (2)

The potential interference is the difference between the permanent plastic

deformation and the minimum displacement of the elastic surface as shown in
Fig. 5.

Contact Stress Analyses

The potential interference acts as a wedge behind the crack tip, creating a
stress intensity at the crack tip. To determine the stresses behind the tip caused
by this wedge, a simple contact stress model of closure was developed. This

Load,~Kmax"-/

Crack-
PhysicTal i p - - ~

L L---_Jp~stie
Zone

Unloading,~AK --J ~ 1

~Kmin= ~iKmax" ~AK I1~-PlastLc

I Zone
Net Displacementat I

CODKmin= CODKmax_ A
Plastic
Zone
FIG. 3-Elastic crack surface displacements at minimum load.

model, symbolized in Fig. 6, uses 25 constant stress elements to idealize the

wedge. Bueckner's weight function approach [14] was used to develop an
influence coefficient matrix for the displacement-stress relationship between
elements. The analysis is iterative so that a solution is determined wherein the
maximum contact stress is limited to the yield stress and there is no tensile
contact stress. A typical example of this analysis is summarized on Fig. 7
showing crack surface displacements and stresses following 0.635 mm of growth
after a single overload. This plot shows the analytical displacements at Kma x and
Kmin, permanent plastic deformation, potential interference, and contact
stresses behind the crack tip.
The stresses determined from interference are used to compute the contact
stress intensity occurring at the minimum load, the effective minimum stress
intensity, and the effective stress intensity range.
Figure 8 shows a comparison of the results obtained using this model and
Elber's test data [1].

Elasttc / - - Ligament
CrackSurface--~ ~J="~ r~T LigamentStress

Elongation
LJ
SuCrrfaCke
J

LigamentStress

Elongation

A~=PlasticElongation---~

Elongation

o~ ~o~176 T ~
inter,..... X C ~(~ ~ LigamentSt....

/ Elongation

FIG. 4-Crack closure phenomenon caused by interference of crack surfaces.

PermanentPlastic
Deformation,5p ~ ' b

Physical
CrackTip

NetDisplacementat MinimumLoad,~Kmin

bwedge =ISp-5Kmin

PotentialInterference,~wedge

FIG. 5-Potential interference at minimum load under constant-amplitude cycling.

f Constant 0 2-*-~
Stress 0.3" ~

Computaton--~ t / / ! - ~ ~,.~\ \ ~--0.uo

~ . ~,1---,ypma, =1.... t \'~,~/-0.05
Ax/~
1.4
J 1.3 I 1.2 I 1.1
I"0
I 10 I 0.9 I 0.8 I 0.7 10.610.51
= I'~ I,~,v,=
I~

x/COtotaI = 10.5 CO I
,~ - Reference Length I
9 25 x 25 influence Coefficient Matrix Derwed Through Bueckner's Weight Function Approach
9 Maximum Stress Limited to Yield
9 No Tensile Stresses
9 Displacements and Stress Intensity Computed for Input Values of Kma x, Kapp, E, and f.

FIG. 6-Contact stress model o f closure.

16
Maxlmuro Elastic Oisplacement,~rnex
J
I l j J
14
f

12
J
J
10
/
[8 / Constant Amplitu Loading
AK = 16.5 MN/m~2
R = 0,05 3/2

~ 6 / Highload Khmax = 24,7 MN/m

Minimum Elastic Displacement,~min

f
\
~ otential nterferen e,~wed
-4

-6
400
JiifT
300

"
8
100
/
J
0
0 0.2 0.4 0,6 0.8 1.0 1.2 1,4 1.6 1.8 2.0
Distance from Crack Tip - mm

FIG. 7-Crack surface displacements and stresses O.635 mm after a single high load.

1.0

0.8

0.6
I
AKef f = Kmllx - Kclolur e

<1 0.4
! .f L ) t~.
0.2

DiIplacm,nen t

I 1
0.2 0.4 0.6 0.8 1.0
Stress Ratio

FIG. 8-Comparison of crack growth models in constant-amplitude loading.

Constant-Amplitude Crack Growth

Input to the growth analyses includes the sigmoidal da/dN-AK curve for a
stress ratio, R, of zero. This curve is subsequently modified to be a da/dN-AKef f
curve, through results of analyses of closure as typified by Fig. 8. Crack growth
for R ratios other than zero are determined using the/~r~eff appropriate for that
R ratio. To estimate the crack growth rate increase which occurs when K m a x
approaches Kc, a factor magnifying the crack growth rate is used

dN _ Kmax (3)

This magnification is similar to the modification made by Forman [15].

The usefulness of the contact stress model of crack closure to explain the
effects of stress ratio is indicated in Fig. 9. This figure shows the correlation of
analytical constant-amplitude crack growth and test results reported by Schijve,
Brock, and de Rijk [16]. The crack growth curves shown are for stress ratios of
0.202, 0.428, and 0.546, Curves 1, 2, and 3, respectively. The large differences
in constant-amplitude growth rates shown in Fig. 9 are due to testing at different
stress ranges, with mean stress held constant and stress amplitude varied between
tests, rather than solely due to stress ratio effects. Currently, the model treats
negative stress ratios as R = 0 with AK = Kmax-

Crack Growth Following High Loads

Figure 10 shows comparisons of predicted crack growth rates using the
contact stress model and test results for constant-amplitude cycling with
multiple high loads. The data of Trebules, Roberts, and Hertzberg [3] were
generated from thin compact tension specimens of 2024-T3 aluminum. These
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
314 FATIGUE CRACK GROWTH

Constant Amphtude

Spectrum Truncat=on
,k
[
1 | | / |
II I

J,/
Ii
' li
E
E 20
=.
==
=,

/',/// ,g
Legend
1
m Test, Ref. 16
m Contact StressModel
(~) R = O.20
(~) R = 0.43
(~) R = 0.55
3 StressAmplitudes
(~) 4 StressAmplitudes
( ~ 5 StressAmphtudes

0 50 100 150 200

I
25O 300
Cycle x tO 3

FIG. 9-Comparison of analysis with constant-amplitude and spectrum truncation tests.

comparisons show good correlation with the delay recorded in test. The analysis
predicts a longer delay in retardation following a single high load than that
recorded in test. The minimum growth rate predicted agrees well with the
empirical curves. For multiple high loads the agreement between analysis and
theory is very good for both delay and minimum growth rate.
As indicated in Fig. 10, current crack propagation models do not explain these
phenomena very well. Wheeler's model [17] is based on plastic zone size
interaction and can be "tuned," through a retardation exponent, to match
results for a particular spectrum and specimen geometry. However, the model
can not vary retardation with the number of high load cycles and does not
predict delayed retardation.
T. R. Porter [18] has reported test results for eight variations in the number
of high loads repeatedly applied during constant-amplitude testing. These tests

Ra 0.1

R a = Crack Growth Rate Rath

da/dN
da/dN (Constant Am, rude}

0.01

V
L
9~ Wheeler Model
~ ~ Contact Strew Model
Test Ref 3

0.001 t I I I
0 0.25 0.50 0.75 0 0.25 0.50 0,75 0 0.25 0,50 0,75
Crack G r o w t h A f t e r High Loads - ram

FIG. l O-Comparison of measured and predicted crackgrowth following high loads.

were performed on center cracked panels of 7075-T6 aluminum. Crack growth

analyses using the contact stress model are shown with the test data in Fig. 11.
Correlation of analysis and test results is consistently good, indicating the ability
of the model to account for the effects of high loads on constant-amplitude
crack growth.

Blocked Spectrum Crack Growth

To determine the ability of the analysis technique to account for the effects of
variations in blocked spectra on crack growth, the test results reported by
Schijve, Brock, and de Rijk [16] were selected. Three test series were performed
using a base spectrum block having three stress amplitudes applied with a
constant mean stress, Curve 4, Fig. 9. The two additional spectra were generated
by adding first a lower stress amplitude, Curve 5, and then a higher stress
amplitude, Curve 6. The test specimens consisted of center cracked panels of
2024-T3 clad aluminum with a 4-ram diameter hole and 1-mm notches used as a
crack starter.
Correlation of the contact stress model analysis with test results is presented in
Fig. 9. Also shown is correlation of the analysis with constant-amplitude results.
Note that the analytical curves for both constant amplitude and spectrum crack
growth were generated using the contact stress model and the same da/dN curve.
The correlation shown in Fig. 9 indicates the usefulness of this analysis
technique for prediction of spectrum crack growth.

Conclusions
A method for prediction of spectrum crack growth behavior has been
developed based on analysis of crack surface contact stresses. Based on the

Constant AmDlitucle High Load

Curve Cycles/Block Cyclw~/Block

=--~-- | (9 o 5O
5O
25
m ~ Contact Streu Model ~ 5o 0

~o__~ 5O
Telt, Ref 18 10
| 6o 6
3
5O I

| | | |

|
/

EE
/l'IIF /
//
I

-i //

//
/,
/
_ I
!

