Metal Fatigue What It Is, Why It Matters Pook 2007

@Seismicisolation
@Seismicisolation
METAL FATIGUE
@Seismicisolation
@Seismicisolation
SOLID MECHANICS AND ITS APPLICATIONS
Volume 145
Series Editor: G.M.L. GLADWELL

Department of Civil Engineering
University of Waterloo
Waterloo, Ontario, Canada N2L 3GI
Aims and Scope of the Series

The fundamental questions arising in mechanics are: Why?, How?, and How much?
The aim of this series is to provide lucid accounts written by authoritative researchers
giving vision and insight in answering these questions on the subject of mechanics as it
relates to solids.
The scope of the series covers the entire spectrum of solid mechanics. Thus it includes
the foundation of mechanics; variational formulations; computational mechanics;
statics, kinematics and dynamics of rigid and elastic bodies: vibrations of solids and
structures; dynamical systems and chaos; the theories of elasticity, plasticity and
viscoelasticity; composite materials; rods, beams, shells and membranes; structural
control and stability; soils, rocks and geomechanics; fracture; tribology; experimental
mechanics; biomechanics and machine design.
The median level of presentation is the first year graduate student. Some texts are
monographs defining the current state of the field; others are accessible to final year
undergraduates; but essentially the emphasis is on readability and clarity.
For a list of related mechanics titles, see final pages.
@Seismicisolation
@Seismicisolation
Metal Fatigue
What It Is, Why It Matters
by
LES POOK
University College London, UK
@Seismicisolation
@Seismicisolation
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-1-4020-5596-6 (HB)

ISBN 978-1-4020-5597-3 (e-book)
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Printed on acid-free paper
All Rights Reserved

© 2007 Springer
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
@Seismicisolation
@Seismicisolation
The author’s first car, a 1932 Riley 9 Falcon. Its many ailments included metal
fatigue failures in a gearbox selector fork, timing gear teeth, a half shaft, and a
suspension spring. Modern cars are much better in that metal fatigue failures
are very unusual.
@Seismicisolation
@Seismicisolation
Leslie Philip (Les) Pook was born in Middlesex, England in 1935. He ob-
tained a BSc in metallurgy from the University of London in 1956. He started
his career at Hawker Siddeley Aviation Ltd, Coventry in 1956. In 1963 he
moved to the National Engineering Laboratory, East Kilbride, Glasgow. In
1969, while at the National Engineering Laboratory, he obtained a PhD in
mechanical engineering from the University of Strathclyde. Dr Pook moved
to University College London in 1998. He retired formally in 1998 but re-
mained affiliated to University College London as a visiting professor. He is
a Fellow of the Institution of Mechanical Engineers and a Fellow of the In-
stitute of Materials, Minerals and Mining. Les married his wife Ann in 1960.
They have a daughter, Stephanie, and a son, Adrian.
@Seismicisolation
@Seismicisolation
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Historical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Experimental Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Fatigue Testing of Components and Structures . . . . . . . 8
2.2.2 Fatigue Testing of Laboratory Specimens . . . . . . . . . . . 9
2.2.3 Investigation of the Mechanisms of Metal Fatigue . . . . 10
2.2.4 Investigation of Fatigue Crack Paths . . . . . . . . . . . . . . . . 10
2.2.5 Fatigue Crack Propagation Rate Testing . . . . . . . . . . . . . 11
2.3 The Modern Era . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 Fracture Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.2 Servohydraulic Testing Equipment . . . . . . . . . . . . . . . . . 12
2.3.3 Influence of Computers . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.4 Standardisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Constant Amplitude Fatigue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 S/N (Wöhler) Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Scatter in Fatigue Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4 Mechanisms of Metal Fatigue . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4.1 Fatigue Crack Initiation . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4.2 Fatigue Crack Propagation . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4.3 Ratchetting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
@Seismicisolation
@Seismicisolation
viii Contents
3.5 Effect of Mechanisms on S/N Curves . . . . . . . . . . . . . . . . . . . . 31
4 Variable Amplitude and Multiaxial Fatigue . . . . . . . . . . . . . . . . . . 35

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Mathematical Description of Fatigue Loadings . . . . . . . . . . . . . 37
4.2.1 Constant Amplitude Fatigue Loadings . . . . . . . . . . . . . . 39
4.2.2 Gaussian Random Loadings . . . . . . . . . . . . . . . . . . . . . . . 39
4.2.3 Block Fatigue Loadings . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Miner’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3.1 Weighted Average Stress Range . . . . . . . . . . . . . . . . . . . 47
4.3.2 Damage Density Functions . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.3 Cycle Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 Standard Load Histories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.1 Applications of Standard Load Histories . . . . . . . . . . . . 54
4.4.2 Development of Standard Load Histories . . . . . . . . . . . . 55
4.5 Multiaxial Fatigue in Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.5.1 Multiaxial Fatigue Tests . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.5.2 Multiaxial Failure Criteria . . . . . . . . . . . . . . . . . . . . . . . . 62
5 Fatigue Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Failure Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3 Situations, Philosophies and Approaches . . . . . . . . . . . . . . . . . . 69
5.3.1 Situations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.2 Philosophies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.3 Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.4 Product Liability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.4.1 The Consumer Protection Act . . . . . . . . . . . . . . . . . . . . . 78
5.4.2 Enforcement Authorities . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.5 Safety Regulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.5.1 The General Product Safety Regulations . . . . . . . . . . . . 80
6 The Uncracked Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.2 Effect of Surface Finish . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.3 Effect of Mean Stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.3.1 Effect of Residual Stresses . . . . . . . . . . . . . . . . . . . . . . . . 88
6.4 Effect of Multiaxial Fatigue Loading . . . . . . . . . . . . . . . . . . . . . 88
6.5 Effect of Notches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.6 Bicycle Crankshaft Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
@Seismicisolation
@Seismicisolation
Contents ix
6.7 Scatter under Variable Amplitude Fatigue Loading . . . . . . . . . 96

6.8 Improvement of Fatigue Life . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7 The Cracked Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.2 Initial Crack Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.3 Final Crack Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.4 Constant Amplitude Fatigue Crack Propagation . . . . . . . . . . . . 105
7.4.1 Determination of Fatigue Crack Propagation Rates . . . 106
7.4.2 Factors Affecting Fatigue Crack Propagation Rates . . . 109
7.4.3 Threshold for Fatigue Crack Propagation . . . . . . . . . . . . 116
7.4.4 Overall Fatigue Crack Propagation Behaviour . . . . . . . . 117
7.4.5 Short Cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.4.6 Fatigue Life in the Presence of a Crack . . . . . . . . . . . . . 122
7.5 Variable Amplitude Fatigue Crack Propagation . . . . . . . . . . . . 124
7.5.1 Weighted Average Stress Range . . . . . . . . . . . . . . . . . . . 124
7.5.2 Interaction Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.6 Fractography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
7.6.1 Fatigue Crack Propagation in a Welded Joint . . . . . . . . 130
8 Fatigue Crack Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.1.1 Classes of Fatigue Crack Propagation . . . . . . . . . . . . . . . 137
8.2 Crack Paths in Two Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.2.1 Mixed Mode Thresholds . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.2.2 Crack Path Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.2.3 Crack Path Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
8.3 Planar Crack Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.3.1 Crack Path Development . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.3.2 Leak before Break . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
8.4 Crack Paths in Three Dimensions . . . . . . . . . . . . . . . . . . . . . . . . 155
8.4.1 Lower Bound Mixed Mode Thresholds . . . . . . . . . . . . . 157
8.4.2 Crack Path Determination . . . . . . . . . . . . . . . . . . . . . . . . 158
9 Why Metal Fatigue Matters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
9.2 Avoidance of Fatigue Failure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
9.3 Research and Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9.4 The Role of Non Destructive Testing . . . . . . . . . . . . . . . . . . . . . 164
9.5 Current Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
@Seismicisolation
@Seismicisolation
x Contents
A Fracture Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
A.2 Notation for Stress and Displacement Fields . . . . . . . . . . . . . . . 168
A.2.1 Crack Surface Displacement . . . . . . . . . . . . . . . . . . . . . . 168
A.2.2 Volterra Distorsioni . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
A.3 Stress Intensity Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
A.3.1 Stress Intensity Factor Solutions . . . . . . . . . . . . . . . . . . . 174
A.3.2 Validity of Stress Intensity Factors . . . . . . . . . . . . . . . . . 180
A.3.3 Effects of Small Scale Yielding . . . . . . . . . . . . . . . . . . . . 182
A.4 Corner Point Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
A.4.1 Crack Front Intersection Angle . . . . . . . . . . . . . . . . . . . . 189
A.4.2 Derivation of Stress Intensity Factors . . . . . . . . . . . . . . . 191
A.5 Stress Intensity Factors for Irregular Cracks . . . . . . . . . . . . . . . 195
A.5.1 Use of Semi Ellipses and Ellipses . . . . . . . . . . . . . . . . . . 196
A.5.2 Projection onto a Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
A.5.3 Crack Front Smoothing . . . . . . . . . . . . . . . . . . . . . . . . . . 199
A.5.4 Use of Crack Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
A.5.5 Interaction between Cracks . . . . . . . . . . . . . . . . . . . . . . . 200
B Random Load Theory and RMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
B.2 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
B.2.1 Random Process Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 204
B.2.2 Fatigue Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
B.3 Some Sinusoidal Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
B.3.1 Narrow Band Random Loading . . . . . . . . . . . . . . . . . . . . 208
B.3.2 Two Parameter Weibull Distribution . . . . . . . . . . . . . . . . 210
B.3.3 Incomplete Distributions and Processes . . . . . . . . . . . . . 212
B.3.4 Non Stationary Narrow Band Random Loading . . . . . . 216
B.4 Broad Band Random Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
B.4.1 Spectral Density Function . . . . . . . . . . . . . . . . . . . . . . . . 218
C Non Destructive Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
C.2 Visual Inspection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
C.3 Magnetic Particle Inspection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
C.4 Dye Penetrant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
C.5 Radiography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
C.6 Ultrasonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
C.7 Electromagnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
@Seismicisolation
@Seismicisolation
Contents xi
C.7.1 Eddy Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

C.7.2 Alternating Current Potential Drop . . . . . . . . . . . . . . . . . 236
C.8 Probability of Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
C.8.1 Determination of Probability of Detection . . . . . . . . . . . 240
C.8.2 Probability of Sizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
@Seismicisolation
@Seismicisolation
Preface
The first major book in English on metal fatigue was Fatigue of Metals
(Gough 1926). Since then numerous books have been published which are
primarily on this important topic. These range from general surveys, for
example Suresh (1998) to books on specific aspects of metal fatigue, for
example Radaj and Sonsino (1999) on assessment of welded joints, and
Murakami (2002) on the effects of small defects. Books on the related topic
of fracture mechanics, for example Broek (1988), usually include extensive
treatment of fatigue crack propagation. There are many other books where
metal fatigue is a major theme. These range from undergraduate texts, for
example Dowling (1993), to a 10 volume compilation on structural integrity
edited by Milne et al. (2003). There do not appear to be any recent books on
metal fatigue which are presented in a format that appeals to engineers, and
which can be recommended to newcomers to the topic.
Metal fatigue has both metallurgical and engineering aspects, and also its
own specialised jargon which newcomers often find confusing. In the last few
decades treatment of most aspects of metal fatigue has become much more
mathematical, with extensive use of computers. This book aims to present the
important ideas in metal fatigue in as straightforward a manner as possible
for the benefit of readers who need to be able understand more advanced doc-
uments on a wide range of metal fatigue topics. Indications on how metal
fatigue problems are solved in engineering practice are included. It is based
on 50 years experience of metal fatigue, and on 15 years experience of intro-
ducing engineering undergraduates to its basic ideas.
The prerequisite knowledge required for readers is a basic understanding
of stress analysis and mathematics covered in engineering undergraduate
courses. No prior knowledge of metal fatigue is assumed. The objectives of
the book are: to explain the terminology used in metal fatigue, to provide a
@Seismicisolation
@Seismicisolation
xiv Preface
brief description of the mechanisms of metal fatigue, to describe the basis of

design against the fatigue failure of components and structures, to provide
examples illustrating various points made, and hence to provide a firm
foundation for readers who wish to study various aspects of metal fatigue
in more detail, including finding solutions to specific fatigue problems. The
symbols used have been chosen to represent widespread practice but it has
not been possible to be entirely consistent. Other symbols are used by some
authors.
Les Pook
August 2006
@Seismicisolation
@Seismicisolation
Notation
I, II, III modes of crack surface displacement

A crack area
A, m constants in Basquin equation
a crack length, half crack length of internal crack, crack radius,
characteristic crack dimension, semi minor axis of ellipse
af final value of a
ai initial value of a
δa increment of crack propagation
B specimen thickness, biaxiality ratio
C, m constants in Paris equation
c half crack surface length, semi major axis of ellipse
E Young’s modulus
E(k) complete elliptical integral of the second kind
F force
f frequency
K stress intensity factor, subscripts I, II, III denote mode
KI∗ value of KI for a small Mode I branch crack
KIc plane strain fracture toughness
KB bending mode stress intensity factor
Kc fracture toughness, plane stress fracture toughness
Kf fatigue strength reduction factor
Kh weighted average range of KI
Kmax maximum value of KI in fatigue cycle
Kmin minimum value of KI in fatigue cycle
Kop crack opening value of KI
Kt stress concentration factor
KλA antisymmetric mode stress intensity measure
@Seismicisolation
@Seismicisolation
xvi Notation
KλS symmetric mode stress intensity measure

Kλ stress intensity measure
K range of KI in fatigue cycle
Ki value of K for ith cycle
Keff effective value of K
Kth threshold
value of K
k = 1 − a 2 /c2
M bending moment
m dimensionless magnetic permeability parameter
N number of cycles
Ni life at ith load level
n number of applied cycles
ni number of applied cycles at ith load level
P pressure
P cyclic pressure range
p(S) probability density of S
q notch sensitivity index
R stress ratio (= σmin /σmax ), stress ratio (= ni /Ni ), relative
damage
r region radius, radius of open crack tip, distance from crack tip
rch characteristic value of r
r, θ polar coordinates
r, θ, φ spherical coordinates
r(S/σ ) damage density of S/σ
S stress, stress range (= 2σa = σmax − σmin ), reference stress,
statistical variable
Sa alternating value of reference stress, peak stress in a Rayleigh
process
Sh weighted average stress range
Sm mean value of reference stress
T time interval, temperature, torque, stress parallel to crack
TR T -stress ratio
TRc critical value of TR
t time
V voltage
W specimen width
x, y, z Cartesian coordinates
u, v, w displacements in x, y, z directions
Y geometric correction factor
@Seismicisolation
@Seismicisolation
Notation xvii
α = a/W
α, β, γ rotations about axes parallel to x, y, z axes
β crack front intersection angle
βc critical crack front intersection angle
γ curvature factor
probe spacing
δ skin depth
δa increment of crack propagation
ε strain, reference strain
θ branch crack propagation angle
λ coefficient defining singularity, biaxial loading ratio
µ magnetic permeability
µ0 magnetic permeability of a vacuum
ν Poisson’s ratio
ρ equivalent stress
σ tensile stress, root mean square value of S, electrical
conductivity
σa alternating stress
σe equivalent stress
σm mean stress
σmax maximum stress in fatigue cycle
σmin minimum stress in fatigue cycle
σt tensile strength
σx , σy , σz stresses in x, y, z directions
σY yield stress
σ0 fatigue limit
σ1 , σ2 , σ3 principal stresses
σ stress range
σi value of σ for ith cycle
τxy , τyz , τzx shear stresses on xy, yz, zx planes
@Seismicisolation
@Seismicisolation
1
Introduction
The term metal fatigue refers to gradual degradation and eventual failure that
occur under loads which vary with time, and which are lower than the static
strength of the metallic specimen, component or structure concerned. The
static strength is the load which causes failure in one application. The loads
responsible are called fatigue loads. These loads are cyclic in nature, but the
cycles are not necessarily all of the same size or clearly discernible. A fatigue
load in which individual cycles can be distinguished is sometimes called a
cyclic load.
Metal fatigue is largely a descriptive subject, and as such it has accumu-
lated an enormous literature (Pook 1983a). Nevertheless, the basic concepts
needed for an understanding of the metal fatigue literature are reasonably
straightforward, and these are described in this book. The descriptions can
be divided into two groups, metallurgical and mechanical. Metallurgical de-
scriptions are concerned with the state of the metal before, during and after
the application of fatigue loads, and are usually taken to include the study of
metal fatigue mechanisms. Mechanical descriptions are concerned with the
mechanical response to a given set of loading conditions, for example the
number of load cycles needed to cause failure. Mechanical descriptions are
more useful from an engineering viewpoint, where service behaviour must be
predicted, and are therefore given more emphasis in this book.
Rigorous definition of exactly what is meant by metal fatigue is difficult
and not particularly helpful for its understanding. An early dictionary defini-
tion is: the condition of weakness in metal caused by repeated blows or long-
continued strain (Murray 1901). A more recent definition is: failure of a metal
under a repeated or otherwise varying load which never reaches a level suffi-
cient to cause failure in a single application (Pook 1983a).
@Seismicisolation
@Seismicisolation
2 Chapter 1: Introduction
Figure 1.1. Car drive line component. National Engineering Laboratory photograph. Repro-
duced under the terms of the Click-Use Licence.
From an engineering viewpoint, metal fatigue matters because it is a ma-

jor potential cause of failure of components and structures, including load
bearing consumer items. In a sense, the problem might appear to have been
largely solved, in that catastrophic failures due to metal fatigue are now rare.
Official inquiries into catastrophic metal fatigue failures, involving loss of life
or major financial loss, usually indicate a clear reason for the failure and often
indicate apparent human negligence. An example is the failure of a fairground
ride in Glasgow, which killed two people (Pook 1998).
Lesser metal fatigue failures are still common and cause a great deal of in-
convenience and expense. In other words, they are a nuisance rather than cata-
strophic. These lesser failures are often unrecognised as being due to metal
fatigue unless they happen to be seen be an expert. Figures 1.1–1.4 show a
selection of nuisance metal fatigue failures. They illustrate the point that what
is meant by failure depends on the function of component.
Figure 1.1 shows part of the car drive line component that had broken off
due to metal fatigue, and this meant that power was no longer transmitted
to the car wheels. The car had covered many miles in mountainous country,
so many power on, power off cycles, probably of the order of 106 , had been
applied. This particular failure led to some acrimonious correspondence with
the manufacturer over whether or not the failure, following use of the car
@Seismicisolation
@Seismicisolation
Metal Fatigue 3
Figure 1.2. Cracking in aircraft jet engine nacelle.
Figure 1.3. Cracked ring spanner. National Engineering Laboratory photograph. Reproduced
under the terms of the Click-Use Licence.
under unusually severe operating conditions, could be regarded as fair wear

and tear.
Figure 1.2 shows cracking in the aircraft jet engine nacelle found during
a routine inspection after 2000 hours flying (Pook 2004). The cracks were in
an area affected by jet noise. The nacelle was made from 0.56 mm thick clad
aluminium 4% Cu alloy. Acoustic fatigue due to jet noise is well known. If
it had been allowed to continue the nacelle would eventually have broken up
and would no longer have provided protection to the engine. The cracking
was due to jet noise so very many cycles had been applied. Failure analysis
suggested that failure had been caused by biaxial bending fatigue loading. In
2000 hours flying very many load cycles (of the order of 1010 ) would have
been applied.
Figure 1.3 shows a ring spanner with a fatigue crack that had propagated
right through the ring. This caused severe loss of stiffness, so the ring sprung
open and slipped when the user tried to tighten a nut. The spanner had been
used a lot but even so only a relatively small number of cycles, probably of
the order of thousands, could have been applied.
@Seismicisolation
@Seismicisolation
4 Chapter 1: Introduction
Figure 1.4. 9.5 mm long brass chain link from weight driven clock. National Engineering
Laboratory photograph. Reproduced under the terms of the Click-Use Licence.
The brass chain link from a weight driven clock, shown in Figure 1.4,
was bent into shape from a piece of brass wire, but the ends of the wire were
not joined. When a fatigue crack had propagated about half way through one
of the links, the section was so reduced that the link distorted due to plastic
deformation. The chain link still withstood the load due to the clock driving
weight, but the distortion meant that it would no longer pass through the clock
mechanism. The clock was wound daily, so one load cycle was applied per
day, with the total number of cycles of the order of 103 .
These examples show that failure can mean that a component no longer
performs its intended function because it has broken into two (or more) parts,
or because of loss of stiffness or distortion. They also illustrate the very wide
variation in the number of load cycles applied during the life of a component.
Fatigue also affects non metallic materials. For example, Figure 1.5 shows
a plastic domestic tap. It was observed to be leaking where it was screwed into
a fitting on the supply pipe. The tap had a fitting for a hose pipe and appeared
to be a replacement for the original brass tap which did not have a hose fitting.
When an attempt was made to unscrew the tap, it failed completely. The two
parts of the broken tap are shown in Figure 1.5. The dark area is where fatigue
crack propagation took place and the light area where the final static failure
took place. The age of the tap at the time of failure was unknown but as one
@Seismicisolation
@Seismicisolation
Metal Fatigue 5
Figure 1.5. Plastic domestic tap.
fatigue cycle is applied each time a tap is turned on and off the number of load
cycles applied was probably of the order of thousands. Pressure containing
components where safety is an issue, such as parts of hydraulic systems, are
often designed to leak before break in order to avoid catastrophic failure. It
is fortunate that the tap did leak otherwise the utility room in which it was
installed would probably have been flooded. The failed tap was replaced with
a brass tap, which has given satisfactory service.
Comparison of the plastic and brass taps showed that the detail design
in the vicinity of the failure was exactly the same. In retrospect the detail
design of the plastic tap should gave been changed to allow for the different
mechanical properties of the plastic. The episode is an example of the danger
of using a different material for a component, which is subject to fatigue
loads, without making appropriate changes to detail design.
@Seismicisolation
@Seismicisolation
This page intentionally blank
@Seismicisolation
@Seismicisolation
2
Historical Background
2.1 Introduction
Early in the history of engineering design, there was a recognition of the need
to know the different ways in which a material or component could fail. Fail-
ure was usually associated with fracture, or with excessive deformation. Fail-
ure under static loads, tensile, compressive and shear, became widely known.
Much early design aimed at making a component or a structure that would
last indefinitely.
Metal fatigue has been of interest for about 170 years. This interest dates
back to the development of the steam engine, mechanical transport, and the
more extensive use of mechanical devices. This mechanisation meant that
many components were subjected to fatigue loads, and fatigue failure was
beginning to become a common occurrence.
The history of metal fatigue, from an engineering viewpoint, is well docu-
mented but early references are often difficult to locate. Most books on metal
fatigue include a historical summary, usually concentrating on mechanical
descriptions. The best recent history, on mechanical descriptions of metal fa-
tigue, is by Schütz (1996). It includes over 500 references, mostly in English
and German.
The first use of the term fatigue in print appears to be by Braithwaite
(1854), although in his paper Braithwaite states that it was coined by a Mr
Field. The general opinion had developed (Frost et al. 1974) that the material
had tired of carrying the load, or that the continual re-application of a load
had in some way exhausted the ability of the material to carry load. The use
of the term has survived to this day. Since fatigue failures also occur in many
non metallic materials the term metal fatigue is often used, as in this book, to
@Seismicisolation
@Seismicisolation
8 Chapter 2: Historical Background
denote the particular kind of fatigue that occurs in metallic materials, com-
ponents and structures.
The first known catastrophic fatigue failure, involving major loss of life,
was the Versailles (France) railway accident in 1842 (Smith 1990). The train
was unusually long, with 17 carriages hauled by two steam engines. The front
axle of the leading, four wheeled engine failed due to metal fatigue and the
body of the leading engine fell to the ground. The second engine smashed it to
pieces. Following carriages passed over the wreck and some were set on fire.
This, and numerous other railway axle failures, led to extensive investigations
into the nature of metal fatigue (Parsons 1947, Smith 1990, Schütz 1996).
The first known, reasonably well documented, metal fatigue failures were
in clock mainsprings (Wayman et al. 2000). The use of uncoiling springs,
rather than descending weights, as a driving force was an important factor in
the development of clocks for general use, and appears to have started in the
early fifteenth century. By the late eighteenth century the technology for the
manufacture of durable watch and clock mainsprings was well established;
a detailed description of the state of the art of making watch springs was
published by Blakey (1780). Even so, high quality watches and clocks were
designed (and still are) so that a broken mainspring could easily be replaced.
This shows that metal fatigue failures were indeed a problem,
2.2 Experimental Work

Understanding of metal fatigue is based largely on the results of laboratory
experiments and on service behaviour. Much information was obtained be-
fore mechanisms of metal fatigue were understood, and there was no under-
lying central theme to correlate the data. Thus isolated sets of data became
almost metal fatigue folklore, and their relationship to other sets of data was
not obvious (Frost et al. 1974). Numerous theories, some seriously mislead-
ing (Schütz 1996), have been developed in attempts to explain experimental
results.
There are five main themes in experimental work on metal fatigue, which
started at various times. Partly for convenience, and partly because it is often
representative of service conditions, most experiments are carried out in air
at ambient temperatures.
2.2.1 FATIGUE T ESTING OF C OMPONENTS AND S TRUCTURES

Fatigue testing of components and structures started in the 1830s. The first
known fatigue test results were published in German by Albert in 1837. These
@Seismicisolation
@Seismicisolation
Metal Fatigue 9
results were cited by Schütz (1996) and by Frost et al. (1974). Albert con-
structed a machine for repeatedly proof-loading welded mine hoist chains,
continuing some tests up to 105 cycles.
The first fatigue test results published in English appear to be those by
Fairbairn (1864) on repeated bending fatigue tests on beams. He used a mech-
anism actuated by a water wheel to apply a load repeatedly to the centres of
6.7 m long wrought iron built up girders. His apparatus is illustrated in Gough
(1926), Timoshenko (1953) and Marsh (1988). The calculated static failure
load of the beam under a central load was 120 kN. Fairbairn found that re-
peated loads of 30 kN were insufficient to cause failure in 3 × 106 cycles,
while if a greater load was applied failure did occur at a lower number of
repeated loads. He concluded that there was a safe repeated load which could
be applied to such a structure. This load would either be sustained indefin-
itely, or the number of repetitions to failure would be so large as to exceed
the normal life of a bridge. Fairbairn estimated 12 × 106 repetitions as being
equivalent to a bridge life of 328 years, assuming that the loading is applied
100 times per day.
2.2.2 FATIGUE T ESTING OF L ABORATORY S PECIMENS
Fatigue testing of specially designed laboratory specimens started in the

1850s. It is generally accepted that the first fatigue tests on laboratory spe-
cimens were carried out by Wöhler. He published the results of his classical
experiments on metal fatigue, using both notched and unnotched specimens,
in a series of papers in German starting in 1858. These results are cited by
Schütz (1996). An account in English is given in Anon. (1871). Wöhler con-
structed various types of fatigue testing machines, and carried out the first
fatigue tests in which strict attention was paid to the magnitude of the applied
fatigue loads. An important point is that appropriate stress analysis had been
developed (Timoshenko 1953) and was available to Wöhler. This enabled
him to relate his experimental results to service stresses in railway axles, and
hence produce design rules which were incorporated in Technical Regulations
for German Railways (Schütz 1996). This did not require detailed knowledge
of the mechanisms of metal fatigue.
In 1870 Wöhler published a final report. This is cited in Schütz (1996) and
contains the following conclusions, known as Wöhler’s laws.
(a) Materials can be induced to fail by many repetitions of stress, all of which
are lower than the static strength.
@Seismicisolation
@Seismicisolation
(b) The stress amplitudes are decisive for the destruction of the cohesion of
the material. (In modern terminology this is the stress range, see Sec-
tion 3.1.)
(c) The maximum stress is of influence only in so far as the higher it is, the
lower are the stress amplitudes which lead to failure. (In modern termin-
ology this means that increasing the mean stress decreases the number of
cycles to failure, see Section 6.3.)
2.2.3 I NVESTIGATION OF THE M ECHANISMS OF M ETAL FATIGUE
Investigation of the mechanisms of metal fatigue started in earnest in the

1900s. During the nineteenth century there was much speculation on the
nature of metal fatigue ‘damage’ (Parsons 1947, Smith 1990). With the be-
nefit of hindsight, some of the theories advanced now appear bizarre. Extens-
ive metallurgical investigations, starting with Ewing and Humphrey (1903)
showed that metal fatigue damage in a plain specimen is a surface phe-
nomenon. They showed that, in a ductile metallic material, bands of slip lines
form on the surface of grains under fatigue loading. These slip lines eventu-
ally turn into cracks. In general only one crack develops to any considerable
depth, and this crack propagates across the material until the section is so re-
duced that static failure takes place. Detailed investigation of the mechanisms
of fatigue crack propagation in metals started in the 1960s. One of the first
significant papers was published by Forsyth (1961).
2.2.4 I NVESTIGATION OF FATIGUE C RACK PATHS
The complete solution of a fatigue crack propagation problem includes de-

termination of the crack path. At the present state of the art the factors con-
trolling the path taken by a propagating fatigue crack are not completely un-
derstood. In general, fatigue crack paths are difficult to predict (Pook 2002a).
Macroscopic aspects of fatigue crack paths have been of industrial interest
since the earliest fatigue investigations (Parsons 1947, Smith 1990). The book
by Cazaud published in 1948 includes an analysis of fatigue crack paths in
both laboratory specimens and industrial components. An English translation
of his book, including additional material, was published in 1953 (Cazaud
1953). In the last four decades there have been substantial advances in the
understanding and prediction of fatigue crack paths, largely through devel-
opments in fracture mechanics and in the application of modern computers.
Despite recent theoretical advances, fatigue crack paths in structures are still
often determined by large scale structural tests (Pook 2000a). At macroscopic
@Seismicisolation
@Seismicisolation
Metal Fatigue 11
scales differential geometry has been used in the interpretation of some fea-
tures of crack paths (Pook 2002a).
The microscopic examination of fatigue fracture surfaces started in the
1950s. The appearance of fatigue fracture surfaces in metals, at low magnific-
ation, had been of interest since the early days of service failure analysis and
of fatigue testing. The study of fracture surfaces is known as fractography.
This term appears to have been coined by Zapffe and Clogg (1945). The
optical metallurgical microscope was first used for the examination, at high
magnification, of fatigue fracture surfaces in metals by Zapffe and Worden
(1951). The use of fractography in metal fatigue research and development
rapidly became routine, for example Forsyth et al. (1959). By 1962 the use
of quantitative fractography in the reconstruction of crack path information
was well developed (Pook 1962) (see Section 7.6). At microscopic scales
there has been interest in the use of fractals (Mandelbrot 1983) and random
process theory (Pook 1976a, 2002a) in the characterisation of fatigue crack
paths.
2.2.5 FATIGUE C RACK P ROPAGATION R ATE T ESTING
Fatigue crack propagation rate testing, in both laboratory specimens and

structures, also started in the 1950s. The importance of fatigue crack propaga-
tion in metallic materials had been appreciated since about 1870. However,
the lack of an appropriate applied mechanics framework meant that little ef-
fort had been devoted to experimental determination of the laws governing
the rate of propagation of fatigue cracks in metals and structures. In fact no
accurate quantitative data had been published until around the time that Head
(1953, 1956) published his theoretically derived relationship between crack
length and number of stress cycles.
Following the catastrophic failures in Comet aircraft in the 1950s, major
structural fatigue tests were carried out on Comet aircraft structures (Anon.
1954) during which the propagation of fatigue cracks was monitored. At that
time it was not possible to relate such data on structures directly to the results
of fatigue crack propagation rate tests on laboratory specimens. However,
it was possible to correlate fatigue crack propagation data, obtained during
fatigue tests on structures, with crack propagation data obtained fractograph-
ically from the fatigue fracture surfaces (Pook 1960, 1962, Troughton et al.
1963). At about this time it also became clear that fatigue crack propaga-
tion does not take place if the applied fatigue load is too low. That is, there
is a threshold for fatigue crack propagation. This threshold exists because
a fatigue crack propagation rate of less than one lattice spacing per cycle is
@Seismicisolation
@Seismicisolation
not possible on physical grounds (Frost et al. 1971). However, because crack
propagation is not necessarily continuous along the whole crack front, av-
erage crack propagation rates of less than one lattice spacing per cycle are
sometimes observed.
2.3 The Modern Era
The modern era in metal fatigue research started around 1970 due to several
major developments (Pook 1983a). By this time the mechanisms of metal fa-
tigue were understood in general, but not necessarily in detail. A very large
amount of data had been accumulated, and there was a good general under-
standing of how to avoid fatigue failures in service. The state of the art of
metal fatigue in the early 1970s was summarised by Frost et al. (1974). At
the time, the book was unusual in that it was written in SI units, and a frac-
ture mechanics approach to fatigue crack propagation was used; both are now
universal.
2.3.1 F RACTURE M ECHANICS
The analysis and application of fatigue crack propagation rate and threshold
data, in both laboratory specimens and structures, became much easier with
the development of fracture mechanics through the pioneering work of Irwin
(Anon. 1965a, Rossmanith 1997). The fracture mechanics parameter stress
intensity factor provides a convenient single parameter description of the
elastic stress field in the vicinity of a crack tip (see Section A.3). One of
the first collections of fatigue crack propagation rate and threshold data for a
wide range of metallic materials, analysed in terms of stress intensity factors,
was published by Frost et al. (1971). Despite the apparent simplicity of labor-
atory fatigue crack propagation rate and threshold testing it was another ten
years before standard test methods started to appear (Anon. 1981a).
2.3.2 S ERVOHYDRAULIC T ESTING E QUIPMENT
The introduction of closed loop servohydraulic testing equipment (Marsh

1988, Schijve 2001) meant that almost any desired load history could be
applied to a component or structure. Service load simulation testing, using
appropriate load histories, is often the most cost effective means of determ-
ining service life and is sometimes a requirement of regulatory authorities.
It first became common in the aircraft industry, but is now used for critical
@Seismicisolation
@Seismicisolation
Metal Fatigue 13
components in a wide range of industries. Much effort has been devoted to the
development of standard load histories for fatigue testing purposes (Schütz
and Pook 1987, Schütz 1996)). A load history is sometimes called a load
spectrum.
2.3.3 I NFLUENCE OF C OMPUTERS
For a long time metal fatigue was largely a descriptive subject (Pook 1983a).
This has changed in the last four decades. The increasing power and soph-
istication of now ubiquitous desktop computers has meant an increase in the
application of numerical methods, sometimes based on sophisticated math-
ematics, to metal fatigue research and development. Much current work on
metal fatigue simply would not be possible without computers. In the Pre-
face to his book Statistics of Extremes (Gumbel 1958), which includes metal
fatigue applications, Gumbel states ‘Graphical procedures are preferred to te-
dious calculations.’ This is not surprising since at that time a typical desktop
calculator was an electro mechanical device which would not even extract
square roots automatically, and mainframe computers were in their infancy.
In the 1960s mainframe computers became more accessible, and started
to be used for metal fatigue related calculations such as Miner’s rule summa-
tions (see Section 4.3). Around 1970 user friendly programmable desk top
calculators, such as the HP 9100B, started to become available. As is well
known, computer development has continued apace and has now reached the
stage where entry level desk top computers are as powerful as many of the
mainframe computers in use in the 1980s. One of the consequences is that
approaches to metal fatigue have become much more mathematical.
2.3.4 S TANDARDISATION
Standard procedures of various degrees of formality are the most satisfactory

way of carrying out a metal fatigue assessment of a proposed design (Pook
1983a). These range from established good practice in a particular design
office to elaborate published procedures. These may be based on analytical
procedures, service experience, the results of fatigue tests, or some combin-
ation of these, and need not have a theoretically sound basis, provided that
they give sufficiently accurate answers. Very simple methods are often used
in the early stages of design. The development of standard procedures, which
can be a lengthy process, is probably the best way of assimilating the res-
ults of metal fatigue research into engineering practice. Most metal fatigue
@Seismicisolation
@Seismicisolation
assessments are based on simplified standard procedures, which despite their

apparent lack of physical validity, are known to give conservative answers.
In the modern era, the use of formal standards in many aspects of metal
fatigue has become more and more widespread. Such standards are often in-
corporated in software and are sometimes insisted upon by regulatory author-
ities. An attempt to list all the documents issued by the British Standards In-
stitution which contained fatigue clauses included more than 300 items (Pook
1989a). Lists of metal fatigue standards and related software are not included
in this book because such lists would date rapidly.
@Seismicisolation
@Seismicisolation
3
Constant Amplitude Fatigue
3.1 Notation
A constant amplitude fatigue loading (or constant amplitude loading) is a fa-

tigue loading in which all the load cycles are identical. A load cycle is usually
called either just a cycle or a fatigue cycle. The notation used is shown in Fig-
ure 3.1. A cycle is the smallest unit of the stress history which repeats exactly.
Cycles are often, but not always, sinusoidal. σa (sometimes σ , as in the figure)
is the alternating stress, σm is the mean stress, σmax is the maximum stress in
the load cycle, and σmin is the minimum stress in the load cycle. Mathem-
atically a load cycle (or stress cycle) is expressed as σm ± σa . Compressive
stresses are taken as negative. It follows that
σmax = σm + σa , (3.1)
σmin = σm − σa , (3.2)
σmax + σmin
σm = . (3.3)
2
The stress range is S = 2σa = σmax − σmin , and the stress ratio, R =
σmin /σmax . The term stress ratio and its symbol, R, are very well established,
and both are often used in the metal fatigue literature without explanation.
3.2 S/N (Wöhler) Curves
Very large numbers of constant amplitude fatigue tests have been carried out
on plain (unnotched) metallic specimens. Test results are sensitive to the sur-
face finish. Compilations of results are available, for example Shiozawa and
Sakai (1996). Many tests were on specimens of circular cross section tested
@Seismicisolation
@Seismicisolation
16 Chapter 3: Constant Amplitude Fatigue
Figure 3.1. Notation for constant amplitude fatigue loading.
in rotating bending. In rotating bending the mean stress is zero (stress ratio,
R = −1), and cycles are always sinusoidal. Results are presented in terms
of the nominal surface stress, which is calculated assuming that a specimen
remains wholly elastic. Figure 3.2 shows a typical specimen. This is designed
for testing in four point bending so that stress conditions are uniform along
the parallel portion. In his original tests Wöhler (see Section 2.2.2) used can-
tilever bending as shown in Figure 3.3. Two specimens were tested simul-
taneously and the load was applied by springs. In modern machines only
one specimen is tested and loads are usually applied by weights. For tests at
other stress ratios tests are usually carried out either on circular cross section
specimens tested in direct stress or on rectangular cross section specimens
tested in plane bending. Sinusoidal load cycles are normally used. Figure 3.4
shows a specimen with threaded ends for testing under direct stress. Tech-
niques are now being developed for the fatigue testing of very small speci-
mens (Connolley et al. 2005).
Conventionally, results are presented as S/N curves. These are plots of al-
ternating Stress versus Number of cycles to failure, with an appropriate curve
fitted through the individual data points. Sometimes, stress range is used; care
is needed when using data to check which convention has been used. Failure
is usually defined as the separation of a specimen into two parts, but other
@Seismicisolation
@Seismicisolation
Metal Fatigue 17
Figure 3.2. Plain fatigue specimen for testing in four point rotating bending.
Figure 3.3. Schematic diagram of Wöhler’s rotating cantilever bending fatigue testing ma-
chine.
Figure 3.4. Plain fatigue specimen for testing in direct stress.
definitions are sometimes used. For example, loss of a specified amount of

stiffness or the appearance of a crack of a specified size. S/N curves are
sometimes called Wöhler curves after August Wöhler (see Section 2.2.2).
The number of cycles to failure is sometimes called the life or the endurance.
It is usually plotted on a logarithmic scale, but the alternating stress may be
plotted on either a linear or a logarithmic scale.
Typical S/N curves are shown in Figures 3.5 and 3.6. As used to be con-
ventional (Frost et al. 1974) these S/N curves are for endurances of less than
108 cycles. The region where failure takes place in less than about 104 cycles
is called low cycle fatigue, and the region for longer endurances high cycle
fatigue. Arrows are attached to symbols to indicate that a specimen was un-
@Seismicisolation
@Seismicisolation
Figure 3.5. S/N curve for carbon steel specimens tested in rotating bending. Arrows attached
to symbols denote specimen unbroken (Frost et al. 1974).
broken when a test was discontinued. The reason why metal fatigue data are
conventionally presented in terms of numbers of cycles, rather than in terms
of time, is that for many metallic materials tested in air at room temperature
the number of cycles to failure is independent of the test frequency (Frost et
al. 1974). These materials are called frequency independent. Frequency inde-
pendence implies that the number of cycles to failure is also independent of
the waveform that connects positive and negative peaks.
It was usually found for steels having tensile strengths up to about
700 MPa that, if a specimen has not broken after 107 cycles, it was most
unlikely to break if tests were continued to longer endurances; tests on these
materials were therefore terminated at about 2 × 107 cycles. Tests on other
materials were often continued up to about 108 cycles. Figure 3.5 shows an
S/N curve obtained from a batch of carbon steel specimens tested in rotating
bending. The data showed that specimens either had a life less than 5 × 106
cycles or were still unbroken when the tests were stopped at 2–4×107 cycles,
and suggested that the line through the points became horizontal. When this
is so, the stress corresponding to the horizontal line is called the fatigue limit.
It was implied that specimens tested at below the fatigue limit would never
break no matter how many stress cycles were applied. If stress levels are plot-
ted on a logarithmic scale, it is often found that the finite life portion of the
S/N curve can be represented as a straight line given by the Basquin equation
(Basquin 1910)
N = Aσ m , (3.4)
@Seismicisolation
@Seismicisolation
Metal Fatigue 19
Figure 3.6. S/N curve for high-strength aluminium alloy specimens tested in rotating bending
(Frost et al. 1974).
where A and m are empirically determined constants. The intersection