,I
30
1/
=: i
NI/ ! /

~ , i/j/ !
/ /
/

o I
0 20 40 60 80 I O0 120 140 160 180 200 220
Load Cycles x 103

FIG. ll-Comparison of analysis and constant-amplitude tests with multiple high loads
repeatedly applied.

correlatiQn between test and analysis presented herein, the method appears to be
capable of explaining:
1. stress ratio effects on constant-amplitude crack growth,
2. effects of single and multiple high loads on constant-amplitude crack
growth, and
3. effects of truncation and other variations of blocked spectra on fatigue
crack growth.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized
DISCUSSION ON SPECTRUM CRACK GROWTH PREDICTION 317

References
[1 ] Elber, W. in Damage Tolerance in Aircraft Structures, ASTM STP 486, American
Society for Testing and Materials, 1971, pp. 230-242.
[2] yon Euw, E. F. J., "Effect of Overload Cycle(s) on Subsequent Fatigue Crack
Propagation in 2024-T3 Aluminum Alloy," Ph.D. thesis, Lehigh University,
Bethlehem, Pa., 1971.
[3] Trebules, V. W., Roberts, R., and Hertzberg, R. W. in Progress in Flaw Growth and
Fracture Toughness Testing, ASTM STP 536, American Society for Testing and
Materials, 1973, pp. 115-146.
[4[ McMillan, J. C. and Pelloux, R. M. N. in Fatigue Crack Propagation, ASTM STP
415, American Society for Testing and Materials, 1967, pp. 505--535.
[5] Probst, E. P. and Hillberry, B. M., "Fatigue Crack Delay and Arrest Due to Single
Peak Tensile Overloads," AIAA Paper No. 73-325, AIAA Dynamics Specialists
Conference, Williamsburg, Va., 19-20 March 1973.
[6] Corbly, D. M. and Packman, P. F., Engineering Fracture Mechanics, Vol. 5, 1973,
pp. 479--497.
[ 7] He, C. L., Buck, O., and Marcus, H. L. in Progress in Flaw Growth and Fracture
Toughness Testing, ASTM STP 536, American Society for Testing and Materials,
1973, pp. 5-21.
[8] lones, R. E., Engineering Fracture Mechanics, VoL 5,1973, pp. 585--604.
[9] Dugdale, D. S., Journal of Mechanics and Physics of Solids, Vol. 8, 1960, pp.
100-104.
[10 ] Westergaard, H. M., Journal o f Applied Mechanics, Vol. 6, 1939, pp. A-49--A-53.
[11] Rice, J. R. in Fatigue Crack Propagation, ASTM STP 415, American Society for
Testing and Materials, 1967, pp. 247-309.
[12] Rice, J. R., International Journal of Solids and Structures, Vol. 8, 1972, pp.
751-758.
[J3] Adams, N. J. I., Engineering Fracture Mechanics, Vol. 4, 1972, pp. 543--554.
[14] Bueekner, H. F., Zeitschrift fuer Angewandte Mathematik und Mechanik, Vol. 50,
1970, pp. 529--546.
[15] Forman, R. C., Kea_rney, V. E., and Engle, R. M., Journal of Basic Engineering,
Trans. ASME, Series D, Vol. 89, No. 3, 1967, pp. 459--464.
[16] Schijve, J., Brock, D., and de Rijk, P., "Fatigue Crack Propagation Under Variable
Amplitude Loading," National Aerospace Laboratory, NLR T. N. M. 2094, Dec.
1961.
[17] Wheeler, O, E., "Spectrum Loading and Crack Growth," ASME Paper No. 71 -
Met-X, 197t.
[181 Porter, T. R., "Method of Analysis and Prediction of Variable Amplitude Fatigue
Crack Growth," Engineering Fracture Mechanics, VoL 4, 1972.

DISCUSSION

Otto B u c k x (written discussion)-We used the Dugdale model too, some time
ago. 2 It uses an infinitely thin strip of material only. That means it does not
describe the real situation at all. So, I wonder why the results you obtained look
so good? Could you comment on that?
H. D. Dill and C. R. S a f f (authors' closure)-We use the Dugdale model to
define plane stress plastic zone size and crack opening displacement because it
incorporates the material properties thought to be important in governingplastic

i Science Center, Rockwell International, Thousand Oaks, Calif. 91360.

2 He, C. L., Buck, O., and Marcus, H. L. in Progress in Flaw Growth and Fracture
Toughness Testing, ASTMSTP536, American Society for Testing and Materials, 1973, p. 5.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
318 FATIGUE CRACK GROWTH

zone size, and it exhibits the expected load-displacement characteristics. In plane

strain conditions, characteristic of thick materials, we use plastic zone sizes and
crack opening displacements shown by Rice 3 to be roughly one half of those
predicted by the Dugdale model. Using these assumptions, the contact stress
model predicts a lower closure stress intensity, less retardation, and less delay in
minimum retardation under plane strain conditions than under plane stress. By
selecting an appropriate assumption of plane stress or plane strain we have been
able to correlate with data from several materials.
C. E. Richards 3 (written discussion)-In the plane stress model you employed,
it is relatively easy to visualize the "sucking in" of material from the free
surfaces to create net tensile strain in regions where the crack has passed whichin
turn leads to compressive stresses in tensile cyclic loading. However, in plane
strain conditions I find it less easy to visualize what happens metallurgically that
creates net tensile strains. Do you have any suggestions?
A member of the audience suggested that the net tensile strains required to
give compressive closure in zero-tension or tension-tension loading may arise
from one or both of the following effects. The net tensile strain may be caused
by nonlinear elasticity in the atom core regions of heavily worked material. This
may result in up to 2 percent strain. The second possibility is that the two
fatigue surfaces behind the crack tip may not "fit" exactly. For example, in
striation fatigue growth there may be "Nil-hilt" interference rather than
"hill-vaUey" matching. Nevertheless, there would be differences metallurgically
between net strains produced under plane stress and plane strain conditions.
Another member from the audience confirmed this and described evidence
that closure stresses in plane stress were about 50 percent of Kmax whereas in
plane strain the value was lower at about 30 percent ofKma x.
11. D. Dill and C. R. Saff (authors" closureJ-J. R. Rice4 has documented an
approach toward understanding the differences in plastic zone plane stress and
plane strain conditions in terms of slip or tensile yielding on discrete surfaces
emanating from the crack tip. He refers to the discussion of Hahn and
Rosenfield s in which plane stress plastic flow ahead of flat through-the-thickness
cracks in thin sheets tends to consist of two intersecting 45 deg shear bands. The
plasticity is then localized to a narrow region of height roughly equal to sheet
thickness. Rice models plane strain plastic flow as in-plane sliding on two
discrete surfaces inclined at angles +0 with the crack line. Perhaps his discussion
will give some insight into the mechanisms of plane stress and plane strain plastic
zone formation.
In the same reference Rice shows that the plastic zone size and crack opening
displacement are roughly half as large for plane strain conditions as compared to
plane stress conditions. We have incorporated this relationship into our model

3 Central Electricity Research Laboratories, Leatherhead, Surrey, England.

4 Rice, J. R. in Fatigue CrackPropagation, ASTMSTP415, AmericanSociety for Testing
and Materials, 1967, pp. 247-309.
s Hahn, G. T. and Rosenfield, A. R.,Acta Metallurgica,Vol. 13, No. 3, 1965.

and results indicate that, under constant-amplitude loading conditions with zero
minimum stress, closure will occur at 46.7 percent of Kmax in plane stress and
33.1 percent in plane strain. This seems to agree with the trends noted.

Spectrum Fatigue Crack Growth

in Lugs

REFERENCE: Impellizzeri, L. F. and Rich, D. L., "Spectrum Fatigue Crack Growth

in Lugs," Fatigue Crack Growth Under Spectrum Loads, ASTM STP 595, American
Society for Testing and Materials, 1976, pp. 320-336.

ABSTRACT: Analytical and experimental investigations were conducted to deter-

mine the crack growth behavior in lug joints subjected to a randomized flight-by-
flight spectrum. Specific objectives were to evaluate capabilities for analyzing
spectrum crack growth in lugs and to assess the effects on crack growth of KIc and
cold working the material surrounding the lug holes. Stress intensity solutions were
developed for cracks radiating from lug holes using Bueckner's weight function
approach for various stress distributions surrounding the lug hole. The stress
distributions were determined using an elastic/plastic finite element analysis.
Included in these analyses were residual stress distributions following either large lug
loads or mandrel hole enlargement. Spectrum crack growth calculations were made
using the resulting stress intensity values and the Wheeler retardation crack growth
model.
Lug specimens were fabricated from two heats of 6A1-4V mill annealed titanium
representative of material having low and high fracture toughness levels. Basic
mechanical properties including ultimate and yield strength, percent elongation, and
reduction in area and fracture characteristics, including KIc and constant-amplitude
crack growth rates, were determined for each heat of material. EDM cuts in the lug
specimens were used as points of stress concentration where preeracks were
developed by lowqevel constant-amplitude fatigue loading. The lug holes of selected
specimens were cold worked to determine the effect of compressive residual stresses
on the crack growth rate. Plots of crack growth versus spectrum hours were obtained
for each specimen by post-failure examination of the fracture surfaces using the
scanning electron microscope.