between this straight line and the horizontal straight line is often referred to as
the knee. It is now widely known (Murakami 2002) that the S/N does not be-
come quite horizontal and that failures can occur in the region of 109 cycles.
This region is known as gigacycle fatigue, and was first observed by Shabalin
in 1958 (Romanovskaya and Botvina 2003). The importance of gigacycle fa-
tigue in practical engineering was pointed out by Gough (1926). He listed
estimated numbers of fatigue cycles applied in the lives of various structures
and components. These ranged up to 1.5 × 1010 cycles for a steam turbine
shaft. The suspension springs of clock pendulums are subjected to very large
numbers of cycles. For a typical long case clock pendulum, 10 years corres-
ponds to about 1.6 × 108 cycles (Matthys 2004); many long case clocks are
still in use after more than two centuries, corresponding to more than 3 × 109
cycles.
Not all metals gave S/N curves which exhibited a definite knee, even
when tests were continued to very long endurances (Frost et al. 1974). An
S/N curve for a high strength 4.5% Cu aluminium alloy is shown in Fig-
ure 3.6; failures were seen to occur at approaching 108 cycles. It is usual in
such cases to specify the fatigue strength of the material at a given endurance,
that is, specimens tested at higher stress levels will break before the stipulated
@Seismicisolation
@Seismicisolation
life, whereas those tested at lower stress levels will be unbroken after a life of
at least the stipulated value. The fatigue strength at the stipulated value is usu-
ally called the endurance limit. The slope of an S/N curve such as that shown
in Figure 3.6 is generally small at endurances in the region of 108 cycles, and
the stress for this life is typically used as an endurance limit.
Fatigue testing is very time consuming. For example, if a testing ma-
chine operates at 30 Hz then 108 cycles takes more than a month. With
the development of very high speed fatigue testing machines, operating at
up to 20 kHz, testing in the gigacycle region became practical (Bathias
2001, Stanzl-Tschegg 2002). It has been found, for a wide of metallic ma-
terials, that fatigue failures can occur at up to around 1010 cycles, although as
1010 cycles is approached the slope of an S/N curve is small.
At high stress levels, yielding takes place, and the S/N flattens, as shown
schematically in Figure 3.7(a). It may then be necessary to plot results in
terms of strain, ε, rather than stress (Figure 3.19(b). Other conventions are
sometimes used; for example, results may be plotted in terms of the maximum
stress in the fatigue cycle, σmax , rather than the alternating stress. For tests on
components, it is sometimes convenient to use the applied fatigue load rather
than a stress.
3.3 Scatter in Fatigue Data
By its very nature, metal fatigue is a random process, and the consequent
scatter of results, even in carefully controlled experiments, complicates both
the analysis of experimental data and their subsequent application to practical
problems. The amount of scatter in data, such as those shown in Figures 3.5
and 3.6, is greater than can be accounted for by experimental error. The de-
termination of an S/N curve involves subjective judgements when fitting a
curve to the individual data points. Statistical methods provide a rational ap-
proach to this type of problem, but do not of themselves either avoid the need
for subjective judgement at some stage or increase the amount of information
present in a given set of data. Statistical theory provides information on the
most efficient use of a limited number of test specimens, and on the num-
ber of test specimens required to give a specified degree of confidence in test
results (Anon. 2003a).
If a number of nominally identical fatigue specimens are tested at the
same stress amplitude and the lives tabulated, a histogram may be plotted
by dividing lives into groups, of a fixed width, distributed about the mean
(average) value. If a large enough number of specimens is tested, the groups
@Seismicisolation
@Seismicisolation
Metal Fatigue 21
Figure 3.7. Schematic diagrams showing S/N curves plotted in terms of (a) stress and
(b) strain.
may be made sufficiently narrow for the histogram to be represented by a

smooth curve. One function that has been found to give an adequate fit to
certain data is the Gaussian distribution or (Normal distribution), which is
given by
2
1 −S
p(S) = √ exp , (3.5)
σ 2π 2σ 2
where S is the variable involved, p(S) is the probability density, and σ is the
root mean square (RMS) value for all values of S, measured from the mean,
usually called the standard deviation. This is a measure of the width of the
scatter band.
Distributions of fatigue lives of metallic specimens do not conform to the
Normal distribution, but it has been shown that, in many cases, the Normal
distribution is a good approximation to the distribution of the logarithms of
lives of metallic specimens (Bastenaire 1963, Frost et al. 1974). This is called
@Seismicisolation
@Seismicisolation
Figure 3.8. Distribution of fatigue lives of plain copper specimens at various stress levels
(Frost et al. 1974). (a) 150 specimens, ±88 MPa. (b) 148 specimens, ±90 MPa. (c) 133 spe-
cimens, ±97 MPa. (d) 200 specimens, ±109 MPa. (e) 350 specimens, ±116 MPa. (f) 100
specimens, ±131 MPa.
the log Normal distribution. Other distributions are sometimes used in mod-
ern work, for example in Murakami (2002) and Schijve (2005). Large num-
bers of specimens have sometimes been tested in attempts to find the precise
form of the probability density function in given circumstances. Some typical
results are shown in Figure 3.8 (Frost et al. 1974).
A basic concept of statistics is that a group of one or more specimens is
merely a sample taken from a large body or population. Such a sample is
considered to be just one of a number of samples that could be tested. The
results obtained from tests on a random sample from the population can be
used to estimate the characteristics of the whole population and to measure
@Seismicisolation
@Seismicisolation
Metal Fatigue 23
Figure 3.9. Schematic P-S-N diagram for three probabilities of failure and log Normal distri-
butions of lives.
the reliability of the estimates. The values of the parameters of the population
can only be estimated from tests on the sample. To obtain exact values would
require the whole population to be tested. These estimates of the behaviour of
the population from tests on a sample, and the confidence that can be placed
on them, are the essence of statistical analysis in general and the statistical
analysis of metal fatigue specimens in particular (Anon. 2003a).
If batches of specimens of a particular metallic material are tested at dif-
ferent stress levels, S/N curves for different probabilities of failure can be
drawn; 50 per cent probability corresponds to the median, that is the middle-
most life, and is the conventional S/N curve. These curves are sometimes
referred to as P-S-N curves. Figure 3.9 shows schematic P-S-N curves for
failure probabilities of 1, 50 and 99 per cent. Scatter in log endurance in-
creases as the fatigue limit is approached, as indicated in the figure.
Experimental data on the scatter of the applied alternating stress required
to cause failure at a specified life cannot be obtained directly. This is be-
cause the applied alternating stress is the independent variable and specimen
life is the dependent variable. However, indirect methods suggest that for
a specified life the distribution of applied alternating stresses is approxim-
ately Normal (Bastenaire 1963, Ford et al. 1963). Further, when alternating
stresses are normalised by the alternating stress for a probability of failure
of 50 per cent their standard deviation is approximately independent of the
specified life. Table 3.1 shows normalised stresses for various probabilities of
failure assuming a Normal distribution and a standard deviation for normal-
@Seismicisolation
@Seismicisolation
Table 3.1. Normalised stresses for various probabilities of failure.

Probability of failure Normalised stress
(Per cent)
0.01 0.702
0.1 0.752
1 0.814
5 0.864
10 0.897
50 1
90 1.103
95 1.132
99 1.186
Table 3.2. Metallurgical feature scales.

Scale (mm) Metallurgical feature
10−6 Ions, electron cloud
10−5 Dislocations
10−4 Subgrain boundary precipitates
10−3 Subgrain slip band
10−2 Grains, inclusions, voids
ised stresses of 0.08, which is a typical value. In design against metal fatigue
very low probabilities of failure are of interest. The important point is that the
normalised stress is much lower for low probabilities of failure.
3.4 Mechanisms of Metal Fatigue
Mechanisms responsible for the failure of metals, and associated metallur-

gical features, can be viewed at a wide range of scales (McClintock and Irwin
1965) as listed in Table 3.2. Metal fatigue mechanisms are relatively simple
when viewed at the largest scale in the table, but are more complex when
viewed at smaller scales. This is typical of a wide range of physical phe-
nomena, where different results are obtained by viewing at different scales
(Mandelbrot 1977).
In essence, the mechanisms involved in the fatigue failure of a plain metal-
lic specimen with a polished surface are simple (Frost et al. 1974), although
the details may be complex, especially when viewed at smaller scales, and
may vary between different metals. Firstly, a fatigue crack is initiated at the
specimen surface, secondly the crack propagates slowly across the specimen
and finally, when the net cross section has been sufficiently reduced, a static
failure takes place on the final load cycle. In the presence of cracks or crack
@Seismicisolation
@Seismicisolation
Metal Fatigue 25
Table 3.3. Stress analysis feature scales.

Scale (mm) Stress analysis feature
10−1 Large plastic strains
1 Elastic-plastic field
10 Stress intensity factor
100 Component or specimen
like defects, such as some types of inclusions, the crack initiation phase is
largely absent.
Much of the scatter observed in fatigue behaviour arises because on a
microscopic scale metal fatigue is a very irregular process. In considering
mechanisms in detail a metallic material cannot be regarded as a homogen-
eous continuum. However, from an engineering viewpoint, a simple approach
to mechanisms is sufficient for the understanding of metal fatigue. Associ-
ated stress analyses, used as an applied mechanics framework for the study of
metal fatigue mechanisms, are usually at the larger scales shown in Table 3.3.
3.4.1 FATIGUE C RACK I NITIATION
Fatigue crack initiation in a ductile metal is a consequence of reversed plas-

ticity within a grain on a scale of 10−3 mm (Table 3.2). Surface grains are
weakest, they deform plastically at the lowest stress and this leads to the
production of a microcrack within a grain. Such microplasticity, due to slip
within grains, can occur at stresses much lower than the tensile yield stress
(Cottrell 1964). Slip can take place only on certain crystallographic planes
within a grain. Resistance to crack initiation depends strongly on surface
roughness, residual stress and environment, all of which are difficult to con-
trol.
Under a uniaxial tensile stress a plane of maximum shear stress is at 45◦ to
the surface, so slip takes place on favourably oriented slip planes at about this
angle. Under a unidirectional stress this leads to steps on the surface, as shown
schematically in Figure 3.10. An important point is that slip takes place on a
number of parallel planes. Slip which takes place when a load is applied is
not simply reversed when the load is removed. Under fatigue loading reverse
slip takes place on nearby planes leading to the development of intrusions
and extrusions (Figure 3.10). Eventually, surface cracks are produced under
these small amounts of reverse plastic strain (Forsyth 1969). Because slip
lines traverse a grain, the cracking is transgranular rather than intergranular.
@Seismicisolation
@Seismicisolation
Figure 3.10. Formation of surface cracks by slip (Pook 1983). Reproduced under the terms of
the Click-Use Licence.
Figure 3.11. Schematic section through a fatigue fracture showing the three stages of crack
propagation.
3.4.2 FATIGUE C RACK P ROPAGATION
In the metal fatigue literature the terms fatigue crack propagation and fatigue
crack growth are both used for the increase in size of a fatigue crack. In this
book fatigue crack propagation is used. In the 1950s it became clear that
fatigue crack propagation in metals is a two stage process (Forsyth 1961,
1969), as shown schematically in Figure 3.11.
In Forsyth’s notation, Stage I crack propagation is an extension of the ini-
tiated microcrack without change of direction (see previous section). Hence a
Stage I crack is a crack propagating within a slip band, which is on a plane of
high shear stress. In fracture mechanics terms it is a mixed mode crack (see
Section A.2.1). Stage I crack propagation is encouraged by plasticity (Pook
2002a). For convenience, the term fatigue crack initiation sometimes includes
Stage I fatigue crack propagation. A Stage I crack becomes a Stage II crack
@Seismicisolation
@Seismicisolation
Metal Fatigue 27
Figure 3.12. Striations on the fracture surface of a Al-4%Cu alloy, direction of propagation
left to right, spacing approximately 2 µm.
when it reaches a critical length, changes direction and propagates normal to

the maximum principal tensile stress. In fracture mechanics terms it is Mode
I (see Section A.2.1). The critical length is strongly dependent on microstruc-
tural features and on stress conditions, and varies widely. It is usually less than
1/4 mm and typically around 0.02 mm. After the transition a Stage II crack
propagates through the majority of the cross section (see Sections 7.1, 8.1
and A.2.1). More descriptive terms for Stage I and II, microcrack and macro-
crack respectively, are sometimes used. In current metal fatigue literature the
terms Stage I crack, Stage II crack, microcrack and macrocrack are usually
used without explanation. Similarly, unqualified references to fatigue crack
propagation in this book and (usually) elsewhere are to macrocrack propaga-
tion. Finally, the cross section is so reduced that conditions for failure in one
load cycle are satisfied. This is sometimes called Stage III and can be due to
crack propagation by brittle fracture or due to ductile collapse.
Under microscopic examination the most striking feature of many of the
fracture surfaces created by a propagating Stage II fatigue crack in a metallic
material is the presence of distinct line markings, parallel to each other, and
normal to the local direction of crack propagation. These lines are called stri-
ations (Pook 1983a, Broek 1988). Figure 3.12 shows striations on the fracture
surface of a Al-4%Cu alloy. Each striation corresponds to one stress cycle,
hence the distance between striations is the amount the crack front has moved
forward during one cycle. This one to one correspondence between striations
and applied stress cycles has been demonstrated experimentally many times
(Pook and Smith 1979). On a microscopic scale fatigue crack propagation in
a metallic material is a very irregular process. The presence of striations is
evidence that a crack was created by fatigue crack propagation. Striations are
not always observed, so their absence does not necessarily imply that fatigue
crack propagation did not take place.
@Seismicisolation
@Seismicisolation
Figure 3.13. Repeated sequence of crack opening and closing under a fatigue loading of 0 to
σmax . The sequence (a) to (e) is repeated for each successive cycle (Pook and Frost 1973).
Reproduced under the terms of the Click-Use Licence.
The mechanism of Stage II fatigue crack propagation, and striation form-

ation involves ductile deformation at the crack tip as it opens and closes.
This mechanism is shown schematically in Figure 3.13 for a crack of initial
length a, and a fatigue loading of 0 to σmax . When a crack is loaded the tip
blunts, and fatigue crack propagation takes place in a ductile metal because
unloading resharpens the crack tip on each cycle, leading to an increment of
crack propagation δa, as shown schematically in the figure. In other words,
fatigue crack propagation is a consequence of irreversible plastic deformation
at the crack tip, and this basic mechanism is responsible for the production
of striations. The increment of propagation in each cycle is related to this
mechanism (see Section 7.4.2.1).
The philosophical basis for theoretical analysis of crack propagation is
that for a crack growth to propagate two conditions need to be satisfied
(Cottrell 1964). Firstly, sufficient energy needs to be available to operate a
crack propagation mechanism (thermodynamic criterion), and secondly crack
tip stresses must be high enough to operate the mechanism (stress criterion).
The mechanism shown in Figure 3.13 satisfies both these conditions. The
nature of the crack tip stress field ensures that adequate stresses are avail-
able to operate the mechanism (see Section A.3). The energy absorbed by
@Seismicisolation
@Seismicisolation
Metal Fatigue 29
Figure 3.14. Schematic appearance of metal fatigue fracture surfaces for lives of around
(a) 104 cycles, (b) 106 cycles and (c) 109 cycles.
plastic deformation at the crack tip, and by creation of new fracture surface,
is provided by the cyclic loading. These criteria at not satisfied at below the
threshold for fatigue crack propagation (see Section 7.4.3).
It is often easy to recognise a metal fatigue failure from the macroscopic
appearance of the fracture surface, and it is important to be able to do so.
General features of metal fatigue fracture surfaces are shown schematically
in Figure 3.14 for a plain specimen and depend upon the number of cycles to
failure (Bathias 2001). The white areas represent fatigue crack propagation,
and these have a smooth appearance. The shaded areas represent the final
static failures; these have a rough appearance and are usually darker. At high
stress levels, with lives of around 104 cycles, fatigue cracks initiate at several
points on the specimen surface, and the individual fatigue cracks eventually
merge (Figure 3.14(a)). At low stress levels, with lives of around 106 cycles,
there is usually just one dominant fatigue crack that initiated at the surface
(Figure 3.14(b)). In the gigacycle region, with lives of around 109 cycles, the
fatigue crack may propagate from an internal defect (Figure 3.14(c)). A small
circular area around the initial defect usually has a dark appearance. This is
called a fish eye (Murakami 2002).
Some examples of metal fatigue service failures are shown in Figures 1.1
and 3.15, and in laboratory tests in Figures 3.16 and 3.17. These illustrate
various possibilities. The fatigue crack in the drive line component (Fig-
ure 1.1) initiated at a stress concentration (see Section 6.5). Failure in the
turbine blade was at mild stress concentration (Figure 3.15). The aircraft spar
boom failed inside the fastening bolt hole on both sides (Figure 3.16). The
rough areas are static fractures from where the spar boom was broken open
for examination after testing. The fatigue crack in the cast steel specimen ini-
tiated internally at a deliberately introduced shrinkage defect (Figure 3.17).
The fracture surface has a darker appearance after the propagating crack in-
tersects the specimen surface.
@Seismicisolation
@Seismicisolation
Figure 3.15. Marine steam turbine blade, width 18 mm. National Engineering Laboratory
photograph. Reproduced under the terms of the Click-Use Licence.
Figure 3.16. Al-Cu alloy aircraft spar boom, width 85 mm.
Figure 3.17. Cast steel specimen, diameter of light area 8 mm. National Engineering Labor-
atory photograph. Reproduced under the terms of the Click-Use Licence.
3.4.3 R ATCHETTING
At high stress levels, with a mean stress of above zero, a phenomenon known
as ratchetting may occur (Dowling 1993, Suresh 1998). What happens is that
@Seismicisolation
@Seismicisolation
Metal Fatigue 31
plastic deformation occurring during loading is not fully reversed during un-
loading and this ratchetting may be repeated on successive cycles. Depending
on circumstances ratchetting may slow down and stop, continue at a constant
rate, or accelerate. If ratchetting slows down and stops it is known as shake-
down. Ratchetting is sometimes observed in watch and clock mainsprings
which, after a long period of use, take a permanent set and no longer provide
sufficient power to drive the clock work (Camm 1941). Ratchetting can also
occur under thermal loading where local heating and cooling results in fa-
tigue stresses due to expansion and contraction of the metal. Ratchetting due
to thermal loading is often observed in cooking utensils, which sometimes
become unusable because of excessive deformation.
3.5 Effect of Mechanisms on S/N Curves
The total fatigue life is the sum of the number of cycles for fatigue crack
initiation, taken as including Stage I fatigue crack propagation, plus the num-
ber of cycles for Stage II fatigue crack propagation (see Section 3.4.2). It is
generally accepted that most of the scatter in the lives of metallic specimens
tested at a given stress amplitude is associated with crack initiation (Frost et
al. 1974). This apparently contradicts the statement that Stage II fatigue crack
propagation is a very irregular process (see Sections 3.4.2, 7.4.1 and 7.6.1).
In tests conducted near the fatigue limit on small specimens, up to about 90
per cent of the fatigue life is occupied by fatigue crack initiation. In tests at
higher stresses, or on large specimens and structures the importance of Stage
II fatigue crack propagation increases. Since the scatter associated with Stage
II fatigue crack propagation is less than that associated with crack initiation,
the overall scatter is reduced.
The distribution of fatigue lives of plain specimens is an additive distri-
bution made up from the distributions associated with fatigue crack initiation
and with Stage II fatigue crack propagation. These two distributions are stat-
istically independent. Data such as those shown in Figure 3.8 must therefore
be interpreted in this light. At high stress levels, the number of cycles needed
to initiate a crack is negligible, so that the S/N curve has the relatively nar-
row scatter in life associated with Stage II fatigue crack propagation. At low
stress levels, the number of cycles needed to propagate the crack is negligible,
so that the S/N curve shows the wide scatter in fatigue life associated with
fatigue crack initiation. At intermediate stresses, crack initiation and crack
propagation are of roughly equal importance and the resultant additive distri-
bution can be bimodal (two-humped). Bimodal distributions are often asso-
@Seismicisolation
@Seismicisolation
Figure 3.18. Schematic diagram showing the S/N diagram for a combination of S/N curves
for fatigue crack initiation and for Stage II crack propagation.
ciated with two different mechanisms. Thus, the data may be interpreted as
shown schematically in Figure 3.18. Unfortunately, it is not possible to sep-
arate an additive distribution into its components without prior knowledge of
the components (Papoulis 1965).
Metallurgical defects are inevitably present in plain specimens, and some-
times Stage II fatigue crack propagation starts directly from a small in-
ternal defect. Figure 3.17 shows an extreme example. This effect is most
noticeable in high strength metallic materials at long endurances (Bathias
2001, Murakami 2002). When this happens, there may be separate S/N
curves associated with fatigue crack initiation, followed by Stage II fatigue
crack propagation, and with Stage II fatigue crack propagation directly from a
defect. Fatigue crack initiation followed by Stage II fatigue crack propagation
is associated with high stresses, and Stage II fatigue crack propagation from
a defect with low stresses. This is shown schematically in Figure 3.19, where
the underlying distributions are assumed to be log Normal. At stress levels
where the scatter bands do not overlap, failure is entirely by the mechanism
corresponding to the shorter life. In the vicinity of the crossover point the
scatter bands overlap, and a proportion of the specimens will survive, and fail
by the mechanism corresponding to longer life. As only a proportion of the
specimens survive to fail by the longer life mechanism, the two distributions
corresponding to the two mechanisms are not statistically independent. Hence
the longer life mechanism has a conditional distribution (Papoulis 1965). The
@Seismicisolation
@Seismicisolation
Metal Fatigue 33
Figure 3.19. Schematic diagram showing S/N curves associated with fatigue crack initiation
followed by Stage II Fatigue crack propagation, and with Stage II fatigue crack propagation
from a small defect.
resulting distribution is therefore skewed, as in the figure, rather than bimodal.

The overall S/N curve has a discontinuity at the crossover point.
In welded joints, which contain small crack-like slag inclusions, the whole
fatigue life may be occupied by Stage II fatigue crack propagation (see Sec-
tions 7.4.6.1 and 7.6.1).
In testing structures it is sometimes found that one fracture mode predom-
inates at high stresses and another at low stresses, and an effect similar to
that shown in Figure 3.18 is observed (Frost et al. 1974). For example, in the
testing of riveted sheet structures, rivet failure is associated with high stresses
and sheet failure with low stresses (Ryman 1962).
@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
4
Variable Amplitude and Multiaxial
Fatigue
4.1 Introduction
Early fatigue tests on structures, components and test specimens were all car-
ried out using constant amplitude fatigue loading. However, many structures
and components are subjected in service to variable amplitude fatigue load-
ing (or variable amplitude loading), with variations following either a regular
or a random pattern (Frost et al. 1974). Variable amplitude fatigue loadings
can divided into two broad classes (Pook 1979): those in which individual
load cycles can be distinguished, such as narrow band random loading (Fig-
ure 4.1), and those in which individual load cycles cannot be distinguished,
such as broad band random loading (Figure 4.2). Modern fatigue testing
equipment makes it possible to apply virtually any desired load history, and
narrow and broad band random loading have been used extensively for some
time, especially in structural testing (Pook 1983a, Marsh 1988). The use of
broad band random loading can make the analysis of test results difficult
(Smith 1965). Concepts taken from random process theory (Papoulis 1965,
Bendat and Piersol 1971, 2000) are used in the characterisation of random
load histories (see Section B.2.1). Conventions used in the fatigue testing lit-
erature sometimes differ from those usual in random process theory. Although
referred to as random, load histories used in tests are usually pseudo random
in that they repeat exactly after a return period.
Many fatigue tests on test specimens, components and structures are car-
ried out using a simple, uniaxial loading. More complex fatigue loadings are
known as multiaxial fatigue loadings (or multiaxial loadings). Many com-
ponents and structures are subjected to multiaxial fatigue loading in service.
A multiaxial fatigue loading may be defined as one in which more than one
stress system is produced at specific points in the object being loaded. This
@Seismicisolation
@Seismicisolation
36 Chapter 4: Variable Amplitude and Multiaxial Fatigue
Figure 4.1. Narrow band random loading, frequency ∼100 Hz, irregularity factor ∼0.99 (Pook
1987a). Reproduced under the terms of the Click-Use Licence.
Figure 4.2. Broad band random loading, irregularity factor 0.410 (Pook 1987a). Reproduced
may be because more than one fatigue load is applied, or because of the nature
of the load. Where more than one fatigue load is applied these are not neces-
sarily proportional, or in phase.
The crankshaft in a bicycle is an example of a component subjected to
multiaxial fatigue loading. Figure 4.3 shows a crankshaft, bearing and hous-
ing assembly removed from a bicycle. The bearings are not visible in the
photograph. When the rider pushes on a pedal, bending moments and torques
are produced in the crankshaft. When the left hand pedal is pushed, a torque
@Seismicisolation
@Seismicisolation
Metal Fatigue 37
Figure 4.3. Bicycle crankshaft assembly, length 155 mm.
Figure 4.4. Bicycle chain wheel and crank.
passes through the crankshaft into the chain wheel. However, when the right
hand pedal is pushed bending moments and torques are produced, but torques
pass directly into the chain because the crank is attached to the chain wheel
(Figure 4.4), and there is no torque in the crankshaft.
The term biaxial fatigue loading (or biaxial loading) is often used in con-
nection with sheets. It is a special case of multiaxial fatigue loading. Fig-
ure 4.5 shows a schematic diagram of a rig for carrying out biaxial fatigue
tests on sheet specimens. The two actuators are controlled independently.
4.2 Mathematical Description of Fatigue Loadings
A precise mathematical description of a fatigue loading is sometimes needed,

especially for some types of variable amplitude fatigue loading. One defin-
ition of metal fatigue is: Failure of a metal under a repeated or otherwise
@Seismicisolation
@Seismicisolation
Figure 4.5. Schematic diagram of biaxial fatigue testing rig for thin sheet specimens, actuators
are controlled independently. National Engineering Laboratory diagram. Reproduced under
the terms of the Click-Use Licence.
varying load which never reaches a level sufficient to cause failure in a single
application (Pook 1983a). This definition was intended to exclude quasi static
tests such as tensile and creep tests, although the boundary between fatigue
tests and other types of test is not always clear (see Chapter 1).
Mathematically, a fatigue loading is an example of a stochastic process
(Papoulis 1965, Bendat and Piersol 1971, 2000), and may be defined in terms
of a varying force, F , which is applied as a function of time, F (t) in some
time interval. Restricting the varying force to cases where there is at least one
maximum and one minimum in the intervals considered excludes most quasi
static tests. Alternatively, a fatigue loading can be expressed in terms of a
reference stress, S, at a reference point or region such as the test section of a
plain test specimen (Figures 3.2 and 3.4). A fatigue loading may be defined in
terms of the time function S(t). In many metal fatigue applications, material
behaviour can be considered to be linearly elastic so that S is proportional to
F , and the two approaches are equivalent (Frost et al. 1974).
@Seismicisolation
@Seismicisolation
Metal Fatigue 39
4.2.1 C ONSTANT A MPLITUDE FATIGUE L OADINGS

In a constant amplitude fatigue loading individual cycles can be distinguished
(see Figure 3.1) and S varies in a regular cyclic manner. Cycles are often
sinusoidal so that
S(t) = Sm + Sa sin(2πf t), (4.1)
where Sm is mean reference stress, Sa alternating reference stress, f fre-
quency of load application and t time. The process is statistically stationary in
that the coefficients Sm , Sa and f are unaffected by a shift of the origin along
the time axis, that is they are independent of time. As the process is cyclic the
continuous function S(t) can be replaced by the continuous function S(N),
where N is the number of cycles and N = t/f . In metal fatigue, it is the
maxima and minima, and the number of cycles (rather than time) which are
the main controlling parameters. The shape of the intervening curve between
a maximum and a minimum is of little importance (Frost et al. 1974). It is
usual to regard a cycle as the interval between two successive minima, and to
consider only integral numbers of cycles. Hence S(N) may be written as the
discontinuous function
S(N) = Sm ± Sa , (4.2)
which is understood to give successive maxima and minima.
4.2.2 G AUSSIAN R ANDOM L OADINGS

Many of the random loadings encountered in fatigue work are statistically
stationary, at least in the short term, and instantaneous values are a close
approximation to the Gaussian distribution or Normal distribution. Charac-
terisation of a Gaussian random loading using random process theory is well
established (Bendat and Piersol 2000). Random process theory is also applic-
able where two or more different Gaussian random loadings are applied in
multiaxial fatigue loadings.
A Gaussian random process, such as that shown in Figure 4.2 may be
described by the function S(t), where S is a random process and t is time. In
metal fatigue, S will be a quantity such as stress or load. Considering the time
interval 0 to T , the root mean square (RMS) value of the process is given by

1 T 2
σ = lim(T → ∞) S (t) dt . (4.3)
T 0
In random process theory, values of S are measured from zero. In metal fa-
tigue values of S are often measured from the mean value of the process.
@Seismicisolation
@Seismicisolation
When the mean value is not zero the two conventions give different answers.
For example the RMS of 2 and 14 is (22 + 142 )/2 = 10. However, if one
starts from the mean value of 2 and 14, which is 8, and takes the RMS of
(2 − 8 = −6) and (14 − 8 = 6) the answer is 6.
In this section, values of S are measured from its mean value. In other
words, without loss of generality, mean values are taken as zero. This is often
done in the metal fatigue literature because it simplifies equations. When a
mean value is non zero its value is stated in accompanying text and equa-
tions are not rewritten. Other definitions of RMS are sometimes used in the
metal fatigue literature, which can make it difficult to compare data produced
by different authors. Some of these definitions of RMS are given in Sec-
tions B.2.1 and B.3.3.1, and various practical points which arise in the cal-
culation of RMS values are discussed there. Numerical methods are usually
required (Kreyszig 1983).
The form of a Gaussian random process is usually characterised by the
probability distribution of its instantaneous values, and also by the spectral
density function, which is a measure of the frequency content, or bandwidth,
of the process (see Section B.4.1). The irregularity factor, which is the ratio
of zero crossings to peaks, and lies in the range 0 to 1, is also used. The irreg-
ularity factor can be obtained directly from a time history, this is usually done
by counting the upward going zero crossings and positive peaks. It can also
be derived from the spectral density function (see Section B.4.1). In metal
fatigue the irregularity factor is often used to characterise bandwidth (Pook
1978).
Probability distributions of Gaussian random processes may be described
by the exceedance, P (S/σ ), which is the probability that it exceeds S/σ . The
cumulative probability is 1 − P (S/σ ), and the probability density, p(S/σ ),
is the derivative of 1 − P (S/σ ). The probability density of a Gaussian distri-
bution is given by
2
S 1 −S
p = √ exp , (4.4)
σ 2π 2σ 2
which is similar in form to Equation (3.5). Its bell shaped form is well known
(Figure 4.6(a)). Integrating Equation (4.4) gives the exceedance
∞
S 2 −S S
P =√ exp 2
d , (4.5)
σ 2π S/σ 2σ σ
which does not have a closed form. The values shown in Figure 4.6(b) are
for the positive half of a Gaussian distribution, and are therefore twice those
given by Equation (4.5). It can be seen that P (S/σ ) is the area under the
@Seismicisolation
@Seismicisolation
Metal Fatigue 41
Figure 4.6. Gaussian distribution. (a) Probability density; (b) Exceedance.
curve of p(S/σ ) between (S/σ ) and infinity, as indicated by the shaded area
in Figure 4.6(a).
A Gaussian narrow band random process (Figure 4.1) occurs when a
broad band random signal is applied to a sharply tuned resonant system. Indi-
vidual sinusoidal cycles appear, whose frequency corresponds to the natural
frequency of the resonant system. The cycles have a slowly varying random
amplitude and the irregularity factor tends to one as the system is increas-
ingly sharply tuned. There is no precise generally agreed definition of what
constitutes narrow band random loading (Pook 1978), although it is some-
times taken as a process where the irregularity factor is at least 0.99.
In metal fatigue, the distribution of peak values is usually of interest. For
a Gaussian narrow band random process the probability density function of
the occurrence of a positive peak of amplitude S is shown in Figure 4.7(a),
and is given by the Rayleigh distribution
@Seismicisolation
@Seismicisolation
2
S S −S
p = exp . (4.6)
σ σ 2σ 2
This equation is an approximation, which becomes exact as the irregular-
ity factor tends to one. It is usually taken as applicable in practical situations
provided that the irregularity factor is at least 0.99. Theoretically the Rayleigh
distribution extends to infinity, but in practice peaks do not exceed a cut off
value of S/σ , know as the clipping ratio. Clipping implies that higher peaks
are reduced to the level given by the clipping ratio; truncation that they are
omitted altogether. The clipping ratio does not usually exceed four or five.
The narrow band random process is symmetrical (Figure 4.1) so there are
corresponding negative peaks, also known as troughs, although this is not
necessarily true for the general case of a Rayleigh distribution. The corres-
ponding exceedance (Figure 4.7(b)) is
2
S −S
P = exp . (4.7)
σ σ2
If the resonant system is not sharply tuned, the peak amplitudes vary more
rapidly and their distribution deviates from Equation (4.6) (Cartwright and
Longuet-Higgins 1965). In the limit as the bandwidth tends to infinity, and
the irregularity factor tends to zero (white noise), the distribution of peaks
tends to the Gaussian distribution (Equations (4.4) and (4.5)).
4.2.3 B LOCK FATIGUE L OADINGS
As applied in a fatigue test, a block fatigue loading (or block loading) is one in
which the loading parameters vary stepwise with time. A regularly repeated
block fatigue loading is sometimes called a programme loading. There is a
very wide range of possibilities, and block fatigue loadings have been used
extensively, both in basic metal fatigue research and in service simulation
testing (Schütz 1996). Loading details are sometimes tabulated (Table B.1),
but a graphical approach is more informative. Figure 4.8 shows, in general
terms, a simple two level block programme in which each block is a con-
stant amplitude fatigue loading. The return period is n cycles. At one time
block programmes based on constant amplitude fatigue loading, and some-
times with large numbers of blocks, were widely used in variable amplitude
fatigue testing (Frost et al. 1974). They have now largely been superseded by
more sophisticated block programmes based on random loadings.
Many service loadings are non stationary random processes, in which
statistical coefficients vary with time, and their characterisation can be diffi-
cult (Bendat and Piersol 2000). An example is the fatigue loading of offshore
@Seismicisolation
@Seismicisolation
Metal Fatigue 43
Figure 4.7. Rayleigh distribution. (a) Probability density; (b) Exceedance.
Figure 4.8. A two level block programme.
structures due to wave action, where sea states vary with time (Pook and
Dover 1989). What is often done is to model a non stationary random loading
as blocks of stationary random loadings. An example is shown in Figure 4.9
@Seismicisolation
@Seismicisolation
Figure 4.9. A load history (C/12/20) for fatigue testing relevant to offshore structures, σ is its
overall RMS (Pook 1976b). Reproduced under the terms of the Click-Use Licence.
(Pook 1976b, 1978). This load history was developed for fatigue testing relev-
ant to offshore structures, and became known as C/12/20. In it, wave loading
is modelled as blocks of narrow band random loading, at four different levels,
arranged as an approximate representation of calms and storms. The return
period is 100,000 cycles, which corresponds to approximately one week in
real time.
4.3 Miner’s Rule
For many years the vast majority of fatigue tests were carried out under con-
stant amplitude fatigue loading (see Chapter 3). Designers are faced with the
problem of how to use constant amplitude fatigue data in the prediction of
fatigue lives under the wide range of variable amplitude load histories en-
countered in service. A variable amplitude load history is sometimes called
a load spectrum. The investigation of metal fatigue under variable amplitude
fatigue loading came to be known as the study of cumulative damage (Frost
et al. 1974), although this term is not now in general use. This was because
of early interest in how metal fatigue damage accumulated at various stress
levels. Many attempts have been made to predict the life of a metallic speci-
men or component under variable amplitude fatigue loading. Much data has
been accumulated for the purpose of either deriving empirical relationships,
or of testing theoretical predictions. Although mechanisms of fatigue are the
same as under constant amplitude fatigue loading (see Section 3.4), in detail
the problem is extremely complicated. In general, only empirical solutions to
particular problems are possible (Schütz 1979).
@Seismicisolation
@Seismicisolation
Metal Fatigue 45
Figure 4.10. Schematic S/N curve.
Palmgren (1924), in predicting the life of ball bearings, assumed that dam-
age accumulated linearly with the number of revolutions. Similarly, Langer
(1937) and Miner (1945), suggested that fatigue damage at a given stress level
could be considered to accumulate linearly with the number of stress cycles.
This idea is known as the Palmgren–Miner rule or, more usually, as Miner’s
rule. It is also sometimes known as the Palmgren–Miner law or Miner’s law.
From Miner’s rule, if a specimen or component stressed at S1 has a life of
N1 cycles, as shown schematically in Figure 4.10, then the damage after n1
cycles will be n1 /N1 , and the damage per cycle will be 1/N1 . Similarly, at S2
the damage per cycle is 1/N2 .
Hence, for the two level block test shown in Figure 4.8, at failure
n1 n2
+ = 1. (4.8)
N1 N2
Similarly, for a multi level fatigue loading
n1 n2 n3 n4 ni
+ + + + ... = = 1, (4.9)
N1 N2 N3 N4 Ni
where the ratio of the number of cycles at a given stress level to the expected
life at the same stress level, ni /Ni , is sometimes called the stress ratio. This
should not be confused with the ratio of minimum to maximum load in a
fatigue cycle (see Section 3.1). The mathematical expression of Miner’s rule
ni
=1 (4.10)
Ni
is sometimes called the linear damage rule. The physical reality of the fatigue
damage at any instant during the fatigue life is not defined. However, in crack
@Seismicisolation
@Seismicisolation
Figure 4.11. Miner’s rule summation values for medium strength steel cruciform welded joints
tested under non stationary narrow band random loading (Pook 1983b). Reproduced under the
terms of the Click-Use Licence.
propagation dominated situations, damage is associated with the propagation

of a fatigue crack, and it is sometimes possible to derive Miner’s rule from
fatigue crack propagation data (see Section 7.5.1).
Numerous authors, including Frost et al. (1974) and Schütz (1996), have
pointed out that experimental data lead to a wide range of (ni /Ni ) values.
These are typically in the range 0.3 to 3, but the range can be as much as 0.1 to
10. Figure 4.11 shows some typical data. These are for structural steel cruci-
form welded joints tested under non stationary narrow band loadingusing the
C/12/20 load history shown in Figure 4.9 (Pook 1983b). Values of (ni /Ni )
less than one are unacceptable since they lead to unsafe predictions if used
in fatigue assessments (Pook 1983b). The problem can be avoided by using
a hypothetical S/N curve (rather than an experimental S/N curve), which
is adjusted so that, for a particular set of circumstances, (ni /Ni ) is always
at least one. This approach was used in the aircraft industry by around 1960
(Troughton et al. 1963), and was later used in standards that specify the use
of Miner’s rule.
A schematic hypothetical S/N curve for welded joints, developed by
Gurney (1979), is shown in Figure 4.12, plotted using logarithmic scales.
The straight line intermediate portion of the experimental S/N curve has a
@Seismicisolation
@Seismicisolation
Metal Fatigue 47
Figure 4.12. Schematic hypothetical S/N curve for welded joints. National Engineering
Laboratory diagram. Reproduced under the terms of the Click-Use Licence.
slope of −1/m (see Equation (3.4)). There is a horizontal portion for static
limitations at short lives, and another for the fatigue limit at long lives. In the
hypothetical S/N curve the fatigue limit is replaced by a line with a slope
of 1/(m + 2) for lives greater than 107 . A series of Miner’s rule summations
showed that (ni /Ni ) was always at least one. This idea is used in various
standards, including BS 5400 (Anon. 1980a).
4.3.1 W EIGHTED AVERAGE S TRESS R ANGE

If all the applied stress ranges in a variable amplitude fatigue loading are
within the region on an S/N curve which is a straight line when plotted on
logarithmic scales (Figure 4.12), then a weighted average stress range (or
equivalent constant amplitude stress range) can be calculated (Paris 1962,
Etube 2001). This weighted average stress range may then be used to read the
corresponding life from an S/N curve, and its use is equivalent to a Miner’s
rule summation (see Section 7.5.1).
For example, consider the annual fatigue loading shown in Table 4.1
where m is 4.13 (Figure 4.12). The weighted average stress range, Sh , is given
by

NS m 1/m
Sh = , (4.11)
N
where m is as shown in Figure 4.12.
Calculating
NS 4.13 and summing (Table 4.2) gives NS 4.13 = 3.208 ×
10 and N = 1.857 × 10 , hence
11 8
@Seismicisolation
@Seismicisolation
Table 4.1. Annual fatigue loading.

Number of cycles, N Stress range, S
per year (MPa)
4 × 103 59.0
5 × 103 46.5
8 × 104 26.3
6 × 105 14.4
1.5 × 107 6.8
1.7 × 108 4.2
Table 4.2. Annual fatigue loading calculations.