KEY WORDS: crack propagation, fatigue (materials), cyclic loads, aircraft, tensile
properties, mathematical prediction, strain energy methods, stress analysis, residual
stress, plastic deformation, inspection, titanium alloys

Structural fatigue very often initiates in areas of high load transfer due to the
superimposed stress concentration caused by the fastener bearing load. In many
structural joints, the magnitude of this fastener bearing load can be minimized
by transferring the load through a row of fasteners and tapering the mating parts
to soften the "hot spot" at the end fastener. This reduces the percentage of total
load transferred by each individual fastener and thereby reduces the ratio of
bearing stress to tension stress. In some structural joints, however, the geometry

1 Section chief-Strength and lead engineer-Strength, respectively, McDonnell Aircraft

Company, McDonnell Douglas Corporation, St. Louis, Mo. 63166.

320
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright9 1976 by ASTM
Downloaded/printed by Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions autho
IMPELLIZZERI AND RICH ON CRACK GROWTH IN LUGS 321

of the design dictates that I00 percent of the load be transferred through a
single fastener or pin. This type of structural joint is, of course, called a lug.
The elastic gross section stress concentration factor for a typical lug is about
five compared to about three for a fastener hole with zero bearing load. The
degradation in crack initiation life due to this high stress concentration factor is
significant. The degradation in crack growth life is even more serious because of
the large diameter of the lug hole. This is explained by the nature of a lug,
transferring 100 percent of the load through a single fastener or pin, which
requires the pin and therefore the lug hole to be large in diameter. Typical
fastener holes in aircraft structure are on the order of 6 mm in diameter, but lug
holes in aircraft structure are very often on the order of 50 mm in diameter. The
stress gradient for the larger diameter lug hole is much more shallow, that is, the
stress remains relatively high for larger distances from the edge of the hole
simply because of the larger diameter. While a 2-mm crack emanating from a
50-mm lug hole would still be in the highly stressed region and therefore
significantly affected by the lug hole, a 2-mm crack emanating from a 6-mm
fastener hole would be much less affected by the presence of the fastener hole.
The purpose of this paper is to present empirical crack growth characteristics
for a 6A1-4V annealed titanium lug subjected to a typical fighter aircraft
flight-by-flight fatigue spectrum, and to compare these spectrum crack growth
test results to analytical calculations.

Test Program
The objectives of the test program were to determine basic material properties,
for example, yield strength, fracture toughness, and da/dN as well as
flight-by-flight spectrum crack growth data for a lug. Test variables included two
toughness levels of 6A1-4V mill annealed titanium, two crack types, and three
mandrel interference levels for cold working to induce compressive residual
stresses around the lug hole. Materials, specimens, fatigue spectrum, and test
results are detailed in the following paragraphs.

Basic MaterialProperties
Two heats of 6AI-4V mill annealed titanium forgings were obtained to
evaluate the effect of fracture toughness on crack growth characteristics. Heat 1
consisted of three pieces of a 67 by 360 by 1220-mm forging which was forged
from a billet having an oxygen content of 0.175 percent and was representative
of low to medium toughness material. Heat 2, representative of high toughness
material, was forged as a single piece 150 by 230 by 910-ram forging from a
billet having an oxygen content of O. 152 percent.
Standard tensile and fracture toughness tests were conducted on specimens
taken from each heat of material. Fracture toughness was determined using
compact tension specimens, 31.75 mm thick, with all testing in conformance
with ASTM Test for Plane Strain Fracture Toughness of Metallic Materials (E

399-72). All test values were valid with the two heats showing a distinct
difference in toughness levels, 66 and 88 MN/m 3/2 . Constant-amplitude crack
growth data were obtained by testing two wedge opening loaded specimens from
each heat. The specimens had a thickness of 31.75 mm and a width of 125 mm.
The constant-amplitude crack growth rates were very similar for the two heats of
material, the only difference being for high ZkK.

Lug Specimen
Figure 1 shows the detailed geometry of the lug specimens. They were
fabricated from the mill annealed 6A1-4V titanium described previously and are

OlO i IZ,ql

.__~ 50.8 } 61.5 ~ EDMFlaws

(2,0) i /f/~A~2.42) [
Specimens
2 and 15

105.2
(4.14)
-

_2_
44 5 (1 75)Dia--/ EDMFlawsExcept
' " Specimens2 and 15"-"
SectionA-A
13.2
(o.s2) ~ . . L__ ~23.2__J___120.7__..I
1 ,466, I, (4"5'/
,I I
(6"145) . '. . .33~1~30)--~C I , 1o.61
.~-'T- I "' "

25.4 (1.0) R Typ"-J 25.4 (1.0) RTyp Unitsarein millimeters(inches)

FIG. 1-Lug specimen.

representative of typical wing spar-to-fuselage bulkhead attachments in fighter

aircraft structure. Twelve lug specimens, including both the male and female
parts, were machined from Heat 1 and three from Heat 2. As indicated in Fig. 1,
the mouth of the female lug is 33.0 mm, while the male lug is 21.8 mm thick,
producing a total 11.2-mm gap. The purpose of the gap is to allow for ease of
wing-to-fuselage assembly in a three-spar, three-bulkhead combination where
manufacturing tolerances can build up. Seven of the twelve Heat 1 specimens
were assembled unsymmetrically with the 11.2-mm gap all on one side. The
remaining five Heat 1 specimens and three Heat 2 specimens were assembled
symmetrically with 5.6-mm gap on each side. The assembly of all 15 specimens
included a 3.2-mm-thick beryllium copper bushing installed in both the male and
female lug. The bushings were installed with a very mild O.lO-mm diametral
interference which, coupled with its relative thinness and low stiffness, produced
no appreciable residual stress. The steel pins were installed with a very slight
clearance providing a "neat fit."
Electrical discharge machined (EDM) flaws were used as points of stress

concentration for precrack development. The EDM flaws, as indicated in Fig. 1,

were located at the hole wall midway between the lug surfaces for one specimen
from each heat, and at the corner of the hole wall and lug surfaces for the other
13 specimens. The comer flaws in the female lugs were located on the inside
surface which is the higher stressed surface because of secondary bending. The
corner flaws in the male lugs were located on both surfaces. The EDM flaws were
approximately 0.50-mm radius quarter-circle and 0.50-mm radius semi-circle for
the corner flaw and midway flaw, respectively. All specimens were precracked
by constant-amplitude cycling at 55 percent design limit load; the precrack size
for each specimen was about 3-mm radius.
Four of the Heat 1 specimens were cold worked after precracking. A split
sleeve process was utilized wherein a steel mandrel is pulled through a
0.46-mm-thick stainless steel sleeve installed in the lug hole. The split in the
sleeve is located at the back side of the lug. The lug hole is reamed to the final
44.45-mm-diameter size after cold working. The amount of material removed by
the reaming operation is nominally 0.60 mm on the diameter. The diametral
interference of the mandrel for cold working was 0.91 mm for two of the
specimens, 0.69 mm for the third specimen, and 0.46 mm for the fourth. The
extent of residual compressive stress introduced by the cold working is given in
the section on analytical techniques.

Flight-by-Flight Spectrum Definition

All of the lug specimens were tested to a fighter aircraft flight-by-flight
spectrum. The positive and negative cumulative peak exceedances per 1000 h are
given in Fig. 2 in terms of percent design limit load. Design limit load (DLL) for

100,000,

li~1 I11 II
-30 -10 10 30 50 70 ~ 10
Design Limit Load -percent
FIG.2-Fatigue spectrum positive and negative cumulative peak exceedances.
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
324 FATIGUF CRACK GROWTH

the lug specimens was 509 330 N which corresponds to a gross section stress of
223 MN/m 2 in the mate lug. The flight-by-flight spectrum includes 365 fighter
missions and 355 attack missions in 1000 h. A ground load is applied between
each flight to simulate take-off and landing loads. The higher positive and
negative cycles are randomly selected for inclusion in a particular mission. The
highest positive cycle applied once per 1000 h is 105 percent DLL; it is applied
approximately one tenth of the way through a 1000 h sequence. The spectrum
includes higher load levels applied less frequently. A 115 percent DLL cycle is
applied after 2000 h and every 4000 h thereafter. A 125 percent DLL cycle is
applied after 8000 h and every 16 000 h thereafter.

Spectrum Test Results

In addition to the 15 lug specimens that are the subject of this paper, lug
specimens of the same geometry were also tested in an earlier program but
without flaws or precracks. Those specimens were cycled without failure for
64 000 h to the spectrum defined in the preceding paragraph, and then statically
failed at approximately 240 percent DLL.
The detailed test results for the 15 precracked lugs are presented in Table 1.
The precrack sizes shown represent post-test measurements made by viewing the
fracture surfaces; it is a measurement giving crack depth perpendicular to the lug
hole surface. As indicated earlier, there was more than one precrack introduced
into each lug specimen. The EDM flaws were located in both the male and
female parts. The precrack size given in Table 1 is for the failure origin, so it is
either in the male or female lug depending on which failed. As shown in Table 1,
the failures were equally divided between the male and female lug suggesting a
rather well balanced design. The gross section design limit tension stress in the
male lug is 223 MN/m 2 , while the comparative tension stress in each ear of the
female lug is 184 MN/m 2. However, the male lug is in double shear, whereas
each ear of the female lug is in single shear which subjects it to bending stresses.
Strain gage measurements on the female lug indicate the combination of tension
plus bending gives 239 MN/m 2 on the critical inside surface.
The data in Table 1 show that, with an 0.5 l-ram EDM flaw, the lug specimen
sustained 50 000 h without failure. This specimen was tested without a visible
precrack to determine the effect of the EDM flaw. The largest precrack
introduced into a non-cold-worked lug specimen was 4.57 mm; that specimen
failed after 4542 h. Table 1 indicates a potent beneficial effect resulting from
the cold-working operation. Three of the four cold-worked specimens were
cycled for 50 000 h without failure and one of them had a 6.35-mm precrack.
The one cold-worked specimen that did fail had a relatively low mandrel
interference. It is shown in the section on analytical techniques that the
0.46-mm interference level produces a rather ineffective level of compressive
residual stress. All of the cold-worked specimens were tested in the unsymmetri-
cal configuration with 11.2-mm gap on one side. The effect of the unsymmetric

TABLE 1 - Spectrum test results.

Mandrel Precrack
Specimen Interference, Size, Crack Growth,
Number mm (in.) mm (in.) Life-Hottrsc

I 0.51 (0.02) 50 000 NF

2a 2.03 (0.08) 12 133 M
3 2.54 (0.10) 7 745 F
4 2.79 (0.11) 7 023 F
5 4.57 (0.18) 4 542 F
6b 2.29 (0.09) 5 139 F
7b 2.79 (0.11) 5 740 M
8b 4.06 (0.16) ? 228 M
9b 0.46 (0.018) 1.52 (0.06) 8 000 F
lO b 0.69 (0.027) 3.05 (0.12) 50 000 NF
ii b 0.91 (0.036) 2.79 (0.11) 50 000 NF
12 b 0.91 (0.036) 6.35 (0.25) 50 000 NF
13 1.78 (0.07) 10 931 M
14 2.79 (0.11) 6 000 M
15 a 3.05 (0.12) 6 000 M

a These specimens had midway crack; all others had corner crack.
b These specimens tested with 11.1 mm (0.44 in.) gap on one side; all others tested
symmetrically with 5.6 mm (0.22 in.) gap on each side.
c NF indicates no failure. M indicates failure in male lug. F indicates failure in female
lug.

loading is minimal on the male lug and only slightly more significant on the
higher loaded ear of the female lug. The crack growth life of the female lugs
tested unsymmetrically was somewhat less than the female lugs tested
symmetrically. For example, Specimen 3 tested symmetrically and with a
2.54-mm precrack sustained 7745 h, while Specimen 6 tested unsymmetrically
and with a 2.29-mm precrack sustained 5139 h.
Specimens 1 through 12 in Table 1 are from Heat t and Specimens 13 through
15 are from Heat 2. As indicated previously, the material Kxc for Heat 1 is 66
MN/m 3/2 while the material KIc for Heat 2 is 88 MN/m 3/2 . The higher fracture
toughness for Heat 2 did not provide slower constant-amplitude crack growth
except for high ZkK. The same trend of no appreciable difference between Heat 1
and Heat 2 is demonstrated in Table 1 for spectrum crack growth. For example,
Specimen 7 from Heat 1 with a 2.79-mm precrack sustained 5740 h. This
compares to Specimen 14 from Heat 2, also with a 2.79-mm precrack, which
sustained 6000 h.

Analytical Techniques
The objective of the analysis presented in this paper was to determine the
adequacy of developed techniques in terms of correlation with spectrum crack
growth data for lugs. The analysis techniques included a combination of
Bueckner's weight function for an edge crack, a geometry correction factor, and
finite element solutions giving lug stress distributions. These were utilized to
obtain the stress intensities for the loads applied in the flight-by-fiight fatigue
spectrum and for those resulting from mandrel hole enlargement. The Wheeler
plastic zone model was used for determining crack growth retardation.

Approximate Weight Function for a Hole

Bueckner's weight function [1] 2 is given by the following relation

m(x,a) = H au(x,a)
2K aa
where H equals E for plane stress and El1 - v2 for plane strain, K is the stress
intensity, and u(x,a) is the crack opening displacement at x for a crack of length
a. The weight function was shown to be unique by Bueckner [1] and by Rice
[2] for a given structural geometry and crack size regardless of the stresses
acting on the structure, that is, it is independent of the loading condition. The
integral of the product of this function and the stress distribution along the
crack boundary gives the stress intensity, or in equation form
K = f a p(x) m(x,a) dx

where p(x) is the stress distribution that would exist along the crack boundary if
the crack were not there. Since the weight function is independent of the
loading condition, it can be determined for one condition and then utilized to
obtain the stress intensity for another.
The present development of the weight function for a hole utilizes the exact
weight function derived by Bueckner [3] for an edge crack in a semi-infinite
plate. This function is modified by a geometry correction factor to obtain the
desired result. The final equation for the approximate weight function for a hole
is
m(x,a) = mec ~1 r ~3
where
mec = (a-x) -':2 [1 + 0.6147(1 _ x ) + 0.2502(1 - x ) ~ ] [ x / ~ ]
a a

~1 = I - 0.6449(_~ + 0.8964(-~-)2 - 0.7327(~) 3

+ o 0 + 0

~2 and ~P3 are given in the Appendix.

2 The italic numbers in brackets refer to the list of references appended to this paper.

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
I M P E L L I Z Z E R I A N D RICH ON C R A C K GROWTH IN LUGS 327
The term rnec is Bueckner's weight function for an edge crack in a semi-infinite
plate, and x is measured from the edge of the plate toward the crack tip. The
term ~1 is the geometry correction factor, and R is the hole radius. The
geometry correction factor was obtained as the ratio of the stress intensity for a
crack emanating from one side of a hole in an infinite plate to the stress
intensity for an edge crack in a semi-infinite plate. The stress intensities for both
of these configurations were of course determined for the same loading, uniform
pressure on the crack faces. The stress intensity determined for the crack
emanating from the hole was obtained numerically [4]. The preceding equation
for qbl was obtained by point to point matching of the two stress intensity
solutions in the range aR < 3; therefore, the weight function can only be
considered accurate for a/R <_3.
It should be recognized that since the geometry correction factor q~l was
determined as the ratio of stress intensities for a particular loading condition,
uniform pressure on the crack faces, that the resulting weight function is not
necessarily accurate for other loading conditions. To evaluate its accuracy, the
weight function mec cb~ was utilized to compute the stress intensity for a crack
emanating from one side of a hole in an infinite plate subjected to uniform
tension applied at a large distance from the hole. This is a significantly different
loading condition than uniform pressure on the crack faces. The resulting curve
of stress intensity versus crack length is compared to Bowie's [5] accepted
standard solution and Grandt's [6] approximate solution in Fig. 3. Both curves
are within 7 percent of the Bowie prediction. Another approximate solution was
developed by Hall et al [7] using superposition principles. It is within 2 percent
of the Bowie curve for one crack and 7 percent for two cracks for a/R>_0.05.

3.5
0

3o t tf

= I

I .O - ~ ~

t
~ Bowie Solution
0.5 . . . . . ProposedSolution
--~ Grandt Solution

0 I I I I I
0 0.5 1.0 1.5 2.0 2.5 3.0
a

FIG. 3-Correlation with Bowie stress intensity solution.