Number of cycles, N Stress range, S NS 4.13
per year (MPa)
4 × 103 59.0 8.235 × 1010
5 × 103 46.5 3.851 × 1010
8 × 104 26.3 5.855 × 1010
6 × 105 14.4 3.649 × 1010
1.5 × 107 6.8 4.115 × 1010
1.7 × 108 4.2 6.375 × 1010
1.857 × 108 3.208 × 1011
1/m
3.208 × 1011
Sh = = 6.080 MPa.
1.857 × 108
If any of the stress ranges are below the fatigue limit then they will be non
damaging. For example if the lowest stress range is 4.13
below the fatigue limit,
then
ignoring damage due to this stress range NS = 2.705 × 1011 and
N = 1.857 × 10 , hence
8
1/m
2.705 × 1011
Sh = = 5.762 MPa.
1.875 × 108

Alternatively, ignoring
the lowest stress range altogether NS 4.13 =
2.705 × 10 , and N = 1.569 × 10 , hence
11 7
1/m
2.705 × 1011
Sh = = 10.48 MPa.
1.569 × 107
The two methods give different values for Sh , but the same final answer for
the life in years provided, that the appropriate value for N is used when
converting from the life in cycles. This is an example where, for a variable
amplitude fatigue loading, quoting fatigue lives in numbers of cycles is po-
tentially misleading. It is sometimes better to quote fatigue lives in time.
@Seismicisolation
@Seismicisolation
Metal Fatigue 49
Figure 4.13. Damage density for three distributions (Pook 1976b). Reproduced under the
4.3.2 DAMAGE D ENSITY F UNCTIONS
In a sinusoidal variable amplitude process, from Miner’s rule, the fatigue

damage caused by each cycle is proportional to (S/σ )m , where S is stress
range, σ the root mean square (RMS) value of the process, and m is as in
Figure 4.12. The damage√relative to a constant amplitude sinusoidal loading
of the same RMS is [S/( 2σ )]m . Define a damage density function as (Pook
1976b)
m
S S S
r = √ p , (4.12)
σ 2σ σ
where r(S/σ ) is the damage density relative to peaks of (S/σ ). Damage dens-
ity curves for m = 3 are shown in Figure 4.13 for three distributions: the
Rayleigh distribution (Figure 4.7), the C/12/20 load history (Figure 4.9) and
the Laplace distribution. The Laplace distribution is given by

S −S S
P = exp =p . (4.13)
σ σ σ
It is a straight line when plotted on log linear coordinates.
Figure 4.13 shows that peaks in the region of S/σ of 2 to 3 cause the
most fatigue damage, whereas peaks of S/σ of less than 1/2 cause very little
damage. In practice some small peaks would be below the fatigue limit, and
@Seismicisolation
@Seismicisolation
Figure 4.14. Damage density curve showing effect of gust velocity on a typical aircraft wing
component.
hence would not cause any fatigue damage. Large peaks also cause very little
fatigue damage, but a rare very large peak could be above the static limita-
tions shown in Figure 4.12, and hence cause immediate failure. Arrows on
Figure 4.13 show values of S/σ for which the exceedance, P (S/σ ), is 10−5 .
The area under a damage density curve gives the relative damage R, which is
the ratio of the life of a constant amplitude fatigue loading to that of a vari-
able amplitude fatigue loading of the same root mean square (RMS) value.
Values of R are 1.33, 1.45 and 2.12 for the Rayleigh, C/12/20 and Laplace
distributions respectively.
An early application of damage density curves was to assess the fatigue
damage due to gust loading of aircraft wings (Heywood 1962). Figure 4.14
shows a typical damage density curve; the load at which most fatigue damage
is produced corresponds to a gust velocity of 2.75 m/s. It was argued that
this load is appropriate for fatigue tests since it provided the best constant
amplitude representation of the gust load distribution.
4.3.3 C YCLE C OUNTING
In using Miner’s rule, the situation is further complicated when individual

cycles cannot be distinguished, as in broad band random loading (Figure 4.2).
A cycle counting method is then needed to reduce the process to discrete
cycles, and hence permit the application of Miner’s rule. Equivalent cycles ex-
tracted from a process may be characterised by using the notation for constant
amplitude cycles (see Section 3.1). A number of empirical cycle counting
methods have been developed (Watson and Dabell 1976, Etube 2001, Steph-
@Seismicisolation
@Seismicisolation
Metal Fatigue 51
Figure 4.15. A simple process in which cycles cannot be distinguished. Reprinted from Linear
Elastic Fracture Mechanics for Engineers. Theory and Applications. LP Pook, WIT Press,
ISBN 1-85312-703-5, 2000.
ens et al. 2001). Detailed descriptions of some of these methods are given in
the references. Development of cycle counting methods is largely on a trial
and error basis, and all have shortcomings. The success of a proposed method
is assessed by comparing the results of Miner’s rule summations, based on the
method, with experimental data.
Results obtained using different cycle counting methods vary widely
(Stephens et al. 2001). This is illustrated by cycle counts for the simple pro-
cess shown in Figure 4.15 (Pook 2000a). Stresses are 45 MPa at peaks 2 and
6, 60 MPa at peak 4, and 30 MPa at troughs 3 and 5. A simple approach is
the peak counting method in which peaks are counted from zero (or some
other reference) load. This has no physical basis in terms of fatigue crack
propagation mechanisms (see Section 3.4). Peak counting gives stress ranges,
measured from zero, of 45 MPa (peak 2), 60 MPa (peak 4) and 45 MPa (peak
6). The corresponding weighted average stress range, Sh (see Section 4.3.1),
with m taken as 3, is 51.01 MPa. A more realistic alternative is range count-
ing in which positive going ranges are counted (Dover 1979). This gives stress
ranges of 45 MPa (range 1 to 2), 30 MPa (range 3 to 4) and 15 MPa (range 5 to
6). It is assumed that there are corresponding negative going ranges nearby,
that is ranges 6 to 7, 4 to 5 and 2 to 3). A justification for this approach is
that it is primarily the positive going half cycles that cause crack propagation,
as indicated by the model shown in Figure 3.13, and it is widely used. The
corresponding value of Sh is 34.34 MPa.
Perhaps the most logically defensible and widely used cycle counting
method is rainflow counting (Murakami 1992a). In this approach individual
hysteresis loops in the plastically deformed material at the tip of a station-
@Seismicisolation
@Seismicisolation
Figure 4.16. Rainflow method of cycle counting (Pook 1983a). Reproduced under the terms
of the Click-Use Licence.
ary crack are identified. It derives its name from the first practical algorithm,
developed by Endo, in which water is imagined to flow down the load time
history, with the axes reversed (Figure 4.16). Flow starts at the beginning of
the record, then at the inside of each peak in the order in which peaks are ap-
plied. It stops when it either meets flow from a higher level, or a point oppos-
ite a peak which is arithmetically greater (or equal to) the point from which it
started, or when the end of the record is reached. Each separate flow is coun-
ted as a half cycle. There is always a complimentary half cycle of opposite
sign, except perhaps for a flow which either starts at the beginning of the re-
cord or reaches the end. The method is fully discussed in Murakami (1992a),
which includes a reprint of Endo’s original Japanese language paper. There
are many variants of the rainflow method (Socie 1992) often differing in the
treatment of the beginning and end of a record. Rainflow counting is now
a generic term for any cycle counting procedure in which hysteresis loops
are identified (Etube 2001). A typical practical algorithm is given by Anzai
(1992). In Figure 4.15 rainflow counting gives stress ranges of 60 MPa (cycle
1-4-7), 15 MPa (cycle 3 -2-3) and 15 MPa (cycle 5-6-5 ). The corresponding
value of Sh is 42.03 MPa, which is smaller than is given by peak counting but
larger than is given by range counting.
@Seismicisolation
@Seismicisolation
Metal Fatigue 53
In these cycle counting methods the number of equivalent cycles is equal

to the number of peaks in the process. Hence analysis of data in terms of
the number of equivalent cycles would appear to be appropriate. However, if
small cycles such as 3 -2-3 and 5-6-5 (Figure 4.15) are neglected, perhaps on
the grounds that they are non damaging, then the number of equivalent cycles
changes, and analysis in terms of equivalent cycles becomes ambiguous. In
these circumstances analysis in terms of time may become appropriate (cf.
Section 4.3.1).
The disadvantage of both range and rainflow counting is that in general
counting has to be done numerically using a sequential listing of peaks and
troughs. Some analytical progress has been made using random process the-
ory. For example, Bishop and Sherratt (1990) have developed a numerical
method of estimating rainflow range statistics, to any desired degree of accur-
acy, directly from the spectral density function of a Gaussian process. Earlier,
Chaudhury and Dover (1985) developed an approximate method for calcu-
lating the weighted average stress range without the need for a cycle count.
The need for a cycle count is sometimes avoided by carrying out tests using
an appropriate standard load history representative of the structure of interest,
for example Austin (1994).
4.4 Standard Load Histories
Standard load histories for fatigue testing, as opposed to ad hoc sequences

used for specific purposes, have quite a long history (Schütz and Pook 1987).
As early as 1939, Gassner (1954) described his eight level block programme,
which was used from about 1948 onwards in Germany and later elsewhere
(cf Figure 4.8). At that time limitations of available fatigue testing equipment
meant that nothing more complex was attainable. The peak distribution used
in Gassner’s programme became known as the LBF Normal distribution. It
has been used as the basis of a very large number of fatigue tests simulating
service loading conditions in which a continuous load (or stress) distribution
is replaced by an eight level block programme. It has been claimed that the
results of such tests have a very good correlation with the service lives of a
wide range of components (Frost et al. 1974).
Computer controlled servohydraulic fatigue test machines first became
available in the early 1960s and within ten years they were in widespread
use, together with servohydraulic actuators in rigs for fatigue testing of struc-
tures. They had the big advantage that there was no limitation whatsoever on
the load histories which could be applied; but that was also their main dis-
@Seismicisolation
@Seismicisolation
advantage (Schütz 1989). Many different load histories have been employed
indiscriminately, sometimes without an adequate documentation of exactly
what had been done. Consequently, fatigue test results were often not usable
by anybody except the author. Moreover, the results of test programmes car-
ried out by different laboratories were then not comparable. This may not be
of importance in ad hoc tests for particular situations, but for general fatigue
investigations it can produce a confusing situation or, worse, it may result
in even qualitatively incorrect conclusions. It is therefore not surprising that
complex standard load histories have been developed for fatigue test purposes
in a wide range of industries, and by the late 1980s were in widespread use
(Watanabe and Potter 1989).
Ideally, a standard load history should cover several different types of
structure or component, and the number of variants should be minimised.
Inevitably, compromises are needed, but some success has been achieved.
For example, FALSTAFF (Fighter Aircraft Loading Standard For Fatigue)
covers several types of tactical aircraft, when flown for peace time training
missions, but is restricted to wing lower surface stresses near the wing to
fuselage joint (Schütz 1989). On the other hand, during the development of
WASH (Wave Action Standard History) it became clear that variants would
be needed to cover different types of offshore structures installed in vari-
ous locations. Software was therefore written using a modular architecture so
that variations could easily be incorporated, but different versions of WASH
would have a family resemblance (Pook and Dover 1989). Later the same
modular approach was used in the successful development of JOSH (Jack up
Offshore Standard load History) (Etube 2001).
4.4.1 A PPLICATIONS OF S TANDARD L OAD H ISTORIES
As noted by Edwards and Darts (1984), standard load histories are needed
when available life prediction methods are not sufficiently accurate to pre-
dict fatigue lives under variable amplitude service loadings. When making a
fatigue assessment of, for example, a new detail, fastening system or fatigue
life improvement method, variable amplitude fatigue loading has to be used.
Frequently, such tests are of general application rather than in support of a
particular project.
If an appropriate standard load history exists then this will usually be the
best choice. Any resulting data can be used as design data, and can also be
compared directly with any other data obtained using the same standard load
history. Experience has shown that, once a standard load history becomes
available, relevant data accumulate quickly. This can greatly increase the
@Seismicisolation
@Seismicisolation
Metal Fatigue 55
technical value of individual test results, and can also reduce the need for
expensive fatigue testing. When standard load histories are used, extensive
fatigue testing programmes can easily be shared between different organisa-
tions. For example, the load sequence shown in Figure 4.9 was developed
specifically for use in the fatigue testing of tubular welded joints under the
United Kingdom Offshore Steels Research Project (Crisp 1974). Testing was
carried out at several different laboratories.
Hence there are numerous applications for which standard loads histories
can be advantageously used. As summarised by Schütz (1989), these include:
(a) Evaluation of the fatigue strength of specimens and components made
from different materials.
(b) Evaluation of data for the preliminary fatigue design of components.
(c) Investigation of the scatter of fatigue lives under variable amplitude fa-
tigue loading, which could well differ from that under constant amplitude
fatigue loading.
(d) Assessment of models, such as Miner’s rule, for the prediction of fatigue
lives.
(e) Round Robin programmes on general metal fatigue problems, under vari-
able amplitude fatigue loading, in which several laboratories participate.
4.4.2 D EVELOPMENT OF S TANDARD L OAD H ISTORIES

By the late 1980s, methods of constructing standard load histories were well
established (Schütz 1989). Sophisticated computer methods for their imple-
mentation are now available (Heuler and Klätsche 2005). The development
of a new standard load history for a particular application can be a lengthy
process. It is best carried out as a collaborative effort between several organ-
isations, preferably from different countries. Ideally, the basis of a standard
load history should be strain or load measurements made in service, prefer-
ably from a number of similar structures, for example several types of trans-
port aircraft. There are two distinct sources of fatigue loads, human actions
and environmental effects. In a fighter aircraft, fatigue loads are largely the
result of the pilot’s actions, whereas in an offshore structure fatigue loads are
largely the result of wave loading (Schütz and Pook 1987).
Once data have been collected, common features then need to be extrac-
ted from the measurements. In particular, similar load probability distribution
shapes need to be sought. What constitutes similarity in this respect is a dif-
ficult question, which needs careful consideration (Pook and Dover 1989).
Once similarity is found, then a representative load distribution must be se-
lected, and a logical sequence of individual cycles must be decided upon.
@Seismicisolation
@Seismicisolation
For example, an aircraft’s flight always begins with taxiing, followed by the
ground to air cycle (transfer of the aircraft’s weight from the undercarriage to
the wings), and so on. Only if the position and size of each and every cycle
is fixed in the sequence will the results obtained by different investigators be
satisfactorily comparable.
If just the load probability distribution were fixed, an infinite number of
load histories could be synthesised, that is reconstituted, from one given dis-
tribution, and different versions could possibly result in different fatigue lives.
This is a weakness of the C/12/20 load history for fatigue testing relevant to
offshore structures shown in Figure 4.9 (Pook 1976b). This was originally de-
veloped for use in one laboratory (Ewing 1986). The root mean square (RMS)
of each block is specified, but the user is left to generate a narrow band ran-
dom loading sequence for each block. Two different implementations were
used in the laboratory in which the load history was developed, but full de-
tails of these implementations were not included in their descriptions (Holmes
and Kerr 1982). The later standard load histories for fatigue testing relevant to
offshore structures, WASH, did specify load sequences unambiguously (Pook
and Dover 1989, Schütz et al. 1990).
By the late 1980s it had become clear that stress distributions due to wave
loading could not necessarily be regarded as narrow band random loading,
and WASH took this into account (Pook and Dover 1989). For example, Fig-
ure 4.17 shows spectral density functions (SDF) obtained for a 0.76 m dia-
meter horizontal member, immersed 10.8 m, on a tubular welded tall platform
in the North Sea (see Section B.4.1). In practice, although water waves have
a dominant wave passing frequency, they are not particularly narrow band
so energy may be available to excite resonances (Pook 1987b). The SDF for
the water surface elevation (Figure 4.17(a)) shows this, with a clearly defined
peak corresponding to the dominant wave passing frequency, but with sig-
nificant energy at other frequencies. To avoid resonances, platforms are de-
signed so that resonant frequencies are substantially higher than the dominant
wave passing frequency. This has been successful for the axial stresses since,
as might be expected, the SDF (Figure 4.17(b)) is of similar form, with no
resonances excited. However, the SDF for the bending stress (Figure 4.17(c))
does show two peaks corresponding to structural resonances.
4.4.2.1 Truncation and Omission Levels
The most difficult aspects of the development of a standard load history are
appropriate selection of a truncation level (see Sections 4.2.2 and B.3.1), an
@Seismicisolation
@Seismicisolation
Metal Fatigue 57
Figure 4.17. Spectral density functions for a 0.76 diameter horizontal member immersed
10.8 m, significant wave height 4.75 m. (a) Water surface elevation, (b) bending stress, (c) axial
stress (Pook 1989b). Reproduced under the terms of the Click-Use Licence.
@Seismicisolation
@Seismicisolation
omission level (see Section B.3.3.2), and a return period, after which the load
history repeats exactly (see Section B.3.3.4) (Schütz and Pook 1987).
The retention of large but infrequent tensile stresses, due to a high trun-
cation level, may actually prolong fatigue life due to the beneficial residual
stresses they cause (see Section 7.5.2). Thus, if a variable amplitude fatigue
test is carried out on a structure or component with too high a truncation level
the test result may lead to an optimistic prediction of service life. This is
the truncation dilemma. One suggested rule of thumb is that the highest load
should occur at least 10 times during a test (Schijve 1985). The position of
high loads in a sequence is important since failure will normally occur at one
of them.
Long life structures may have fatigue lives up to the gigacycle fatigue re-
gion (see Section 3.2). For example, an offshore structure in the North Sea
has approximately 108 wave loading cycles applied during a service life of
20 years. This is too many for any economically feasible standard load his-
tory. A typical service load sequence contains a large number of small loads,
so omission of loads that are too small to cause fatigue damage can have a
dramatic effect on the total number of loads in a standard load history. For
example, omitting loads below 50 per cent of the fatigue limit may reduce the
number of cycles by 90 per cent (Heuler and Seeger 1986). If the omission
level is too high then optimistic lives may be obtained in tests on structures
and components. This is the omission dilemma. Unfortunately, the only safe
way of determining the maximum permissible omission level, for a particu-
lar set of circumstances, is to carry out comparative fatigue tests at various
omission levels. In the C/12/20 load history (Figure 4.9) each block is a nar-
row band random loading, so some of the cycles in each of the three level
lower blocks will be higher than some cycles in a higher level block. Omit-
ting the two lower level blocks would reduce the number of cycles by 66 per
cent. In this case low level cycles would not be omitted altogether, but their
proportion would be reduced.
The length of the return period is critical, and it has to be chosen in con-
junction with truncation and omission levels. On one hand, the sequence
needs to be repeated at least several times before failure occurs, otherwise
the various loads would not be included in their correct proportions, and the
fatigue life would not be accurately defined. On the other hand, too short a re-
turn period means that infrequent but high loads would not be included in the
load history, whereas they do occur in service and could well affect fatigue
life. This is the truncation dilemma in reverse.
@Seismicisolation
@Seismicisolation
Metal Fatigue 59
The practicalities of fatigue testing can be another constraint. The C/12/20

load history (Figure 4.9) was designed for use on tests carried out in sea wa-
ter, so tests had to be carried out in real time using a typical wave passing
frequency of 1/6 Hz. For convenience the return period chosen is 105 cycles,
which is approximately one week in real time. The clipping ratio is the ratio
between the highest load in a sequence of its root mean square value (see Sec-
tions 4.2.2 and B.3.1). In a narrow band random process the expected clip-
ping ratio depends on the length of the sample (see Section B.3.3.4). From
Equation (B.20) the expected clipping ratio for the 4000 cycles at 1.960σ
is 2.98, which gives an expected overall clipping ratio for C/12/20 of 5.84
whereas the desired clipping ratio for the underlying distribution correspond-
ing to P (S/σ ) = 10−5 is 7.7 (see Section B.3.2). In practice, for the original
implementation, a computer generated sample of narrow band random load-
ing was selected such that the overall clipping ratio was 8 (Holmes and Kerr
1982). This is typical of the compromises needed when constructing load
histories with specified properties.
4.5 Multiaxial Fatigue in Metals

4.5.1 M ULTIAXIAL FATIGUE T ESTS
The fatigue test specimen shown in Figure 3.4 is designed so that, under a uni-
axial fatigue loading, a uniaxial stress is produced in the 8.3 mm diameter test
section. More generally, a uniaxial loading can result in multiaxial stresses in
regions of interest. For example in the cylindrical specimen with a circum-
ferential notch there is a biaxial stress system in an element at the notch root
(Figure 4.18). When considering material response it is convenient, albeit not
strictly accurate, to refer to such situations as biaxial loading or, more gener-
ally, as multiaxial loading. Another example is the cabin of an aircraft which,
when pressurised, is subjected to a biaxial loading consisting of hoop stresses
and longitudinal stresses.
The rig shown in Figure 4.5 is designed so that biaxial loading produces
biaxial stresses in the central portion of a thin sheet specimen (Pook and
Holmes 1976). Since the two actuators are at right angles principal stress
directions are always the same whatever loads are applied. In a proportional
fatigue loading (or proportional loading) the ratio of the loads applied by the
two actuators is always the same whereas in a non proportional fatigue load-
ing (or non proportional loading) it is not. Non proportional fatigue loadings
are of two types: those in which principal stress directions change with time,
and those in which they do not.
@Seismicisolation
@Seismicisolation
Figure 4.18. Biaxial stress system at a notch root element in a cylindrical specimen with a
circumferential notch under uniaxial loading.
Figure 4.19. Bicycle crankshaft, length 155 mm.
In publications on multiaxial fatigue the introduction usually states cor-

rectly that nearly all components and structures are subject to multiaxial fa-
tigue loading in service. However, in practice one stress component often
dominates. Hence, for analysis purposes, a multiaxial loading can be reduced
to a uniaxial stress state (Zenner 2004). For example, in an internal com-
bustion engine crankshaft there are several critical regions, in each of which
fatigue failure could occur, and each of these regions can be associated with
a stress component. One of the purposes of service simulation fatigue testing
(Marsh 1988) is to identify critical regions.
Multiaxial stress states in service can be very complex, even for appar-
ently simple situations. For example, consider the crankshaft in a bicycle (see
Sections 4.1 and 6.6). Figure 4.3 shows a typical bicycle crankshaft assembly,
and Figure 4.19 the crankshaft removed from its housing. The 16.65 mm dia-
meter shaft is supported by 6.35 mm diameter ball bearings which run in the
outer radii of the two shoulders, as indicated by the arrow in Figure 4.20.
Forces applied by the rider to the pedals are transferred through the cranks to
the squared ends of the crankshaft (Figure 4.3). The most significant stresses
are those due to bending and torsion of the shaft, and the most highly stressed
regions are where the outer radii of the shoulders blend into the crankshaft.
The stress concentration factors (SCF) for these regions are 1.32 for bending
and 1.14 for torsion.
@Seismicisolation
@Seismicisolation
Metal Fatigue 61
Figure 4.20. Close up of bicycle crankshaft.
Figure 4.21. Schematic diagram showing torques and bending moments at crankshaft outer
radii. (a) Left radius. (b) Right radius.
Figure 4.21 is a schematic diagram showing the fatigue loading (torques

and bending moments) at the outer radii of the crankshaft. It is assumed that
a force equal to the rider’s weight (typically 700 N) is applied to the centre
of a pedal during the downward half cycle, and that the force on the pedal is
zero during the upward half cycle. The bending moment is constant during
the downward half cycle and zero during the upward half cycle. Its value is
equal to the pedal force times the axial distance from the bearing to the centre
@Seismicisolation
@Seismicisolation
of the pedal. For a bicycle with a single front chain wheel (Figure 4.4) this
is typically 120 mm on both sides. For a bicycle with a double or triple front
chain wheel the distance is greater on the right hand side and is typically
140 mm. The torque varies sinusoidally during the downward half cycle. It
is zero at the beginning and end of the half cycle and is maximum when the
crank is horizontal; its value is then the pedal force times the crank length,
typically 170 mm. The torque is zero during the upward half cycle. In practice
pedal forces will vary with road and traffic conditions, and with the rider’s
inclinations.
At the left hand radius (Figure 4.21(a)) the torques and bending moments
are due to the force on the left hand pedal. The torques and bending moments
are in phase but the fatigue loading is non proportional because of the differ-
ent waveforms. At the right hand radius (Figure 4.21(b)) the torques are due
to the force on the left hand pedal. Torques from this pedal pass through the
crankshaft into the chain wheel, whereas torques from the right hand pedal
pass directly into the chain wheel (Figure 4.4). Bending moments are due to
the force on the right hand pedal. If the pedal to radius distance is greater on
the right hand side, then the bending moments will be greater, as is shown
schematically in the figure. The torques and bending moments are 180◦ out
of phase. Overall, the fatigue loading on the crankshaft is non proportional in
that the forces on the two pedals are applied 180◦ out of phase.
Development of standard load histories for multiaxial fatigue testing is
less advanced than for uniaxial fatigue testing. The necessary mathematical
background needed for the application of non proportional random loading
is well understood (Kam and Dover 1989), and some multiaxial non propor-
tional random standard load histories have been developed for use in the mo-
tor industry (Bruder et al. 2004). Petrone and Susmel (2003) have proposed a
simplified standard load history for the acceptance fatigue testing of welded
handlebar stems used in mountain bikes. This consists of 105 cycles of con-
stant amplitude torsion followed by 105 cycles of constant amplitude bending.
This is equivalent to a bicycle life of 40,000 km. The simplified standard load
history was verified for several different types of handlebar stem by compar-
ative tests using realistic non proportional random loading. Use of the simpli-
fied load history shortened test times by a factor of 450 compared with non
proportional random loading.
4.5.2 M ULTIAXIAL FAILURE C RITERIA
The usual purpose of a multiaxial failure criterion is to calculate an equivalent

tensile stress, ρ, which can then be compared with the uniaxial mechanical
@Seismicisolation
@Seismicisolation
Metal Fatigue 63
properties of a metal. For example, in a yield criterion, ρ is compared with the

tensile yield stress in order to predict the onset of yielding under a multiaxial
stress state. The equivalent tensile stress can be either scalar or a vector. If
it is a vector then the maximum value defines a critical plane, normal to the
maximum value of the vector, on which failure is predicted to take place.
Equivalent stresses are sometimes shear stresses, in which case the plane of
the maximum value is a critical plane.
Multiaxial failure criteria are used extensively in the analysis of mul-
tiaxial fatigue loadings, and numerous criteria have been developed (Suresh
1998, Stephens et al. 2001, Sonsino et al. 2004). Most of these have been used
successfully in the analysis of various experimental data sets. Some criteria
require extensive numerical calculations for their implementation, and have
been made possible by the widespread availability of powerful modern com-
puters. Multiaxial failure criteria, as applied to multiaxial fatigue loadings are
a rapidly developing field, but despite extensive work it is not yet possible to
make definitive statements on which criteria are most appropriate in particu-
lar circumstances. Some criteria used for multiaxial fatigue loadings include
empirical material constants determined experimentally (Suresh 1998), which
means that they are not of general application. A particular problem is that,
even when the same failure criterion is used, software written by different
authors sometimes leads to different results (Zenner 2004).
4.5.2.1 Scalar Criteria

Scalar criteria for the onset of yielding have been of interest for about 150
years, and a number of different criteria are available (Dowling 1993). Most
of the yield criteria which have been developed are now only of historical in-
terest because they contain fundamental flaws (Hill 1950). Two which are still
in general use are the Tresca criterion and the von Mises criterion (or Mises
criterion). Both are based on the principal stresses, σ1 , σ2 and σ3 . The Tresca
criterion was developed in 1864 on the basis of a long series of experiments
in which Tresca measured the loads required to extrude metals through dies
of various shapes (Hill 1950). It states that yielding takes place when
σ1 − σ3 = σY , (4.14)
where σ1 ≥ σ2 ≥ σ3 , and σY is the yield stress. That is, the equivalent stress,
σe is given by
σe = σ1 − σ3 . (4.15)
The von Mises yield criterion was developed in 1913, but the underlying idea
can be traced back to 1904 (Hill 1950). Mathematical derivations of the von
@Seismicisolation
@Seismicisolation
Mises criterion tend to be unconvincing, for example that given by Derby et

al. (1992), but its use is justified by its widespread success in predicting the
onset of yielding. It can be written as

σ12 + σ22 + σ32 − σ1 σ2 − σ2 σ3 − σ3 σ1 = σY . (4.16)
That is

σe = σ12 + σ22 + σ32 − σ1 σ2 − σ2 σ3 − σ3 σ1 , (4.17)
where σe is now the von Mises equivalent stress.
Expressed in terms of Cartesian stress field components Equation (4.17)
becomes

(σx − σy )2 + (σy − σz )2 + (σz − σx )2
σe = + 3(τxy
2 + τ2 + τ2 ) .
yz zx (4.18)
2
This is a convenient form when most of the stress components are zero. For
example, for a combination of a uniaxial stress and a shear stress it becomes

σe = σx2 + 3τxy 2 . (4.19)
The von Mises equivalent stress is sometimes used in the analysis of pro-
portional fatigue loadings, but it cannot be used for non proportional fatigue
loadings (Suresh 1998, Zenner 2004). The Tresca criterion leads to the ap-
proximately correct prediction that the fatigue limit (at zero mean stress) in
torsion (that is, under shear) is approximately half that under uniaxial stress.
In some modern scalar criteria for analysis of multiaxial fatigue loadings, an
integral approach is used in which σe is calculated as an average value of
some stress feature for all possible directions (Zenner 2004).
4.5.2.2 Critical Plane Approaches
Failure criteria in which the equivalent stress is a vector are usually known as
critical plane approaches. The first application of a critical plane approach
to metal fatigue appears to have been by Bacon (1930). The simplest critical
plane approach is the fracture criterion proposed by Lamé in 1831 (Zenner
2004). This maximum normal stress criterion states that failure is expected
when the maximum principal stress reaches the uniaxial tensile strength of
the material. That is
σe = σ1 , (4.20)
where σ1 ≥ σ2 ≥ σ3 . This has had its greatest success in predicting the
fracture of brittle materials (Dowling 1993). It can be adapted to multiaxial
@Seismicisolation
@Seismicisolation
Metal Fatigue 65
fatigue loadings by finding the critical plane on which the normal stress range
is a maximum. This would work for the load history shown in Figure 4.21(a)
where the maxima and minima of the torques and bending moments are in
phase. For the load history shown in Figure 4.21(b) maxima are out of phase.
Hence, values of σe at maxima and associated normal plane orientations are
significantly different for torsion and bending fatigue loadings. In this sort
of situation an appropriately weighted averaging procedure has to be used
to identify both an effective value of σe and the critical plane orientation
(Carpinteri et al. 2004). The situation is further complicated when two (or
more) applied fatigue loadings are random. The difficulty is that, in general,
detailed procedures have to be justified by reference to appropriate experi-
mental data and hence are not of general applicability. For low cycle fatigue,
where analysis in terms of strain is appropriate, critical plane approaches are
based on strain rather than stress (Suresh 1998).
The KoNoS hypothesis is a typical critical plane approach which was used
for aluminum tube to plate welded joints tested under combined constant
amplitude fatigue bending and torsion loading (Kueppers and Sonsino 2004).
Tests were carried out at zero mean load, with the bending and torsion fatigue
loads in phase or 90◦ out of phase. Values of σe were calculated for all pos-
sible planes using the normal and shear stresses on each plane using the von
Mises criterion (Equation (4.19)). A maximum value of σe was obtained for in
phase fatigue loading and a different maximum value for out of phase fatigue
loading. These maxima identified critical planes which had different orienta-
tions for the two cases. Values of σe were then recalculated. For the in phase
case σe was recalculated from the stress components in a plane perpendicular
to the weld using Equation (4.19). For the out of phase case, this value was
then multiplied by the ratio of the maxima (out of phase to in phase) obtained
earlier. This gave a satisfactory prediction for the out of phase experimental
results. It was pointed out that the method did not work for steels.
@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
5
Fatigue Design
5.1 Introduction
The term fatigue design is a misnomer, what is meant is assessment of an ex-

isting design against the possibility of fatigue failure. The possibility of fail-
ure due to metal fatigue is only one of the factors that have to be considered
during the design of an engineering component or structure. Usually, fatigue
behaviour is checked only after basic design decisions have been taken. There
are many successful structures which have been designed without any specific
attention to the possibility of fatigue failure. A fatigue assessment sometimes
shows that an initial design has to be modified in order to ensure that the
product performs as required for its intended lifetime. Where the possibility
of fatigue failure is a major preoccupation, design and fatigue assessments
are integrated in an iterative approach.
When a fatigue failure does occur, under product liability legislation, any-
one connected with the design, manufacture, supply, servicing, repair or op-
eration of the product might find themselves in the position of being held leg-
ally responsible for personal injury and material damage caused by the failure
(Anon. 2001a). Modifications carried out without consulting the original de-
signer are a frequent cause of fatigue failure, for example Pook (1998). Many
authors, for example Jones (2003), have drawn attention to the importance
of analysing fatigue failures so that future designs may be modified to avoid
similar failures.
In broad terms, if not in detail, the mechanisms of metal fatigue are not
very complex (see Section 3.4), and a century and a half of metal fatigue test-
ing means that large amounts of experimental data are available. Despite this,
even rough estimates of fatigue strength at the preliminary design stage can
present problems. As pointed out by Frost (1975) this is the inevitable result
@Seismicisolation
@Seismicisolation
68 Chapter 5: Fatigue Design
of the complexity and variety of realistic engineering design situations, to-

gether with the difficulty of generalising from known fatigue data and then ap-
plying these generalisations to specific engineering problems. Unfortunately,
there is in general no simple relationship between a metal’s fatigue resistance
and its tensile strength, so simple substitution of a stronger material is seldom
the answer to a fatigue problem. At times there seems to be little correlation
between what appear to be flaws of one sort or another and service perform-
ance (Pook 1983a).
Most books on metal fatigue contain advice on its alleviation, for example
Stephens et al. (2001). Much advice amounts to hints and tips on how to re-
duce the level of stresses due to fatigue loads at actual or potential failure
sites. Traditionally, many metal fatigue problems have been solved by taking
some action to improve fatigue performance on what amounts to a trial and
error basis (Whyte 1975). However, knowledge and experience are required
in order to avoid apparent solutions which actually make matters worse, espe-
cially when welding is involved (Radaj and Sonsino 1999). In the last 30 years
or so much effort has been devoted to the analysis of experimental metal fa-
tigue data so that it can be incorporated in standards such as BS 5400 (Anon.
1980a) (see Section 2.3.4).
5.2 Failure Analysis
There are numerous publications which include examples of failure analysis

involving metal fatigue. These include Parsons (1947), Anon. (1954, 1958),
Pugsley (1966), Whyte (1975), Pook (1983a), Schijve (2001) and Jones
(2003). One of the earliest documented examples of metal fatigue failure
analysis is in the records of the Institution of Mechanical Engineers (Anon.
1958). In 1850 members were concerned with the fracture of wrought iron
railway axles. John Ramsbottom, with his characteristic good sense, pointed
out that all axles failed (by fatigue) at the shoulder on the axle used to loc-
ate the wheels. He rightly stated that it was more important to redesign the
axle to eliminate this shoulder and so reduce stresses, than to argue about
the mechanisms of metal fatigue. From a practical design viewpoint his ad-
vice still holds good. A century later, the Proceedings of the Institution of
Mechanical Engineers’ Conference on Fatigue, held in 1956, includes about
100 examples of fatigue failures, all intended to illustrate some point on the
alleviation of fatigue (Anon. 1958).
Many metal fatigue failures, perhaps the majority, are never diagnosed.
Failures such as those shown in Figures 1.1, 1.3 and 1.4 are recognised as
@Seismicisolation
@Seismicisolation
Metal Fatigue 69
being due to metal fatigue only if they happen to be seen by a specialist. It

is at times difficult to establish the precise sequence of events leading up to a
particular failure (Pook 1998). The reasons range from simple forgetfulness,
through lack of adequate documentation, to a human desire to avoid blame.
The results of failure analyses can leave designers with difficult decisions
(Pook 1983a).
Using defective in the legal (see Section 5.4) rather than the intuitive
sense, defects leading to metal fatigue failures may be divide into two broad
classes. Firstly, design defects where the original design was inadequate for
the envisaged use, including cases where the loading was more severe than
anticipated by the designer; and secondly, manufacturing defects where the
failure occurred because the designer’s intentions were not followed during
manufacture, including faulty repairs and the use of material that is incorrect
or below specification. Relatively few fatigue failures are found to be due to
the use of faulty or incorrect materials (Pook 1983a).
5.3 Situations, Philosophies and Approaches
In approaching a mechanical engineering design problem in general, or a

fatigue assessment in particular, a designer must first select, perhaps uncon-
sciously, the approach to be used (Pook 1983a, Zerbst et al. 2003). It was
pointed out some time ago by Bignall et al. (1977) that failure to adopt an ap-
propriate overall strategy, including allowance for human fallibility, is the one
theme common to diverse catastrophic failures. The topic is frequently dis-
cussed in the literature, for example by Grandt (2004). Arguments over the
selection of an appropriate overall strategy are at the heart of the controversy
over the safety of nuclear power stations (Steele and Stahlkopf 1980, Anon.
2005a). In the UK the Nuclear Installations Inspectorate’s Safety Assessment
Principles were first published in 1979, and are currently being revised (Anon.
2004). At a more mundane level, much time and expense can often be saved
by choosing an appropriate strategy early in the job. The planning and modi-
fication and repairs should be taken as seriously as the original design. Too
often it is left to skilled tradesmen who lack the necessary expertise (Pook
1983a).
5.3.1 S ITUATIONS
There are four basic situations in mechanical engineering design which will
often include a fatigue assessment (Fuchs 1980). The first is the in house
@Seismicisolation
@Seismicisolation
tool. A tool engineer given the job of designing a welding fixture for some
new device will probably think about it, sketch several possible solutions,
select one by intuition or with the help of some quick analysis, and proceed
to draw the tool which will do the desired job. In effect, the design is based
on previous experience of broadly similar jobs, together with the assumptions
that any failure will not be catastrophic, and also that any problem which
arises when the tool is first used can be resolved quickly.
Second is the mass product. The product engineer in charge of a new di-
gital door lock will probably build several prototypes, test them extensively,
and analyse and optimise the design as much as possible. Reliability is im-
portant from both customer and manufacturer viewpoints. Wherever possible
standards will be used in metal fatigue assessments (Pook 1997). The cost of
design and development can be amortised over a large number of products, so
it is worth evaluating several ideas thoroughly. The eventual solution will be
documented in detail. Figures 1.1, 1.3, 1.5, 3.15, 4.4 and 4.19 are all examples
of components from mass products.
Thirdly, there is the major project. The designer of a steel offshore struc-
ture cannot afford prototypes. As much as possible must be learnt from ex-
isting field experience, and much analysis be carried out, even though this is
complex and expensive (Etube 2001). The cost of design and analysis will
normally be small compared with rectification of faults at a later stage.
The fourth is the code design, which uses a standard procedure. Standards
whose use in design are a legal requirement are often referred to as codes. The
designer of a boiler will usually have to conform to a code, in contrast to the
other designers mentioned, and hence is concerned with legalistic study of the
code. Ideally, the designer ought to be able to assume that code requirements
are the real requirements, and will lead to a satisfactory design. In practice,
this does not always happen, and a designer may need to ensure that a design
meets the real requirements, as well as the code requirements. In the past
25 years the imposition of codes through legislation has become much more
widespread in many countries, not always with satisfactory results.
Most mechanical design jobs fall somewhere between the four extreme
situations, for example a motor car is a mass product, but is subject to various
codes.
5.3.2 P HILOSOPHIES
The two basic philosophies of design against metal fatigue are safe life and
fail safe (Anon. 1986). They were originally developed in the aircraft industry
(Troughton 1960). In a structure designed on safe life principles, the objective
@Seismicisolation
@Seismicisolation
Metal Fatigue 71
is to ensure that the structure will not fail within the design life, which may
be indefinite. When the design life is reached, the component or structure
concerned is discarded. The design life can be expressed either as the fatigue
loading service, or in time based on the maximum likely rate of usage. For
example, lift cables are replaced at specified time intervals.
The safe life may be extended by periodic inspection for cracks. If one
is found, the component is discarded or repaired. In this case, a fatigue as-
sessment must have demonstrated that the largest crack likely to be missed
by the inspection technique used must not lead to failure before the next in-
spection. Steps to ensure that adequate inspection is feasible need to be taken
at the design stage. If the part concerned is cheap, such as the rotating head
of a centrifuge, The cost of inspection may well be greater than the cost of
replacing the part when the deign life is reached.
In a fail safe structure, alternative load paths are provided, and the struc-
ture is said to be redundant. This means that failure of any one member does
not lead to failure of the whole structure. Once the structure has lost redund-
ancy owing to failure of a member, it is less safe. Provision must be made,
perhaps by regular inspection, for any loss of redundancy to become known.
In practice, many structures are designed using a combination of fail safe and
safe life philosophies, for example tubular offshore structures (Kallaby and
Price 1978). A variation on fail safe design is to ensure that failed parts are
constrained before they can cause injury or excessive damage. For example,
failure of the rotating head of a centrifuge, if its contents are innocuous, is
merely an expensive nuisance, provided that the broken parts are contained
within the machine.
The incorporation of fail safe features makes the fatigue assessment of
safe life less crucial, but too much reliance should not be placed on fail safe
features. The metal fatigue failure of a cross bracing member on the semi sub-
mersible accommodation rig Alexander L. Kielland resulted in a catastrophic
capsize because, in the event, the structure proved to be non redundant (Anon.
1981b). Sister rigs were later modified by additional cross bracing in order to
ensure redundancy.
The damage tolerance approach to aircraft design was developed as a sup-
plement to the safe life and fail safe philosophies (Anon. 1974a, Kirkby et al.
1980). It requires that fatigue assessments be carried out using the assump-
tion that cracks of certain specified sizes, are present at all critical locations
such a rivet holes. It therefore directs particular attention to the determination
of fatigue crack propagation rates in structures.
@Seismicisolation
@Seismicisolation
5.3.3 A PPROACHES
There are three possible approaches to fatigue assessment; an analytical ap-
proach, service loading testing, and reliance on satisfactory experience with
previous similar designs. In practice, some combination of these is normally
used. Often, analytical methods are best used to extrapolate experimental data
to broadly similar situations. Where innovation is involved, ensuring the in-
tegrity of structures subject to fatigue loading can be an extremely expens-
ive, time consuming business. The United Kingdom Offshore Steels Research
Project is an example (Anon. 1974b). It was set up to collect fatigue and frac-
ture information relevant to tubular structures in the North Sea where wave
loading is very severe. Twenty-five years later it was still an active research
and development area (Etube 2001).
Whichever approach is used, allowance must be made for the inevitable
scatter in the fatigue life of specimens and structures (see Sections 3.3, 3.4
and 4.4.1), and factors such as uncertainty in service load history and stress
analysis (Pugsley 1966). Methods range from a safety factor based on ex-
perience (Pook 1983a, Dowling 1993), through the statistical methods used
to develop the design curves in BS 5400 (Gurney 1979, Anon. 1980a), to
the elaborate procedures that have been used in the assessment of nuclear
pressure vessels (Smith 1979). The degree of confidence required obviously
depends on the consequences of a failure. For example, the failure of the ro-
tating head of a centrifuge would be more serious if it were being used in a
medical laboratory and dangerous organisms escaped.
5.3.3.1 Analytical Approach

An analytical approach makes use of information on service loads, material
properties and applied mechanics. Such an approach requires expert know-
ledge. The fatigue crack initiation and fatigue propagation phases must be
considered separately, although in practice one or the other normally pre-
dominates. This is usually the crack propagation phase. The exceptions are
generally carefully made, highly stressed components where stress concen-
trations are kept to a minimum (Stephens et al. 2001). Typically, elaborate
calculations are necessary, and the approach may fail because not all the large
amount of detailed information required is available (Schütz 1979). The de-
sire to put fatigue assessment on an analytical basis has stimulated much of
the enormous amount of academic work which has been carried out on metal
fatigue.
Sometimes simplifications are possible, provided that they can be shown
to give conservative estimates of fatigue lives. For example, the fatigue life
@Seismicisolation
@Seismicisolation
Metal Fatigue 73
of the bicycle shaft shown in Figures 4.3 and 4.19 could be assessed using
estimates of the number of revolutions in the design life of the bicycle, the
fatigue loading shown schematically in Figure 4.21, the stress concentration
factors given in Section 4.5.1, and S/N data for the shaft material together
with an allowance for scatter taken from Table 3.1. In practice, the number
of revolutions is of the order of millions, so design would be based on the
fatigue limit of the material (see Section 3.2).
From the designer’s viewpoint, standardised analytical procedures of vari-
ous degrees of formality are undoubtedly the most satisfactory. These range
from informally established good practice in a particular design office to elab-
orate published codes, often imposed by regulatory authorities. If they give
sufficiently accurate answers they need only have an empirical basis. Dif-
ferences in detail design requirements, and in the stringency of quality con-
trol required, usually account for the different stresses permitted for similar
structures by different codes (Gerlach 1980, Zerbst et al. 2003). Standard-
ised methods have the advantage that little or no expert knowledge is re-
quired, and they can be conveniently incorporated in software packages for
computers. This facilitates the assessment of complex structures, for example
(Sumi 2005). Available packages for fatigue design should be carefully ex-
amined to ensure that the basis on which they operate is suitable for the in-
tended application (Sonsino et al. 2004).
Most metal fatigue assessments are based on simplified, standard proced-
ures which, despite their apparent lack of physical validity, are known to give
conservative answers. From time to time assessments are revised using later
versions of standards. The need to carry out a fatigue re-assessment of a struc-
ture sometimes arises when it has reached the end of its design life and an
extension is desired. This can cause problems when the structure was con-
structed using materials which are not permitted by current standards.
5.3.3.2 A Fatigue Re-assessment
Flow measurement faculties at the National Engineering Laboratory included

a unique high pressure gas flow primary standard test rig (Anon. 1979). The
main air storage vessel for this rig was a 1.8 m diameter welded steel cylinder
6.3 m long which had a volume of 11.5 m3 , and was intended to hold air pres-
sures up to 138 bar. The consequences of failure were therefore potentially
catastrophic. A fatigue re-assessment was needed because the fatigue loading
used during its design differed from the estimated fatigue loading.
The vessel was manufactured BS 1515 (Anon. 1965b). This was writ-
ten for Imperial units and these are therefore given with conversions to ISO
@Seismicisolation
@Seismicisolation
units added. The design life for the vessel was 15 years and the estimated
fatigue loading in terms of internal pressure was 1500 full pressure cycles of
0–2000 lbf/in2 (0–137.9 bar), 15000 fluctuating cycles of 1400–2000 lbf/in2
(96.5–137.9 bar).
Perusal of the manufacturer’s drawings and calculations during commis-
sioning showed that the vessel met the requirements of BS 1515, except that
the full pressure cycles had not been taken into account, and also that the
number of fluctuating cycles was taken as 21900 cycles. The manufacturer
had carried out a simplified fatigue assessment in accordance with BS 1515
Appendix B. In this the permitted number of fatigue cycles is quoted as

1160(3000 − T ) 2
N= , (5.1)
4S − 14500
where T is the temperature in degrees C, and S is the fatigue stress range.
The value of S is based on the nominal hoop stress. Stress concentrations due
to penetrations, etc. are allowed for in the code, and a corrosion allowance
is deducted from the actual vessel thickness. For this particular vessel the
internal diameter was 72 in (1.83 m) and the net thickness 2.75 in (69.9 mm).
Hence S = 13.614P , where P is the cyclic pressure range. For T =
50◦ C and the fluctuating loading, Equation (5.1) gives N = 35430. As this
was greater than 21900 the manufacturer had concluded, in accordance with
Appendix B, that the fatigue strength of the vessel was satisfactory.
To check this conclusion, further calculations were carried out to BS 1515
using the estimated fatigue loading. Where more than one fatigue loading
is involved, Appendix B requires that a Miner’s rule summation (see Sec-
tion 4.3) in the form
ni
≤ 1.0 (5.2)
Ni
be satisfied, where ni is the expected number of cycles for the ith load level,
and Ni is the permitted number of cycles for the ith load level calculated
using Equation (5.1).
For the fluctuating cycles ni /Ni = 0.423 and for the full pressure cycles
ni /Ni = 1.42, showing immediately that the vessel did not meet the require-
ments of Appendix B. A calculation for the expected fatigue loading using
Equations (5.1) and (5.2) showed that the vessel would have had a safe life
of only 9.6 years. This was not acceptable. However, keeping the fluctuating
pressure range (600 lbf/in2 , 41.4 bar) the same, but reducing the maximum
working pressure, provided a potentially acceptable solution. A further cal-
culation showed that the desired 15 year design life could be achieved by
reducing the maximum working pressure to 1499 lbf/in2 (103 bar).
@Seismicisolation
@Seismicisolation
Metal Fatigue 75
BS 1515 was obsolescent, so a further re-assessment was carried out using

the more recent BS 5500 Appendix C (Anon. 1976). Appendix C contains a
simplified fatigue assessment procedure, which was deliberately intended to
be very conservative. A calculation in accordance with Appendix C, using the
same fluctuating pressure range as before, showed that the maximum working
pressure should be 75.7 bar. Detail differences in the general requirements of
the two standards mean that this is not directly comparable with the higher
figure obtained from BS 1515, to which the vessel was designed. Neverthe-
less, it was decided that the maximum working pressure should be limited to
76 bar. This reduced maximum working pressure was not a serious operating
constraint.
Both standards allowed the fatigue clauses to be disregarded provided that
evidence of adequate fatigue life, such as more detailed calculations or prac-
tical experience of usage, was available. Details of the methods to be used
were not specified in the standards, and were left to agreement between the
manufacturer and the purchaser. Such escape clauses were at one time com-
mon in British Standards, but are not now permitted (Anon. 2005b).
5.3.3.3 Service Loading Testing
Modern servo hydraulic equipment permits the application of virtually any

load history (see Section 4.4) and can therefore be used to determine service
life where an analytical approach fails. Service loading testing of prototypes
has for some time been widely used for critical structures in various branches
of engineering (Marsh 1988). It is sometimes a requirement of regulatory
authorities. Service loading testing has the advantage that its basis is easily
understood by laymen. Acquiring the necessary information on service loads
may be a major problem, and expert judgement is needed on how represent-
ative the loading applied to the structure needs to be. Actual load histories
in service cover a very wide range of possibilities. They can be difficult to
characterise because they are usually statistically non stationary (see Sec-
tions 4.2.3 and 4.3). Service loading testing also has the advantage that weak
points in a design can be identified and rectified at the prototype stage. It
is sometimes the only way that fatigue crack paths and crack propagation
rates can be determined (Etube 2001). Simple comparative tests are some-
times useful in establishing whether a proposed modification does result in a
worthwhile increase in fatigue strength.
@Seismicisolation
@Seismicisolation
5.3.3.4 Acceptance Testing
Because of cost and time constraints, simplified service loading testing is used
as much as possible for acceptance testing purposes, in which prototypes are
required to withstand a specified fatigue loading without failure (Petrone and
Susmel 2003). A striking feature of an analysis of documents issued by the
British Standards Institution is the large number which include clauses re-
quiring acceptance fatigue testing of components (Pook 1997). Most of them
are product standards. Some examples are given below.
BS 2A 241: 2005. General requirements for steel protruding head bolts
of tensile strength 1250 MPa (180000 lbf/in2 ) or greater (Anon. 2005c). This
gives requirements for high strength steel bolts for aerospace use. A bolt is a
ubiquitous engineering component. It is not surprising that a long established
standard for high quality high strength steel bolts includes an acceptance fa-
tigue test. Sophisticated statistical criteria are used to determine the acceptab-
ility of a particular batch of bolts. The constant amplitude direct stress fatigue
loading specified is obviously not intended to represent any particular service
loading.
BS ISO 10771-1: 2002. Hydraulic fluid power. Fatigue pressure testing of
metal pressure containing envelopes (Anon. 2002). This was prepared as the
result of research, during which it became apparent that both the frequency
and the waveform have a pronounced effect on the internal pressure fatigue
life of metal hydraulic fluid power components made from materials that are
normally frequency independent (see Sections 3.2 and 7.4.2.3). The wave-
form of the pressure cycle, the maximum permissible viscosity of the pres-
surizing medium, and the maximum permissible pressure cycling rate are all
specified. Hydraulic system components are examples of safety related in-
dustrial components. The constant amplitude fatigue loading specified does
not represent any particular service loading. The standard uses fatigue tests
to define a fatigue pressure rating rather than precise acceptance criteria.
BS AU 50-2c: 1996. Tyres and wheels. Wheels and rims. Specification for
road wheels manufactured wholly or partly of cast light alloy for passenger
cars (Anon. 1996). This includes a radial fatigue acceptance test. For this test
the wheel is fitted with an appropriate tyre and a constant radial force, which
rotates around the wheel, is applied. A car wheel is an example of a safety
related vehicle component. It is not surprising that a long established standard
for car wheels includes an acceptance fatigue test. A constant amplitude fa-
tigue loading is representative only of a vehicle running at constant speed on
a straight and level road. The number of cycles specified is small compared
@Seismicisolation
@Seismicisolation
Metal Fatigue 77
with the number of service cycles, but to compensate for this a high fatigue
load is specified.
BS EN 60669-1: 2000, BS 3676-1: 2000. Switches for household and sim-
ilar fixed electrical installations. General requirements (Anon. 2000a). This
dual numbered standard includes a normal operation acceptance test. In this
test switches make and break a resistive load equal to their rated current, at
their rated voltage, in a substantially non inductive alternating current circuit.
Switches are operated for a specified number of operations. The acceptance
criteria are that a switch must remain operational, and mechanically and elec-
trically sound, throughout a test. An electrical switch is an example of an
electrical component with mechanical parts which might fail in fatigue due
to repeated operation. The test specified is defined in terms of a number of
normal operations. It is effectively a constant amplitude fatigue test on the
mechanical components of the switch.
BS EN 12983-1: 2000 Cookware. Domestic cookware for use on top of a
stove, cooker or hob (Anon. 2000b). This was originally prepared in response
to accident statistics which demonstrated that serious accidents can occur as
the result of the premature failure of handles of domestic cookware. It sets
levels of performance for cookware for use on top of a stove, cooker or hob
by the accelerated simulation of hazards experienced in normal use. The ac-
ceptance fatigue test specified involves continuously raising and lowering a
loaded item of cookware from a level surface once per minute by means of
its handle; this is a constant amplitude fatigue test. A cookware handle is an
example of a safety related item in domestic use. The acceptance criterion is
that there must be no permanent distortion or loosening of the handle or its
fixing system.
5.4 Product Liability
In general the law on product liability is complex and varies widely from
country to country. Anyone likely to be adversely affected by product liab-
ility legislation should seek legal advice, and perhaps take out appropriate
insurance cover. Unfortunately, there are no precise, internationally accep-
ted definitions of the various legal terms involved. For product liability to be
established a product has to be defective. A product that is not defective is
sometimes called a safe product. There is usually a distinction between crim-
inal liability and civil liability. For criminal liability to be established, that is
for a person to be held to have committed a criminal offence for which he (or
@Seismicisolation
@Seismicisolation
she) may be punished, it is usually necessary to prove that he (or she) was
negligent in some way in his (or her) association with the product.
For civil liability to be established, that is for compensation to be pay-
able to a person who is injured or suffers loss, it is sometimes necessary
to prove negligence. However, under strict liability it is merely necessary to
show that the product was defective, but some defences are possible. One pos-
sible defence is that a product conforms with relevant safety regulations (see
Section 5.5). The concept of strict liability now covers many situations, espe-
cially in personal injury cases. It is therefore of importance to those concerned
with all types of products. A variant of strict liability is absolute liability (or
no fault liability) where there is no defence.
Misuse of a product which is not defective can also lead to legal liability.
For example, the operator of a fairground ride who deliberately exceeded
the manufacturer’s recommended maximum speed had to pay compensation
when this resulted in catastrophic metal fatigue failure (Pook 1983a).
From the viewpoint of a designer the meaning of defective is obviously
of crucial importance. In the special case of metal fatigue, it can be helpful
to differentiate between design defects and manufacturing defects where the
designer’s intentions have not been followed during manufacture (see Sec-
tion 5.2). Once a metal fatigue failure has occurred, with the benefit of hind-
sight it is always possible to point to some action which would have pre-
vented the failure. However, it is difficult to reconcile the essentially random
nature of metal fatigue with any definition of defective which requires precise
prediction of future events (see Sections 3.3, 3.4, 4.4.1 and 5.3.3). This em-
phasizes the importance of selecting an appropriate strategy in design against
metal fatigue (see Section 5.3). If necessary, a designer must be able to show
that care had been taken.
5.4.1 T HE C ONSUMER P ROTECTION ACT
The UK Consumer Protection Act 1987 is a typical example of product liabil-

ity legislation (Anon. 2001a). The Act implements the European Community
directive on product liability, which provides a similar degree of protection
for people throughout the European Union. The aim of the Act is to help
safeguard the consumer from products that do not reach a reasonable level of
safety. Liability under the Act applies to all consumer goods, and to goods
used at a place of work. Under the Act, people injured by defective products
may have the right to sue for damages. The Act provides the same rights to
anyone injured by a defective product, whether or not the product was sold to
them. People can also sue for damage to private property. For civil liability
@Seismicisolation
@Seismicisolation
Metal Fatigue 79
to be established the plaintiff must be able to show that, on the balance of

probabilities, the defect in the product caused the harm (injury or damage).
For criminal liability this must be proved beyond reasonable doubt.
The Consumer Protection Act imposes strict liability on harm caused by
defective products. Hence the plaintiff does not have to prove negligence on
anyone’s part. Under the Act a defective product is defined as one where the
safety of the product is not such as persons are generally entitled to expect.
This definition provides an objective test of defectiveness and refers neither
to the harm caused by the product nor to the particular product. When de-
ciding whether a product is defective, a court takes into account all relevant
circumstances, including the manner in which a product is marketed, any in-
structions or warnings that are given with it, and what might reasonably be
expected to be done with it.
Clearly the definition of defective product in the Consumer Protection
Act, and similar definitions in other legislation, include products where cata-
strophic metal fatigue failures have occurred during normal use. It is obvi-
ously acceptable from the consumer viewpoint. However, from the point of
view of those responsible for a product it is unsatisfactory, in that it could be
very difficult indeed to decide in advance whether a particular product might
be regarded as defective.
5.4.2 E NFORCEMENT AUTHORITIES
Product liability legislation makes provision for enforcement authorities who

enforce the legislation. Complaints about defective products can be made to
the relevant enforcement authority. In most of the UK enforcement of the
Consumer Protection Act is primarily the responsibility of trading stand-
ards officers of local authorities, who act as enforcement officers (Anon.
2001a). Under the act, enforcement officers have, inter alia, authority to make
test purchases, seize goods, enter premises for the purpose of ascertaining
whether there has been a breach of the Consumer Protection Act, and bring
prosecutions.
5.5 Safety Regulations
Products may be subject to safety regulations whose purpose is to ensure

that products are safe under normal or reasonably foreseeable conditions of
use. The aim of safety regulations is to ensure that only safe products are
marketed. This is in contrast to product liability legislation which aims to
@Seismicisolation
@Seismicisolation
identify defective products (see Section 5.4). Design guidance in safety reg-
ulations must be followed by designers whenever this is a legal requirement,
a contractual requirement, or both. It is sometimes very detailed. Preventat-
ive measures in safety regulations often form part of quality systems such as
those specified in the BS EN ISO 9000 family of standards (Anon. 2005e).
Safety regulations are overseen by regulatory authorities. For example, a
passenger carrying aircraft is not allowed to fly without a certificate of air-
worthiness from an appropriate regulatory authority. Another example is the
periodic inspection of motor vehicles by a regulatory authority that is a legal
requirement in many countries.
Unfortunately, categorising products as either safe or unsafe means that
safe products are sometimes wrongly categorised as unsafe products, when
what is meant is that they have not been proved to be safe. The resulting
expense causes much resentment among consumers.
5.5.1 T HE G ENERAL P RODUCT S AFETY R EGULATIONS
The UK General Product Safety Regulations 2005 are a typical example of

product safety regulations (Anon. 2005d). They cover both new and second
hand goods intended for, or likely to be used, by consumers.
The General Product Safety Regulations were made under the European
Communities Act 1972, and transpose a European General Product Safety
Directive into UK law. The Directive pursues its principal objective of en-
suring consumer product safety by specifying that products must be safe,
defining a safe product, imposing obligations on producers and distributors
consistent with marketing safe products, laying down a framework for assess-
ing safety, and requiring that enforcement authorities be empowered to take
action to protect consumers from unsafe products. Under the Regulations it
is an offence for a producer to place a product on the market, or supply a
product, unless it is safe.
Where specific product legislation covers exactly the same ground as the
General Product Safety Regulations the specific legislation applies. The UK
Department of Trade and Industry has overall policy responsibility for the
Regulations, but responsibility for the safety of some consumer products rests
with other Departments. For example, the Vehicle Operator Services Agency
leads on motor vehicle safety.
In the General Product Safety Regulations a safe product is defined as any
product which under normal, or reasonably foreseeable conditions of use,
presents no risk, or only minimum risk compatible with the product’s use,
@Seismicisolation
@Seismicisolation
Metal Fatigue 81
and which is compatible with a high level of protection for consumers. Un-
der the Regulations, the safety of a product is assessed using various consid-
erations including the product’s characteristics, packaging, instructions for
assembly, maintenance, use and disposal, labelling and other information
provided for the consumer, and the categories of consumer at risk when using
the product. The safety of some products depends on how they have been in-
stalled and maintained. These services are an essential feature of the safety of
the product, and may form part of the contract to supply the product. As such
they are taken into account when judging whether a product is a safe product.
@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
6
The Uncracked Situation
6.1 Introduction
In plain metallic specimens, such as those shown in Figures 3.2 and 3.4, or in
the presence of a mild notch, such as that shown schematically in Figure 4.18,
fatigue lives are dominated by fatigue crack initiation, rather than by fatigue
crack propagation (Frost et al. 1974). If fatigue crack initiation is defined as
the development of a small crack, say 0.5 mm deep, then typically 80 per cent
of the fatigue life is occupied by fatigue crack initiation.
Very few components are of uniform cross section; most contain some
form of change of cross section resulting from a discontinuity such as a fil-
let, a hole, or an external groove or notch. The bicycle crankshaft shown in
Figure 4.19 is an example. For convenience, in metal fatigue, any of these
discontinuities is generally referred to as a notch, irrespective of its geomet-
ric shape. Failure analysis of components often shows that a fatigue crack
had initiated at some point at a notch root. A crack initiates here because
the fatigue stresses at, or near, the notch root are higher than the nominal
fatigue stresses away from the notch. The car drive line component shown
in Figure 1.1 and the marine steam turbine blade shown in Figure 3.15 are
examples of fatigue failures originating at a notch. The brass chain link, Fig-
ure 1.4, is an example of a fatigue failure in a region of uniform cross section.
All three are examples of crack initiation dominated situations, that is un-
cracked situations.
In crack initiation dominated situations a fatigue assessment using an
analytic approach is relatively straightforward, especially if an appropriate
standard method is available (see Section 5.3.3.1). In this chapter some of the
factors affecting the fatigue strength of metals in crack initiation dominated
situations are discussed and illustrated by examples.
@Seismicisolation
@Seismicisolation
84 Chapter 6: The Uncracked Situation
Over the years much effort has been devoted to the analysis of exper-
imental metal fatigue data for plain and mildly notched specimens with a
view to their presentation in forms that can be used in analytic approaches to
fatigue assessments (Gough 1926, Frost et al. 1974, Stephens et al. 2001).
6.2 Effect of Surface Finish
Fatigue crack initiation in a metal is, in general, associated with a free sur-
face (see Sections 3.4.1 and 3.4.2). Hence, the fatigue strength of plain spe-
cimens, especially at long endurances depends on the surface finish, includ-
ing the condition and roughness of specimen surfaces. This in turn depends
on the techniques used for specimen manufacture such as grinding, turning,
forging and extrusion. There are three main reasons why manufacturing tech-
niques used to prepare specimen surfaces may affect fatigue strength. First,
notch like surface irregularities may have been created, for example machin-
ing marks. Secondly, the condition of the material at the surface may have
been changed, for example it might have been hardened by cold work. Fi-
nally, residual stresses might have been introduced into the surface layers.
An estimate of the magnitude of the surface irregularities created by a
particular machining process can be obtained from a profile record of the
specimen surface. However, these records do not necessarily provide an ac-
curate record of the surface profile because the probe cannot explore a groove
narrower than itself. The roughness is often expressed by a single figure; this
is the centre line average roughness (CLA) over the distance traversed by the
probe. The higher the figure, the rougher the surface. The CLA is frequently
used to specify surface finishes on workshop drawings. Surfaces are normally
much smoother when measured parallel to the final direction of machining
than they are when measured perpendicular to this direction. Measured per-
pendicular to the direction of machining a mechanically fine turned or ground
finish may have a CLA value of 0.125 µm, and a rough ground or turned fin-
ish a value of 0.75–1.25 µm (Frost et al. 1974). Unfortunately, the CLA is not
an adequate parameter for evaluating the effect of surface finish on fatigue
strength; the pitch. that is the distance between successive grooves must also
be taken into account (Murakami 2002). Nevertheless, experimental data do
provide qualitative guidance on the magnitude of effects.
In evaluating the effects of surface finish on fatigue strength under con-
stant amplitude fatigue loading the concept of intrinsic fatigue strength is
sometimes used (Frost et al. 1974). This is the fatigue strength for carefully
polished specimens in which care has been taken to avoid the introduction of
@Seismicisolation
@Seismicisolation
Metal Fatigue 85
Table 6.1. Surface factors for a 0.4% C steel, tensile strength 570 MPa.
Condition Fatigue limit (MPa) Surface factor
Polished, 0000 emery ±280 1
Buffed with red lead ±276 0.99
Polished, 0 emery ±272 0.97
Ground, 120 wheel (fine) ±268 0.96
Ground, 46 wheel (medium) ±258 0.92
Ground, 30 wheel (coarse) ±232 0.83
residual stresses, or hardening or softening the surface layers. Final polishing

is in the longitudinal direction. Specimens are sometimes electropolished to
remove surface layers that might have been modified during specimen manu-
facture.
The effect of surface finish can conveniently be characterised by the sur-
face factor. This is the ratio of the fatigue strength for a particular surface
finish to the intrinsic fatigue strength, and is normally, as would be expected,
less than 1. Some typical surface factors for constant amplitude fatigue load-
ing at zero mean stress are shown in Table 6.1 (Frost et al. 1974). Dowling
(1993) gives surface factors for the fatigue limit of a range of steels. These
are shown as plots of surface factor against tensile strength for several qual-
itatively described surface conditions. In general the surface factor decreases
as the tensile strength increases.
6.3 Effect of Mean Stress
Figure 3.1 shows the constant amplitude fatigue loading σm ± σa , where σm

is the mean stress and σa is the alternating stress. The fatigue strength of a
metallic material, both at given endurances and at the fatigue limit, depends
on the mean stress. In general for a tensile mean stress, the fatigue strength,
expressed in terms of the alternating stress, decreases as the mean stress is
increased. In order to avoid the need to carry out large numbers of tests,
attempts have been made, starting in the 19th century, to find relationships
between fatigue strength and mean stress (Frost et al. 1974). In his pioneering
experiments Wöhler investigated the effects of mean stress on fatigue strength
(Anon. 1871). A contemporary analysis showed that Wöhler’s tensile mean
stress data for fatigue limits conformed to a parabola, having as end points the
fatigue limit at zero mean stress, ±σ0 , and the tensile strength of the material,
σt (Gerber 1874). The Gerber parabola is given by
@Seismicisolation
@Seismicisolation
Figure 6.1. Gerber diagram.
Figure 6.2. Goodman diagram.
2
σm
±σ = ±σ0 1 − . (6.1)
σt
The corresponding Gerber diagram is shown in Figure 6.1.
The Goodman diagram, shown in Figure 6.2, is a straight line given by

σm
±σ = ±σ0 1 − (6.2)
σt
is now the most widely used relationship for both fatigue limits and fatigue
strengths at given endurances. In its original form σ0 in Equation (6.2) was
taken as σt /3 (Goodman 1899). Consequently, Figure 6.2 is sometimes called
the modified Goodman diagram.
The Goodman line is more conservative than the Gerber line (cf. Fig-
ures 6.1 and 6.2). In practice most experimental data lie between the Good-
man and Gerber lines, and some typical data are shown in Figures 6.3 and 6.4.
This means that the Goodman diagram is usually conservative. The Goodman
@Seismicisolation
@Seismicisolation
Metal Fatigue 87
Figure 6.3. Effect of mean stress on the fatigue strength at 5 × 108 cycles of two wrought
aluminium alloys (Frost et al. 1974).
Figure 6.4. Effect of mean stress on the fatigue strength at 5 × 107 cycles of a wrought 5.5%
Zn-magnesium alloy (Frost et al. 1974).
diagram has been criticised, for example by Cazaud (1953), as being insuf-
ficiently accurate, and it should not be used unless it has been established as
satisfactory for a given set of circumstances.
To ensure that neither yielding nor fatigue failure occurs, a diagram of
similar form to the Goodman diagram has been proposed in which the cri-
terion of failure at zero alternating stress is taken as the yield stress (or 0.1
@Seismicisolation
@Seismicisolation
per cent proof stress) instead of the tensile strength. The straight line joining
this to σ0 is sometimes called the Soderberg line (Frost et al. 1974).
More elaborate expressions are sometimes used. For example, Heywood
(1962) derived an experimental relationship from analysis of available exper-
imental data which can be written in the form

σ σm σ0 σ0
= 1− +γ 1− , (6.3)
σt σt σt σt
where for steels

σm σm
γ = 2+ (6.4)
3σt σt
and for aluminium alloys
−1
σm σt log N 4
γ = 1+ , (6.5)
σt 2200
where stresses are in MPa and N is the number of cycles at which the fatigue
strength is estimated. The parameter γ can be regarded as characterising the
curvature on an alternating stress versus mean stress diagram. For γ = 0
Equation (6.5) reduces to Equation (6.1) but with ± signs omitted.
6.3.1 E FFECT OF R ESIDUAL S TRESSES

When residual stresses are present in a metallic specimen, component or
structure they are superimposed on applied fatigue loads. Hence, in general,
residual stresses affect fatigue behaviour in the same way as an applied mean
stress. It follows that in crack initiation dominated situations residual stresses
at a surface are favourable if compressive, but detrimental if tensile. In prin-
ciple assessment of the effect of residual stresses is therefore straightforward,
but in practice the very wide range of possibilities, together with the diffi-
culty of determining the levels of residual stress, mean that empirical methods
have to be used. The deliberate introduction of surface compressive residual
stresses by shot peening is a well known and widely used method of improv-
ing fatigue strength (see Section 6.8).
6.4 Effect of Multiaxial Fatigue Loading

A number of criteria have been proposed for failure under in phase multiaxial
fatigue loading (see Section 4.5.2). Experience has shown that, in crack ini-
tiation dominated situations, the von Mises criterion (Equation (4.17)) some-
times works well for metallic materials. In situations where fatigue cracks
@Seismicisolation
@Seismicisolation
Metal Fatigue 89
Figure 6.5. Fatigue limit of 3.5% Ni steel under multiaxial loading (Frost et al. 1974).
are initiated at a free surface only the biaxial surface stresses need to be con-
sidered, and the von Mises criterion (Equation (4.18)) becomes

σe = σx2 − σx σy + σy2 + 3τxy 2 , (6.6)
where σe is the von Mises equivalent stress, and σx , σy and τxy are Cartesian
stress field components.
The von Mises criterion predicts
√ that the ratio of the fatigue limit in torsion
(shear) to that in tension is 1/3 (≈ 0.577). Data for 89 metallic materials
gave an average value for this ratio of 0.559 (Frost et al. 1974). Fatigue limit
data for two steels tested under combined in phase (proportional) tension and
torsion are shown in Figures 6.5 and 6.6. Scales are chosen so that the von
Mises criterion (solid lines) becomes a straight line. The dashed straight lines
are fits to the experimental points. Agreement with the von Mises criterion
is good for both steels. Some experimental data for proportional multiaxial
fatigue loading do not conform to the von Mises criterion, and it should only
be used in situations where it is known to give satisfactory results. The von
Mises criterion cannot be used for non proportional fatigue loadings (see Sec-
tion 4.5.2.1).
6.5 Effect of Notches
The effect of a notch on metal fatigue behaviour may be illustrated by con-

sidering the cylindrical specimen with a circumferential notch, under uniaxial
@Seismicisolation
@Seismicisolation
Figure 6.6. Fatigue limit of Cr-V steel under multiaxial loading (Frost et al. 1974).
loading, shown in Figure 4.18. If it is subjected to a static tensile load which

gives rise to a nominal stress on the net cross section at the notch root of,
say, 80 per cent of the yield stress, it will not break, even though the metal in
the highly stressed region in the vicinity of the notch root will have deformed
plastically. However, if the specimen is subjected to a fatigue load in which
the nominal stress range on the net section is equal to 80 per cent of the plain
fatigue limit fatigue, cracks will be initiated at the notch root because the
local stress range is greater than the fatigue limit of the material. Once initi-
ated, fatigue cracks then propagate across the net section and cause complete
failure.
The need to provide design data has led to widespread constant amplitude
fatigue testing of notched specimens. Just as it is possible to estimate fatigue
strengths (fatigue limit or strengths at given endurances) for plain specimens,
it is also possible to estimate fatigue strengths for notched specimens. The
ratio of a fatigue strength for a plain specimen, to the corresponding fatigue
strength in the presence of a given notch, based on nominal stresses rather
than stresses at the notch root, is called the fatigue strength reduction factor,
Kf .
Figure 6.7 shows the S/N curve for notched mild steel cylindrical speci-
mens tested in direct stress at zero mean stress. The circumferential V-notch
was 5 mm deep with an included angle of 55◦ , and a root radius of 0.13 mm.
Nominal stresses are based on the net cross section. The results show that the
notched specimen fatigue limit is about ±58 MPa. It would have a different
@Seismicisolation
@Seismicisolation
Metal Fatigue 91
Figure 6.7. S/N curve for notched mild steel specimens tested in direct stress at zero mean
stress. Arrows indicate unbroken specimens (Frost et al. 1974).
value if specimens containing a notch of a different geometry were tested,

or if stresses were based on the gross cross section rather than the net cross
section. That is, the notched fatigue limit depends on specimen geometry and
method of analysis, as well as on the material. The severity of a notch is
usually characterised by the stress concentration factor, Kt . This is the ratio
of the maximum elastic stress at a notch root and the nominal elastic stress.
Compilations of stress intensity factors for a wide range of geometries are
available, for example Pilkey (1997). In using stress concentration factor data
it is important to check exactly how the nominal stress was calculated. Tradi-
tionally, stress concentration factors were presented graphically for a range of
related geometries (Peterson 1953, 1974). This is convenient for preliminary
calculations.
While the fatigue behaviour of any particular notched configuration can
always, given sufficient effort, be analysed in detail, extracting generalisa-
tions of reasonably wide applicability is very difficult. It might appear obvi-
ous that if failure due to metal fatigue is to be avoided then the alternating
stress at a notch root must not exceed the plain specimen fatigue strength
of the material, in other words Kf = Kt . However, experimental data show
@Seismicisolation
@Seismicisolation
that Kf may be less than Kt , and sometimes very much less. This has led to
the description of materials as notch sensitive and notch insensitive. This is
sometimes expressed numerically by the notch sensitivity index, q, given by
(Peterson 1974)
Kf − 1
q= . (6.7)
Kt − 1
If Kf = Kt then q = 1, and a material is said to be fully notch sensitive,
whereas if Kf = 1, q = 0, and a material is said to be fully notch insensit-
ive. Unfortunately, q is not a material constant, and this classification can be
seriously misleading. This has been demonstrated by the analysis of a large
amount of experimental data for the fatigue limits of metallic materials, which
may be summarised as follows (Frost et al. 1974).
(a) For low values of Kt , Kf may equal Kt , but in general it is somewhat less.
(b) Different geometries producing the same Kt may give differing Kf values.
(c) For high values of Kt , Kf is often very much less than Kt ,
(d) For a given material and certain notch geometries, there appears to be
a particular value of Kt at which Kf reaches a maximum value; higher
values of Kt result in no further increase in Kt .
There are two major factors which affect metal fatigue behaviour in the pres-
ence of a notch. First, a stress concentration factor gives only the maximum
stress at a notch root, it does not describe the stress field or its degree of biaxi-
ality. For the cylindrical specimen with a circumferential notch under uniaxial
loading shown in Figure 4.18, there is a biaxial stress system at the notch root.
According to the von Mises criterion (Equation (6.6)) the hoop stress at the
notch root can increase the fatigue stress required for failure by up to about
13 per cent, depending on the degree of constraint (Frost et al. 1974). Differ-
ences in stress gradients around notches can also have a significant effect on
material behaviour (Schijve 1980). The nature and extent of any yielding at
a notch is strongly dependent, both on the notch root stress field, and also on
material properties (Dowling 1993).
Secondly, the fatigue life of a notched specimen is the sum of the number
of cycles required to initiate a fatigue crack, and the number of cycles required
to propagate the crack to failure. For a blunt notch (small Kt ) fatigue life is, as
for plain specimens, crack initiation dominated (see Sections 6.1 and 6.4). By
contrast, for a sharp notch (large Kt ) cracks may initiate quickly at the notch
root, even at low nominal stress levels, and their propagation across the speci-
men may then occupy the major part of the fatigue life. The fatigue life is then
crack propagation dominated. In extreme situations the phenomenon of non
@Seismicisolation
@Seismicisolation
Metal Fatigue 93
propagating cracks appears. That is, for sufficiently sharp notches, initiated
fatigue cracks propagate for a short distance, of the order of a millimetre, and
then stop (see Sections 7.4.3 and 7.4.5).
The fatigue limit data shown in Figure 6.8 were obtained from 43 mm
diameter cylindrical mild steel specimens containing circumferentional V-
notches, 1.3 mm deep, tested in rotating bending, that is at zero mean stress.
The notched specimen fatigue limits, based on nominal gross cross sec-
tion stresses, are shown plotted against the stress concentration factor Kt .
The curved line represents the plain specimen fatigue limit of the mater-
ial (260 MPa) divided by Kt . When Kt exceeds a value of about 3 the
notched specimen fatigue limit, based on complete failure, is constant at about
90 MPa, as shown by the horizontal line. The junction between the two lines
is sometimes called the branch point. Above the branch point, the necessary
and sufficient criterion for complete failure is the initiation of a fatigue crack,
and this is correctly predicted by the notch root fatigue stresses. Below the
branch point, fatigue crack initiation is correctly predicted by notch root fa-
tigue stresses, but this is not a sufficient condition for complete failure, and
non propagating cracks may be present in unbroken specimens after testing
to very long endurances.
Most books on metal fatigue, starting with Gough (1926), draw attention
to the need to avoid inadvertent stress raising notches, such as machining
marks and accidental scratches. Gough also remarks ‘A number of sudden
failures in connecting rods and the valve springs of Diesel engines have res-
ulted from the practice – which cannot be too strongly condemned – of stamp-
ing an inspector’s mark upon these components.’ This advice still holds good.
Deep machining marks, transverse to fatigue stresses, are significant notches
(Murakami 2002). As an example, Figure 6.9 shows the fatigue failure of a
hammer head that originated at transverse machining marks. In a hammer
head, tensile fatigue loads arise from reflected shock waves at impact. This
particular hammer was a favourite tool that had been used by a mechanic for
many years.
6.6 Bicycle Crankshaft Analysis
Simplified fatigue assessments are often helpful at the preliminary design

stage in order to check that a proposed design is feasible from the metal fa-
tigue viewpoint. Analysis of the bicycle crankshaft shown in Figures 4.3 and
4.19 provides an example. The 16.65 mm diameter crankshaft is supported
by 6.35 mm diameter ball bearings, which run in the outer radii of the two
@Seismicisolation
@Seismicisolation
Figure 6.8. Nominal stress at fatigue limit versus stress concentration factor for notched mild
steel specimens tested in rotating bending (Frost et al. 1974).
shoulders, as indicated by the arrow in Figure 4.20. Forces applied by the

rider to the pedals are transferred through the cranks to the squared ends of
the crankshaft. The crank length is 170 mm, and the distance from the centre
of a pedal to a bearing is 120 mm. The most significant stresses are those due
to bending and torsion of the crankshaft, and the most highly stressed regions
are where the outer radii blend into the crankshaft (see Section 4.5.1).
A convenient and conservative simplifying assumption for the fatigue
loading of the crankshaft is that a vertical force equal to the rider’s weight is
applied to the centre of a pedal during the downward half cycle of the pedal,
and that the force is zero during the upward half cycle (see Section 4.5.1). The
resulting bending moments and torques at the outer radii are shown schemat-
ically in Figure 4.21.
The bending moment, M, at the nearer outer radius on the crankshaft is
constant during a downward half cycle of a pedal. Its value is equal to the
pedal force (700 N) times the axial distance from an outer radius to the centre
@Seismicisolation
@Seismicisolation
Metal Fatigue 95
Figure 6.9. Fatigue failure of a hammer head from machining marks. National Engineering
Laboratory photograph. Reproduced under the terms of the Click-Use Licence.
of a pedal (120 mm), so is 84 Nm. The bending moment is zero during an

upward half cycle, and is zero throughout a cycle at the outer radius further
from the pedal. The direct stress, σx , due to the bending moment, M, is given
by
32M
σx = , (6.8)
π d3
where d is the crankshaft diameter, and is 185.37 MPa. Multiplying by the
stress concentration factor (1.32) gives σx = 244.69 MPa.
A force on the left hand pedal produces a torque during a downward half
cycle. This torque passes through the crankshaft to the chain wheel. Hence, it
appears at both outer radii on the crankshaft. The torque varies sinusoidally.
It is zero when the crank is vertical, and has its maximum value when the
crank is horizontal. The maximum value is equal to the pedal force (700 N)
times the crank length (170 mm) so is 119 Nm. The torque is zero during
the upward half cycle. The right hand crank is attached directly to the chain
wheel (Figure 4.4) so torque due to a force on the right hand pedal passes dir-
ectly into the chain wheel, and the torque in the crankshaft is zero throughout
a cycle. At the left hand outer radius the bending moment and torque are
@Seismicisolation
@Seismicisolation
in phase, whereas at the right hand outer radius they are 180◦ out of phase
(Figure 4.21). The shear stress, τxy , due to the torque, T , is given by
16T
τxy = (6.9)
π d3
and is 131.30 MPa. Multiplying by the stress concentration factor (1.14) gives
τxy = 149.69 MPa.
Taking the design life of the bicycle as 40000 km, and the average distance
travelled for each turn of the pedals as 8 m, the design fatigue life for the
crankshaft is 5 × 106 cycles. This large number of cycles means that fatigue
limits may be used to select an appropriate material for the crankshaft.
It is difficult to generalise on whether in phase or out of phase biaxial
fatigue loadings are the more severe. However, for fatigue limit design un-
cracked situations fatigue crack initiation must be avoided, and it is reason-
able to assume that the in phase fatigue loading is the more severe. Hence
σx and τxy may be combined using Equation (4.19) to obtain the von Mises
equivalent stress, which is 356.5 MPa. Since the minimum bending moment
and torque are both zero the design values of the alternating stress, σa , and
the mean stress, σm , are both 356.5/2 = 178.25 MPa (Figure 3.1).
Assume that a steel with a tensile strength of 700 MPa and a fatigue limit,
at zero mean stress (stress ratio R = −1), of 300 MPa, is being considered for
the crankshaft. The fatigue limit when the mean and alternating stresses are
equal may be found by using a Goodman diagram (Figure 6.2). Entering val-
ues in Equation (6.2), which is the equation for a Goodman diagram, gives the
fatigue limit as 210 MPa. This is greater than 178.25 MPa, so the proposed
design appears to be satisfactory. However, conventional fatigue limits are
for a probability of failure of 50 per cent. Table 3.1 lists normalised stresses
for various probabilities of failure. The present calculations give a normal-
ised stress of 178.25/210 = 0.849. Comparing this with Table 3.1 gives an
estimated probability of failure of about 5 per cent, which might not be ac-
ceptable. It may therefore be concluded that the proposed design is probably
satisfactory, but that a more detailed fatigue assessment is needed.
6.7 Scatter under Variable Amplitude Fatigue Loading
Scatter of fatigue lives under constant amplitude fatigue loading of metal-

lic specimens has been studied extensively for many years (see Section 3.3).
Much experimental data has been accumulated, and methods of analysis are
well established. The situation for variable amplitude fatigue loading is quite
@Seismicisolation
@Seismicisolation
Metal Fatigue 97
different. Very little experimental data is available and, in general, it is pos-