The present solution is slightly on the conservative side, that is, for the same
crack length it predicts a somewhat higher stress intensity than Bowie. The fact
that the stress intensity calculated herein compares favorably with the accepted
standard for a loading distribution significantly different from the one for which
it was developed suggests that reasonable accuracy could be expected for other
loading distributions.
The quantity rnec ~1 represents the weight function for a crack emanating
from one side of a hole in an infinite plate. The term ep2 provides the necessary
adjustment factor to account for the case of cracks emanating from both sides of
a hole in an infinite plate. The term ~3 gives the correction factor for finite
width effects. The accuracy of the product of these factors was determined by
comparison to Newman's [8] solution for cracks emanating from both sides of a
hole in finite width plates with width/diameter = 2 and 4. The present solutions
were within 4 percent of both of Newman's solutions.

Calculations of Lug Stress Intensity

The analytical technique described in the previous section provides a method
of computing the stress intensity for cracks in holes for any loading condition.
The additional information required for this analysis is the lug stress distribution
that would exist along the crack boundary if the crack were not there. Figure 4

~>'1e, 3.0

2.0

1.0

0 I I I I I I
0,2 0.4 0.6 0.8 1.0 1.2 1.4
R

FIG. 4-Elastic stress distribution in loaded lug.

gives that distribution based on a two-dimensional finite element solution for a

pin loaded lug with the geometry of the titanium specimens discussed in the first
part of this paper. The through-the-thickness variation in stress is considered to
be small due to the relatively large diameter of the lug hole compared to its
thickness. The steel "neat fit" pin was also modeled for the finite element

analysis to produce the correct pin bearing pressure on the inside surface of the
lug hole. The resulting stress intensity versus crack length is given in Fig. 5. It
should be noted that the stress intensity given in Fig. 5 includes a 1.13 front face
correction factor which corresponds to Bueckner's [3] derivation for an edge
crack.

6.0

5.0

4.0

v 3.0

2.0--

1.0

0
o 0.2 0.4 0.6 0.8 1.0 1.2 1.4
a

FIG. 5-Stress intensity solution for a through crack in a lug.

Four of the lug specimens were cold worked prior to testing with a split sleeve
technique as mentioned earlier. The resulting compressive residual stresses
produced by the three different levels of mandrel interference are given in Fig. 6.
These stress distributions were obtained by elastic/plastic analysis [9] assuming

x - inches
3000" 0.2 0.4 0.6 0.8 1.0 1,2
40
200 | ! ~ I
z
100 /-/ - 20

o '
0
-100
~--I = 0.46 10.0181 - -20
.~ -200
-300 - -40
E -400 -60
.~ -600
-80
~ -611(I
* Mandrel d i a m e t r a l
~
e,-
-700 interference in mm (inch)
- -100
-800 I I I t I -120
5 10 15 20 25 3O
x - millimeters
FIG. 6-Residual stresses after cold-work expansion o f holes.

a "donut shaped" configuration with the outside diameter equal to the width of
the lug. An elastic/plastic finite element solution using the actual lug geometry
was also obtained for the 0.91-mm mandrel interference which verified the
validity of the donut-shape assumption. The residual stress intensities versus
crack length for the three different mandrel interference levels are given in Fig.
7. It is of interest to note that the residual stress intensity remains negative

Crack Length - inches

0.2 0.4 0.6 0.8 1,0 1.220
20 -- 1 I I 1 I

10 10
r

,-10 -10
~ -2o = . , q
-20 !
| -30 I = 0.91 (0.036) _ -30
"~ -40
.==
"El
40~
n- - 5 0
* Mandrel diarnetral -50
interference in mm (inch)
-60
I I I I I -60
0 5 10 15 20 25 30
Crack Length - millimeters

FIG. 7-Residual stress intensity for through cracks in cold.worked lugs.

farther out than the residual stress in Fig. 6. This is because the stress intensity is
obtained by integration, and the negative residual stresses predominate in this
computation until larger cracks are developed.
It should be emphasized at this point that the analysis in this paper is not
extended beyond linear elastic fracture mechanics. Although an elastic/plastic
analysis technique was utilized to obtain the residual stress distributions in Fig.
6, the residual stresses themselves are elastic and were simply used as the
distribution p(x) required for Bueckner's elastic weight function approach to
compute stress intensities. The basic idea for the cold-worked specimens is that
Krnax and Kmin for each cycle in the flight-by-flight spectrum are determined
by adding the stress intensity from Fig. 7 to the stress intensities from Fig. 5 for
each cycle's maximum and minimum stress. This is simply a superposition
technique valid for elastic systems. As long as the sum of the residual stresses
and the stresses produced by the externally applied loads do not cause yielding,
the solution should be reasonably accurate. In the absence of any compressive
residual stresses, the highest loads in the flight-by-flight spectrum produce
stresses only slightly greater than tension yield at the edge of the lug hole.
Therefore, when this tension stress field is added to the large compressive
residuals produced by the mandrel operation, the sum is below the elastic limit.

The stress intensities versus crack length given in Figs. 5 and 7 are for
through-the-thickness cracks as indicated on the graphs. Throughout most of the
spectrum crack growth life of the subject lug specimens, the crack was not
through the thickness but rather a quarter circle or semi-circle corresponding to
the shape of the corner flaw or midway flaw, respectively, introduced by
electrical discharge machining (EDM). The flaw shape parameter, Q, [10] was
utilized to account for this difference. It was assumed that the stress intensity
was equal to the quantity I / V ~ multiplied times the through-the-thickness stress
intensities given in Figs. 5 and 7; a value of Q = 2.47 was used. Although this
assumption is only accurate for short crack lengths, it provides a conservative,
that is, greater than actual, estimate of the stress intensity. Comparison with a
solution by Kobayashi [11] for an open hole with aiR = 0.2, and using this
assumption resulted in about 4 percent disagreement. The slight variations in Q
due to O/ay s variations were included in the specfrum crack growth analysis.

Spectrum Crack Growth Analysis

The ZkK value for each cycle in the flight-by-flight spectrum was determined
using the stress intensity computation procedures detailed in the previous
section. These were then used to enter a curve of da[dN versus zkK to obtain the
crack extension for each cycle. The crack growth was thereby linearly summed
on a cycle-by-cycle basis.
Stress ratio adjustments were made using Forman's equation [12]. There are
many cycles in the spectrum where the valley is compression, but the stress ratio
was assumed to be zero in these instances rather than negative. This is because
the critical area in a lug is not put into compression during reversed loading, but
simply put into a nearly zero stress condition. The stress ratio for all cycles in
the spectrum is therefore either zero or a positive value. This is the case except
for the cold-worked specimens. The stress ratio for many of the cycles is
negative for these specimens because of the large residual stress intensity, as
shown in the preceding section, especially for the 0.91-mm mandrel interference.
Crack growth retardation, due to periodically applied higher load levels, was
taken into account, using the Wheeler [13] plastic zone model. The size of the
plastic zone in front of the crack tip was recomputed on every cycle. The
magnitude of the Wheeler retardation parameter m used in the spectrum crack
growth analysis was 2.98 based on the best fit of crack growth data from mill
annealed 6A1-4V titanium unloaded hole specimens subjected to the same
flight-by-flight fighter aircraft spectrum as the lug specimens.

Correlation with Fracture Surface Crack Growth Measurements

The fracture surfaces of the lug specimens were viewed with the scanning
electron microscope to match individual striation spacings with particular load
levels in the spectrum. This was done by knowing the sequence of load level
magnitudes in the spectrum and relating that to the observed sequence of
striation spacing widths. The resulting spectrum crack growth data for the male
lugs are presented in Fig. 8 and for the female lugs in Fig. 9. The crack growth
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
332 F A T I G U E C R A C K GROWTH

1.10
0
Predicted crack growth A A
- 1.00
25
. . . . . . . . . . . . . Kc= 110 MN/rn3/2 (100 ksi ~r 9 0~ o
KC= 88MN/m 3/2 (80i<si~m.) l 0 - 0.90
- - Kc= 66 MN/m 3/2 (60ksi~/~'n.) /
- 0.80
2O
I O
0.70~
Specimen I n~/; 0 A
-

Symbol number IA 0 0 - 0.60 =

E15 - /~ ::
o 2 A 0 K
,~ 7 / / o - o.soB
o 6 /OJ zx ~
_ A 13 /,d../~o o - 0.40 G
(J O 14 ~ ,~00
0 15 /4_o'~']"~ 0.30

-- 0,20

-- 0.10

--"--1- I I I l I 1 I l l l I I I I i 0
1 2 3 4 6 6 7 8 9 10 11 12 13 14 15 16 17
Hours - thousands

FIG. 8-Crack growth in male lugs.