sible to make only qualitative observations on the amount of scatter in various
situations, for example Schijve (2005). This situation is unexpected in view of
the practical importance of the variable amplitude fatigue loading of metallic
components and structures. However, there does seem to be a consensus that
the amount of scatter under variable amplitude fatigue loading is usually less
than for constant amplitude fatigue loading. This would imply that carrying
out a Miner’s rule summation (see Section 4.3) using a P -S-N curve for a
low probability of failure could be expected to lead to a safe result.
Miner’s rule summations are usually based on conventional S/N curves,
for
which the probability of failure is 50 per cent, and it is assumed that
(ni /Ni ) = 1 (Equation (4.10)). In practice experimental data lead to a
wide range of (ni /Ni ) values (see Section 4.3). These values are typically
in the range 0.3 to 3, but the range can be as much as 0.1 to 10. Although
it is not usually done, these data could simply be regarded as scatter and
analysed statistically. Implicitly, the use of a hypothetical S/N curve does
take this scatter into account (see Section 4.3). Analysing data in terms of
stress, instead of life, a range of 0.3 to 3 becomes a range of around 0.7 to
1.4. Exact values depend on the shape of the S/N curve. This range is still
wide, but nevertheless comparable with variations in service load histories
that can occur. The car drive line component failure shown in Figure 1.1 was
associated with an unusually severe service load history.
As pointed out earlier (see Section 5.1), many practical metal fatigue prob-
lems have been solved by taking some action to improve fatigue performance
on what amounts to a trial and error basis. What seems to have happened over
scatter under variable amplitude fatigue loading is that accumulated lore has
been taken into account when writing standards and regulations on design
against metal fatigue, but that the underlying data used by the writers has
not been made readily available. This sometimes happens when confidential
information, (whose source cannot be acknowledged) is used by standards
writers. Requirements to use particular methods in design against metal fa-
tigue thus mean that, in effect, scatter under variable amplitude fatigue load-
ing is indeed taken into account (see Section 4.4.1).
6.8 Improvement of Fatigue Life
The results of fatigue assessments and failure analyses often reveal the need
to improve the fatigue lives of existing, or proposed, components and struc-
tures. Basically, there are only two ways in which fatigue lives can be im-
@Seismicisolation
@Seismicisolation
proved. The first is to reduce fatigue stresses at critical locations, and the
second is to improve the fatigue strength of the material used. The first is the
province of the engineer and the second the province of the metallurgist. The
development of materials with improved fatigue strength is a popular topic in
the metal fatigue literature. If a material with better fatigue strength is avail-
able, then changing the material can be a straightforward way of improving
the fatigue strength of components and structures, but will usually involve
increased material costs, and often involve increased manufacturing costs.
Fatigue stresses can be reduced by reducing the applied fatigue loading.
For example, flying an aircraft around severe turbulence, rather than through
it, reduces gust loading of the aircraft’s wings, as well as improving the com-
fort of passengers and crew. Sometimes, fatigue loadings are reduced by im-
posing operational constraints (see Section 5.3.3.2).
Much published advice on the alleviation of metal fatigue amounts to hints
and tips on how to reduce the level of fatigue stresses at actual or potential
failure sites (see Section 5.1). This is often easier said than done. For ex-
ample, consider the bicycle crankshaft shown in Figures 4.19 and 4.20 (see
Section 6.6). Fatigue stresses at the outer radii of the shoulders could be re-
duced by increasing the diameter of the crankshaft, but this would impose
a cost and weight penalty, which might not be acceptable. One way to re-
duce fatigue stresses is to increase radii at shoulders on crankshafts. Made
up ball bearings are sometimes used to support bicycle crankshafts. For these
bearings to fit snugly against the shoulders there is a maximum permissible
radius at a shoulder. This means that there may be a limit to the extent to
which fatigue stresses can be reduced by increasing a radius.
In crack initiation dominated situations, metal fatigue is usually a surface
phenomenon (see Section 3.4). Hence surface hardening is sometimes used
to improve overall fatigue strength by making fatigue crack initiation more
difficult. Refinements of surface hardening techniques to optimise them for
particular situations is a popular topic in the metal fatigue literature. The in-
troduction of compressive residual stresses at the surface, hence reducing the
severity of the applied fatigue stresses is another method of improving fatigue
strength (see Section 6.3.1). The surface compressive residual stresses are, of
course, balanced by tensile residual stresses elsewhere.
There are three main metallurgical techniques commonly used to produce
a hardened surface layer on steel (Frost et al. 1974): induction hardening or
flame hardening, carburising, and nitriding. Induction and flame hardening
consist of heating the surface above the critical temperature of the steel, and
then quenching to produce a hard martensitic surface layer. Both carburising
@Seismicisolation
@Seismicisolation
Metal Fatigue 99
and nitriding produce a hardened layer on the surface of suitable steels. Car-
burising consists of heating the steel in a carbon bearing environment and
then quenching. Nitriding consists of heating in an ammonia environment so
that nitrogen combines with certain elements in the steel. It has the advantage
that a quench is not needed, so minimising any distortion, but it is benefi-
cial to lap or hone the finally hardened surface in order to remove a brittle
surface skin. All three processes have the additional benefit of introducing
compressive residual stresses at the surface.
Shot peening is a widely used and low cost method of introducing bi-
axial compressive residual stresses into the surface of a metallic material,
and hence improving its fatigue strength. In shot peening the surface is bom-
barded by small, hard spheres, typically 0.1 to 1 mm in diameter. The mech-
anisms by which the residual stresses are introduced are surprisingly complex
(Kobayashi et al. 1998). Typically, shot peening produces biaxial compress-
ive residual stresses in a layer 0.1 to 0.5 mm deep. It has been shown to
significantly increase fatigue strength in a wide range of metals (Frost et al.
1974). A method of measuring the intensity of shot peening for quality con-
trol purposes is well established (Almen and Black 1963, Anon. 2001b). This
is based on shot peening a thin steel Almen strip (3 in (76 mm) long ×0.75 in
(19 mm) wide), which is bolted to a stiff support. After the bolts have been
removed, the Almen strip is curved due to the residual stresses induced by the
shot peening. The amount of curvature is taken as a measure of the intensity
of the shot peening.
Other methods of introducing surface compressive residual stresses can
be effective in increasing fatigue strength. If a notched cylindrical specimen
(Figure 4.18) is loaded in tension at a load high enough to deform the material
plastically at the notch root, then compressive residual stresses are induced at
the notch root on unloading. For example, applying a static tensile stress of
390 MPa to 4% Cu aluminium alloy specimens containing a circumferential
V-notch 1 mm deep and 0.1 mm root radius, increased the rotating bending
fatigue strength, at 5 × 106 , cycles from ±58 MPa to ±123 MPa (Frost et
al. 1974). This effect is sometimes exploited by the proof loading of safety
critical components, such as chains used for lifting purposes, for example
as specified in Anon. (1997). In proof loading, a specified load higher than
the maximum working load is applied before a component is put into service.
Conversely, applying a static compressive stress can introduce tensile residual
stresses, and hence decrease the fatigue strength.
In a thick walled cylinder, internal pressurisation leads to higher stresses
at the bore than elsewhere. In autofrettage a pressure high enough to cause
@Seismicisolation
@Seismicisolation
yielding at the cylinder bore is applied (Hill 1950). On removal of the pressure
there are residual compressive stresses at the bore. Autofrettage is used to
improve the fatigue strength of internally pressurised cylinders such as large
gun barrels (Newhall 1999).
@Seismicisolation
@Seismicisolation
7
The Cracked Situation
7.1 Introduction
Because of the nature of metal fatigue mechanisms, the fatigue lives of speci-
mens, components and structures are sometimes dominated by fatigue crack
initiation, and sometimes by fatigue crack propagation (see Sections 3.4, 6.1
and 6.5). From a practical viewpoint, probably the most significant advance
in the understanding of metal fatigue behaviour was the general realisation,
some 30 years ago, that many components and structures are crack propaga-
tion dominated (Frost 1975). Cracks, or crack like flaws, may be introduced
during manufacture, especially if welding or casting is used, or cracks may
form early on during service (Pook 2000a, Murakami 2002). As examples of
cracked situations, Figure 7.1 shows multiple fatigue cracking from crack like
flaws in a 25 mm thick structural steel cruciform welded joint, and Figure 7.2
fatigue cracking from shrinkage in a 30 × 35 mm cast steel bar.
In the presence of an actual or postulated crack, determining the fatigue
life requires finding the number of cycles needed to propagate a fatigue crack
from the initial crack size, ai , to the final crack size, af , at which static failure
takes place. This is usually at or near the maximum load in a fatigue cycle.
Methods of fatigue life determination in the presence of cracks and crack like
flaws are now well established, for example Anon. (2005f). Fracture mech-
anics, especially the concept of stress intensity factor, provides the necessary
applied mechanics framework (see Appendix A).
In this chapter, some of the factors affecting the fatigue strength of metals
in crack propagation dominated situations are discussed and illustrated. It is
assumed that the crack is a Stage II crack propagating in Mode I (see Sec-
tions 3.4.2, 8.1 and A.2.1). It is also assumed that the crack path is known. If
@Seismicisolation
@Seismicisolation
102 Chapter 7: The Cracked Situation
Figure 7.1. Fatigue cracking in a 25 mm thick structural steel cruciform welded joint showing
programme markings, specimen A2/5 (Pook 1983b). Reproduced under the terms of the Click-
Use Licence.
Figure 7.2. Fatigue cracking from shrinkage in 30 × 35 mm cast steel bar. National Engineer-
ing Laboratory photograph. Reproduced under the terms of the Click-Use Licence.
the crack path is not known a priori then it has to be determined as part of the
solution (see Chapter 8).
7.2 Initial Crack Size
The initial crack size is an important variable in the determination of fatigue

crack propagation life (see Section 7.4.6). The initial crack size can either be
for an actual crack or for a postulated crack.
In a failure analysis the position, dimensions and shape of an initial crack
can usually be determined directly from a fracture surface of a broken speci-
men or component. However, secondary damage to fracture surfaces some-
times makes this difficult or impossible. Uncertainties are relatively small.
@Seismicisolation
@Seismicisolation
Metal Fatigue 103
Information on cracks that may be present, either at the start of service or

at periodic inspections during service, can be obtained by a variety of non de-
structive testing (NDT) techniques (see Appendix C). There is always some
uncertainty, and this has to be taken into account in fatigue assessments. Ac-
curate sizing of cracks using NDT is expensive, so cost versus benefit analysis
is sometimes needed. Limitations of NDT techniques for the detection and
sizing of cracks mean that postulated cracks are sometimes used in fatigue
assessments. If a specimen or component is inspected using a particular NDT
technique, it is assumed that cracks of the minimum size that can be reliably
detected are present. The damage tolerance approach to fatigue assessments
does not rely on NDT. It is assumed that cracks of certain specified sizes are
present at critical locations (see Section 5.3.2).
Because of the important role of non destructive testing in the determ-
ination of initial crack size, NDT and fracture mechanics are complement-
ary disciplines that influence each other (Pook 1992). Both disciplines are
used in the measurement of material properties in the presence of a crack,
and both are used in fatigue assessments of components and structures in the
presence of actual or postulated cracks. Taken together the two disciplines
answer questions such as: ‘Is the minimum crack size that can be found by an
NDT technique acceptable?’, ‘Is a more sophisticated (and expensive) NDT
technique needed?’, and ‘Should the part be redesigned (perhaps lowering its
structural efficiency) so that a larger initial crack is acceptable?’. It is widely
recognised that, for critical structures, allowance for appropriate NDT must
be made at the design stage.
7.3 Final Crack Size
Final crack sizes can found by analysis of laboratory and service static fail-
ures. However, in general final crack sizes have to be determined analytic-
ally. Consequently, various practical procedures have been devised and stand-
ardised. These procedures are usually presented as the calculation of static
strength in the presence of a crack of a given size.
There are two extreme situations. One extreme situation is failure by
plastic collapse at a limit load. A limit load is normally higher than the load
for general yielding. Limit load compilations for various configurations are
available, for example (Miller 1988). The other extreme situation is brittle
fracture, where failure occurs by rapid crack extension at a nominal stress be-
low the yield stress, that is under essentially elastic conditions. Modern ana-
@Seismicisolation
@Seismicisolation
Figure 7.3. Effect of specimen thickness on fracture toughness of Ti6 Al6 V2.5 Sn, σY =
1200 MPa. Reprinted from Linear Elastic Fracture Mechanics for Engineers. Theory and
Applications. LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
lytical procedures are essentially interpolations between these two extremes

(Zerbst et al. 2003, Anon. 2005f).
In the presence of a Mode I crack, brittle fracture takes place when the
Mode I stress intensity factor, KI , exceeds the fracture toughness, Kc , of the
material (see Sections A.2.1 and A.3). In general Kc is a function of mater-
ial thickness. Under plane strain conditions, Kc has a minimum value, KIc ,
called the plane strain fracture toughness, which is a material property (see
Section A.3.3.1). Figure 7.3 shows an example of the effect of thickness on
fracture toughness (Srawley et al. 1967). In practice, failure is not completely
abrupt, and in standard test methods the fracture toughness is defined as the
value of KI to cause a small, specified amount of crack propagation (Anon.
2005g). Because of this, fracture toughness values calculated from the crack
size at the end of fatigue life, and the maximum load in the fatigue cycle, are
usually slightly different from those measured using a standard test method.
End of life values are sometimes called the fatigue fracture toughness.
@Seismicisolation
@Seismicisolation
Metal Fatigue 105
7.4 Constant Amplitude Fatigue Crack Propagation
For fatigue crack propagation under constant amplitude fatigue loading the
fatigue cycle is usually described by the stress intensity factor range, K
(Pook 2000a). This is given by
K = Kmax − Kmin , (7.1)
where Kmax and Kmin are the maximum and minimum values of the Mode I
stress intensity factor, KI , calculated from the corresponding values of σmax
and σmin (Figure 3.1). Compressive stresses simply close a crack so if Kmin
is negative it is taken as zero in the calculation of K (see Section A.3). It
has been shown experimentally that, in general, K has the major influence
on fatigue crack propagation rates in metallic materials (Frost et al. 1974).
In particular, if K is constant then the fatigue crack propagation rate is
constant.
Limitations on the validity of stress intensity factors apply to Kmax (see
Section A.3.2). In particular, yielding at the crack tip must be small scale.
Corrections to Kmax for the effect of the crack tip plastic zone may be made
(see Section A.3.3.2). However, these corrections are usually very small so
are not normally used. If σmax is such that extensive yielding occurs, it may
still be possible to use stress intensity factors in the analysis in the analysis
of fatigue crack propagation data. If the stress range, σ , is not too large
then unloading from σmax to σmin is essentially elastic, and for some metallic
materials it is possible to calculate meaningful values of K, even though
meaningful values of Kmax cannot be calculated (Frost et al. 1971).
The Paris equation (Paris 1962) is often used to characterise fatigue crack
propagation rates in metallic materials. It is given by
da
= C(K)m , (7.2)
dN
where a is crack length, N is number of cycles, and C and m are empirically
determined constants. Strictly speaking, N is a discontinuous variable but it
is usually large, and it is conventionally regarded as continuous. The Paris
equation can be integrated to give the number of cycles for a crack to propag-
ate from an initial size to a final size. The constant C in Equation (7.2) has
peculiar dimensions, which depend on the value of the exponent m. To avoid
confusion, it is usually best to work in MN-m units, that is with a in metres
√
and K in MPa m. The Paris equation is sometimes called the Paris law,
but because of its empirical basis the latter term should be avoided.
@Seismicisolation
@Seismicisolation
Figure 7.4. Centre cracked tension specimen, width 2W .
7.4.1 D ETERMINATION OF FATIGUE C RACK P ROPAGATION R ATES
In principle, determination of fatigue crack propagation rates under constant

amplitude fatigue loading is straightforward. Crack length, a, versus number
of cycles, N, data are obtained for a fatigue crack propagating in a convenient
specimen, such as the pin loaded centre cracked tension specimen shown in
Figure 7.4 (see Section A.3.1.1). Some typical data for mild steel are shown in
Figure 7.5. In the figure the individual data points are averages for both sides
of the sheet and for both ends of the crack. The slope of the resulting curve
is then taken, at convenient intervals, to obtain values of the fatigue crack
propagation rate, da/dN, and corresponding values of K are calculated.
As in Figure 7.6 it is conventional to plot da/dN versus K on logarithmic
scales.
Data from several specimens may be combined to give a scatter band
for the material. Figure 7.6 shows data for mild steel. These include data
derived from Figure 7.5. Most of the scatter band is straight, corresponding to
Equation (7.2). For the centre line of the band C in the equation is 2.4×10−12 ,
and m is 3.3. The dashed lines at the top of the scatter band in Figure 7.6
indicate a region where, because of extensive yielding, it is not possible to
@Seismicisolation
@Seismicisolation
Metal Fatigue 107
Figure 7.5. Fatigue crack propagation curve for a centre crack, length 2a, in 0.76 m wide ×
2.5 mm thick mild steel sheet specimen, nominal stress 108 ± 31 MPa (Frost et al. 1974).
calculate meaningful values of Kmax , but it is possible to calculate meaningful

values of K (see previous section).
Despite the apparent simplicity of determining fatigue crack propagation
rates in metallic materials, difficulties arise in practice, and it took a long
time for standard test methods to appear (see Section 2.3.1). Problems arise
because, on a microscopic scale, fatigue crack propagation in metallic ma-
terials is a very irregular process (see Section 3.4.2). Further, fatigue crack
propagation rates of down to the order of one lattice spacing per cycle (about
3 × 10−10 m/cycle) are of interest (see Section 7.4.3). This means, in general,
that the amount of crack propagation per cycle is very much less than the
sensitivity of available crack size measurement systems (see Appendix C).
In consequence, the apparent amount of scatter in fatigue crack propagation
rate data, such as those shown in Figure 7.6, is a strong function both of the
test method used, and also of the data reduction technique used. A particular
practical problem is that accurate numerical differentiation of a curve defined
by discrete data points is difficult (Kreyszig 1983).
The amount of scatter in experimental da/dN versus K data can be
very large, especially when viewed on the da/dN axis. The scatter band in
Figure 7.6 includes about 90 per cent of the data points. Unfortunately, it is
not possible to separate inherent scatter in da/dN data from that due to ex-
perimental inaccuracies. It can be shown that, neglecting measurement errors,
@Seismicisolation
@Seismicisolation
Figure 7.6. Fatigue crack propagation rates for mild steel, 56 tests, stress ratio R = 0.06−0.74
the amount of scatter in da/dN data is approximately inversely proportional

to the square root of the crack length increment over which an average crack
propagation rate is calculated, and also approximately inversely proportional
to the square root of specimen thickness (Pook 1976a).
In standard fatigue crack propagation rate test methods, the intervals at
which crack length measurements are taken, and methods of numerical dif-
ferentiation, are both carefully specified. This means that it is possible to
make qualitative judgments about the relative amount of scatter in fatigue
crack propagation rate data for different metallic materials. In some standards
for fatigue crack propagation rate determination, for example Anon. (2003b),
there is also a requirement that da/dN versus K data be presented in tab-
ular form, as well as graphically. The point is that fitting a curve through the
data points could give a misleading impression of the amount of scatter in the
data.
@Seismicisolation
@Seismicisolation
Metal Fatigue 109
Frost et al. (1974) include a compilation of fatigue crack propagation rate

data for a range of metallic materials. This has been reproduced in some
later publications, for example Derby et al. (1992). However, extensive re-
cent compilations of fatigue crack propagation rate data do not appear to be
available. What appears to have happened is that carefully validated upper
bound fatigue crack propagation rate data for relevant materials are included
in fatigue assessment standards such as Anon. (2005f) (see Section 7.4.3).
An upper bound to a scatter band is appropriate for fatigue design purposes,
whereas the mean values quoted by Frost et al. (1974) are appropriate for
failure analysis.
7.4.2 FACTORS A FFECTING FATIGUE C RACK P ROPAGATION R ATES
The mechanism of fatigue crack propagation in ductile metallic materials

shows that it is primarily a deformation controlled, rather than a stress con-
trolled, process (see Section 3.4.2). This has important consequences for fa-
tigue crack propagation behaviour. In particular, fatigue crack propagation in
metallic materials tends to be insensitive to a material’s strength, and also to
metallurgical factors. Some of the factors affecting fatigue crack propagation
rates are described below.
7.4.2.1 Effect of Young’s Modulus
Fatigue crack propagation takes place because unloading resharpens the crack
tip on each cycle, as shown schematically in Figure 3.13, and Young’s modu-
lus, E, is an important factor. Fatigue crack propagation is a consequence of
irreversible plastic deformation at the crack tip, and this was used as the basis
of a continuum mechanics deformation theory of fatigue crack propagation
(Pook and Frost 1973). The increment of crack propagation on each cycle,
for plane stress and a zero to maximum load (σmax ), can be estimated by cal-
culating the crack profile at the maximum σmax using Equation (A.3). In their
theory Pook and Frost assumed that the part of the crack profile subjected to
tensile stresses greater than the yield stress, σY , retains its length on unload-
ing, as shown in Figure 3.13. For ductile metals, E/σY is of the order of 103 ,
and the assumption leads to

da 9 K 2
= . (7.3)
dN π E
For plane strain, the equation becomes
@Seismicisolation
@Seismicisolation
2
da 7 K
= . (7.4)
dN π E
Equations (7.3) and (7.4) were derived for the fatigue cycle of 0 to σmax ,
that is the stress ratio, R = 0 (see Section 3.1). However, the resharpening
mechanism ensures that they also apply to R > 0. It had been shown that any
continuum mechanics deformation theory of fatigue crack propagation leads
to a (K/E)2 dependence, unless a characteristic dimension is introduced
(Rice 1967). Hence refinements to Pook and Frost’s theory would simply
alter the numerical factors in Equations (7.3) and (7.4). The exponent, m,
in the Paris equation (Equation (7.2)) is usually greater than the theoretical
value of 2 in Equations (7.3) and (7.4). This is largely because fatigue crack
propagation mechanisms differ in detail at different levels of K.
Despite its limitations, Pook and Frost’s theory does correctly predict gen-
eral features of fatigue crack propagation behaviour in metals, especially the
dependence on Young’s modulus rather than strength. Data for materials of
widely different Young’s moduli may be collapsed onto one scatter band by
plotting da/dN against K/E rather than K. For example, Figure 7.7
shows data for tungsten (E = 390 GPa) and dispersion strengthened lead
(E = 16 GPa) (Speidel 1982). The stress ratio, R, is zero.
The theory also correctly predicts that fatigue crack propagation rates in
mild steel are largely independent of the stress ratio. Figure 7.6 includes data
from 56 test specimens with R values ranging from 0.06 to 0.74. Such ma-
terials are called mean stress insensitive. However, some materials are mean
stress sensitive and crack propagation rates increase significantly as R in-
creases. This is because fatigue crack propagation mechanisms differ in detail
at different values of R (Frost et al. 1974). As an example, Figure 7.8 shows
data for a 5.5% Zn aluminium alloy. Mean stress sensitivity appears at high
crack propagation rates as static failure is approached. This is simply because
short burst of brittle fracture, controlled by the maximum load in the fatigue
cycle, occur.
7.4.2.2 Effect of Thickness
Fatigue crack propagation rates in metallic materials sometimes depend on

sheet or plate thickness, so tests should be carried out using material of the
thickness of interest. There are two effects of thickness that can affect fa-
tigue crack propagation rates. Firstly, the fracture toughness, Kc , in general
decreases as the thickness increases (see Section 7.3). Secondly, the trans-
ition from square to slant fatigue crack propagation, that sometimes occurs
@Seismicisolation
@Seismicisolation
Metal Fatigue 111
Figure 7.7. Fatigue crack propagation rates for tungsten (E = 390 GPa) and dispersion
strengthened lead (E = 16 GPa), R = 0. Reprinted from Linear Elastic Fracture Mech-
anics for Engineers. Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-703-5,
2000.
in thin sheets, but this does not happen in thick plates (see Section A.3.3.3).
Figure 7.9 shows the main features of the transition.
Fatigue crack propagation rates tend to increase as the maximum stress
intensity factor in the fatigue cycle, Kmax , approaches Kc (Pook 2000a). Con-
sequently, at high values of stress intensity factor range in the fatigue cycle,
K, fatigue crack propagation rates increase as thickness increases as shown,
for example, in Figure 7.10. Equation (7.2) is sometimes modified to allow
for this effect.
Fatigue crack propagation rates usually decrease following the transition
to slant fatigue crack propagation. The mechanisms involved in the transition,
and in slant fatigue crack propagation, are not clear, and only qualitative the-
oretical explanations appear to be possible (Pook 2000a, 2002a). However, a
large amount of fatigue crack propagation data has been accumulated, and for
practical engineering purposes the phenomenon is well understood (Zuidema
1995). The transition takes place only in thin sheets.
In the transition region the transition starts with the appearance of shear
lips at the sheet surfaces (Figure 7.9). The initiation of shear lips is sensit-
ively dependent on precise initial conditions, and cannot be predicted theor-
@Seismicisolation
@Seismicisolation
Figure 7.8. Effect of stress ratio R on fatigue crack propagation rates for 5.5% Zn aluminium
alloy (Frost et al. 1974).
etically, but it does appear to be associated with the crack tip plastic zone
(see Section A.3.3.2). The start of the transition appears to be associated with
the attainment of a minimum value of the range of stress intensity factor in
the fatigue cycle, K, or equivalently a minimum value of the fatigue crack
propagation rate, da/dN). Hence this is a necessary condition for the trans-
ition to take place. However, the transition is sometimes suppressed, so it is
not a sufficient condition.
The shear lips increase in size as fatigue crack propagation proceeds until
they either reach a maximum size, or if the specimen is sufficiently thin, they
meet, completing the transition. After the transition the crack propagation
surface is flat, and its inclination to the sheet surface is approximately 45◦ .
Crack fronts are roughly straight. It is possible to reverse the transition by
reducing the fatigue loading (Schijve 1974).
@Seismicisolation
@Seismicisolation
Metal Fatigue 113
Figure 7.9. Transition from square to slant fatigue crack propagation in thin sheets. The arrow
shows the direction of fatigue crack propagation (Pook 1983a). Reproduced under the terms
7.4.2.3 Crack Closure
Conventionally, a Mode I crack is assumed to be open when the applied stress,

σ , is tensile, and closed when σ is zero or compressive. However, this con-
vention is correct only for the linearly elastic situation (see Section A.3). In
particular, it is implicit in the sequence shown in Figure 3.13, that for a min-
imum stress in the fatigue cycle, σmin , which is greater than zero, a fatigue
crack is open when σ > σmin and closed when σ ≤ σmin . In terms of the
Mode I stress intensity factor, KI , this means that, for a propagating fatigue
crack, the crack is open when KI > Kmin , where Kmin is the minimum value
of KI in the fatigue cycle.
In a ductile metal the presence of a crack tip plastic zone means that a
plastic wake of plastically deformed material is left adjacent to the surfaces
of a propagating fatigue crack (see Section A.3.3.2). It follows that the crack
does not open unless KI > Kop where Kop is the crack opening value of KI
where Kop > Kmin . In consequence the effective value of the range of KI in
the fatigue cycle, Keff , is given by
Keff = Kmax − Kop , (7.5)
@Seismicisolation
@Seismicisolation
Figure 7.10. Effect of thickness on fatigue crack propagation rates for RR 58 aluminium alloy
where Kmax is the minimum value of KI in the fatigue cycle. The value of
Keff given by Equation (7.5) is less than the value of K given by Equa-
tion (7.1), as shown schematically in Figure 7.11.
This phenomenon is known as crack closure. Its existence was demon-
strated experimentally by Elber (1970), but it had been noted much earlier
(Paris et al. 2006). Elber showed that fatigue crack propagation data for mean
stress sensitive metallic materials, for example as is shown in Figure 7.8,
could be collapsed into a single scatter band by plotting in terms of Keff ,
instead of K. Various empirical expressions have been proposed for the cal-
culation of Keff (Suresh 1998). However, because behaviour depends on nu-
merous factors, these expressions are all of limited applicability. Fortunately,
fatigue crack propagation rate behaviour for a particular metallic material is
usually sufficiently consistent for K (rather than Keff ) to form a satis-
factory basis for the analysis of fatigue crack propagation rate data and their
subsequent application to the solution of practical problems.
The development of a plastic wake means that as an initially stress free
crack propagates, the value of Kop increases, with concomitant reductions in
Keff , until a stable state is reached. This is illustrated by some finite element
calculations for constant K, which are shown in Figure 7.12 (Nakagaki and
@Seismicisolation
@Seismicisolation
Metal Fatigue 115
Figure 7.11. Definition of Keff in the presence of crack closure at KI = Kop .
Atluri 1979). In the figure the crack length is normalised as the ratio of current
crack length to initial crack length (a/ai ).
The development of a plastic wake does not always account for crack clos-
ure. It can also be induced by several other mechanisms. Suresh (1998) sum-
marises these: oxide induced crack closure, roughness induced crack closure,
transformation induced crack closure, and viscous fluid induced crack clos-
ure.
Oxide induced crack closure occurs when corrosion products (usually ox-
ides) form on the crack surfaces; these have a greater volume than the ori-
ginal metal. Roughness induced crack closure occurs when microscopic ir-
regularities on opposite crack surfaces interfere with each other as a result of
complex deformation patterns. Transformation induced crack closure occurs
when stress or strain induced phase transformations in material in the vicin-
ity of the crack tip lead to a net increase in the volume of the transformed
material. These mechanisms are all material dependent. They are sometimes
deliberately invoked in the development of fatigue crack propagation resistant
metallic materials.
@Seismicisolation
@Seismicisolation
Figure 7.12. Change in Keff due to development of a plastic wake. Constant amplitude
fatigue loading, R = 0. Reprinted from Linear Elastic Fracture Mechanics for Engineers.
Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
Fluid induced crack closure is a special case in that it is of mechanical

rather than metallurgical origin. In a viscous fluid environment, the fluid can-
not flow freely into and out of the crack as it opens and closes during a fatigue
cycle, and this leads to crack closure. Under internal pressure fatigue this can
lead to frequency dependence in materials that are normally frequency inde-
pendent. This is because Keff decreases as the frequency increases (Pook
and Short 1988, Davis and Ellison 1989) (see Sections 3.2 and 5.3.3.4).
7.4.3 T HRESHOLD FOR FATIGUE C RACK P ROPAGATION
The Paris equation (Equation (7.2)) implies that any value of the range of
Mode I stress intensity factor, K, no matter how small, will result in a pos-
itive value of fatigue crack propagation rate, da/dN. However, experimental
data show that the crack propagation rate tends to zero at some critical value
of K, which is the threshold value, Kth . The existence of a threshold for
fatigue crack propagation is responsible for the phenomenon of non propagat-
ing cracks (see Sections 6.5, 7.4.5 and 8.2.1). The threshold is not neces-
sarily sharply defined (Pook 1983a). Unlike fatigue crack propagation rates,
thresholds are sensitive to metallurgical factors, and also to the stress ratio, R.
Kth tends to decrease as R increases. Compilations of Kth values for vari-
@Seismicisolation
@Seismicisolation
Metal Fatigue 117
ous metallic materials are available (Frost et al. 1974, Taylor 1989). However,
extensive, recent compilations do not appear to be available. What appears to
have happened is that carefully validated lower bound fatigue crack threshold
data for relevant materials are included in fatigue assessment standards such
as Anon. (2005f). This is similar to the situation for fatigue crack propagation
rate data (see Section 7.4.1).
There are several possible formal definitions of the fatigue crack propaga-
tion threshold (Pook and Greenan 1976, Pook 1989c). Thresholds can be ob-
tained using several techniques. Usually, all give essentially the same result
(Pippan et al. 1994). The most obvious technique is simply to follow down-
wards a curve of fatigue crack propagation rate (da/dN) versus range of
Mode I stress intensity (K). However, unless this is done very carefully, the
threshold can be seriously overestimated (Pook 2000a). Some standards spe-
cify what is essentially a refined version of this method, for example Anon.
(2003b). Figure 7.13 shows some data for ASTM A517 Grade F (T-1) steel,
√
R = 0.1 (Paris et al. 1972). The threshold is about 4.4 MPa m.
The threshold exists because fatigue crack propagation rates of less than
about one lattice spacing per cycle are not possible on physical grounds. If the
lattice spacing is taken as a typical value of 3 × 10−10 m, then experimental
values of the threshold, Kth , correspond fairly well with values of K cal-
culated from the Paris equation (Equation (7.2)), with da/dN taken as one
lattice spacing per cycle (Frost et al. 1971). This means that the stress cri-
terion and the thermodynamic criterion are not both satisfied at K < Kth
(see Section 3.4.2). Fatigue crack propagation at average rates of much less
than one lattice spacing per cycle is sometimes observed. When this happens
the crack cannot be propagating along the whole crack front on every cycle
(Pook 1983a).
7.4.4 OVERALL FATIGUE C RACK P ROPAGATION B EHAVIOUR
Overall fatigue crack propagation behaviour for metallic materials under con-
stant amplitude fatigue loading is summarised schematically in Figure 7.14
(Pook 2000a). When plotted on logarithmic scales the overall curve of fatigue
crack propagation rate, da/dN, versus range of Mode I stress intensity factor,
K, is usually sigmoidal. In the intermediate Paris region the curve becomes
a straight line and fatigue crack propagation rates are given by the Paris
equation (Equation (7.2)). In general, the microstructure and the stress ra-
tio, R, have little influence. In the threshold region fatigue crack propagation
rates become vanishingly small as the threshold for fatigue crack propaga-
tion, Kth , is approached. In this region fatigue crack propagation rates may
@Seismicisolation
@Seismicisolation
Figure 7.13. Fatigue crack propagation data for ASTM A517 Grade F (T-1) steel, stress ra-
tio R = 0.1. Reprinted from Linear Elastic Fracture Mechanics for Engineers. Theory and
Applications. LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
be strongly influenced by metallurgical factors, and also by the stress ratio.

In the static failure region, fatigue crack propagation rates increase rapidly as
the conditions for static failure, at the maximum load in the fatigue cycle, are
approached. Fatigue crack propagation rates are strongly influenced by the
stress ratio. They may be strongly influenced by metallurgical factors, and
also by specimen thickness (see Section 7.4.2.2).
7.4.5 S HORT C RACKS
Conventionally, fatigue crack propagation data, such as those shown in Fig-

ures 7.5–7.8, 7.10 and 7.13 are obtained from specimens containing relat-
ively long cracks (a > 10 mm). Short cracks (say a less than about 1 mm)
are often of practical importance. For some time there has been consider-
@Seismicisolation
@Seismicisolation
Metal Fatigue 119
Figure 7.14. Overall fatigue crack propagation behaviour for metallic materials under constant
amplitude fatigue loading. Reprinted from Linear Elastic Fracture Mechanics for Engineers.
Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
able interest in the acquisition and application of fatigue crack propagation

data for short cracks (Miller 1982, Miller and de los Rios 1992). In a given
metallic material the threshold for fatigue crack propagation from a short
crack can be lower than for a long crack (Murakami 2002). Similarly, fatigue
crack propagation rates can be greater. Unfortunately, behaviour varies in de-
tail from material to material. In general, it is not possible to extract useful
generalisations, and practical problems have to be solved on an ad hoc basis.
Another source of difficulty with short cracks can be the failure of stress in-
tensity factors to provide an adequate description of the crack tip stress field
(see Section A.3.2). Test methods for short cracks have not yet been stand-
ardised.
Short crack effects can have several different origins (Miller 1982, Suresh
1998). These origins include the irregular nature of many short cracks, and
also the influence of metallurgical features such as inclusions and grain
boundaries, whose scale is comparable with the crack size (see Section 3.4).
The plastic wake left by a propagating fatigue crack is sometimes responsible
for crack closure (see Section 7.4.2.3). A plastic wake can only reach a stable
state if the range of the Mode I stress intensity factor, K, for each cycle
differs little from its predecessor, that is d(K)/da is small (Pook 2000a).
@Seismicisolation
@Seismicisolation
Differentiating Equation (A.4) with respect to the crack length a, and norm-
alising with respect to the Mode I stress intensity factor, KI , leads to

1 dKI 1
= . (7.6)
KI da 2a
Hence, for a short crack, d(K)/da may not be small. The plastic wake may
then never reach a stable state. This corresponds to the minimum value of the
effective value of K, Keff , which is shown in Figure 7.12. The amount
of crack closure would then be reduced when compared with that for a long
crack, leading to an increase in Keff , and a concomitant increase in fatigue
crack propagation rates.
What can happen with short cracks propagating in fatigue under constant
amplitude fatigue loading is shown schematically in Figure 7.15. For a spe-
cific specimen, and a constant amplitude fatigue loading, the range of Mode
I stress intensity factor, K, is a measure of the crack length, a. If the crack
is sufficiently long then overall fatigue crack propagation behaviour is the
same as in Figure 7.14. This is shown by the curve marked ‘long crack’. For
a short crack there are two possibilities. Firstly, fatigue crack propagation can
slow down and arrest, as shown by the solid line marked ‘short crack’, lead-
ing to a non propagating crack (see Sections 6.5 and 7.4.3). Alternatively, a
propagating fatigue crack can slow down and then accelerate, blending with
long crack behaviour, as shown by the dashed line.
In a cracked metallic specimen the fatigue limit (see Section 3.2) is some-
times controlled by the crack length, together with the threshold for fatigue
crack propagation. The relationship, for a metallic material, between the fa-
tigue limits of specimens containing cracks of various sizes and the fatigue
limit of an uncracked (plain) specimen may be summarised by means of a
Kitagawa diagram (Kitagawa and Takahashi 1976). In a Kitagawa diagram,
shown schematically in Figure 7.16, fatigue limits are plotted against crack
length, both on logarithmic scales.
In the diagram the fatigue limit of a plain specimen is shown by the hori-
zontal line. If cracks are below a critical size then they have no effect on the
fatigue limit. The other line, which has a slope of −0.5, shows fatigue limits
calculated from the crack length and the long crack fatigue crack propaga-
tion threshold using an appropriate expression for the Mode I stress intens-
ity factor, such as Equation (A.6). Actual material behaviour, shown by the
dashed line, is a smooth blend between the two straight lines. This blend can
be interpreted as summarising the threshold behaviour of short cracks.
As an example of the relationship between crack length and the fatigue
crack propagation threshold, Kth , some data for NiCr steel specimens con-
@Seismicisolation
@Seismicisolation
Metal Fatigue 121
Figure 7.15. Short crack and long crack fatigue crack propagation behaviours for metallic
materials under constant amplitude fatigue loading.
Figure 7.16. Kitagawa diagram showing relationship between crack length and fatigue limits.
Reprinted from Linear Elastic Fracture Mechanics for Engineers. Theory and Applications.
LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
taining edge cracks between 0.25 and 6 mm long tested were fitted by an
empirical expression (Frost et al. 1974). This empirical expression implies
that Kth ∝ a 1/6 , where a is crack length. The point of a Kitagawa diagram
is that the fatigue crack propagation threshold, expressed in terms of stress,
decreases as crack length decreases, even though Kth increases.
@Seismicisolation
@Seismicisolation
7.4.6 FATIGUE L IFE IN THE P RESENCE OF A C RACK
If the initial crack size, ai , the final crack size, af , and the crack path are all
known then calculation of the number of cycles, N, for a crack to propag-
ate from ai to af is straightforward. If, at the initial crack size, the range
of Mode I stress intensity factor, K, is less than the threshold for fatigue
crack propagation then the fatigue life is indefinite (see Section 7.4.3). Other-
wise, an expression for the stress intensity factor is substituted into a fatigue
crack propagation equation and integrated. In general, integration has to be
carried out numerically (Anon. 2005f). Some stress intensity expressions for
common fatigue crack propagation specimens, which permit analytical in-
tegration, in conjunction with the Paris equation, are available (Jones and
James 1996). The simple approach below, based on the Paris equation (Equa-
tion (7.2)), is adequate for some purposes. It illustrates the general approach
to fatigue crack propagation life calculations.
In carrying out fatigue crack propagation life calculations it is sometimes
assumed, as an approximation. that the Paris region, shown in Figure 7.14,
extends from the threshold region to the static failure region (Pook 2000a).
This simplification, shown schematically in Figure 7.17, is safe in the vi-
cinity of the threshold region, where it leads to an underestimate of fatigue
crack propagation life. However, it is unsafe in the vicinity of the static fail-
ure region, where it leads to an overestimate of fatigue crack propagation life.
This overestimate may be unimportant because in this region fatigue crack
propagation rates are high, and relatively few cycles are involved.
Scatter in fatigue crack propagation rate data has to be taken into account
in calculations (see Section 7.4.1). This is often done by taking the upper
bound to a fatigue crack propagation data scatter band, such as those shown
in Figures 7.6 and 7.8. For failure analysis calculations mean values are used.
For a constant amplitude fatigue loading the general expression for the
stress intensity factor (Equation (A.6)) may be written in the form
√
K = SY π a , (7.7)
where S is stress range in the fatigue cycle (see Section 3.1)), Y is a geometric
correction factor (of the order of one), and a is the crack length. Inverting the
Paris equation (Equation (7.2)) gives
dN 1
= . (7.8)
da C(K)m
Combining Equations (7.7) and (7.8)
dN 1
= . (7.9)
da CY m S m π m/2 a m/2
@Seismicisolation
@Seismicisolation
Metal Fatigue 123
Figure 7.17. Simplified overall fatigue crack propagation behaviour for metallic materials
under constant amplitude fatigue loading. Reprinted from Linear Elastic Fracture Mechanics
for Engineers. Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
Assuming that Y is constant then the number of cycles, N, for a crack to

propagate from its initial length, ai , to its final length, af , is given by
af
1 1
N= m m m/2 m/2
. (7.10)
CY S π ai a
Hence

2 1 1
N= − . (7.11)
(m − 2) × CY m S m π m/2 (m−2)/2
ai
(m−2)/2
af
For example, for S = 119 MPa, ai = 1.3 mm, af = 51 mm, Y = 1.12,
m = 3.23, and C = 2.88 × 10−12 . Equation (7.11) becomes
2
N=
1.23 × 2.88 × × 1.123.23 × 1193.23 π 1.615
10−12

1 1
× − . (7.12)
0.00130.615 0.0510.615
C is in MN-m units so crack lengths are entered in metres, and N = 649,700
cycles.
If af is very large compared with ai then the second term in the square
brackets may be neglected and Equation (7.11) becomes
@Seismicisolation
@Seismicisolation
2
N= (7.13)
(m − 2) × CY m S m π m/2 a (m−2)/2
and N becomes 725,700 cycles. This is an increase of 3.2 per cent. For com-
parison, decreasing S by 1 per cent also increases N by 3.2 per cent.
7.4.6.1 Welded Joints
In a welded joint small crack like flaws may be present at the toe of a weld.
These are often slag inclusions. In consequence virtually the whole fatigue
life may be occupied by fatigue crack propagation from these flaws (Gurney
1979). In a welded joint, in the as welded condition, tensile residual stresses
of yield point magnitude may be present in the vicinity of the weld (Maddox
1991). A stress relief heat treatment is often used to relieve residual stresses in
welded joints. In unstressrelieved (as welded), welded joints residual stresses
are superimposed on any applied fatigue loading. Detailed information on the
distributions of residual stresses is available for some types of welded joints,
and this facilitates calculation of fatigue crack propagation life in the presence
of residual stresses (Lee et al. 2005). In the absence of detailed information,
it is conventional to assume that the effective stress cycle is from the yield
stress downwards. This means that a crack may remain open throughout the
stress cycle, even when the minimum applied stress is compressive. Hence,
for unstressrelieved welded joints it has become conventional to use the whole
stress range in the calculation of K (Pook 2000a). Stress relieving a welded
joint may result in a substantial improvement in its fatigue life.
7.5 Variable Amplitude Fatigue Crack Propagation
7.5.1 W EIGHTED AVERAGE S TRESS R ANGE
The simplest approach to estimating fatigue crack propagation rates under

variable amplitude fatigue loading is to assume that each cycle causes the
same amount of fatigue crack propagation as if it were part of a constant
amplitude fatigue loading. This approach leads to the use of the concept
of a weighted average stress range (see Section 4.3.1). Its use to charac-
terise the level of a variable amplitude fatigue loading, was first proposed
by Paris (1962). It is defined so as to have physical significance for fatigue
crack propagation, and also in such a way that it can be used in fatigue crack
propagation life calculations as if it were a constant amplitude stress range.
If necessary, equivalent cycles are extracted from a load history in which
@Seismicisolation
@Seismicisolation
Metal Fatigue 125
individual cycles cannot be distinguished by using cycle counting (see Sec-

tion 4.3.3).
Interaction effects due to changes in load levels which affect the crack tip
plastic zone size and the plastic wake are not take into account (see Sec-
tion 7.5.2). The implicit assumption that interaction effects are negligible
is sometimes justified, for example for some offshore structures (Pook and
Dover 1989).
To derive the weighted average stress range first write a generalised ex-
pression for the Mode I stress intensity factor, KI in the form
KI = σ f(a), (7.14)
where σ is stress and a crack size. Hence if the stress range for the ith cycle
is σi , then the corresponding stress intensity factor range, Ki , is given by
Ki = σi f(a). (7.15)
Assuming that fatigue crack propagation rates are given by the Paris equation
(Equation (7.2)), then the increment of crack propagation, δa, corresponding
to Ki is given by
δa = C(Ki )m (7.16)
and the average increment of crack propagation over N cycles is

C(Ki )m
δa = . (7.17)
N
If the weighted average range of KI , Kh , is defined by

C(Ki )m 1/m
Kh = , (7.18)
N
then the Paris equation becomes
da
= CKhm . (7.19)
dN
This means that it is possible to calculate fatigue crack propagation rates for
variable amplitude fatigue loading by replacing K in the Paris equation by
Kh . It follows that for constant amplitude fatigue loading Kh = K.
The Paris equation and Equation (7.14) are both nonlinear, which at first
sight appears to invalidate the argument. However, Eastabrook’s theorem
(Eastabrook 1981) ensures that, within wide limits, it is possible to define
a corresponding weighted average stress range as

(σi )m 1/m
σh = . (7.20)
N
@Seismicisolation
@Seismicisolation
Table 7.1. Stress cycles during each operation.

Number of cycles Stress range (MPa)
1 285
4 195
10 95
This equation provides a direct means of calculating Kh from an appropriate

expression for the stress intensity factor, and hence fatigue crack propagation
rates. It is mathematically equivalent to the Miner’s rule summation given
by Equation (4.11) (see Section 4.3.1). In effect it is a derivation of Miner’s
rule in which fatigue damage is taken as increments of crack propagation
(see Section 4.3). The approach can be extended to more elaborate fatigue
crack propagation expressions, but calculations become less straightforward
(Austin 1994).
It has been found experimentally, for example by Hibberd and Dover
(1977), that for situations where interaction effects are not significant, the use
of Kh collapses fatigue crack propagation rate data for constant amplitude
and variable amplitude fatigue loading onto a single scatter band.
For a variable amplitude fatigue loading Equation (7.11) becomes

2 1 1
N= − , (7.21)
(m − 2) × CY m Shm π m/2 ai(m−2)/2 af(m−2)/2
where Sh is the weighted average stress range. For the loading shown in
Table 7.1 the number of cycles per operation is 15 and Sh , calculated us-
ing Equation (4.11), is 162.9 MPa. If the initial crack depth ai = 1 mm, the
final crack depth af = 3.98 mm, the geometric correction factor Y = 1.12,
and the Paris equation constants are m = 3.23 and C = 2.88 × 10−12 MN-m
units, then Equation (7.21) becomes
2
N= −12
1.23 × 2.88 × 10 × 1.123.23 × 162.93.23 × π 1.615

1 1
× − (7.22)
0.0010.615 0.0039880.615
and N = 177,023 cycles, hence the number of operations to failure is
177,023/15 which is approximately 11,800.
√
If the threshold for fatigue crack propagation, Kth = 9 MPa m, and
this is taken into account, then a different result is obtained. First calculate
the crack depth at which each stress range becomes damaging by entering
Kth and stress ranges into Equation (7.7). This leads to the results shown in
Table 7.2.
@Seismicisolation
@Seismicisolation
Metal Fatigue 127
Table 7.2. Damaging crack depths.

Stress range (MPa) Damaging crack depth (mm)
285 0.25
195 0.54
95 2.28
Since 0.25 mm and 0.54 mm are both less than ai the 285 MPa and
195 MPa stress ranges are damaging throughout the fatigue crack propaga-
tion life. The 95 MPa stress range becomes damaging at a crack depth of
2.28 mm. The fatigue crack propagation life integration is carried therefore
out in two increments.
For the first increment calculate the weighted average stress range, Sh , as-
suming that the lowest stress range is non damaging (see Section 4.3.1). Thus
Sh = 156.7 MPa, ai = 0.001 m and af = 0.0028 m and N = 139,398 cycles.
For the second increment all three stress ranges are damaging and Sh =
162.9 MPa, ai = 0.00228 m, af = 0.00398 m and N = 54,045 cycles. Hence
the number of operations to failure = (139,398 + 54,045)/15 = 12,900, a
significant increase of about 9 per cent.
7.5.2 I NTERACTION E FFECTS
Interaction effects affecting fatigue crack propagation rates in metallic ma-

terials can occur when load levels change during a block fatigue loading
(see Section 4.2.3). Within a block fatigue loading each block consists of
a constant amplitude fatigue loading. At the new load level the fatigue crack
propagation rate may not be the same as it would be for the same fatigue load
under constant amplitude fatigue loading. Numerous tests have been carried
in attempts to quantify interaction effects, including tests using the simple
two level load histories shown schematically in Figures 7.18–7.21.
Some generalisations are possible (Pook 1983a, 2000a). It is often found
that a crack will propagate, at the rate for constant amplitude fatigue loading,
if the load is increased, as shown in Figure 7.18, but will be temporarily
retarded to below the expected rate if the load is reduced (Figure 7.19) or
following a single overload (Figure 7.20). These effects can be explained
by considering changes in crack closure due to changes in the plastic wake
(Nakagaki and Atluri 1979) (see Section 7.4.2.3).
On the other hand a single underload, as shown in Figure 7.21, can tem-
porarily increase crack propagation rates. A practical load history will often
start and finish at zero, as shown schematically in Figure 7.22. When this load
history is repeated there is effectively a periodic underload.
@Seismicisolation
@Seismicisolation
Figure 7.18. Two level load history, load increased.
Figure 7.19. Two level load history, load decreased.
In detail, interaction behaviour is too complicated for the results of simple

tests, such as those based on Figures 7.18–7.22, to be used to predict fa-
tigue crack propagation behaviour under more complicated load histories
such as that shown in Figure 4.9 (Schütz 1979, Pook 1983a, 2000a). A num-
ber of models to account for interaction effects have been proposed. The best
known is Wheeler’s model (Wheeler 1972). He considered that fatigue crack
propagation would be retarded if the calculated plastic zone for the current
cycle was within that due to some previous cycle (see Section A.3.3.2). All
the models, including Wheeler’s, make use of empirically determined con-
stants. The models are therefore restricted to making predictions for situations
similar to those used to determine the constants. In practice, calculations have
to be carried out using computer programs. Sophisticated software, validated
for various situations, is now available.
@Seismicisolation
@Seismicisolation
Metal Fatigue 129
Figure 7.20. Two level load history, single overload.
Figure 7.21. Two level load history, single underload.
7.6 Fractography
The study of fracture surfaces is known as fractography (see Section 2.2.4).

For more than 40 years it has been used to reconstruct fatigue crack propaga-
tion data from the fracture surfaces of metallic specimens and components
(see Section 2.2.5). Striations, such as those shown in Figure 3.12, can be
used to reconstruct crack length versus number of cycles curves by the meas-
urement of striation spacing. This was an early application of fractography
(Pook 1962). Under programme loading, programme markings such as those
visible in Figure 7.1, may appear (see Sections 4.2.3 and 4.4). These pro-
gramme markings can also be used to reconstruct fatigue crack propagation
curves (Ryman 1962). Further, programme markings provide information on
crack front shape development during fatigue crack propagation.
@Seismicisolation
@Seismicisolation
Figure 7.22. Load history starting and finishing at zero load.
Fractography is important in the failure analysis of both laboratory and

service failures in metallic materials. It is particularly valuable when, for any
reason, fatigue crack propagation has not been monitored during the fatigue
life. A practical difficulty is that striations and programme markings cannot
always be observed, and even when they are present, it can be difficult to
resolve them. They are sometimes damaged mechanically during subsequent
fatigue crack propagation, and are also sometimes damaged by corrosion.
7.6.1 FATIGUE C RACK P ROPAGATION IN A W ELDED J OINT
A fractographic and fracture mechanics analysis was carried out on some me-
dium strength structural steel cruciform welded joints (Pook 1982b, 1983b.
Fatigue tests were carried out as part of the United Kingdom Offshore Steels
Research Project, which was set up to obtain fatigue and fracture data rel-
evant to tubular structures in the North Sea (Anon. 1974b). Details of the
specimens, test techniques used, and fatigue test results are given by Holmes
(1980a, 1980b) and by Holmes and Kerr (1982). The specimens were full
penetration welded joints of the type shown in Figure 7.23. They were made
from BS 4360: 1979 grade 50D steel (modified), assembled using manual
metal arc welding, and were not stress relieved after welding (see Sec-
tion 7.4.6.1). Tests were carried out joints under direct stress (axial) fatigue
loading at zero mean stress, using the load history shown in Figure 4.9.
In all specimens several individual fatigue cracks were nucleated at the
weld toe on both sides of the specimen, as shown in Figure 7.1. There was
usually more fatigue crack propagation on one side of the specimen than on
the other. As fatigue crack propagation proceeded individual cracks merged
@Seismicisolation
@Seismicisolation
Metal Fatigue 131
Figure 7.23. 25 mm thick structural steel cruciform welded joint (Pook 1983b). Reproduced
into a single fatigue crack of large aspect ratio (ratio of crack surface length
to depth. (In the literature aspect ratio is sometimes taken as the ratio of crack
depth to surface length).
Fatigue crack propagation data were obtained from measurements of the
programme marking spacing made with a travelling microscope capable
of being read to 0.01 mm. Measurements were made along a line passing
through one of the multiple origins, and at or near the greatest fatigue crack
depth. The programme markings were sometimes difficult to resolve and il-
lumination had to be carefully adjusted. Only limited data were obtained for
some specimens. Programme markings below a certain size were not resolv-
able, even if the magnification was increased. In addition, few data were ob-
tained for crack depths of less than 1 mm because of mechanical damage
to the fracture surfaces. This apparently occurred during the later stages of
fatigue tests.
Figure 7.24 shows the extensive data which were obtained for the lar-
ger fatigue crack in Specimen A2/5 (Figure 7.1 bottom). The programme
markings on this crack surface were unusually clear. The nominal root mean
square (RMS) stress applied during the fatigue test was 25.6 MPa, and the
fatigue life was 4.798 × 106 cycles (47.98 programmes). The programme
@Seismicisolation
@Seismicisolation
Figure 7.24. Fatigue crack propagation in a 25 mm thick structural steel cruciform welded
joint, specimen A2/5 (Pook 1983b). Reproduced under the terms of the Click-Use Licence.
markings suggest that fatigue crack initiation took place at about 1.1 × 106
cycles (11 programmes). Data not shown in Figure 7.24 suggest that for the
shallower crack (Figure 7.1 top) initiation was at about 3.7 × 106 cycles (37
programmes).
For comparison with the fractographic results theoretical predictions were
carried out using stress intensity factor solutions and fatigue crack propaga-
tion data recommended by Det Norske Veritas (Anon. 1977). This informa-
tion was in N-mm units, so these units were used in calculations rather than
the more usual MN-m units. The Mode I stress intensity factor, KI , for the
joint design used is given by
KI = 3.41σ a 0.3 , (7.23)
where σ is nominal stress and a is crack depth. The fatigue crack propagation
rate is given by the Paris equation (Equation (7.2)) with C = 3.1 × 10−13 and
m = 3.1. The Paris equation was integrated numerically (see Section 7.5.1).
Det Norske Veritas did nor recommend a value for the threshold for fatigue
√
crack propagation so a typical value of 200 N/mm3/2 (6.32 MPa m) was
used (Frost et al. 1974). Two sets of calculations were carried out and the
results are shown in Figure 7.24. In one set it was assumed that whole fatigue
cycles are damaging, and in the other set only the tensile half cycles were
assumed to be damaging.
For shallow cracks (<1.5 mm deep) the results shown in Figure 7.24 ap-
proach the prediction in which it was assumed that whole load cycle are dam-
@Seismicisolation
@Seismicisolation
Metal Fatigue 133
aging, in accordance with the conventional view of the effect of tensile resid-
ual stresses in welded joints (see Section 7.4.6.1). The prediction assuming
that only the tensile half cycles were damaging provides a good fit at me-
dium crack depths. This implies that the effect of tensile residual stresses
has died away, in agreement with residual stress measurements on welded
joints (Lee et al. 2005). Fatigue crack propagation rates are underestimated
for deep cracks (>5 mm deep). This appears to be because Equation (7.22)
underestimates stress intensity factors for crack depths greater than one fifth
of the plate thickness (Pook 1983b). Similar results were obtained from other
welded joints analysed (Pook 1982b, 1983b).
Cruciform welded joints are notorious for scatter in fatigue test results
(Gurney 1976). From the above results it is clear that this is associated with
fatigue crack propagation events within 0.5 mm of the weld surface. This
emphasises the importance of the precise weld quality achieved. However,
the established view that, for unstressrelieved welded joints, fatigue life is
controlled by stress range irrespective of mean stress is, for many purposes, a
convenient generalization (Gurney 1979).
@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
8
Fatigue Crack Paths
8.1 Introduction
Macroscopic aspects of fatigue crack paths, corresponding to the top two

scales in Table 3.3, have been of industrial interest since the earliest fatigue
investigations (see Sections 2.2.4 and 5.3.3.3). Microscopic aspects are of
interest in the development of fatigue crack propagation resistant materials.
Macroscopic aspects are emphasised in this chapter. Much current interest
in fatigue crack paths is in the collection of data for the solution of spe-
cific engineering problems. The fatigue crack path in a critical structure in
aerospace, automotive, offshore and other industries can determine whether
a fatigue failure is benign or catastrophic (Grandt 2004). There is relatively
little information on how to ensure that a fatigue crack path is benign. Know-
ledge of potential crack paths is also needed for the selection of appropriate
non destructive testing procedures.
In the last four decades there have been substantial advances in the under-
standing of macroscopic aspects of fatigue crack paths and in their prediction.
This is largely through developments in fracture mechanics and computer
software (Carpinteri and Pook 2005). Theoretical prediction of fatigue crack
paths in three dimensions is an active area of research. Nevertheless, despite
recent theoretical advances, fatigue crack paths in structures are still often
determined by large scale structural tests. For example, Figure 8.1 shows an
experimentally determined fatigue crack path in a fighter aircraft centre sec-
tion. It was tested using a load history representative of service loads. The
times shown are equivalent flying hours. Where two times are shown the first
is when the crack was observed to reach a rivet, and the second when it was
observed to emerge from the other side of the rivet.
@Seismicisolation
@Seismicisolation
136 Chapter 8: Fatigue Crack Paths
Figure 8.1. Fatigue crack path in a fighter aircraft centre section. Times are in flying hours.
It has been pointed out by numerous authors that on a macroscopic scale,

and under essentially elastic conditions, most fatigue cracks in isotropic
metallic materials tend to propagate in Mode I (Figure A.2). These authors
include Broberg (1987), Cotterell (1966), Knauss (1970), Minoshima et al.
(2000), Parton (1992), Pook (1983a, 2000a, 2002a) and Vinas-Pich et al.
(1996). Hence attention is often confined to this mode (see Sections 7.1 and
A.2.1). In metal fatigue this applies to macrocracks, which are Stage II in
Forsyth’s notation (see Section 3.4.2).
A Mode I crack is not necessarily straight, and crack paths are not readily
determined. It is usually assumed in the literature that criteria developed for
the path taken by a Mode I brittle fracture crack are also applicable to a Mode
I fatigue crack, and vice versa. The tendency to Mode I crack propagation,
for both brittle fracture and fatigue cracks, can at best be regarded as a useful
generalisation based on observation. It appears in the literature in a number of
forms, usually in a brittle fracture context, but does not appear to be capable
@Seismicisolation
@Seismicisolation
Metal Fatigue 137
of proof in any strict sense of the word (Ayer 1956). For example, in Cotterell
(1965) it appears as a self evident axiom for a perfectly isotropic material,
with a statistical argument to justify its use in practical situations where crack
direction and isotropy can only be defined in a macroscopic sense. An equiv-
alent form is the criterion of local symmetry (Goldstein and Salganik 1974).
This takes as self evident that a crack tends to propagate such that the elastic
crack tip stress field tends to become symmetrical, and hence Mode I.
The tendency to Mode I crack propagation cannot be justified by a ther-
modynamic criterion alone because satisfaction of an appropriate thermody-
namic criterion is a necessary, but not a sufficient, criterion for crack propaga-
tion (see Section 3.4.2).
8.1.1 C LASSES OF FATIGUE C RACK P ROPAGATION
There are two fundamentally distinct classes of crack propagation (Miller and
McDowell 1999a). The first is maximum principal stress dominated crack
propagation, which is in Mode I (Figure A.2), and is the class most often
observed in metal fatigue. The fatigue cracks shown in Figures 1.1–1.4, 3.15–
3.17, 6.9, 7.1, 7.2, 8.2, 8.7, 8.8, 8.12 and 8.13 are all maximum principal
stress dominated. A Mode I crack is usually, but not always, approximately
perpendicular to the maximum principal stress in the uncracked situation,
hence the term maximum principal stress dominated. As a generalisation, it
is sometimes stated that fatigue cracks tend to propagate perpendicular to the
maximum principal tensile stress (Pook 1983a).
The second class is shear dominated crack propagation. It usually takes
place on planes of maximum shear stress and is an important exception to
the tendency to Mode I fatigue crack propagation. Shear dominated crack
propagation is often observed when the crack tip plastic zone becomes large
(Miller and McDowell 1999b). In metal fatigue this applies to microcracks,
that is to Stage I cracks in Forsyth’s notation (see Section 3.4.2). It also ap-
plies in some special situations such as spot welded joints at high fatigue
loads (Pook 2002a). A shear dominated fatigue crack propagates in either
Mode II or in Mode III, or in a combination of the two.
A transition from square (Mode I) to slant crack propagation is sometimes
observed in thin sheets (see Section A.3.3.3). It is a third class of fatigue crack
propagation, and is another exception to the tendency to Mode I fatigue crack
propagation. Slant crack propagation is sometimes stated to be mixed Modes
I and III, but this is true only for the sheet centre line. Away from the centre
line it is mixed Modes I, II and III (see Section A.3.3.3). It is sometimes
@Seismicisolation
@Seismicisolation
called shear crack propagation, on the grounds that it takes place on planes
of maximum shear stress in an uncracked sheet, but this is a misnomer.
8.2 Crack Paths in Two Dimensions
Two dimensional linearly elastic analyses were used in early theoretical work
on fatigue crack paths. In two dimensions only Modes I and II stress intensity
factors, K and KII , can have non zero values. The Mode III stress intens-
ity factor, KIII , is zero. It is assumed here, as is usual in the literature, that
crack path criteria originally developed for crack propagation under a static
loading (brittle fracture) are also applicable to fatigue loadings. Associated
experimental work was on thin sheets of constant thickness, which were re-
garded as quasi two dimensional.
In the analysis of sheets and plates of constant thickness the use of two
dimensional stress intensity factor solutions is usually satisfactory for fa-
tigue crack propagation from an initial Mode I crack (see Sections A.3.1 and
A.3.3.3). If, in a sheet or plate the initial crack front is curved, as in Figure 8.2,
then this curvature merely leads to uncertainties in values of K. However, for
an initial nominal mixed Modes I and II crack, crack front curvature (Fig-
ure 8.3) not only introduces uncertainty into values of K and KII , but also
introduces, usually unwanted, non zero values of KIII . In addition in a two
dimensional approach the influence of corner point singularities is not taken
into account (see Section A.4). Neglect of these three dimensional effects
means that the use of two dimensional stress intensity factors is sometimes
unsatisfactory for mixed mode fatigue loadings.
8.2.1 M IXED M ODE T HRESHOLDS
For a Mode I fatigue loading the threshold for fatigue crack propagation,
Kth , is the value of the range of Mode I stress intensity factor, K, below
which fatigue crack propagation does not take place (see Section 7.4.3). More
precisely, it is the value of K below which Stage II crack propagation does
not take place (see Section 3.4.2). Under mixed mode fatigue loading the
situation is more complicated (see Section A.2.1). There are at least three
events which could, in principle, be used to define a mixed mode threshold
for fatigue crack propagation (Pook 1994b).
(a) Crack propagation in the plane of the initial crack.
(b) Mode I branch crack formation at or near the tip of the initial crack.
(c) Mode I branch crack propagation.
@Seismicisolation
@Seismicisolation
Metal Fatigue 139
Figure 8.2. Fracture surface of 19 mm thick aluminium alloy fracture toughness test specimen
(Pook 1968). Reproduced under the terms of the Click-Use Licence.
Figure 8.3. Fracture surface of DTD 5050 aluminium 5.5 Zn aluminium alloy fracture tough-
ness test specimen. The fatigue precrack fronts are slightly curved (Pook 1971). Reproduced
The precise definition of a mixed mode threshold for fatigue crack

propagation can have a significant effect on numerical values, and there is
some semantic confusion in the literature on what should be regarded as a
‘true’ mixed mode threshold. This is easily resolved through the use of fail-
ure mechanism maps such as Figure 8.4.
In two dimensions, only Modes I and II are possible. For Mode I (Stage II)
fatigue crack propagation to take place from an initial mixed mode crack, a
small Mode I branch crack must first be formed at the tip of the initial crack,
(or main crack) as shown schematically in Figure 8.5. Mode I branch crack
formation is a necessary, but not sufficient condition, for Mode I fatigue crack
propagation to take place. Occasionally, Mode I branch cracks form behind
the initial crack tip. Sometimes, Mode I branch crack formation is preceded
@Seismicisolation
@Seismicisolation
Figure 8.4. Failure mechanism map for mixed Modes I and II (Pook 1985a). Reproduced
by a limited amount of mixed mode Stage I crack propagation (up to about

0.5 mm) in the initial crack plane (Suresh 1998).
A Mode I branch crack will not propagate unless K, for the branch
crack, exceeds Kth . In a metallic material, mixed mode fatigue crack
propagation threshold behaviour is sometimes controlled by Mode I branch
crack formation, and sometimes by branch crack propagation. Failure to
make this distinction has led to some confusion in the literature.
For a combination of Modes I and II on the initial crack (Figure 8.5) the
Mode I stress intensity factor for the branch crack, KI∗ , is given approximately
by (Pook 1994b)
@Seismicisolation
@Seismicisolation
Metal Fatigue 141
Figure 8.5. Quasi two dimensional mixed Modes I and II initial crack with a small Mode I
branch crack (Pook 1989a). Reproduced under the terms of the Click-Use Licence.

θ θ 3
KI∗ = cos K cos2 − KII sin θ , (8.1)
2 2 2
where K and KII are Modes I and II stress intensity factors for the initial
crack, and θ is the branch crack propagation angle given by
KI sin θ = KII (3 cos θ − 1) (70.50◦ ≤ θ ≤ −70.50◦ ). (8.2)
In the derivation of Equations (8.1) and (8.2) it is assumed that a branch crack
is within the core region of a K-dominated region (Figure A.13).
Experimental mixed Modes I and II fatigue crack propagation threshold
data were used to construct the failure mechanism map shown in Figure 8.4
(Pook 1994b). Ranges of Modes I and II stress intensity factors, K and
KII , are normalised by Kth . In region C Stage II fatigue crack propagation
is possible, but does not necessarily take place. Once Stage II fatigue crack
propagation starts it usually continues to complete failure. Experimental data
show considerable scatter. It has been demonstrated that this is due to differ-
ences in Mode I branch crack formation. The conditions for Mode I branch
crack formation are unclear, but it appears to be facilitated by precrack front
curvature, as in the fracture toughness test specimen shown in Figure 8.3, and
also by metallurgical discontinuities (Pook 1982a).
In chaos theory terms, Mode I branch crack formation is a chaotic event
which is strongly dependent on initial conditions (Hall 1992). The theoretical
lower bound shown in Figure 8.4 is based on Equation (8.1), and the the-
oretical upper bound on the assumption that when K for the initial crack
exceeds Kth a Mode I branch crack must form and propagate (region D). In
@Seismicisolation
@Seismicisolation
Figure 8.6. Directionally stable fatigue crack propagation. Reprinted from Crack Paths. LP
Pook, WIT Press, ISBN 1-85312-927-5, 2002.
regions C and D Stage II fatigue crack propagation is sometimes preceded by

a limited amount of Stage I fatigue crack propagation. In region B, Stage I fa-
tigue crack propagation has been observed, but fatigue cracks usually become
non propagating. The mechanism by which these non propagating cracks de-
velop is not clear (cf. Section 7.4.5). In region A, fatigue crack propagation
does not take place. Any of the boundaries between regions could be used to
define a fatigue crack propagation threshold.
8.2.2 C RACK PATH S TABILITY
Two dimensional linearly elastic analyses are normally used in the consid-
eration of crack path stability. Related experimental work is usually carried
out on sheets or plates of constant thickness, which are regarded as quasi two
dimensional. From this viewpoint a fatigue crack propagating in Mode I may
be regarded as directionally stable if, after a small random deviation, it re-
turns to its expected, ideal crack path, as shown in Figure 8.6 (Pook 2002a).
A directionally unstable crack does not return to the ideal path following a
small random deviation; its path is a random walk, which cannot be prede-
termined. These ideas are not easily given rigorous mathematical form (Pook
2002a). For example, arbitrary limits have to be placed on what is regarded
as returning to the ideal crack path.
Deciding whether or not a particular crack path is stable is a practical dif-
ficulty in the analysis of experimental crack path stability results. A practical
definition of crack path stability needs to be associated with a finite amount
of crack propagation. The British Standard for fatigue crack propagation rate
testing, which uses directionally stable specimens, states that a crack path is
acceptable only if it lies within a validity corridor defined by planes 0.05W
on either side of the plane of symmetry containing the crack starter notch root
(Anon. 2003b). Here, W is the specimen width, or half width for a specimen
containing an internal crack. This criterion may be adapted as a crack path
@Seismicisolation
@Seismicisolation
Metal Fatigue 143
stability criterion by defining a stable crack as one which remains within the
validity corridor. This criterion is easy to apply, but has the disadvantage that
it does not take into account changes in stability as a crack propagates.
The T -stress criterion is based on a two dimensional linearly elastic ana-
lysis of a Mode I crack (see Sections A.2.1 and A.3). In consequence, crack
tip plasticity and three dimensional effects are not taken into account (see pre-
vious section). The elastic stress field in a cracked body may be expanded as
a series. The first term is the stress intensity factor, which dominates the crack
tip stress field, and is a singularity. Its coefficient is the Mode I stress intens-
ity factor, K. Other terms are non singular. For a Mode I crack, the second
term is a stress parallel to the crack, usually called the T -stress. Values of the
T -stress are available for a range of configurations (Sherry et al. 1995). The
third and higher terms can usually be neglected in the vicinity of a crack tip.
It has been argued that the directional stability of a Mode I crack in an
isotropic material under essentially elastic conditions is governed by the T -
stress (Cotterell 1966). If the T -stress is compressive and there is a small
random crack deviation, perhaps due to microstructural irregularity, then the
direction of Mode I crack propagation is towards the initial crack line (Fig-
ure 8.6). The analysis is a special case of crack propagation from a mixed
Modes I and II initial crack (see next section). A Mode I crack in a centre
cracked sheet loaded in uniaxial tension (Figure 7.4) is directionally stable in
this sense. Repeated random deviations mean that the crack follows a zigzag
path about the ideal crack path. In nonlinear dynamics terms, the ideal crack
path is an attractor (Pook 2002a).
When the T -stress is tensile a crack is directionally unstable, and follow-
ing a small random deviation, it does not return to the ideal crack path path.
A fatigue crack propagating in a double cantilever beam specimen is direc-
tionally unstable in this sense. Typical crack path behaviour is shown schem-
atically in Figure 8.7 (Pook 2000a). An initial random deviation can be either
above or below the centre line, so there are two possible crack paths. These
are shown as solid and dashed lines in the figure. The directional stability of
a crack may change as it propagates, and a stable Mode I crack may follow a
curved path. Cracks tend to be attracted by boundaries, as in Figure 8.7, and
are increasingly stable as a boundary is approached.
The biaxiality ratio, B, is a non dimensional function of the T -stress,
which is widely used (Sherry et al. 1995). It is given by
√
T πa
B= , (8.3)
K
@Seismicisolation
@Seismicisolation
Figure 8.7. Crack paths in a double cantilever beam specimen. Reprinted from Linear Elastic
Fracture Mechanics for Engineers. Theory and Applications. LP Pook, WIT Press, ISBN 1-
85312703-5, 2000.
where a is the crack length (half crack length for an internal crack), and K is
the Mode I stress intensity factor.
In practice, it is sometimes found that cracks are directionally stable even
when the T -stress is tensile (or the biaxiality ratio is positive). In particular,
the T -stress is tensile for the widely used compact tensile specimen. This
is specified in fracture mechanics based Mode I testing standards such as
Anon. (2003b) and in practice fatigue cracks in compact tension specimens
are usually directionally stable. The crack path shown in Figure 8.8 is an
exception. Initially, the fatigue crack propagated in the plane of the crack
starter notch but then turned upwards, and its path left the validity corridor.
The light area on the left is static fracture where the specimen was broken
open for examination.
An alternative approach is to consider the direct stresses, parallel to the
crack, and near the crack tip. That due to the T -stress is simply T . The stress,
σx , due to the Mode I stress intensity factor, on the crack line and ahead of
the crack, is given by (cf. Equation (A.2))
KI
σx = √ , (8.4)
2π r
where r is the distance from the crack tip.
The T -stress ratio, TR , may now be defined as the ratio of the T -stress to
σx , given by Equation (8.4), at some characteristic value of r, rch . Provided
that rch is small TR may be regarded as a crack tip parameter which is within
the K-dominated region (Figure A.13). Since the T -stress criterion is based
on the idea of random crack path perturbations due to microstructural irreg-
ularities, rch should be of the same order of size as microstructural features.
Taking rch = 0.0159 . . . mm leads, using MN-m units, to the convenient ex-
pression
@Seismicisolation
@Seismicisolation
Metal Fatigue 145
Figure 8.8. Unstable fatigue crack path in a high strength aluminium alloy compact tension
specimen.
0.01B
TR = √ . (8.5)
πa
For a particular material, there should be a critical value of TR , TRc , below
which a fatigue crack path is directionally stable.
As an example of the calculation of TR consider an infinite panel contain-
ing a centre crack, length 2a, with a uniaxial tensile stress, σ , perpendicular
to the crack (Figure 8.9). From the Westergaard solution for this configuration
(Westergaard 1939) (see Equation (A.4)), the Mode I stress intensity factor,
K, is given by
√
K = σ πa (8.6)
and
T = −σ. (8.7)
Substituting Equations (8.6) and (8.7) into Equations (8.3) and (8.5) gives
B = −1 and
−0.01
TR = √ . (8.8)
πa
@Seismicisolation
@Seismicisolation
Figure 8.9. Centre crack in an infinite sheet under uniaxial tension (Frost et al. 1974).
This equation shows that there is a size effect. That is for geometrically sim-
ilar configurations, TR decreases in absolute value as the crack size increases.
Taking a as a typical value of 25 mm, Equation (8.8) gives TR = −0.0357.
A biaxially loaded square sheet, edge 2W , containing a centre crack,
length 2a, (Figure 8.10) is sometimes used for crack path studies (Pook
2000a). The biaxial loading ratio, λ, is the ratio of the stress parallel to the
crack to the stress perpendicular to the crack. Figure 8.11 shows a typical
specimen, it was tested using the rig shown in Figure 4.5. The biaxiality ratio
is given by (Pook 2002a)
B = −[1 + 0.085(a/W )] + λ[1.029 + 0.115(a/W ) − 2.87(a/W )2
+ 4.829(a/W )3 − 3.125(a/W )4 ] (0.1 ≤ a/W ≤ 0.6). (8.9)
Table 8.1 shows values of the biaxiality ratio for λ values of −1, 0, 1, 2
and 3 calculated using Equation (8.9). Also shown are exact limiting values
for a/W = 0. For λ values of 0 and 1, B has the correct limiting values, but
for other values there appear to errors of the order of a few per cent. Values
of TR , taking W as 50 mm are shown in Table 8.2.
@Seismicisolation
@Seismicisolation
Metal Fatigue 147
Figure 8.10. Biaxially loaded square sheet. Reprinted from Linear Elastic Fracture Mechanics
for Engineers. Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-703-5, 2000.
Figure 8.11. Fatigue crack path in a Waspaloy sheet under biaxial fatigue load, specimen
MFGT 7. The grid is 0.1 inch (2.54 mm). National Engineering Laboratory photograph. Re-
produced under the terms of the Click-Use Licence.
As an example, some tests were carried out at room temperature on

Waspaloy, a nickel based gas turbine material, in order to determine the con-
ditions under which a fatigue crack path became unstable under biaxial fa-
@Seismicisolation
@Seismicisolation
Table 8.1. Values of biaxiality ratio, B, for square, centre cracked sheet.
a/W λ = −1 λ=0 λ=1 λ=2 λ=3
0 –2 –1 0 1 2
0.1 –2.025 –1.009 0.008 1.024 2.040
0.2 –1.988 –1.017 –0.046 0.925 1.895
0.3 –1.936 –1.026 –0.115 0.795 1.705
0.4 –1.879 –1.034 –0.189 0.656 1.501
0.5 –1.820 –1.043 –0.265 0.512 1.289
0.6 –1.754 –1.051 –0.348 0.355 1.058
Table 8.2. Values of the T -stress ratio TR , for square, centre cracked sheet, W = 50 mm.
a/W λ = −1 λ=0 λ=1 λ=2 λ=3
0 –∞ –∞ 0 ∞ ∞
0.1 –0.1616 –0.0805 0.0006 0.0817 0.1628
0.2 –0.1122 –0.0574 –0.0026 0.0522 0.1069
0.3 –0.0892 –0.0472 –0.0053 0.0366 0.0786
0.4 –0.0750 –0.0413 –0.0075 0.0262 0.0599
0.5 –0.0649 –0.0372 –0.0095 0.0183 0.0460
0.6 –0.0571 –0.0342 –0.0113 0.0116 0.0344
tigue loading (Pook and Holmes 1976). The specimens were 254 mm square
and 2.6 mm thick. The material had been cross rolled during production to en-
sure that its properties were reasonably isotropic. Tests were carried out using
sinusoidal constant amplitude fatigue loading at a stress ratio (ratio of min-
imum to maximum load in fatigue cycle), R, of 0.1. In each test the fatigue
load perpendicular to the crack was kept constant. Cracks were first propag-
ated from each end of an initial slit under uniaxial fatigue loading (λ = 0).
Then λ was increased by applying an in phase fatigue load parallel to the
crack, and crack path behaviour observed. Figure 8.11 shows the fatigue crack
path in one of the specimens. The results obtained were re-analysed, and the
re-analysis indicated that the critical value of TR , TRc , at which a fatigue crack
path became unstable in Waspaloy is about 0.022 (Pook 2002a).
8.2.3 C RACK PATH P REDICTION
A centre cracked sheet with an inclined crack, loaded in uniaxial tension (Fig-
ure 8.12), provides an example of Mode I fatigue crack path behaviour un-
der mixed Modes I and II fatigue loading on the initial crack (Pook 2002a).
A Mode I fatigue crack is Stage II crack in Forsyth’s notation (see Sec-
tion 3.4.2). Some experimental results for the configuration shown in the fig-
ure, obtained using thin sheets, show that branch crack propagation angles
from the initial crack vary widely (Lam 1993). This arises because Mode I
@Seismicisolation
@Seismicisolation
Metal Fatigue 149
Figure 8.12. Centre cracked sheet loaded in uniaxial tension with mixed Modes I and II initial
crack. Reprinted from Crack Paths. LP Pook, WIT Press, ISBN 1-85312-927-5, 2002.
branch crack propagation angles given by expressions such as Equation (8.2)

are not sharply defined (Pook 2002a). Consequently, under fatigue loading,
minor deviations from isotropy due to microstructural irregularities can be ex-
pected to have a significant influence on the branch crack propagation angle.
After branch crack formation a Mode I crack follows a curved path which
tends towards a line of symmetry, as shown schematically in Figure 8.3, irre-
spective of the initial direction of Mode I fatigue crack propagation. This line
of symmetry is an attractor (see previous section). Behaviour is in accord-
ance with the well known observation that fatigue cracks tend to propagate
perpendicular to the maximum principal tensile stress (see Section 8.1.1).
In two dimensions, the crack tip is a point and the crack path is a line.
Two dimensional Mode I crack path predictions have been carried out by
a number of authors, using the same general scheme, for example Portela
(1993). Calculations are carried out numerically using small increments of
@Seismicisolation
@Seismicisolation
straight crack propagation. The direction taken by each increment is selected

using the criterion that the increment is pure Mode I, that is the Mode II
stress intensity factor, KII , is zero. Care is needed to ensure that the crack path
approaches the correct limit as the increment size is reduced. In general, the
limit is a curve. Agreement between theoretical predictions and experimental
data obtained using thin sheets is variable (Pook 1995). This is not surprising,
since neither the effects of the crack tip plastic zone (see Sections 7.4, 7.4.2.2,
7.4.2.3 and 8.1.1), nor three dimensional effects are taken into account (see
Section 8.2). Two dimensional predictions have now largely been superseded
by three dimensional predictions (see Section 8.4.2).
Experimentally observed fatigue crack paths are either straight lines or
smooth curves, and are often remarkably reproducible (Broberg 1987). Cal-
culation of stress intensity factors for observed Stage II crack paths usually
confirms that fatigue crack propagation is indeed in Mode I. Occasionally,
non zero values of KII appear to be present, but this is an artefact due to crack
surface interference (Smith 1984). It is sometimes assumed, incorrectly, that
a curved surface implies that Mode II displacements must have been present
(Pook 1994b). A complication is that interference between crack surfaces in
the presence of Mode II displacements can increase the Mode I stress intens-
ity factor, K, and decrease KII .
Geometric features can have a strong influence on fatigue crack paths,
for example the fatigue crack path shown in Figure 8.1 is influenced by rivet
holes. As another example, during routine maintenance in 2002, one of the
two burners in the gas fired domestic central heating boiler in the author’s
house was found to be cracked due to thermal loading (see Section 3.4.3). The
fatigue crack is shown in Figure 8.13. The boiler was about 12 years old so,
assuming it fired about 10 times per day, about 44,000 thermal fatigue cycles
had been applied. The burner consists of a steel box with a series of small
and large holes on top to distribute the gas to the flame above the box. The
larger holes have reinforced perimeters. An internal wire mesh, just visible
in Figure 8.13, helped to distribute the gas evenly. Fatigue cracks appear to
have initiated at three places on the perimeter of a smaller hole, propagated
into two larger holes, with a small triangular piece becoming detached. Then
two fatigue cracks propagated across most of the width of the box, resulting
in improper combustion. The designer did not appear to have appreciated the
point that stress concentration factors are largely independent of hole size.
The reinforcement had prevented fatigue crack initiation at the large holes,
but its absence had allowed crack initiation at a small hole. Annual inspection
@Seismicisolation
@Seismicisolation
Metal Fatigue 151
Figure 8.13. Fatigue crack in burner from domestic central heating boiler.
was recommended by the boiler manufacturer. This ensured that the cracking
was detected before it became dangerous, and the burner was replaced.
8.3 Planar Crack Paths
It is well known that fatigue cracks in metallic materials often tend to propag-
ate in Mode I, that is approximately perpendicular to the maximum principal
tensile stress (see Sections 8.1.1 and 8.2.3). This means that fatigue crack
propagation is, on a macroscopic scale, often confined to a particular plane,
or in some complex geometries, to a particular slightly curved surface. Fig-
ure 7.1 is an example of the latter.
8.3.1 C RACK PATH D EVELOPMENT
Fatigue crack path prediction for a planar crack is essentially two dimen-
sional. It reduces to making the assumption that crack propagation is perpen-
dicular to the crack front, and then calculating increments of crack propaga-
tion along the crack front, using an appropriate fatigue crack propagation
expression such as Equation (7.2). Some highly refined numerical algorithms
have now been developed (Dhondt 2005, Kolk and Kuhn 2005). These take
into account the change in the nature of the crack tip singularity at a corner
point, shown in Figure 8.14, where a crack front intersects a surface (see Sec-
tion A.4). Surface cracks are sometimes modelled as semi elliptical cracks
@Seismicisolation
@Seismicisolation
Figure 8.14. Semi elliptical surface crack under uniaxial tension (Frost et al. 1974).
(see Section A.5.1). Surface cracks are sometimes called part through cracks.
If it is assumed that the crack remains semi elliptical as it propagates then
crack path predictions may be made by calculating increments of fatigue
crack propagation for the deepest point of the crack and for the corner points.
This simple approach is sometimes sufficient.
A crack has some analogies with a crystal dislocation (see Section A.5.3).
In particular, the elastic stress fields associated with a crack front and a dislo-
cation are both singularities. The associated energy means that a dislocation
has a line tension which controls its shape under an applied stress field. Sim-
ilarly, a crack front may be regarded as having a line tension which controls
its shape, but with the important difference that the motion of a crack front
is, in general, irreversible; that is a crack can only propagate, not contract.
Secondly, the crack tip stress field, and hence the line tension, may vary along
a crack front. At a corner point the corner point singularity provides a point
force which balances the line tension in a direction corresponding to the crack
front intersection angle, as shown in Figure 8.15 (Pook 2000a).
The line tension concept may be used qualitatively to account for two
well known aspects of fatigue crack behaviour (Pook 2002a). First, on a mac-
roscopic scale, a crack front is smooth, and any initial sharp corners rapidly
disappear (see Section A.5.3). This is not true if the scale of observation is re-
duced to a level at which microstructural effects are important; the crack front
@Seismicisolation
@Seismicisolation
Metal Fatigue 153
Figure 8.15. Crack front intersection angle. Reprinted from Linear Elastic Fracture Mech-
anics for Engineers. Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-703-5,
2000.
Figure 8.16. Convergence to stable crack front shape for a plate under pure bending. Reprinted
from Crack Paths. LP Pook, WIT Press, ISBN 1-85312-927-5, 2002.
then shows local irregularities within the overall smooth shape. Secondly, in a
particular set of circumstances a crack front tends to a particular stable shape.
As an example, Figure 8.16 shows some theoretical results based on finite
element analysis, for plates containing part through cracks loaded in fatigue
under pure bending (Lin and Smith 1999). The aspect ratio a/c where a is the
crack depth and 2c is the crack surface length (Figure 8.17) is plotted against
a/B, where B is plate thickness (Pook 2002a). (This definition of aspect ratio
differs from that in Section 7.6.1.) Irrespective of the initial aspect ratio, as
fatigue crack propagation proceeds, values converge to the same trend line.
@Seismicisolation
@Seismicisolation
Figure 8.17. Semi elliptical surface crack.
8.3.2 L EAK BEFORE B REAK
If a pressure vessel is subjected to fatigue loading, and a fatigue crack propag-

ates through its wall, it is important to know whether or not this will lead to a
catastrophic failure. There are two extreme situations. In the first, the vessel
disintegrates catastrophically due to brittle fracture before the fatigue crack
has propagated right through the wall, with complete depressurisation and
loss of the contained fluid. In the second, the fatigue crack propagates right
through the wall, and leakage of the contained fluid takes place without disin-
tegration of the vessel. In the short term there is no significant loss of pressure
or of the contained fluid. Further fatigue crack propagation is needed before
the vessel fails by brittle fracture, or the net section is so reduced that it fails
by plastic collapse. This is known as leak before break, and is an example
of the fail safe philosophy of design (see Section 5.3.2). Pressure containing
components, which are critical from a safety viewpoint, are often designed to
leak before break in order to avoid catastrophic failure (Darlaston and Har-
rison 1977, Pook 2000a, Zerbst et al. 2003). Figure 1.5 shows a domestic tap
which did, fortuitously, leak before break.
The propagation of an initial crack from the inside of a pressure vessel
wall is shown schematically in Figure 8.18. When the initial crack penet-
rates the vessel wall it rapidly becomes a through the thickness crack with
approximately straight crack fronts The initial crack is in plane strain (see
Section A.3.3.1). The stress state of the through the thickness crack (plane
strain, plane stress, or an intermediate stress state) depends on the wall thick-
ness, the applied stress, and the yield stress of the material.
There are three main possible sequences of events as a fatigue crack, ori-
ginating at the inner surface, propagates through the pressure vessel wall. In
the first the plane strain fracture toughness, KIc , is reached before the fatigue
crack has propagated right through the wall, and the vessel fails catastroph-
ically (see Sections A.3 and A.3.3.1). In the second sequence KIc is reached
before the fatigue crack has propagated right through the wall, but the through
the thickness crack is in plane stress. The plane stress fracture toughness, Kc ,
is greater than KIc so it is possible for the brittle fracture crack to stop. Hence
@Seismicisolation
@Seismicisolation
Metal Fatigue 155
Figure 8.18. Fatigue crack propagating from inner surface of pressure vessel wall.
depending on circumstances, the pressure vessel may either leak before break
or fail catastrophically. In the third sequence the fatigue crack propagates
right through the wall before KIc is reached and, if the stress intensity factor
for the through the thickness crack is less than the fracture toughness, the
vessel will leak before break.
The margin of safety introduced by leak before break depends on the dif-
ference between the through crack length at which leakage can be detected,
and that at which the pressure vessel fails catastrophically. Determining these
crack lengths is therefore a key factor in carrying out a leak before break as-
sessment, together with determining the length of the through the thickness
crack which results when a crack has propagated right through the vessel wall.
Because of the numerous variables involved, none of these are readily determ-
ined. This makes it difficult to establish whether or not leak before break can
be relied on. Ensuring that an aircraft pressure cabin will leak before break is
a major preoccupation of aircraft designers (Wanhill 2003, Grandt 2004).
8.4 Crack Paths in Three Dimensions
It is well known that fatigue cracks in metallic materials often tend to propag-
ate in Mode I, that is approximately perpendicular to the maximum principal
tensile stress (see Sections 8.1.1 and 8.2.3). This creates a fundamental dif-
ficulty in the study of the general three dimensional case of Mode I fatigue
crack propagation from an initial mixed mode crack. Differential geometry
considerations show that the crack propagation surface for a Mode I crack
must be smooth (Pook 2002a). However, in the presence of Mode III on the
initial crack an element of a Mode I branch crack can intersect the initial
crack front at only one point (Figure 8.19).
@Seismicisolation
@Seismicisolation
Figure 8.19. Mixed Modes I, II and III initial crack with a smooth branch crack and an element
of a Mode I branch crack (Pook 1989a). Reproduced under the terms of the Click-Use Licence.
What happens when a fatigue loading is applied in the presence of Mode