1.10
Predicted crack growth r'l O
Z~ -- 1.00
25 Z~
".-. . . . . . . . Kc = 110 MN/m 3/2 (100 ksi V'~'~.)
m~mKc= 88MN/m 3/2 (80ksi~/iE) ~ r'l -- 0.90

Kc = 66 MN/m 3/2 (60 ksi VTm.) z~ o , -- 0.80

/~ Predictionfor -- 0.70
.E Specimen I ~ i~ g / 0.46 mm Mandrel , }
Interference 0 . 6 0 "7
~16 Symbol number I d :
o 3 /~,O ' oA / --i 0 . 5 0
g ,, 4 / . .',' o-e."

~10 - --I 0,40

<> 6 ./r~." oa ~,"
9 9 /~n~ 9 ,-" - 0.30
/,r OA.. ~
- 0.20

0
0 1 2 3 4 5 6
?,,,,
7 8 9 10 11
1
12 13
1
14
l I
15
I
16 17
0.10

Hours - thousancls

FIG. 9-Crack growth in female lugs.

data for each specimen are plotted starting with its precrack depth entered on
the predicted curve. The variation in crack depth at failure indicated in both
Figs. 8 and 9 is primarily due to the variation in the failing load resulting from
the different load levels in the fatigue spectrum.
The three crack growth prediction curves in the two graphs of spectrum crack
growth show the significant effect of the K e assumption, at least in the later

stages of growth. The question of what K e value to assume is always a difficult

one. It should be emphasized that this question does not relate primarily to what
is the Kic value, but rather what is the extent of plane stress in the failure mode
resulting in a K e level significantly higher than the material plane strain fracture
toughness. The K c computed for the failing load of the male lugs for Heat 1
averaged 90.7 MN/m 3/~ and for Heat 2 averaged 97,1 MN/m 3/2 . The K c
computed for the failing load of the female lugs for Heat 1 averaged 103.2
MN/m 3:2 . None of the Heat 2 specimens failed in the female lug. This is simply
because the EDM flaws and precracks were, by chance, larger in the male lugs.
The K c computed for the female lugs was, as expected, greater than that
computed for the male lugs because they are thinner and plane stress conditions
would be more operable. Although plane stress conditions were more predomi-
nant in the female lugs, plane stress was also active on the top and bottom edges
of the male lug causing tunneling of the crack through the midptane of the male
lug. This factor probably contributed to the average K c value being greater than
Klc.
The one cold-worked specimen that failed had a rather low 0.46-mm mandrel
interference; the residual stress intensity for that specimen is shown in Fig. 7. Its
spectrum crack growth data are represented by the solid circular symbols in Fig.
9. Using the superposition approach mentioned in the preceding section, that is,
adding the residual stress intensity to the stress intensities due to the applied
loading, gives the crack growth prediction curve shown in Fig. 9. Curves of
precrack size versus predicted crack growth life for different mandrel interfer-
ence levels are presented in Fig. 10. The dramatic benefit provided by the cold

20I
0.8

Mandre(
Specimen interference 0.7
Symbol number mm (inch)
[] 9 0,46 (0,018)
15 0.6
A 10 0.69 (0.027)
O 11 0.91 (0.036)
0 12 0.91 (0.036) 0.5

N
~=10 0,4
b5

0.3

" I = 0-46 mm

0 I I I I
0 10 20 30 40
Hours - thousands

FIG. lO-Precrack size versus crack growth life for cold-worked specimens.

working, especially at the 0.gi-mm mandrel interference level, is demonstrated

in this graph and the test data correlate favorably with the curve.

Summary
Precracked lug specimens fabricated from mill annealed 6A1-4V titanium were
tested to a flight-by-flight fighter aircraft fatigue spectrum. A number of the
specimens were cold worked by mandrel hole enlargement to produce
compressive residual stresses around the lug hole to retard crack growth.
Comparisons of the crack growth life for the lug specimens indicated a potent
beneficial effect due to cold working. For example, a non-cold-worked specimen
with a 2.54-mm precrack failed after 7745 spectrum hours compared to a
cold-worked specimen with a 6.35-mm precrack tested for 50 000 spectrum
hours without failure.
An approximate weight function for a hole was developed to compute stress
intensities for different stress distributions surrounding the lug hole. These stress
intensity solutions were utilized to enter daMN versus AK curves for annealed
6A1-4V to obtain data for spectrum crack growth analysis. The Wheeler plastic
zone model was used for determining crack growth retardation due to
periodically applied higher load levels. The magnitude of the retardation
parameter m used in the analysis was 2.98 based on the best fit of crack growth
data from unloaded hole specimens subjected to the same spectrum as the lugs.
The resulting crack growth predictions correlated reasonably well with the
spectrum test data obtained by scanning electron microscope examinations of
the lug fracture surfaces.

Acknowledgment
The authors wish to extend special thanks to a colleague at McDonnell
Aircraft Company, W. T. Fujimoto, who developed the approximate weight
function for a hole utilized in this paper.

Appendix
The terms alp: and ap~ are used to adjust the weight function for double cracks
and to account for finite width effects, respectively. They are given by the
following equations:
~2 = 1 (single crack)

(double crack)
1~Rf~"~'~_2tan-,
l [ 2R-a 1 +2J 1-'-2R-a~t
2u 2I . #2R+2)~-(2R-a)~J n-- 2R+a ~

~3 = /sect( 2R + a~(ir~sin~[2R Sbal('~-~l~

] ~W-al~21) [~N-al~ICl) (singlecrack)
sec 21~ lr 2R+a 2a

"3 = / s e c l ( 2 R i ~ ~)(~) } ' (double crack)

The term 12 was derived using an assumption that the crack or cracks and hole
can be represented by a single effective through-the-thickness crack equal in
length to the diameter of the hole plus the length of the crack or cracks. Stress
intensities were then determined using this assumption with a uniform pressure
applied over the original crack area [14]. The term ~2 then is simply a ratio of
the solution for a crack emanating from two sides of the hole and the solution
for a crack emanating from one side of the hole.
The term ~3 was determined using Isida's [15] finite width correction factors
and the effective through-the-thickness crack length as just defined. Since the
expression m e c ~ l gives the appropriate stress intensity for short cracks, the
term ~3 is normalized by dividing the stress intensity for crack lengths of 2R+a
and 2R+2a, by the stress intensity for a crack of length 2R. An important factor
inherent in term q53 is the correction which allows compilation of finite width
effects for the crack emanating from only one side of the hole.

References
[1] Bueckner, H. F., "A Novel Principle for the Computation of Stress Intensity
Factors," Zeitschrift fur Angewandte Mathernatik und Mechanik, Vol. 50, No. 9,
1970, pp. 529--546.
[2] Rice, J. R., International Journal of Solids and Structures, Vol. 8, 1972, pp.
751-758.
[3] Bueekner, H. F., "Weight Functions for the Notched Bar," Zeitschrift fur
Angewandte Mathematik und Mechanik, Vol. 51, 1971, pp. 97-109.
[4] Bueckner, H. F. and Giaever, I., "The Stress Concentration of a Notched Rotor
Subjected to Centrifugal Forces," Zeitschrift fur Angewandte Mathematik und
Mechanik, Vol. 46, 1966, pp. 265-273.
[5] Bowie,O. L., Journal of Mathematics and Physics, Vol. 35, 1956, pp. 60-71.
[6] Grandt, A. F., "Stress Intensity Factors for Some Thru-Cracked Fastener Holes,"
AFML-TR-74-71,July 1974.
[ 7] Hall,L. R., Shah, R. C., and Engstrom, W. L., "Fracture and Fatigue Crack Growth
Behavior of Surface Flaws and Flaws Originating at Fastener Holes," AFFDL-TR-
74-47, Vol. I, May 1974.
[8] Newman, J. C., Jr., "An Improved Method of Collocation for the Stress Analysis of
Cracked Plates with Various Shaped Boundaries," NASA TN D-6376, Aug. t971.
[ 9] Hoffman, O., Introduction to the Theory of Plasticity for Engineers, McGraw-Hill,
New York, 1953, pp. 80--94.
[10] Brown, W. F., Jr., and Srawley, John E., Plane Strain Crack Toughness Testing of
High Strength Metallic Materials, ASTM STP 410, American Society for Testing and
Materials, Dec. 1967, p. 102.

[11] Kobayashi, A. S., "A Simple Procedure for Estimating Stress Intensity Factor in
Region of High Stress Gradient," AD767404, Sept. 1973.
[121 Forman, R. G., Kearney, V. E., and Engle, R. M. "Numerical Analysis of Crack
Propagation in Cycle-Loaded Structures," American Society of Mechanical Engi-
neers, Paper No. 66-WA-MWT-4,1966.
[13] Wheeler, O. E., "Spectrum Loading and Crack Growth," American Society of
Mechanical Engineers, Paper No. 71-Met-X, 1971.
[14] Tada, H., Paris, P., and Irwin, G., The Stress Analysis of Cracks Handbook, Del
Research Corporation, Hellertown, Pa., 1973, p. 5.11.
[15] Isida, M., Transactions, American Society of Mechanical Engineers, Series E,
Journal of Applied Mechanics, Vol. 33, 1965, pp. 674.

Glossary
A Constant
a Crack length (half crack length in center-cracked specimens, or crack
depth for surface flaws)
ao Initial crack length
af Final crack length
z2~a* Overload affected crack length increment
&a/z2UVor
da/aN Rate of fatigue crack growth
B Specimen thickness
BN Net thickness in face-grooved specimen
C Compliance
c Half crack length at the specimen surface of surface-flawed specimens
E Young's modulus
I Frequency
G Strain energy release rate
K Stress intensity factor
KI K for the opening mode (Mode I)
Kc Critical stress intensity for failure or fracture toughness
Klc Plane-strain fracture toughness
Klscc Apparent threshold K for stress corrosion cracking
gm ax Maximum stress intensity factor in a loading cycle
gmean Mean value of stress intensity factor in a loading cycle
Km in Minimum stress intensity factor in a loading cycle
Kol Overload K
Ko p Crack opening K
Krms Root mean square stress intensity factor for a distribution
Kul Underload K
&K Range of stress intensity factor in a loading cycle; &K = Kmax -
Kmin
L~tK'ef f Effective range of stress intensity factor in a loading cycle
z:~xgth Threshold &K for fatigue crack growth
z2kKth* Overload affected &Kt~
N Number of cycles
1/ Exponent in crack growth rate expression
No Number of cycles of delay
P Applied load
PmBx Maximum applied load in a loading cycle
Pm in Minimum applied load in a loading cycle
Applied load above which the crack is fully open; crack opening load
337
Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Copyright 9 1976 by
Downloaded/printed by ASTM Intemational www.astm.org
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
338 FATIGUE CRACK GROWTH

Pr ~ zSat~ Range of applied load in a loading cycle; Pr = ZkP = Pmax - Pmin

Pr~ Distribution o f P r
Prm Modal value o f P r
Q Shape factor for surface flaws
R Stress ratio; R = Smin/Sma x = g m i n / g m a x
ry Plastic zone correction factor; ry = , ~ (K/oys) 2
S Applied (gross-section) stress; S = P/B W
Sc Closure stress
Smax Maximum applied stress in a loading cycle
Smean Mean value of applied stress in a loading cycle
Stain Minimum applied stress in a loading cycle
Sop Crack opening stress above which the crack tip is fully open
Srd Distribution o f S r
Srm Modal value of stress range S r
A~Sor S r Range of applied stress in a loading cycle; S r = ~ = Sma x - S m i n
/xSer~ Effective stress range
t Time or hold time (time at load)
T Temperature
U Effective stress or stress intensity range ratio; U = ASeff-- = Z~'eff
W Specimen width AxS z~
c Strain
p Poisson's ratio
o Stress
o u Ultimate strength
~ Yield strength
,7- Shear stress

Subscripts
Elemental Load Spectra
h High-load excursions (Khmax, Khmin , ~Kh, Nh, th, t h',fh, Rh)
s Low-load (Ks ,K~min , ~ , N ~ , f ~ , R s

~.~ t].~iolOad~!
W
SINGLE HIGH-LOAD MULTIPLE HIGH-LOAD PERIODIC HIGH-LOAD
EXCURSION EXCURSIONS EXCURSIONS

Copyright by ASTM Int'l (all rights reserved); Mon Dec 21 11:17:56 EST 2015
Downloaded/printed by
University of Washington (University of Washington) pursuant to License Agreement. No further reproductions authorized.
GLOSSARY 339
Block Programmed Loads
m block number
n load level within a given block
Typical Usage
~kg l,l,l , l~1
stress intensity range for the n th load level in the m tk load block

II mth load block ~!