III on the initial crack is illustrated by the fracture surface of the mild steel
angle notch specimen shown in Figure 8.20 (Pook 1985b). The rough area
is static fracture where the specimen was broken open for examination. The
specimen used was tested in three point bending, and is shown schematic-
ally in Figure 8.21. The expected tendency to Mode I crack propagation was
observed on two distinct scales. On a scale of 1 mm crack fronts were approx-
imately straight, and initially crack propagation was mixed mode, as shown
schematically by the smooth branch crack in Figure 8.19. As the fatigue crack
propagated the crack front rotated until it was perpendicular to the specimen
surfaces, and crack propagation was in Mode I. On this scale the crack fol-
lows a curved path which tends towards a plane of symmetry in the uncracked
specimen. Differential geometry considerations show that this curved path is
not a permissible Mode I crack path (Pook 2002a). The plane of symmetry is
an attractor (see Sections 8.2.2 and 8.2.3).
On a smaller scale of 0.1 mm the tendency to Mode I fatigue crack
propagation results in the production of what is known as a twist crack (Lawn
and Wilshaw 1975). A twist crack consists of narrow Mode I facets connec-
ted by irregular, predominantly Mode III cliffs as shown schematically in Fig-
ure 8.22 (Pook 2002a). The Mode I facets gradually merge as, when viewed
@Seismicisolation
@Seismicisolation
Metal Fatigue 157
Figure 8.20. Twist crack fracture surface of mild steel angle notch specimen MFSH 35. Na-
tional Engineering Laboratory photograph. Reproduced under the terms of the Click-Use Li-
cence.
Figure 8.21. Mild steel angle notch specimen (Pook 1985a). Reproduced under the terms of
on the larger scale, the crack propagation surface becomes perpendicular to

the specimen surfaces (Figure 8.20).
8.4.1 L OWER B OUND M IXED M ODE T HRESHOLDS
The theoretical lower bound for the Stage II fatigue crack propagation
threshold under mixed Modes I and II fatigue loading, shown in Figure 8.4,
can be extended to the general three dimensional case in which all three
modes are present (Pook 1994b). For a combination of Modes I, II and III
@Seismicisolation
@Seismicisolation
Figure 8.22. Schematic section through a twist crack. Reprinted from Crack Paths. LP Pook,
WIT Press, ISBN 1-85312-927-5, 2002.
on the initial crack the stress intensity factor, KI∗ , for an element of a Mode I
branch crack (Figure 8.19) is given approximately by

K(1 + 2ν) + K 2 (1 − 2ν)2 + 4KIII 2
∗
KI = , (8.10)
2
where K is the value of KI∗ from Equation (8.1), KIII is the Mode III stress
intensity factor for the initial crack and ν is Poisson’s ratio. The presence of
Mode III does not affect the value of θ given by Equation (8.2) (Figure 8.5).
The lower bound threshold for fatigue crack propagation, shown in Fig-
ure 8.4, is extended to the general case of mixed Modes I. II and III loading
in Figure 8.23. The figure is based on Equation (8.10) with ν = −1/3. Val-
ues of the ranges of Modes I, II and III stress intensity factors, K, KII ,
and KIII , are normalised by the Stage II fatigue crack propagation threshold,
Kth . Experimental data confirm this theoretical lower bound for pure Modes
II and III fatigue loadings on the initial crack, for mixed Modes I and II fa-
tigue loadings, and mixed Modes I and III fatigue loadings (Pook 1994b).
For a long initial crack, Stage I crack propagation is sometimes observed at
below the lower bound, but this does not lead to complete failure. Hence it is
possible to use the envelope for design purposes, although at times it would
be very conservative.
8.4.2 C RACK PATH D ETERMINATION
The development of twist cracks (Figure 8.20) in the presence of Mode III
on the initial crack complicates the numerical determination of fatigue crack
paths in the general three dimensional case. At the present state of the art the
best strategy appears to be to consider the crack path at a scale of the order of
1 mm (see Section 8.4). At this scale the crack path is smooth, but, in general,
the crack path is mixed mode. This strategy was used in the development of an
advanced finite element fatigue crack path prediction program (Schöllmann
et al. 2003).
@Seismicisolation
@Seismicisolation
Metal Fatigue 159
Figure 8.23. Theoretical lower bound for Stage II fatigue crack propagation threshold (Pook
1989a). Reproduced under the terms of the Click-Use Licence.
In the program, increments of fatigue crack propagation are calculated

using the range of an equivalent stress intensity factor, Keq , and a fatigue
crack propagation equation such as Equation (7.2). The directions of these in-
crements are determined for points along the crack front, using an appropriate
expression such as Equation (8.2). The increments then define a ruled surface.
Overall, the increments of fatigue crack propagation, provided that they are
sufficiently small, can be taken as defining a smooth crack path. Using this
sort of approach good agreement has been obtained between theoretical pre-
dictions and crack paths observed on service (Richard et al. 2006).
The following useful approximation functions have been developed
(Richard et al. 2005).
2
|KII | |KII |
θ = ∓ 140 − 70 , (8.11)
KI + |KII | + |KIII | KI + |KII | + |KIII |
where θ is in degrees and K, KII , and KIII are the Modes I, II and III stress
intensity factors.
K 1
Keq = + K 2 + 4(1.155KII )2 + 4KIII , (8.12)
2 2
where K, KII , and KIII are the Modes I, II and III stress intensity factor
ranges.
@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
9
Why Metal Fatigue Matters
9.1 Introduction
In the past century and a half much has been written on the problem of metal
fatigue, and it has only been possible to describe some of the more important
metal fatigue topics in this book. One might argue from the infrequency of
catastrophic metal fatigue failures in structures and components that the prob-
lem is no longer serious. Clear explanations are usually found for the major
failures that do occur, with human error often being involved. However, lesser
metal fatigue failures, often unrecognised unless they happen to be seen by
a fatigue specialist, are a common and expensive nuisance. General aware-
ness of the dangers of metal fatigue has greatly increased. By and large the
problem can be contained, if not solved; but the price is eternal vigilance.
Metal fatigue is very much a descriptive subject. Metallurgical descrip-
tions are concerned with the effect of fatigue loading on the state of the ma-
terial. Mechanical descriptions are concerned with matters such as the num-
ber of cycles to failure, or the rate of propagation of a fatigue crack, and are
the more useful from an engineering viewpoint. The multi volume work on
structural integrity, edited by Milne et al. (2003) includes encyclopaedic cov-
erage of many metal fatigue topics. These are mostly written as state of the
art reviews, and inevitably some of the material has already been outdated.
It is probably impossible to write a comprehensive standard text on metal
fatigue. Perhaps the closest approach is the book by Suresh (1998). On well
established topics, his exposition is in straightforward textbook style. In more
controversial areas, sections are written as state of the art reviews, with oppos-
ing views given equal prominence. Where the author expresses a preference,
this is usually the conventional wisdom rather than a minority view.
@Seismicisolation
@Seismicisolation
162 Chapter 9: Why Metal Fatigue Matters
It must always be remembered that fatigue is only one of many, often

conflicting requirements with which a designer must contend. Innovation fre-
quently brings the risk of fatigue failure, but because metal fatigue is not a
suddenly occurring phenomenon, and also because of the need to pursue de-
tail, it is only possible to progress slowly as experience is acquired in fresh
areas. The transfer of metal fatigue technology from producer to user is often
a frustrating and expensive process.
9.2 Avoidance of Fatigue Failure
Even rough estimation of fatigue strength at the preliminary design stage

presents problems. At first sight this is surprising, since large amounts of ex-
perimental data are available and, in general, the basic mechanisms of metal
fatigue failure are relatively well understood and documented. Problems in
estimating fatigue strength arise because of the complexity and variety of
realistic engineering design situations. This makes it difficult to generalise
on known metal fatigue data and then apply these generalisations to specific
engineering problems. In practice, many advances in design against fatigue
have been made on what amounts to a trial and error basis. A considerable
research and development effort on metal fatigue is usually needed whenever
the service life of a fundamentally new type of structure or component needs
to be predicted. Long-term background work on metal fatigue is needed as it
is likely to decrease the effort needed for a particular new situation, and also
means that a pool of expertise is kept available.
In principle, an analytical approach, based on materials data and applied
mechanics, can be used to predict the service life of a structure or compon-
ent. The fatigue crack initiation and fatigue crack propagation phases must be
treated separately, although in practice one or the other usually predominates.
An analytical approach requires expert knowledge, and often fails because of
lack of the detailed information required. Simplified calculations can be use-
ful when an approximate answer is adequate. Simplified fracture mechanics
calculations have proved particularly valuable in failure analysis. As new in-
formation becomes available, the range of situations that can be analysed is
increasing.
Standard procedures of various degrees of formality are perhaps the most
satisfactory from the designer’s viewpoint. They may be based on analytical
procedures, service experience, the results of structural tests, or on some com-
bination of these. They need not have a theoretically sound basis, provided
they give sufficiently accurate answers, and ideally no expert knowledge
@Seismicisolation
@Seismicisolation
Metal Fatigue 163
is required. These methods fail where no appropriate standard procedure is

sufficiently accurate for the situation being considered. The development of
standard procedures, which can be a lengthy process, is probably the best way
of assimilating the results of research into engineering practice. In practice,
most metal fatigue assessments are based on simplified standard procedures,
which despite their apparent lack of physical validity, are known to give con-
servative answers. For critical structures and components the use of specific
standard procedures is often enforced by legislation.
Service loading testing is sometimes the most cost effective method of de-
termining the service life of a component or structure. Modern servohydraulic
equipment permits the application of virtually any desired load history. This
kind of testing, which first became established in the aircraft industry, is now
used for critical structures and components in a wide range of industries.
Such tests are often used to calibrate approximate analytical methods so that
these methods can be used to extrapolate the results of service loading tests to
broadly similar situations. Much effort has been devoted to the development
of appropriate standard load histories for service loading testing.
When a structural failure due to metal fatigue (or other causes) leads to
catastrophe, society as a whole takes an interest through its institutions. In a
typical catastrophic metal fatigue failure, the immediate cause of the failure is
usually easily ascertained, but it is more difficult to discover how procedures
designed to ensure safety came to fail. Recommendations of official enquiries
into catastrophic failures are often mainly concerned with plugging proced-
ural loopholes.
9.3 Research and Development
Basic research on metal fatigue has always had a strong practical bias in that
the motivation is usually to acquire data which will help to avoid service fail-
ures. In academic circles, it is sometimes assumed that an analytical approach
is the ideal to be aimed for. Much early metal fatigue work was concerned
with the experimental determination of the fatigue strength of plain, or relat-
ively mildly notched, specimens. Numerous variables can affect the fatigue
behaviour of a particular metallic material, including mean stress, surface
finish, the environment, and special situations such as biaxiality of loading.
Further complications arise when realistic load histories are introduced. Con-
sequently, no data collection, however large, can be comprehensive in the
sense that the need for fatigue testing during product development can def-
initely be eliminated. Fatigue testing is time consuming and expensive, but
@Seismicisolation
@Seismicisolation
164 Chapter 9: Why Metal Fatigue Matters
claimed shortcuts to the acquisition of data should be treated with caution, as

they are sometimes based on empirical correlations of limited applicability,
or on an oversimplified view of the nature of metal fatigue.
Perhaps the most significant advance in metal fatigue was the general real-
isation that many structures and components contain crack like flaws which
are either introduced during manufacture, especially if welding is used, or
form early in service. Virtually the whole service life is then occupied by fa-
tigue crack propagation from these flaws. An understanding of fatigue crack
propagation is therefore essential for the understanding and prediction of the
fatigue behaviour of many structures and components. One consequence is
that design often has to be for a specific, limited life, rather than for an indef-
inite life. Another consequence is that the presence of a crack does not neces-
sarily mean that a structure or component is defective in a legal sense. The
widespread use of the fracture mechanics parameter stress intensity factor to
analyse fatigue crack propagation data means that behaviour can be predicted
for any cracked body for which a stress intensity factor expression is avail-
able. Hence a designer is not limited to situations similar to those used to
generate experimental fatigue crack propagation data.
9.4 The Role of Non Destructive Testing

In the context of metal fatigue the usual meaning of non destructive testing
(NDT) is the detection and sizing of cracks and crack like flaws in compon-
ents, structures and laboratory specimens. NDT and fracture mechanics (the
applied mechanics of crack propagation) are complementary disciplines that
influence each other. Both are used in the measurement of material properties
in the presence of a crack, and both are used in fatigue assessments of com-
ponents and structures in the presence of actual or postulated cracks. Taken
together they answer questions such as ‘Is the minimum crack size that can
be found by an NDT technique acceptable?’, ‘Is a more sophisticated (and
expensive) NDT technique needed?’, and ‘Should the part be redesigned (per-
haps lowering its structural efficiency) so that a larger crack is acceptable?’.
It is widely recognised that, for critical structures and components, allowance
for appropriate NDT must be made at the design stage.
9.5 Current Trends
Despite increasing use of non metallic materials in various components and

structures, the extensive use of metallic materials will continue indefinitely.
@Seismicisolation
@Seismicisolation
Metal Fatigue 165
Hence, research and development aimed at the resolution of specific prac-

tical metal fatigue problems will also continue. Many critical structures and
components are subject to legislation, which often means that designers are
constrained in the approaches to fatigue assessment that can be used. This
trend will continue, with further codification of existing theoretical and ex-
perimental data. Another legislative trend is to ban potentially harmful metals
such as lead. This is leading to the development of lead free solders, with the
concomitant need to asses their fatigue properties.
In the 1980s work on metal fatigue as an academic discipline, appeared
to have passed its peak. Much research work on fatigue and related topics
was concerned with obsessive pursuit of detail, involving much labour, but
yielding little in the way of deeper understanding. Metallurgical descriptions
of metal fatigue mechanisms were usually at scales of the same order as sizes
of metallurgical features, that is 10−4 to 10−2 mm. These mechanisms were
often referred to as micromechanisms. The descriptions were usually qualit-
ative, rather than quantitative, in nature. The applied mechanics framework,
including fracture mechanics, which formed the basis of mechanical descrip-
tions dealt with macroscopic aspects at scales of 10−1 mm and above. Frac-
ture mechanics descriptions of fatigue crack propagation at these scales was
highly developed. What has happened, in the intervening two decades, is that
developments in instrumentation, computers and software have made it pos-
sible to extend metallurgical and mechanical descriptions to smaller scales,
with as associated increase in understanding. This has led to remarkable im-
provements in the fatigue resistance of some metallic materials.
Finally, a very welcome trend is the increased international exchange of
information on metal fatigue that has followed political changes in a number
of countries.
@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
A
Fracture Mechanics
A.1 Introduction
The applied mechanics framework for study of the behaviour of cracked bod-
ies under load is known as fracture mechanics. The application of fracture
mechanics to fatigue crack propagation is well established, and most mod-
ern books on metal fatigue include an introduction to the topic. The account
below is based on Frost et al. (1974) and Pook (2000a, 2002a). These books
include numerous references. The book by Frost et al. was the first on metal
fatigue in which a fracture mechanics approach was used throughout. It was
reprinted in 1999.
Fracture mechanics does not provide any information about the processes
involved in fatigue crack propagation. It does provide the descriptive and ana-
lytic framework needed for their characterisation, and for the application of
fatigue crack propagation data to practical engineering problems.
Simplifying assumptions have become conventional in much present day
fracture mechanics, and these are satisfactory for many purposes. The mater-
ial is assumed to be a homogeneous isotropic continuum, and its behaviour
is assumed to be linearly elastic. Crack surfaces are assumed to be smooth,
although on a microscopic scale they are generally very irregular. Modifica-
tions are made to basic linear elastic fracture mechanics theory to allow for
the actual behaviour of real materials. The basic ideas in linear elastic frac-
ture mechanics are straightforward. The mathematics involved is often for-
midable, but does lead to the useful and easily applied key concept of stress
intensity factor, which describes the elastic stress and displacement fields in
the vicinity of a crack tip. A stress intensity factor has the dimensions of
√ √
stress × length. The most widely used units are MPa m. These units ap-
√
pear in many standards and are therefore to be preferred. The use of MPa m
@Seismicisolation
@Seismicisolation
168 Appendix A: Fracture Mechanics
Figure A.1. Notation for crack tip stress field (Frost et al. 1974).
is not particularly convenient since crack sizes are normally

√ measured in mm,
√
and N/mm3/2 units are sometimes used. 1 MPa m = 1000 N/mm3/2 ≈
31.62 N/mm3/2 .
Some figures which appear in the main text are repeated in order to make
the appendix self contained.
A.2 Notation for Stress and Displacement Fields
The conventional notation for the position of a point relative to the crack tip,
and for the stresses at this point, is shown in Figure A.1. The point on the
crack tip is the origin of the coordinate system and the z axis lies along the
crack tip. Displacements of points within the cracked body when the body is
loaded are u, v, w in the x, y, z directions. The terms crack tip and crack front
are synonymous. Crack tip tends to be used for two dimensional situations
and crack front in three dimensions.
A.2.1 C RACK S URFACE D ISPLACEMENT
A fundamental fracture mechanics concept is that of crack surface displace-

ment. In fracture mechanics the interest is in what happens in the vicinity
@Seismicisolation
@Seismicisolation
Metal Fatigue 169
Figure A.2. Notation for modes of crack surface displacement (Frost et al. 1974).
of the crack tip, so it is sometimes referred to as crack tip surface displace-

ment. If a load is applied to a cracked body, then the crack surfaces move
relative to each other. For points on opposing crack surfaces that were ini-
tially in contact there are three possible modes of crack surface displacement
(Figure A.2); Mode I where opposing crack surfaces move directly apart in
directions parallel to the y axis; Mode II where crack surfaces move over each
along the x axis, that is, perpendicular to the crack tip; and Mode III where
crack surfaces move over each other in directions parallel to the z axis, that
is, parallel to the crack tip. By superimposing the three modes, it is possible
to describe the most general case of crack surface displacement. The terms
Mode I, Mode II and Mode III are usually capitalised, and are often used in
the metal fatigue literature without explanation. The descriptive terms; open-
ing mode, edge sliding mode, and shear mode are sometimes used for Modes
I, II and III respectively. The term mixed mode means that at least one mode,
other than Mode I, is present. The modes of crack surface displacement may
also be used to characterise crack propagation.
A particular type of elastic crack tip stress field is associated with each
mode of crack surface displacement (Paris and Sih 1965). These stress fields
are characterised by stress intensity factors, symbol K. Subscripts I, II and
III are used to denote mode. Where there is no subscript, Mode I is usually
implied; this convention is sometimes used in the text. Corresponding dis-
placement fields permit calculation of crack tip surface displacements.
It is matter of observation that, when viewed on a macroscopic scale,
and under essentially elastic conditions, cracks in metals tend to propagate
in Mode I, so attention is largely confined to this mode (see Sections 3.4.2,
@Seismicisolation
@Seismicisolation
Figure A.3. Square section ring element around crack tip. Reprinted from Linear Elastic Frac-
ture Mechanics for Engineers. Theory and Applications. LP Pook, WIT Press, ISBN 1-85312-
703-5, 2000.
7.1 and 8.1). At smaller scales, cracks are generally very irregular and differ-
ent modes of crack surface displacement may be observed.
A.2.2 VOLTERRA D ISTORSIONI
If a crack surface is considered as consisting of points then the three modes of

crack surface displacement (Figure A.2) provide an adequate description of
the movements of crack surfaces when a load is applied. However, if the sur-
face is regarded as consisting of infinitesimal elements, then element rotations
must also be described, and Volterra distorsioni (distortions) are appropriate.
Volterra distorsioni are also applicable in the theory of crystal dislocations
(Nabarro 1967).
In his analysis of distorsioni Volterra considered the simplest multiply
connected body, that is a cylinder with a central hole. An account in English
is given in Zastrow (1985). The cylinder, free of body forces, surface forces,
or any initial stress, is made simply connected by a cut along a radial plane.
The cut surfaces are regarded as completely rigid, but the remainder of the
cylinder is elastic. The cut surfaces may be moved relative to each other in
six different ways, so there are six distinct Volterra distorsioni.
The conventional approach to Volterra distorsioni needs to be modified
for the description of crack tip surface movements (Pook 2000a). Consider
a pair of infinitesimal elements, A and B, which are in the xy plane and are
situated on the upper and lower surface of an unloaded crack respectively
(Figure A.3). Their initial coordinates are (r, 180◦ ) and (r, −180◦ ). The ele-
ments A and B are connected by a ring element of infinitesimal width. For
clarity, this ring element is shown as having a square cross section and the
elements are separated in the figure. Such ring elements correspond to Vol-
terra’s cylinders.
Three of the six Volterra distorsioni, correspond to the three modes of
crack tip surface displacement, Figure A.4 (Pook 2002a). Under a Mode I
@Seismicisolation
@Seismicisolation
Metal Fatigue 171
Figure A.4. Volterra distorsioni, modes of crack surface dislocation. Reprinted from Crack
Paths. LP Pook, WIT Press, ISBN 1-85312-927-5, 2002.
loading element A moves a distance v in the y direction, and element B a dis-

tance −v (Figure A.4 left). For consistency with crack tip surface displace-
ment notation this is called a Mode I dislocation. The ring element remains
within the xy plane, that is, initially plane sections, perpendicular to the crack
tip remain plane. Distortion of the ring element is symmetrical about the xz
plane. Under a Mode II loading element A moves a distance u in the x dir-
ection, and element B a distance −u, and this is a Mode II dislocation (Fig-
ure A.4 centre). The ring element remains within the xy plane, that is plane
sections remain plane, but its distortion is not symmetrical about the xz plane.
Under a Mode III loading element A moves a distance −w in the z direction,
and element B a distance w (Figure A.4 right). This is a Mode III dislocation.
The ring element does not remain within the xy plane, that is plane sections
perpendicular to the crack front do not remain plane. Distortion of the ring
element has rotational symmetry about the x axis.
The remaining three Volterra distorsioni involve rotation of the elements A
and B, and are usually called disclinations. The notation used here is based on
the idea that each mode of dislocation and disclination is associated with the
same coordinate axis (Pook 2002a). Hence in a Mode I disclination, element
A rotates through an angle β about an axis parallel to the y axis, and element
B rotates through an angle −β (Figure A.5 right). In a Mode II disclination,
element A rotates through an angle α about an axis parallel to the x axis
and element B rotates through an angle −α (Figure A.5 centre). In a Mode III
disclination, element A rotates through an angle γ about an axis parallel to the
z axis and element B rotates through an angle −γ (Figure A.5 left). Modes II
and III disclinations cannot exist in isolation because of interference between
elements A and B. Any dislocation mode may be decomposed into a dipole
of equal and opposite disclinations of either of the other two modes.
@Seismicisolation
@Seismicisolation
Figure A.5. Volterra distorsioni, modes of crack surface disclination. Reprinted from Crack
Paths. LP Pook, WIT Press, ISBN 1-85312-927-5, 2002.
If movements of the elements A and B are identical, then a dislocation

becomes a translation and a disclination a rotation. For convenience, con-
tinuous distributions of dislocations are usually referred to as dislocations,
and similarly for disclinations, translations and rotations.
A.3 Stress Intensity Factors
It is intuitively obvious that a large crack is more severe than a small crack.
A basic requirement for the study of the behaviour of cracked bodies is to
express this numerically. The presence of a crack dominates the stress field in
the vicinity of the crack tip, and some results are not intuitively obvious.
The key concept of stress intensity factor, for Mode I and for Mode II,
arises from a two dimensional linearly elastic analysis for a straight crack.
This follows the usual methods of elastic stress analysis in which strains and
distortions are assumed to be small, and conditions of equilibrium and com-
patibility must be satisfied (Gere and Timoshenko 1991). The use of a two
dimensional analysis simplifies the mathematics, and also simplifies some
descriptions. For example, what is meant by crack length is unambiguous.
Mode III is not possible in two dimensions, so for this mode a quasi two
dimensional anti plane analysis is used.
Stress intensity factors may be used to characterise the mechanical prop-
erties of cracked specimens in just the same way that stresses are used to
characterise the mechanical properties of uncracked specimens. They help to
quantify the rather elusive concept of a material’s toughness. One conveni-
ent definition is this: resistance to crack propagation (including fatigue crack
propagation). For example the fracture toughness, Kc , of a metallic mater-
ial may be defined as the value of the Mode I stress intensity factor, KI , for
@Seismicisolation
@Seismicisolation
Metal Fatigue 173
failure under a static load. In practice, failure is not abrupt, and the frac-
ture toughness is defined for a small, specified amount of crack propagation
(Anon. 2005g).
The stress field in vicinity of a crack tip is dominated by the leading term
of a series expansion of the stress field. For a particular mode of crack surface
displacement this leading term is always of the same general form. Individual
√
stress components are proportional to K/ r where K is the stress intensity
factor and r is the distance from the crack tip (Figure A.1). A stress intensity
factor is a singularity of order −1/2; as the crack tip is approached, stresses
tend to infinity. A formal definition of the Mode I stress intensity factor is
√
KI = lim(r → 0)σy 2π r , (A.1)
where σy is √ the stress perpendicular to the crack along the x axis. The numer-
ical factor, 2π , in the equation is the usual convention. Other conventions
are occasionally encountered, especially in early work, leading to numerically
different values for stress intensity factors. There are corresponding equa-
tions, in terms of shear stresses, for the other two modes. Stress intensity
factors of the same mode may be combined by algebraic addition.
Once K is known, stress and displacement fields in the vicinity of the
crack tip are given by standard equations (Pook 2000a). In Modes I and II,
stresses are independent of the stress state, but displacements are a factor
(1 − ν 2 ) less for plane strain than for plane stress (ν is Poisson’s ratio). For
example, for Mode I on the x axis in front of the crack
KI
σy = √ . (A.2)
2π r
Also, for the upper crack surface

2KI 2r
v= (plane stress), (A.3)
E π
where v is the displacement in the y direction, that is, perpendicular to the
crack, and E is Young’s modulus. The equation implies that for Mode I a
crack opens up into a parabola (see Section A.4). Displacements are also
parabolic for Modes II and III. Equation (A.3) also implies that Mode I crack
surface displacement is a combination of a Mode I dislocation and a Mode
III disclination (Figures A.4 and A.5). Similarly, Mode III crack surface dis-
placement is a combination of a Mode III dislocation and a Mode I disclina-
tion. However, Mode II crack surface displacement is just a Mode II disloca-
tion.
KI must be positive, since a compressive load simply holds a crack closed.
However, the Modes II and III stress intensity factors, KII and KIII can be
@Seismicisolation
@Seismicisolation
either positive or negative, and the sign used is a matter of convention. It is

usually taken as positive, but careful attention to sign is needed when calcu-
lating stresses and displacements.
It is possible to obtain stress field equations, using Kirchoff plate bending
theory, in terms of a bending mode stress intensity factor, KB (Paris and Sih
1965). At a plate surface KB is equivalent to KI . Corresponding crack sur-
face displacements are a Mode II disclination (Figure A.5 centre). Because
of crack surface interference this is not physically realistic so KB is not now
used.
A.3.1 S TRESS I NTENSITY FACTOR S OLUTIONS
Stress intensity factors are available for numerous configurations, for ex-
ample Murakami (1987, 1992b, 2001) and this facilitates practical applica-
tions. Solutions for test specimens are included in appropriate standards, for
example Anon. (2003a). A solution for a particular configuration is some-
times called a K-calibration or a compliance function. Where a solution is
presented as an equation fitted to numerical results, care must be taken not
to use it outside its specified range. Conventionally, two dimensional solu-
tions are used for sheets and plates of constant thickness subjected to in plane
loads. This is usually satisfactory. To illustrate the general form of solutions
for the Mode I stress intensity factor, KI , some examples are given below.
These are all for loads perpendicular to the crack. Stresses parallel to a crack
have no effect.
A.3.1.1 Two Dimensional Solutions
For a centre crack, length 2a (it is conventional to take the length of an in-
ternal crack as 2a) under a remote uniaxial tension σ (Figure A.6)
√
KI = σ π a . (A.4)
For a small edge crack, length a, in a sheet under uniaxial tension, shown
in Figure A.7 (Pook 2000a).
√
KI = 1.12σ π a . (A.5)
Equation (A.5) also applies to a crack at a blunt notch if the local stress
is used (Figure A.8), and to a crack at a sharp notch if a is taken as the crack
length plus the notch depth shown in Figure A.9 (Pook 2000a).
Solutions are sometimes presented in the form
√
KI = σ Y π a , (A.6)
@Seismicisolation
@Seismicisolation
Metal Fatigue 175
Figure A.6. Centre crack in an infinite sheet under uniaxial tension (Frost et al. 1974).
where Y is a geometric correction factor, usually of the order of 1, and a

is a characteristic crack dimension. Hence in Equation (A.5) Y = 1.12. To
facilitate calculations other definitions of Y are sometimes used as in Equa-
tion (A.7), below.
Test specimens of standard design are included in various fracture mech-
anics based standards. For example, for the three point bend single edge notch
specimen shown in Figure A.10 (Anon. 2003b),
FY
KI = √ , (A.7)
B W
where F is force, B specimen thickness and W specimen width. Y is given
by
√
6 α [1.99 − α(1 − α)(2.15 − 3.93α + 2.7α 2 )]
Y = , (A.8)
(1 − 2α)(1 − α)3/2
where α = a/W and the equation is valid for 0 ≤ α ≤ 1. In Anon. (2003b)
Equation (A.7) includes a factor 101.5 so that with force measured in kN, and
√
specimen dimensions measured in mm, KI values are in MPa m.
@Seismicisolation
@Seismicisolation
Figure A.7. Small edge crack in a sheet under uniaxial tension. Reprinted from Linear Elastic
85312-703-5, 2000.
For a centre crack, length 2a, with point forces F per unit thickness on
the crack faces (Figure A.11)
F
KI = √ . (A.9)
πa
KI usually increases with increasing crack length. This is one of the few cases
in which KI decreases with increasing crack length.
A.3.1.2 Three Dimensional Solutions

A circular crack in an infinite body is often called a penny shaped crack.
Under uniaxial tension, σ , perpendicular to the crack

a
KI = 2σ , (A.10)
π
where a is the crack radius. Hence Y in Equation (A.6) is 2/π . With central
point loads F on the crack faces (cf. Figure A.11)
F
KI = . (A.11)
(π a)3/2
For an elliptical crack in an infinite body under uniaxial tension the maximum
stress intensity factor is at the ends of the minor axis and is given by
@Seismicisolation
@Seismicisolation
Metal Fatigue 177
Figure A.8. Small edge crack at a blunt notch in a sheet under uniaxial tension. Reprinted
from Linear Elastic Fracture Mechanics for Engineers. Theory and Applications. LP Pook,
WIT Press, ISBN 1-85312-703-5, 2000.
√
σ πa
KI = , (A.12)
E(k)
where a is the semi minor axis of the elliptical crack and E(k) is the complete
elliptic integral of the second kind. This is given by
π/2
E(k) = 1 − k 2 sin2 d, (A.13)
0
where

a2
k= 1− (A.14)
c2
and c is the semi major axis of the ellipse. An approximation for E(k) is
a 1.65
E(k) = 1 + 1.464 (0 ≤ a/c ≤ 1). (A.15)
c
For c = a Equation (A.12) reduces to Equation (A.10) and for c a to
Equation (A.4), this is also the solution for a tunnel crack, width 2a, in an
infinite body.
Many of the cracks observed in service are surface cracks (Figure A.12).
They are often called part through cracks. The aspect ratio of a surface crack
is the ratio of crack surface length to crack depth. Other definitions of aspect
ratio are sometimes used (see Sections 7.6.1 and 8.3.1). Surface cracks with
@Seismicisolation
@Seismicisolation
Figure A.9. Small edge crack at a sharp notch in a sheet under uniaxial tension. Reprinted
from Linear Elastic Fracture Mechanics for Engineers. Theory and Applications. LP Pook,
WIT Press, ISBN 1-85312-703-5, 2000.
Figure A.10. Three point single edge notch bend specimen (Frost et al. 1974).
an aspect ratio of more than two are usually approximated as semi elliptical
cracks (see Section A.5.1). For such a semi elliptical surface crack in a semi
infinite body under uniaxial tension the maximum stress intensity factor is at
the deepest point of the crack, and its value is dominated by the crack depth
rather than the crack surface length. For a semi infinite body approximate
values of the stress intensity factor at the deepest point are given by
@Seismicisolation
@Seismicisolation
Metal Fatigue 179
Figure A.11. Centre crack in an infinite sheet with point forces on crack surfaces (Frost et al.
1974).
Figure A.12. Semi elliptical surface crack in a semi infinite body under uniaxial tension (Frost
et al. 1974).
√
1.12σ π a
KI = . (A.16)
E(k)
For c a Equation (A.16) reduces to Equation (A.5), for c = 2.89a to
Equation (A.4), and for a semi circular surface crack (a = c) it becomes
√
KI = 0.713σ π a . (A.17)
@Seismicisolation
@Seismicisolation
A.3.1.3 Effect of Residual Stresses
The effect of introducing a crack into a body containing large scale residual
stresses is to relieve the residual stresses on the crack plane. Provided that the
crack is not too large the corresponding stress intensity factor is the same as
if stresses, equal in magnitude but opposite in sign, were applied to the crack
surfaces. For example Equation (A.5), for a small edge crack (Figure A.7)
becomes
√
KI = 1.12P π a , (A.18)
where P is the pressure on the crack surfaces. The presence of residual
stresses of unknown magnitude can be a serious limitation in the practical
application of fracture mechanics.
A.3.2 VALIDITY OF S TRESS I NTENSITY FACTORS
The application of stress intensity factors to practical engineering problems

involving cracks has been a spectacular success over the past 40 years or so.
Nevertheless there are some limitations on their validity that arise from the
linearly elastic stress analyses on which they are based. Accumulated exper-
ience has shown how these limitations can be managed, and what steps need
to be taken to mitigate their effects.
A stress intensity factor provides a reasonable description of the crack tip
stress field in a K-dominated region at the crack tip, radius r ≈ a/10, where
a is the crack length, Figure A.13 (Pook 2000a). An apparent objection to
the use of the stress intensity factor approach is the violation, in the imme-
diate vicinity of the crack tip, of the initial linearly elastic assumptions, in
that strains and displacements are not small. However, as the assumptions are
violated only in a small core region, radius r, the general character of the
K-dominated region is, to a reasonable approximation, unaffected. Similarly,
by this small scale argument, small scale nonlinear effects due to crack tip
yielding, microstructural irregularities, internal stresses, irregularities in the
crack surface, the actual fracture process, etc., may be regarded as within the
core region. If a crack is too short then it may not be possible to use this small
scale argument (see Section 7.4.5).
Elastic stress fields at the tips of sharp notches are, with due attention to
detail, similar to those for cracks. By the small scale argument, stress intensity
factors for cracks can be used to describe the elastic stress field at the tips of
sharp notches. The elastic stress at the tip of a sharp V-notch can be described
by stress intensity factors, provided that the included angle does not exceed
@Seismicisolation
@Seismicisolation
Metal Fatigue 181
Figure A.13. K-dominated and core regions at a crack tip. Reprinted from Linear Elastic
85312-703-5, 2000.
30◦ . Negative values of KI are possible for sharp notches and also for open
cracks, that is, cracks where the surfaces are separated by a small amount.
When yielding is not small scale, stress intensity factors do not provide
a reasonable description of the crack tip stress field, and other less versatile
fracture mechanics parameters become appropriate. Failure to check whether
large scale yielding might be occurring is the commonest error in the practical
application of stress intensity factors. One early approach was to ensure that
the nominal net section stress did not exceed 80 per cent of the yield stress,
σY , and this is still a useful check. In stress intensity factor based metallic
material test standards, minimum acceptable test specimen dimensions are
specified in order to avoid large scale yielding. Actual minimum values de-
pend on values of the stress intensity factors and the material’s yield stress,
for example Anon. (2003b, 2005g).
Use of a two dimensional stress intensity factor solution for plates and
sheets implicitly assumes that the crack front is straight through the thickness
and also perpendicular to the surfaces. In practice, crack fronts are usually
curved. For example Figure A.14 shows the fracture surface of a 19 mm thick
aluminium alloy fracture toughness test specimen, the fatigue precrack front
is clearly curved. Standards place limits on the permissible amount of fatigue
crack front curvature, so that a two dimensional stress intensity factor solution
can be used, for example Anon. (2003b, 2005g).
@Seismicisolation
@Seismicisolation
Figure A.14. Fracture surface of 19 mm thick aluminium alloy fracture toughness test speci-
men (Pook 1968). Reproduced under the terms of the Click-Use Licence.
Extension of the essentially two dimensional concept of stress intensity

factor to three dimensions makes two implicit assumptions. The first is that
the line defining a crack front is smooth so that stress intensity factors do not
change abruptly along the crack front. This is generally, but not always, true
for cracks observed in practice when these are viewed on a macroscopic scale,
and does not usually cause difficulties. The second assumption is that a crack
front is continuous, that is, it is a closed curve. This is sometimes true, for
example for an internal elliptical crack. It is not true at a corner point where
a crack front intersects a free surface (Figure A.15). The nature of the crack
tip singularity changes in the vicinity of a corner point. This can usually be
neglected on the basis of the small scale argument, but sometimes has to be
taken into account (see Section A.4).
A.3.3 E FFECTS OF S MALL S CALE Y IELDING
Small scale yielding in the high stress region at a crack tip has two main prac-
tical consequences. First, it leads to a practical definition of plane strain in the
presence of a crack which differs from that usual in the theory of elasticity.
Secondly, the relaxation of stresses within the crack tip plastic zone means
that, to maintain equilibrium, stresses outside the plastic zone increase, and
the effective crack length is increased.
@Seismicisolation
@Seismicisolation
Metal Fatigue 183
Figure A.15. Semi elliptical surface crack under uniaxial tension showing corner points (Frost
et al. 1974).
A.3.3.1 Definition of Plane Strain