STP 91-A-1963
No ratings yet
STP 91-A-1963
94 pages
Environmentally Assisted Cracking - Predictive Methods For Risk Assessment and Evaluation of Materials, Equipment, and Structures (ASTM Special Technical Publication, 1401) PDF
100% (2)
Environmentally Assisted Cracking - Predictive Methods For Risk Assessment and Evaluation of Materials, Equipment, and Structures (ASTM Special Technical Publication, 1401) PDF
491 pages
STP 560-1974
No ratings yet
STP 560-1974
263 pages
Tada, Paris, Irwin The Stress Analysis of Cracks Hadbook - Section
100% (1)
Tada, Paris, Irwin The Stress Analysis of Cracks Hadbook - Section
100 pages
Scientific Study of Mechanical Relaxation of Residual Stress PDF
100% (1)
Scientific Study of Mechanical Relaxation of Residual Stress PDF
120 pages
DS5S2 - (1969) An Evaluation of The Yield, Tensilde, Creep and Rupture Strengths of Wrought 304, 316, 321, and 347 Stainless Steels at Elevated Temperatures
100% (1)
DS5S2 - (1969) An Evaluation of The Yield, Tensilde, Creep and Rupture Strengths of Wrought 304, 316, 321, and 347 Stainless Steels at Elevated Temperatures
74 pages
The Stregth, Fracture Toughnees and Low Cycle 17-4PH PDF
No ratings yet
The Stregth, Fracture Toughnees and Low Cycle 17-4PH PDF
11 pages
ACI 349-06 Code Requirements For Nuclear Safety-Related Concrete Structures and Commentary - MyCivil - Ir PDF
100% (1)
ACI 349-06 Code Requirements For Nuclear Safety-Related Concrete Structures and Commentary - MyCivil - Ir PDF
157 pages
STP 415-1967
50% (2)
STP 415-1967
551 pages
STP 536-1973
No ratings yet
STP 536-1973
504 pages
Fatigue LifA State-Of-The-Art Review On Fatigue Life Prediction Methods For Metal Structures
No ratings yet
Fatigue LifA State-Of-The-Art Review On Fatigue Life Prediction Methods For Metal Structures
14 pages
STP1256 Eb.1415051 1 PDF
100% (1)
STP1256 Eb.1415051 1 PDF
842 pages
Stp1231-Eb 20688 PDF
No ratings yet
Stp1231-Eb 20688 PDF
668 pages
STP 648-1978
No ratings yet
STP 648-1978
300 pages
STP 1546-2012
No ratings yet
STP 1546-2012
390 pages
STP 559-1974
No ratings yet
STP 559-1974
319 pages
STP 590-1976
No ratings yet
STP 590-1976
506 pages
ASTM - STP 588 - Manual On Statistical Planning and Analysis
No ratings yet
ASTM - STP 588 - Manual On Statistical Planning and Analysis
158 pages
STP 556-1974
No ratings yet
STP 556-1974
192 pages
Crashworthiness Ductile Failure Jones Wierzbicki
No ratings yet
Crashworthiness Ductile Failure Jones Wierzbicki
522 pages
STP 518-1983
No ratings yet
STP 518-1983
173 pages
ASTM - STP 407 - Temper Embrittlement in Steel 1968 PDF
100% (1)
ASTM - STP 407 - Temper Embrittlement in Steel 1968 PDF
260 pages
Murakami, Endo - Effects of Defects, Inclusions and Inhomogeneities On Fatigue Strength
No ratings yet
Murakami, Endo - Effects of Defects, Inclusions and Inhomogeneities On Fatigue Strength
20 pages
Reporting Uniaxial Strength Data and Estimating Weibull Distribution Parameters For Advanced Ceramics
No ratings yet
Reporting Uniaxial Strength Data and Estimating Weibull Distribution Parameters For Advanced Ceramics
17 pages
STP1425 Eb.1415051 1 PDF
100% (1)
STP1425 Eb.1415051 1 PDF
445 pages
Fretting Fatigue
100% (1)
Fretting Fatigue
445 pages
STP1303 Eb.1415051 1 PDF
No ratings yet
STP1303 Eb.1415051 1 PDF
157 pages
Fatigue Behavior of Dented 1-42
No ratings yet
Fatigue Behavior of Dented 1-42
42 pages
Little R E Mechanical Reliability Improvement Probability and Statistics For Experimental Testing Marcel Dekker 2001 PDF
No ratings yet
Little R E Mechanical Reliability Improvement Probability and Statistics For Experimental Testing Marcel Dekker 2001 PDF
595 pages
STP1318 Eb.1415051 1 PDF
100% (1)
STP1318 Eb.1415051 1 PDF
236 pages
Seminar On Brittle and Ductile Fracture
100% (4)
Seminar On Brittle and Ductile Fracture
29 pages
STP 527-1973
No ratings yet
STP 527-1973
119 pages
STP780 Eb.1415051 1 PDF
No ratings yet
STP780 Eb.1415051 1 PDF
191 pages
DS63 - (1985) References and Conference Proceedings Towards The Understanding of Fracture Mechanics
100% (1)
DS63 - (1985) References and Conference Proceedings Towards The Understanding of Fracture Mechanics
58 pages
John J. Friel - Practical Guide To Image Analysis (2000, ASM International)
100% (1)
John J. Friel - Practical Guide To Image Analysis (2000, ASM International)
299 pages
STP 601-1976
100% (1)
STP 601-1976
593 pages
Fatigue and Fracture Testing ASTM PDF
No ratings yet
Fatigue and Fracture Testing ASTM PDF
164 pages
A Review of Failure Modes and Fracture Analysis of Aircraftcomposite Materials
No ratings yet
A Review of Failure Modes and Fracture Analysis of Aircraftcomposite Materials
13 pages
Gear Tooth Rooth Fatigue Behaviour PDF
No ratings yet
Gear Tooth Rooth Fatigue Behaviour PDF
12 pages
Aerospace Aluminum PDF
No ratings yet
Aerospace Aluminum PDF
6 pages
Stress Fatigue Principles
No ratings yet
Stress Fatigue Principles
12 pages
Shaft Fatigue - 11838-44955-1-PB
No ratings yet
Shaft Fatigue - 11838-44955-1-PB
8 pages
Fatigue Polymers
No ratings yet
Fatigue Polymers
1 page
Other Causes of Fastener Failures
No ratings yet
Other Causes of Fastener Failures
3 pages
Lecture 4 - Stress-Life Approach
No ratings yet
Lecture 4 - Stress-Life Approach
97 pages
Fatigue of Threaded Fasteners: Alex Hudgins Brad James, Fasm
No ratings yet
Fatigue of Threaded Fasteners: Alex Hudgins Brad James, Fasm
5 pages
(ASTM Special Technical Publication, 1201) C R Brinkman_ S F Duffy_ ASTM Committee C-28 on Advanced Ceramics._ American Ceramic Society._ ASTM International - Life Prediction Methodologies and Data
No ratings yet
(ASTM Special Technical Publication, 1201) C R Brinkman_ S F Duffy_ ASTM Committee C-28 on Advanced Ceramics._ American Ceramic Society._ ASTM International - Life Prediction Methodologies and Data
422 pages
STP 619-1977
100% (1)
STP 619-1977
74 pages
Fatiguehandbook Small PDF
100% (1)
Fatiguehandbook Small PDF
532 pages
STP 338-1963
No ratings yet
STP 338-1963
274 pages
Astm E1426.11123 PDF
No ratings yet
Astm E1426.11123 PDF
6 pages
STP 600-1976
No ratings yet
STP 600-1976
263 pages
Derivation of Endurance Curves From Fatigue Test Data, Including Run-Outs
No ratings yet
Derivation of Endurance Curves From Fatigue Test Data, Including Run-Outs
39 pages
STP768 Eb.1415051 1 PDF
No ratings yet
STP768 Eb.1415051 1 PDF
182 pages
STP624-EB.28676-1 Nondestructive Testing Standards... 1977
100% (2)
STP624-EB.28676-1 Nondestructive Testing Standards... 1977
355 pages
Nasa Rolling-Element Bearings - A Review of State of Art
100% (2)
Nasa Rolling-Element Bearings - A Review of State of Art
87 pages
STP 516-1981
No ratings yet
STP 516-1981
329 pages
Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized
No ratings yet
Downloaded/printed by Universidad de Chile Pursuant To License Agreement. No Further Reproductions Authorized
224 pages
STP 569-1975
No ratings yet
STP 569-1975
336 pages
Fatigue and Crack Initiation - George C. Sih
100% (1)
Fatigue and Crack Initiation - George C. Sih
386 pages
Fatigue Thresholds
100% (3)
Fatigue Thresholds
436 pages
STP 572-1974
No ratings yet
STP 572-1974
299 pages
STP 598-1976
No ratings yet
STP 598-1976
346 pages
STP 543-1974
No ratings yet
STP 543-1974
273 pages
STP 547-1973
No ratings yet
STP 547-1973
79 pages
STP 576-1976
100% (1)
STP 576-1976
316 pages
STP 571-1975
No ratings yet
STP 571-1975
294 pages
STP 538-1973
No ratings yet
STP 538-1973
239 pages
STP 540-1988
No ratings yet
STP 540-1988
140 pages
STP 1594-2016
No ratings yet
STP 1594-2016
156 pages
STP 1590-2015
100% (1)
STP 1590-2015
264 pages
STP 266-1960
No ratings yet
STP 266-1960
253 pages
STP 1589-2017
No ratings yet
STP 1589-2017
306 pages
STP 491-1971
No ratings yet
STP 491-1971
90 pages
STP 169a-1966
No ratings yet
STP 169a-1966
582 pages
The Digitisation of The Sputter Deposition Process of Transparent Conductive Oxides by Implementing Artificial Intelligence
No ratings yet
The Digitisation of The Sputter Deposition Process of Transparent Conductive Oxides by Implementing Artificial Intelligence
153 pages
Mathematica Code For Critically Damped Oscillator
No ratings yet
Mathematica Code For Critically Damped Oscillator
1 page
Tifr Classical Mechanics
No ratings yet
Tifr Classical Mechanics
25 pages
RC2-2k17 (End Semester)
No ratings yet
RC2-2k17 (End Semester)
3 pages
CET205 SURVEYING AND GEOMATICS, December 2020
No ratings yet
CET205 SURVEYING AND GEOMATICS, December 2020
3 pages
Design 7
No ratings yet
Design 7
6 pages
Lab1 Vibration Mechanics
No ratings yet
Lab1 Vibration Mechanics
5 pages
Lecture 01 PDF
No ratings yet
Lecture 01 PDF
13 pages
1586 - Your Second SOLIDWORKS Flow Simulation Tips, Tricks, and Frequent Features
No ratings yet
1586 - Your Second SOLIDWORKS Flow Simulation Tips, Tricks, and Frequent Features
7 pages
Lesson 2.1 Arithmetic Sequence
No ratings yet
Lesson 2.1 Arithmetic Sequence
35 pages
Dokumen.pub Quantum Supereld Supersymmetry 9783030681357 9783030681364
No ratings yet
Dokumen.pub Quantum Supereld Supersymmetry 9783030681357 9783030681364
160 pages
Assumptions of The Kinetic Theory of Gases
No ratings yet
Assumptions of The Kinetic Theory of Gases
37 pages
Cat33 e
No ratings yet
Cat33 e
35 pages
VEM Standard LV K21 Series
No ratings yet
VEM Standard LV K21 Series
17 pages
CH 14 - Chemical Reaction Engineering
No ratings yet
CH 14 - Chemical Reaction Engineering
12 pages
Chemistry XI CH 01 Progressive Test
No ratings yet
Chemistry XI CH 01 Progressive Test
2 pages
Jaske 2000
No ratings yet
Jaske 2000
8 pages
ME251-Lab #333 WSU Fall 20143 Creep Test
No ratings yet
ME251-Lab #333 WSU Fall 20143 Creep Test
4 pages
Table of Content Radiation Physics and Chemistry - Vol 205, April 2023 - ScienceDirect - Com by Elsevier
No ratings yet
Table of Content Radiation Physics and Chemistry - Vol 205, April 2023 - ScienceDirect - Com by Elsevier
8 pages
NHAZCA InSAR Feasibility Tool
No ratings yet
NHAZCA InSAR Feasibility Tool
9 pages
Kailas Sarode
No ratings yet
Kailas Sarode
12 pages
CIVE1144 - Structural Analysis Lec 1 (2019)
No ratings yet
CIVE1144 - Structural Analysis Lec 1 (2019)
67 pages
Choosing An Optimum Sand Control Method PDF
No ratings yet
Choosing An Optimum Sand Control Method PDF
10 pages
The Performance of a Small-Scale Refrigeration Unit Lab 3334b
No ratings yet
The Performance of a Small-Scale Refrigeration Unit Lab 3334b
8 pages
4.1 - Understand Matrices
No ratings yet
4.1 - Understand Matrices
12 pages
Form 4 (Daily Lesson Plan)
No ratings yet
Form 4 (Daily Lesson Plan)
39 pages
MHR ONSci9 TR Unit2
No ratings yet
MHR ONSci9 TR Unit2
97 pages
Case Studies in Construction Materials: Najib N. Gerges, Camille A. Issa, Samer A. Fawaz
No ratings yet
Case Studies in Construction Materials: Najib N. Gerges, Camille A. Issa, Samer A. Fawaz
16 pages
Nonlinear Partial Differential Equations for Scientists and Engineers Second Edition Lokenath Debnath - The latest ebook is available for instant download now
No ratings yet
Nonlinear Partial Differential Equations for Scientists and Engineers Second Edition Lokenath Debnath - The latest ebook is available for instant download now
31 pages