From a theory of elasticity viewpoint an uncracked plate loaded in uniaxial
tension is in a state of plane stress, that is the stress in the thickness direction
is zero. This is still so in the bulk of a plate when a Mode I crack is introduced,
but highly stressed material adjacent to the crack tip is constrained by the less
highly stressed surrounding material, and stresses are induced in the thickness
direction in the interior of the plate in the vicinity of the crack tip.
This situation is often referred to as plane strain in fracture mechanics.
It must not be confused with the conventional theory of elasticity definition,
which is used in the theoretical analysis of crack tip surface displacements
(see Sections A.2.1 and A.3). Limited plastic flow due to yielding of the ma-
terial adjacent to the crack tip does not affect the situation in the interior
of a thick plate (Figure A.12). In this context a plate is said to be thick if
the thickness is at least 2.5(KI /σY )2 where KI is the Mode I stress intensity
factor and σY is the yield stress. This practical definition of plane strain arose
from consideration of the results of tests to determine the fracture toughness,
Kc . It was observed that, in general, Kc decreased as the thickness increased,
but that for most metallic materials it reached a constant minimum value if
the specimen thickness was at least 2.5(Kc /σY )2 (see Section 7.4.2.2). This
minimum value can be regarded as a material constant, and is known as the
@Seismicisolation
@Seismicisolation
Figure A.16. Plane strain, plate thickness ≥ 2.5(KI /σY )2 (Frost et al. 1974).
plane strain fracture toughness, KIc . In the special case of this symbol, the
subscript I denotes both a Mode I crack and plane strain. The minimum speci-
men thickness requirement of 2.5(Kc /σY )2 is included in plane strain fracture
toughness test standards, for example Anon. (2005g). When a crack front is
curved, as in a semi elliptical surface crack (Figure A.12) there is a high de-
gree of constraint along the crack front, except at the corner points, and such
cracks can usually be regarded as being in plane strain.
When the plate thickness is very much less than 2.5(K/σY )2 , then the
crack tip plastic zone size becomes comparable with the thickness, and yield-
ing can take place on 45◦ planes. This relaxes the through thickness stresses,
so that the whole plate is in a state of plane stress (Figure A.17). The sym-
bol Kc is sometimes reserved for the plane stress fracture toughness which
is then regarded as a material constant. At intermediate thicknesses the stress
state is uncertain and Kc is a function of thickness.
A.3.3.2 Effective Crack Length
The increase in effective crack length over the physical crack length due to
yielding at the crack tip is shown schematically in Figure A.18. A first estim-
ate of the plastic zone size may be obtained by substituting von Mises’ cri-
terion of yielding (see Section 4.5.2.1) into the elastic crack tip stress fields.
For Mode I, plane stress this leads to
@Seismicisolation
@Seismicisolation
Metal Fatigue 185
Figure A.17. Plane stress, plate thickness 2.5(KI /σY )2 (Frost et al. 1974).
2
1 KI
rY = (plane stress), (A.19)
2π σY
where rY is the plastic zone size measured in the crack direction, KI is the
Mode I stress intensity factor, and σY is the yield stress. The actual size and
shape of a crack tip plastic zone depends on the flow properties of the metal,
but its dimensions are always proportional to (KI /σY )2 . Typically, a plastic
zone size is about twice that given by Equation (A.19), so rY is interpreted as
the plastic zone radius. The effective crack length becomes a + rY , as indic-
ated in Figure A.18, and the corresponding stress intensity factor is calculated
iteratively. Under plane strain conditions the plastic zone radius is about one
third of that given by Equation (A.19), and

1 KI 2
rY = (plane strain). (A.20)
6π σY
The plastic zone corrections given by Equations (A.19) and (A.20) are often
very small and hence unnecessary. At one time they were quite popular, but
are now rarely used. Plastic zone corrections do not appear to be specified in
any standards.
@Seismicisolation
@Seismicisolation
Figure A.18. Physical and effective crack length, rY is the plastic zone radius.
A.3.3.3 Slant Crack Propagation in Thin Sheets

The transition from square (Mode I) to slant crack propagation sometimes
observed in thin sheets under both static and fatigue loading, as shown schem-
atically in Figure A.19, is an exception to the observation that fatigue cracks
in metals tend to propagate in Mode I (see Sections 7.4.2.2, 8.1.1 and A.2.1).
Slant crack propagation is sometimes stated to be mixed Mode I and Mode
III, but this is true only for the sheet centre line. Away from the centre line,
it is mixed Mode I, Mode II and Mode III. This has been confirmed by finite
element analysis (Pook 1993). It is sometimes called shear crack propaga-
tion, on the grounds that it takes place on planes of maximum shear stress in
an uncracked sheet, but this is a misnomer. In the calculation of stress intens-
ity factors it is usual to treat slant crack propagation, and crack propagation in
the transition region, as if they were Mode I crack propagation, and to use a
two dimensional stress intensity factor solution. This is difficult to justify by
the small scale argument (see Section A.3.2), but it does not cause difficulties
in practice (see Section A.5.2).
A.4 Corner Point Singularities

The analyses on which the concept of stress intensity factor is based are es-
sentially two dimensional in nature, and the crack front is a point (see Sec-
tion A.3.2). When analysis is extended to three dimensions, the crack front
@Seismicisolation
@Seismicisolation
Metal Fatigue 187
Figure A.19. Transition from square to slant crack propagation in thin sheets. The arrow shows
the direction of fatigue crack propagation (Pook 1983a). Reproduced under the terms of the
Click-Use Licence.
becomes a line. Derivations then include the implicit, and usually unstated,
assumption that a crack front is continuous. This is not the case at a corner
point, where a crack front intersects a free surface. The crack front shown by
the dashed line in Figure A.15 intersects the surface at two corner points. As
is well known, the nature of the crack tip singularity changes in the vicinity
of a corner point. For corner point singularities, the polar coordinates (r, θ)
in Figure A.1 are replaced by spherical coordinates (r, θ, φ) with origin at the
corner point. The angle φ is measured from the crack front.
The stress intensity measure, Kλ , is used to characterise corner point
singularities, where λ is an exponent defining the corner point singularity.
Stresses are proportional to Kλ /r λ and displacements to Kλ r 1−λ , where r is
measured from the corner point. For a crack surface intersection angle, γ of
90◦ , defined as in Figure A.20, there are two modes of stress intensity meas-
ure. These are the symmetric mode, KλS , where crack tip surface displace-
ments are Mode I (Figure A.2), and the antisymmetric mode, KλA , which is a
combination of Modes II and III displacements. In other words, the presence
of one of these modes of crack tip surface displacement always induces the
other. For the special case of λ = 0.5, stress intensity factors are recovered.
KλS becomes KI , and KλA a combination of Modes II and III stress intensity
factors, KII and KIII . For the symmetric mode, and Poisson’s ratio, ν = 0.3,
@Seismicisolation
@Seismicisolation
Figure A.20. Definition of crack surface intersection angle, γ . Reprinted from Crack Paths.
LP Pook, WIT Press, ISBN 1-85312-927-5, 2002.
Figure A.21. Definition of the crack front intersection angle, γ . Reprinted from Pook (1994a),
Copyright 1994, with permission from Elsevier.
the theoretical value of λ is 0.452 for a crack front intersection angle of 90◦ ,
Figure A.21 (Pook 2002a), whereas for the antisymmetric mode λ is 0.598.
For λ = 0.5, Mode I crack surface displacements are given, for plane
stress, by Equation (A.3), and the crack opens up into a parabola (Figure A.22
centre). The radius at the tip of the loaded crack, r, is given by
4KI2
r= , (A.21)
π E2
where KI is the Mode I stress intensity factor and E is Young’s modulus.
When λ = 1/2 stresses and displacements cannot, in general, be calculated
in detail because of lack of information. However, when λ < 0.5 r = 0
(Figure A.22 top) and, from Equation (A.21), stress intensity factors tend to
zero as a corner point is approached. Conversely, when λ > 0.5 r = ∞ and
stress intensity factors tend to infinity (Figure A.22 bottom).
There is only limited information available on the size of the corner region
(boundary layer) in which the crack tip stress field is dominated by the stress
intensity measure, although it must be associated with some characteristic
dimension, such as sheet thickness.
As with stress intensity factors, an apparent objection to the use of the
stress intensity measure approach is the violation, in the vicinity of the crack
@Seismicisolation
@Seismicisolation
Metal Fatigue 189
Figure A.22. Crack profiles for loaded Mode I crack. Crack tip radius r = 0 for λ < 0.5. For
λ = 0.5 r is finite and for λ > 0.5 r = ∞. Reprinted from Pook (1994a), Copyright 1994,
with permission from Elsevier.
tip, of the initial assumption on which linearly elastic analyses are based (see
Section A.3.2). However, as the assumptions are violated only in a small core
region, the general character of the corner point singularity dominated region
in the vicinity of the crack tip is unaffected, as is shown for stress intensity
factors in Figure A.13. Similarly, small scale nonlinear effects may be re-
garded as within the core region inside a corner point singularity dominated
region. In turn the corner point singularity dominates only within a limited
region, so in some circumstances a corner point singularity dominated region
may lie within a K-dominated region, as shown schematically for a surface
plane in Figure A.23 (Pook 2002a).
For practical engineering purposes the use of stress intensity measures
is usually unnecessary. They do not appear in standards which make use of
stress intensity factors, for example Anon. (2005f). However, they are two
situations in which corner point singularities have an important influence,
and these are discussed in the next two sections.
A.4.1 C RACK F RONT I NTERSECTION A NGLE
The coefficient defining a stress intensity measure, λ, is a function of Pois-

son’s ratio, ν, and the crack front intersection angle, β (Figure A.21). At a
@Seismicisolation
@Seismicisolation
Figure A.23. K-dominated, corner point singularity dominated and core regions at a surface
plane. Reprinted from Crack Paths. LP Pook, WIT Press, ISBN 1-85312-927-5, 2002.
critical crack front intersection angle, βc , λ = 0.5 and stress intensity factors
then have finite values in the corner point region. For β < βc , λ < 0.5, and
for β > βc , λ > 0.5.
For the symmetric mode βc is given approximately by (Pook 1994a)

−1 (ν − 2)
βc = tan , (A.22)
ν
where ν is Poisson’s ratio. For ν = 0.3, βc = 100.4◦ .
It has been argued, from energy and other considerations, that the crack
front intersection angle must be βc . Intersection angles of about this value
may be observed for Mode I cracks, and in consequence crack fronts in plates
of constant thickness are often curved, for example Figure A.14.
For the antisymmetric mode βc is given approximately by

−1 (1 − ν)
βc = tan . (A.23)
ν
For ν = 0.3, βc = 67.0◦ .
When the crack surface intersection angle γ = 90◦ , a crack at a corner
point is always in a combination of Modes I, II and III crack tip surface dis-
placements. For any given value of γ there are two possible values of βc . For
γ = 45◦ and ν = 0.3 these are 108◦ and 60◦ . There is a corresponding value
@Seismicisolation
@Seismicisolation
Metal Fatigue 191
of the ratio KI : KII : KIII for each value of βc . KI , KII and KIII are the Modes
I, II and III stress intensity factors.
A.4.2 D ERIVATION OF S TRESS I NTENSITY FACTORS
Numerical schemes for the calculation of stress intensity factors for cracks in
three dimensional bodies, explicitly make the assumption that stress intensity
factors provide a description of the crack tip stress field. This is true only
for a continuous crack front, for example an internal crack, and for a crack
front which intersects the surface at the critical crack front intersection angle,
βc . In consequence, such schemes cannot adequately reproduce the behaviour
of stress intensity factors in the vicinity of a corner point. In practice values
obtained are finite, and usually of the same order as elsewhere along the crack
front. At a corner point, values are, in effect, extrapolations which depend on
details of the numerical scheme used (Pook 1994a, 2000c).
At a corner point Mode II and Mode III displacements cannot exist in isol-
ation. The presence of one of these modes always induces the other (see Sec-
tion A.4). Numerical calculations usually show induced values of the Mode
II stress intensity factor, KII , or the Mode III stress intensity factor, KIII , as
would be expected, in the vicinity of a corner point. Two dimensional numer-
ical schemes are widely used in the determination of stress intensity factors
for quasi two dimensional specimens of constant thickness. Only Mode I
and Mode II stress intensity factors are possible in two dimensions, so such
schemes cannot reveal induced KIII values at corner points.
Many three dimensional finite element calculations are carried out for a
Poisson’s ratio of about 0.3, and crack front and crack surface intersection
angles of 90◦ (Figures A.20 and A.21). Hence theoretically KI should tend to
zero, and KII and KIII should tend to infinity, as corner points are approached.
As the corner point is approached, the ratio KIII /KII tends to a finite limiting
value, which is a function of the crack front inclination angle, β, and √ Pois-
◦
son’s ratio, ν. For β = 90 and ν = 0.3 it is 0.5 and for ν = 0.5 it is 0.5.
From an engineering viewpoint the finite stress intensity factor values ac-
tually obtained at corner points do not matter provided both that results are
reasonably consistent, and also that difficulties do not arise in practical ap-
plications. In practice, the situation for Mode I is indeed satisfactory, and
numerous standards make use of Mode I stress intensity factors without any
mention of corner point singularities, for example Anon. (2005f). The situ-
ation appears to be reasonably satisfactory for Mode II, but it is not satisfact-
ory for Mode III where there are inconsistencies in reported KIII values.
@Seismicisolation
@Seismicisolation
Figure A.24. Through thickness variation of KI for 20 mm square models under Mode I
loading (Pook 2000c).
A.4.2.1 Stress Intensity Factors for Square Models
Figures A.24–A.28 show some typical results for through thickness distribu-
tions of stress intensity factors for Mode I, Mode II and Mode III loadings
(Pook 2000c). These were obtained from finite element analysis of a 20 mm
square model, with a crack extending from the middle of one side of the
square to its centre. Loadings and boundary conditions were chosen to give
large K-dominated regions, and Poisson’s ratio was taken as 0.3. For Mode I
and Mode II loadings, the results are normalised by stress intensity factors ob-
tained from two dimensional calculations, whereas for the Mode III loading
they are normalised by centre line values.
The results for the Mode I loading of a 4 mm thick plate and a 40 mm long
bar (Figure A.24) show the well known increase in the Mode I stress intens-
ity factor, KI , at the centre line compared with the corresponding two dimen-
sional solution. The decrease in KI towards the surface is also well known,
and suggests the existence of a corner point singularity dominated region.
The presence of a corner point singularity dominated region was confirmed
by estimating values of λ from crack surface displacements at the model sur-
face. These estimates gave λ = 0.452 for both the plate and the bar, which
agrees with the theoretical value (see Section A.4). The displacements indic-
@Seismicisolation
@Seismicisolation
Metal Fatigue 193
Figure A.25. Through thickness variation of KII for 20 mm square models under Mode II
ated that the size of the corner point singularity dominated region (boundary
layer) was about 0.2 mm for the plate, and about 1 mm for the bar.
Through thickness distributions of KII for Mode II loading are shown in
Figure A.25. At the centre line, KII for the 40 mm long bar is lower than for
the corresponding two dimensional solution. It increases towards the surface,
again suggesting the existence of a corner point singularity dominated region.
This was confirmed by the Mode II crack surface displacements at the surface,
which gave λ = 0.560 compared with the theoretical value of 0.598. They
also indicated a corner point singularity dominated region size of about 3 mm.
The results for the 4 mm thick plate show a different trend. KII is nearly
constant through the thickness, except for a sharp decrease at the surface,
the corner point singularity dominated region size is about 3 mm, and λ =
0.535. Figure A.26 shows the distribution of induced values of KIII . These
are zero at the centre line (by symmetry) for both the plate and the bar, and
increase towards the surface, except that for the plate there is a sharp decrease
at the surface. The Mode III crack surface displacements at the surface did not
define corner point singularity dominated regions, and it was not possible to
@Seismicisolation
@Seismicisolation
Figure A.26. Through thickness variation of KIII for 20 mm square models under Mode II
Figure A.27. Through thickness variation of KIII for 20 mm square bar under Mode III loading
(Pook 2000c).
derive values of λ. Also, it was not possible to derive a limiting value for the
ratio KIII /KII .
The through thickness distribution of KIII for Mode III loading of the
40 mm long bar is shown in Figure A.27. KIII is constant over the central
part of the bar, but there is a marked decrease as the surface is approached,
@Seismicisolation
@Seismicisolation
Metal Fatigue 195
Figure A.28. Through thickness variation of KII for 20 mm square bar under Mode III loading
(Pook 2000c).
rather than the theoretical increase towards infinity. The Mode III crack sur-
face displacements at the surface indicated a corner point singularity domin-
ated region size of about 0.2 mm, but it was not possible to derive a value
for λ. Induced values of KII are zero at the centre line, decrease slightly, and
then increase towards the surface (Figure A.28). The Mode II crack surface
displacements at the surface indicate a corner point singularity region size of
about 0.3 mm, and λ = 0.570.
The inconsistencies in the results for Modes II and III loadings (Fig-
ures A.25–A.28) in the vicinity of corner points are typical of results ob-
tained from finite element analyses. Scatter increases in the vicinity of a
corner point, and what must be regarded as nominal values of KII and KIII are
strongly dependent on details of numerical calculation methods. The incon-
sistencies arise partly from the use of linearly elastic finite element analyses.
In principle, in a linearly elastic analysis KIII at a surface must be zero be-
cause shear stresses perpendicular to a free surface must be zero. This effect
shows up clearly in Figure A.27, but not in Figure A.26.
A.5 Stress Intensity Factors for Irregular Cracks
In three dimensions there are numerous possible crack configurations (Pook

1986). In general, cracks three dimensional cracks observed in service cannot
be approximated by two dimensional stress intensity factor solutions.
Naturally occurring cracks, and crack like flaws, are often irregular in
shape, for example the casting defects in Figures 3.17 and 7.2. The three
@Seismicisolation
@Seismicisolation
Figure A.29. Crack types.
main types of cracks encountered in practice are shown schematically in Fig-

ure A.29. These are surface cracks, internal cracks, and through the thickness
cracks. Methods of estimating stress intensity factors for irregular cracks have
been of interest for many years (Paris and Sih 1965, Anon. 1980b, Chang
1982, Pook 1982a, Murakami 2002). Normally, stress intensity factors vary
along a crack front, and it is the largest value of the Mode I stress intensity
factor, KI , that is usually of interest. Adjacent cracks, as in Figure 7.1, can
increase stress intensity factors through interaction effects. These are difficult
to assess when cracks are not coplanar.
In the analysis of laboratory and service failures, the size and shape of
cracks can usually be ascertained in detail by the examination of fracture
surfaces. However, the results of non destructive testing will usually supply
only a limited amount of information. For example, only the surface length
and maximum depth of a surface crack might be available (see Appendix C).
Initially, an intuitive approach was used by various authors to estimate
stress intensity factors for irregular cracks. This approach was based on gen-
eral knowledge of stress intensity factors for similarly shaped regular cracks
(Paris and Sih 1965). Fifteen years later standardised estimation procedures
for various situations started to appear (Anon. 1980b), and are now in wide-
spread use (Anon. 2005f). Standardised procedures should be used whenever
possible. Some of the ideas used in estimating stress intensity factors for ir-
regular cracks are described below. The ideas sometimes have a sound the-
oretical basis, but the main justification for their use is that they have been
found to work well in practice, and also that they do not lead to unconservat-
ive results.
A.5.1 U SE OF S EMI E LLIPSES AND E LLIPSES
A flat irregular surface crack under Mode I loading is nearly always mod-
elled by a semi ellipse of the same surface length and depth. The Mode I
stress intensity factor, KI , is greatest at the deepest point of the semi ellipse
(see Section A.3.1.2). This intuitive approach is satisfactory for cracks that
@Seismicisolation
@Seismicisolation
Metal Fatigue 197
Figure A.30. Modelling of an irregular surface crack as a semi ellipse.
Figure A.31. Modelling of a more irregular surface crack as a semi ellipse.
Figure A.32. Modelling of a more irregular surface crack as a semi ellipse inscribed in a
containment rectangle.
are close to this shape, as shown schematically in Figure A.29 left, and in
Figure A.30. The approach does not work well for some more irregular sur-
face cracks such as shown in Figure A.31. What is sometimes done is to first
construct a containment rectangle around the crack, and then inscribe a semi
ellipse in the rectangle (Figure A.32). This results in a longer semi ellipse
with a concomitantly higher value of KI at the deepest point. A similar idea
is sometimes used for internal cracks in bodies of rectangular cross section,
as shown in Figure A.33. The containment rectangle sides are parallel to the
body surfaces. A containment rectangle may be used as it stands for a through
the thickness crack (Figure A.29 right).
A.5.2 P ROJECTION ONTO A P LANE
Cracks are not necessarily flat, and are not necessarily oriented so that they
are in Mode I. One approach is to project them onto a plane so that they be-
come equivalent Mode I cracks. The plane chosen is a plane of maximum
principal tensile stress in the uncracked body. After projection, a crack can
then be modelled as in the previous section. As an example, the method works
quite well for the specimen shown in Figure A.20 when this is loaded in three
@Seismicisolation
@Seismicisolation
Figure A.33. Modelling of an internal crack as an ellipse inscribed in a containment rectangle.
Figure A.34. Quasi two dimensional mixed Modes I and II crack with a small Mode I branch
crack (Pook 1989a). Reproduced under the terms of the Click-Use Licence.
point bending (Pook and Crawford 1990). Another example is treating slant
crack propagation in thin sheets (Figure A.19) as if it were Mode I (see Sec-
tion A.3.3.3). In effect, the slant crack and the transition region are projected
onto the plane indicated by dashed lines in the figure.
One theoretical justification is that under mixed mode loading a small
Mode I branch crack may form at the initial crack tip, as shown in Figure A.34
(see Section 8.2.1). Projection of the initial mixed mode crack onto an appro-
priate plane can provide a method of estimating KI for such a branch crack
(Pook 1989c).
@Seismicisolation
@Seismicisolation
Metal Fatigue 199
Figure A.35. Crack front smoothing.
A.5.3 C RACK F RONT S MOOTHING
A crack has some analogies with a crystal dislocation (Pook 2002a). In partic-
ular, the elastic stress fields associated with a crack front and with a disloca-
tion are both singularities. The associated energy means that a dislocation has
a line tension, which controls its shape under an applied stress field (Cottrell
1964). Similarly, a crack front may be regarded as having a line tension which
controls its shape, but with the important difference that the motion of a crack
front is irreversible; that is a crack can propagate, but in general cannot con-
tract. The line tension concept explains why, on a macroscopic scale, a fa-
tigue crack front is smooth and any initial sharp corners rapidly disappear
as the crack propagates. Overall, an initially irregular Mode I crack rapidly
becomes convex, as shown by the dashed line in Figure A.35: at a re-entrant
region (x on the figure) the Mode I stress intensity factor, KI , is much higher
than elsewhere on the crack front, leading to rapid fatigue crack propagation
towards a convex shape. Under fatigue loadings, stress intensity factors for
initially irregular cracks may be approximated by first enclosing them by a
convex outline, as in the figure, and then using the methods in the previous
section.
A.5.4 U SE OF C RACK A REA
For irregular cracks that are small compared with other dimensions it is pos-
sible to use the crack area, A, as a characteristic crack dimension (Chang
1982, Murakami 2002). A is calculated after projection onto a plane, fol-
lowed by crack front smoothing (see Sections A.5.2 and A.5.3). For very
slender cracks the crack length is truncated to 10 times the width before cal-
culating A.
@Seismicisolation
@Seismicisolation
Figure A.36. Interaction between two semi circular surface cracks. (a) No interaction. (b)
Imaginary third crack inserted.
For an internal crack (Figure A.29 centre) the maximum Mode I stress
intensity factor along the crack front, KI , is given approximately by

√
KI ∼= 0.5σ π A , (A.24)
where σ is the stress perpendicular to the crack. The equation applies to
cracks whose length is up to about 5 times the width.
For a surface crack

√
KI ∼
= 0.65σ π A. (A.25)
For a very shallow surface crack the crack surface length is truncated at 10
times the crack depth before calculating A, and for a deep surface crack the
crack depth is truncated at 2.5 times the crack surface length.
For a very shallow surface crack Equation (A.25) becomes
√
KI ∼
= 1.16σ π a , (A.26)
where a is crack depth, and it is close to Equation (A.5), which is the equiva-
lent two dimensional solution.
A.5.5 I NTERACTION BETWEEN C RACKS

When two cracks are close to each other the interaction between them in-
creases their stress intensity factors compared with those for isolated cracks.
Unfortunately, this interaction effect cannot be expressed by a simple equa-
tion, partly because of the numerous possible configurations. Various approx-
imations have been proposed for a wide range of configurations but these tend
@Seismicisolation
@Seismicisolation
Metal Fatigue 201
to be inconsistent, partly because different authors introduce different degrees

of conservatism.
The usual approach is first to define the crack separation between two
cracks below which interaction occurs, and then in some way to define an
equivalent single crack for which stress intensity factors are calculated. For
example, the following rules are suggested for two adjacent semi circular
surface cracks of different sizes (Murakami 2002). If there is enough space
between the cracks to insert an additional crack of the same size as the smal-
ler crack (Figure A.36(a)) then the interaction effect is negligibly small, and
A in Equation (A.25) is taken as A1 . If the space between the two cracks is
too small to insert a crack of the same size as the smaller crack then under fa-
tigue loading the cracks coalesce rapidly. An imaginary semi circular crack is
inserted between the two cracks and the areas of all three cracks are summed.
That is insert in Equation (A.25) A = A1 + A2 + A3 (Figure A.36(b)).
@Seismicisolation
@Seismicisolation
B
Random Load Theory and RMS
Notation
A separate notation is included because many of the symbols listed are used
only in this appendix.
a, b constants in two parameter Weibull distribution
f frequency
G(f ) power spectral density
H wave height
H1/3 significant wave height
I irregularity factor
m0 , m2 , m4 moments of spectral density function
N number of cycles, return period
P (H1/3) exceedance of H1/3
P (S) exceedance of S
P (S/σ ) exceedance of S/s
p(S) probability density of S
p(S/σ ) probability density of S/σ
R(τ ) autocorrelation function
S random process
s instantaneous value of S
S̄/σ expected value of S/σ
Sc /σ clipping ratio
Sm mean value of S
So value of S below which peaks are omitted
T total time, wave period
t time
γ Euler’s constant = 0.5772 . . .
@Seismicisolation
@Seismicisolation
204 Appendix B: Random Load Theory and RMS
ε spectral bandwidth
σ standard deviation (random process theory), root mean square
(fatigue)
σp root mean square of peaks
σp,c root mean square of peaks after clipping of high peaks
σp,o root mean square of peaks after omission of low peaks
σp,t root mean square of peaks after truncation of high peaks
σr root mean square of ranges
σ2 variance
τ time interval
φ root mean square (random process theory)
B.1 Introduction
In this appendix the application of random process theory (Papoulis 1965,

Bendat and Piersol 2000) to fatigue loading is discussed. From an engineer-
ing viewpoint, it might appear that some of the points made are unimportant
and pedantic. However, lack of attention to detail can result in difficulty in
interpreting fatigue test data. In reporting random loading fatigue data it is
important that the precise conventions used in calculations be clearly stated.
No one set of conventions is of universal applicability. Some equations and
figures which appear in the main text are repeated in order to make the ap-
pendix self contained.
B.2 Basic Definitions
B.2.1 R ANDOM P ROCESS T HEORY
Figure B.1(a) shows a random process in which load is plotted against time.
This may be described by the function S(t), where S is a random process and
t is time. In metal fatigue S will be a quantity such as stress or load.
Assume that S(t) is statistically stationary and ergodic. Stationary means
that statistical parameters characterising the process are independent of time.
Ergodic means, broadly, that different samples of the same process yield the
same values for statistical parameters. Only stationary random processes can
be ergodic, and in practice most are. Considering the time interval 0 to T the
mean value of S, Sm is given by

1 T
Sm = lim(T → ∞) S(t) dt (B.1)
T 0
@Seismicisolation
@Seismicisolation
Metal Fatigue 205
Figure B.1. Broad band random process, irregularity factor 0.410, spectral bandwidth 0.912.
(a) Time history. (b) Spectral density function (Pook 1987). Reproduced under the terms of
and the mean square value φ 2 by

1 T 2
φ = lim(T → ∞)
2
S (t) dt. (B.2)
T 0
@Seismicisolation
@Seismicisolation
Hence the root mean square (RMS) value, φ, is given by

1 T 2
φ = lim(T → ∞) S (t) dt . (B.3)
T 0
The positive square root is understood in Equation (B.3) and subsequent equa-
tions.
The RMS can equally well be calculated for periodic processes such as a
sine wave. The use of RMS first became popular in electrical engineering be-
cause it can be used directly in calculations involving power. For convenience
it is sometimes used in metal fatigue (Pook 1987a).
Carrying out calculations from the mean rather than from zero gives the
variance, σ 2 , where

1 T
σ 2 = lim(T → ∞) {S(t) − Sm }2 dt (B.4)
T 0
and the standard deviation, σ , is given by

1 T
σ = lim(T → ∞) {S(t) − Sm }2 dt . (B.5)
T 0
The quantities given by Equations (B.1)–(B.5) are related through the expres-
sion
φ 2 = σ 2 + Sm2 . (B.6)
Hence, for zero mean the RMS and standard deviation are numerically equal.
Instantaneous values of S(t) may be characterized by probability distribu-
tion functions. The exceedance, P (S), is the probability that a value exceeds
S. The cumulative probability, 1 − P (S), is the proportion of values up to
S. The probability density, p(S), is the derivative of the cumulative probab-
ility. For convenience, S is often normalised by σ . The instantaneous values
of many ‘naturally occurring’ random processes are statistically stationary, at
least in the short term, and approximate to the Gaussian distribution (or Nor-
mal distribution), which theoretically extends from −∞ to +∞. The probab-
ility density of a Gaussian distribution (Figure B.2(a)) for a process with zero
mean is given by
2
S 1 −S
p = √ exp (B.7)
σ 2π 2σ 2
and the exceedance (Figure B.2(b)) by
∞
S 2 −S S
P =√ exp 2
d . (B.8)
σ 2π S/σ 2σ σ
@Seismicisolation
@Seismicisolation
Metal Fatigue 207
Figure B.2. Gaussian distribution. (a) Probability density. (b) Exceedance (Frost et al. 1974).
This integral does not have an explicit solution. The values shown in Fig-
ure B.2(b) are for the positive half of a Gaussian distribution and are there-
fore twice those given by Equation (B.8). P (S/s) is the area under the curve
of p(S/σ ) between (S/σ ) and infinity, as indicated by the shaded area in
Figure B.2(a).
B.2.2 FATIGUE L OADING
Conventions used in the metal fatigue literature sometimes differ from those
used in random process theory. The random processes encountered in metal
fatigue are usually symmetrical, in a statistical sense, about the mean value,
Sm , and all calculations are then carried out using values of S measured from
Sm . Mathematically, this is equivalent to treating only cases where Sm is zero.
It follows that there is no numerical difference between root mean square
(RMS) and standard deviation, and in metal fatigue the term standard devi-
@Seismicisolation
@Seismicisolation
ation is not normally used in the characterisation of random processes. (Al-

ternatively it might be said that RMS is used where standard deviation is
meant.) This convention is used in what follows unless the context indicates
otherwise. In particular, references to zero are understood to include the mean
value of a process with non zero mean. Retaining the symbol σ for what was
called standard deviation and is now called RMS Equation (B.5) becomes

T
1
σ = lim(T → ∞) S 2 (t) dt . (B.9)
T 0
The apparent lack of rigour is justified because RMS, as given by Equa-
tion (B.9), has no physical significance in metal fatigue (see Section 4.3.3).
B.3 Some Sinusoidal Processes
B.3.1 NARROW BAND R ANDOM L OADING
In general a narrow band random process (Figure B.3) results when a random
input is applied to a sharply tuned resonant system (Papoulis 1965, Pook
1983b, 1984, Bendat and Piersol 2000). Individual sinusoidal cycles appear
whose frequency corresponds to the centre frequency of the resonant system.
They have a slowly varying random amplitude. The probability density func-
tion for the occurrence of a positive peak of amplitude S (Figure B.4(a)) is
given by the Rayleigh distribution
2
S S −S
p = exp . (B.10)
σ σ 2σ 2
As the process is statistically symmetrical, corresponding negative peaks also
appear. The exceedance (Figure B.4(b)) is given by
2
S −S
P = exp . (B.11)
σ σ2
Equations (B.10) and (B.11) become exact only as the bandwidth tends to
zero (see Section 4.2.2). Used in its general sense Rayleigh distribution does
not imply the existence of a corresponding narrow band random process, and
parameters in Equations (B.10) and (B.11) may differ. A narrow band random
process is Gaussian, so instantaneous values do follow the Gaussian distribu-
tion (Equations (B.7) and (B.8)).
Conventionally, in discussion of the Rayleigh and related distributions,
only positive peaks are described and shown in diagrams such as Figures
A2.4, it being understood that the negative peaks, with due attention to sign,
@Seismicisolation
@Seismicisolation
Metal Fatigue 209
Figure B.3. Narrow band random process, frequency ≈100 Hz, irregularity factor ≈0.99
(Pook 1987a). Reproduced under the terms of the Click-Use Licence.
Figure B.4. Rayleigh distribution. (a) Probability density. (b) Exceedance (Frost et al. 1974).
are also included. Negative peaks are sometimes called troughs. Theoretically
the Rayleigh distribution extends to infinity, but in practice peaks do not ex-
ceed a cut off value of S/σ , known as the clipping ratio. Clipping implies that
@Seismicisolation
@Seismicisolation
higher peaks are reduced to the level given by the clipping ratio; truncation
that they are omitted altogether. The clipping ratio does not usually exceed
four or five.
As fatigue damage depends on the peak values of cycles, and is largely
independent of waveform (see Section 3.2), the root mean square (RMS)
value of √ peaks, σp , is sometimes used. For narrow band random loading:
σp = 2 σ . The RMS of the ranges between positive and negative √ peaks,
σr , is also in use for narrow band random loading, σr = 2σp = 2 2 σ .
B.3.2 T WO PARAMETER W EIBULL D ISTRIBUTION
The two parameter form of the Weibull distribution has a variety of engin-
eering applications; some of its general properties are discussed by Lipson
and Sheth (1975). For metal fatigue purposes it is convenient to write its ex-
ceedance in the form (Pook 1984)

a
S −b S
P = exp , (B.12)
σ a σ
where a and b are adjustable constants (parameters) used to fit the equation
as needed.
A functional relationship between b and a can be obtained by assum-
ing that Equation (B.12) gives the distribution of peaks of a sinusoidal pro-
cess which is symmetrical about zero. There is no closed form relationship
between b and a. Values of b for a in the range 0.5 to 3 are tabulated in Pook
(1984). The expression
b = (1 − 0.076(a 2 − 3a + 2) (B.13)
provides a satisfactory fit for a in the range 0.71 to 2.36. Putting a = 2,
b = 1 and a = 1, b = 1 gives as special cases the Rayleigh distribution
(Equation (B.11)) and the Laplace distribution (or Exponential distribution)

S −S S
P = exp =p . (B.14)
σ σ σ
Exceedances for a range of a values are shown in Figure B.5. A logarithmic
scale is used for exceedances in order to emphasise detail at low values. As
a result the curve for the Rayleigh distribution has a different appearance
from Figure B.4(b) where a linear scale is used. The peaks of the C/12/20
load history, shown in Figure 4.9, can be fitted approximately by the two
parameter Weibull equation with a = 1.2715 (Pook 1987a).
Differentiating Equation (B.12) gives the probability density of the two
parameter Weibull distribution as
@Seismicisolation
@Seismicisolation
Metal Fatigue 211
Figure B.5. Exceedances for the two parameter Weibull distribution (Pook 1987a). Repro-
duced under the terms of the Click-Use Licence.
a−1
a
S b S −b S
p = exp . (B.15)
σ a σ a σ
If only the distribution of peaks is specified by a probability distribution, then
in a computer generated process both the order in which peaks are applied
and the waveform connecting them need to be specified. Usually, a negative
peak is made arithmetically equal to the preceding positive peak. The value
of σ depends on the waveform used to connect the peaks, but both σp and σr
are independent of waveform. When √ peaks are connected
√ by sine waves to
give a sinusoidal process, σp = 2 σ and σr = 2 2 σ . In general, computer
generated processes are do not follow the Gaussian distribution. However, for
the special case of a sinusoidal process with a Rayleigh distribution of peaks,
@Seismicisolation
@Seismicisolation
and with constant frequency, the process is Gaussian, irrespective of the order
in which peaks are applied.
B.3.3 I NCOMPLETE D ISTRIBUTIONS AND P ROCESSES
Distributions and processes encountered in metal fatigue practice are always

incomplete in some way. In consequence root mean square (RMS) values
differ from those of theoretical forms which extend to infinity. Provided that
a difference in RMS values is less than about 1 per cent it can usually be
neglected. Examples illustrating some of the issues involved are given below.
B.3.3.1 Truncated and Clipped Distributions
The term clipping ratio is used to cover both truncation and clipping. This
is because, in a physical system involving a narrow band random process, it
is nonlinearities rather than clipping or truncation that limit the peaks which
appear, and it may not be possible to maintain a clear distinction between
truncation and clipping (see Section B.3.1). However, in a process generated
by first generating positive, and corresponding negative peaks, which follow
some distribution, then joining the positive peaks and adjacent negative peaks
with an appropriate waveform, a distinction has to be made to avoid ambigu-
ity. In any process used for fatigue testing the maximum load applied has to
be limited to meet physical limitations.
If the distribution of peaks is known in terms of the root mean square
(RMS) value of the process, then the RMS of the peaks, σp , is given by

∞ 2
S S S
σp = p d . (B.16)
0 σ σ σ
Hence, if the process is truncated at a clipping ratio, Sc /σ , the RMS of the
truncated distribution of peaks, σp,t , is given by

Sc /σ S 2 p S d S
σp,t = σ σ
Sc σ , (B.17)
0 1 − P σ
where the term {1 − P (Sc /σ )} corrects for the reduction in the total number
of peaks, and σ is the RMS of the complete process.
If the process is clipped, then the RMS of the clipped distribution of peaks,
σp,c , is given by
@Seismicisolation
@Seismicisolation
Metal Fatigue 213
σp,c = (B.18)

2 Sc −σ 2
S S Sc S S S
P c c
+ 1−P p d ,
σ σ σ 0 σ σ σ
where the first term under the square root sign represents the peaks that that
have been reduced to Sc /σ . Clipping has less effect on RMS than trunca-
tion. Equations (B.17) and (B.18) may be used to calculate the change in the
RMS of peaks due to truncation and clipping of different versions of the two
parameter Weibull distribution (Equation (B.12)). Results show that for the
Rayleigh distribution (a = 2 in Equation (B.12)) the effects are negligible
when the clipping ratio exceeds about 3.5, and for the Laplace distribution
(a = 1) when it exceeds about 8. Further terms appear in Equations (B.17)
and (B.18) if positive and negative peaks are not truncated or clipped sym-
metrically.
As an example of what can happen, consider the construction of a si-
nusoidal process, whose peaks follow the two parameter Weibull distribu-
tion, with a taken as 0.5. Assume that the complete process will be trun-
cated to give a desired clipping ratio of 5. From Equation (B.17), taking
S/σ = 5, σp,t = 0.7704σ , and the clipping ratio for the truncated process
is 5/0.7704 = 6.490. For the clipping ratio of the truncated process to be 5,
the clipping ratio applied to the complete process would have to be 3.243.
For a process to remain sinusoidal, clipping has to be carried out correctly,
as shown in Figure B.6. Form (a) is an original unclipped half cycle. Reducing
instantaneous values of the process to the clipping ratio results in form (b),
which is not sinusoidal. For the process to be sinusoidal the half cycle has to
be reshaped, as in form (c). Truncation or clipping of a process that follows
the Gaussian distribution renders it non Gaussian. However, it can reasonably
be regarded as Gaussian if the percentage change in RMS is negligibly small.
B.3.3.2 Omission of Low Loads

Peaks below an omission level, σo , are sometimes omitted to reduce fatigue
testing times, on the grounds that they cause negligible fatigue damage. The
root mean square (RMS) value of the remaining peaks, σp,o, is given by

∞ S 2p S d S
σp,o = σ
σ σ . (B.19)
So /σ P Sσo
In practice, large numbers of cycles are omitted, so there is always a signific-
ant effect on RMS. Omission is always combined with truncation or clipping.
@Seismicisolation
@Seismicisolation
Figure B.6. Clipping a half cycle. (a) Original half cycle. (b) Clipped. (c) Clipped and re-
shaped (Pook 1987a). Reproduced under the terms of the Click-Use Licence.
B.3.3.3 One Sided Narrow Band Random Loading
Tests are sometimes carried out using a modified narrow band random load-
ing from which negative peaks have been removed, to give a one sided pro-
cess (Sherratt and Edwards 1974). Three ways of doing this are shown, for
a constant amplitude sinusoidal process, in Figure B.7. Part (a) of the figure
shows the original process. In Figure B.7(b) the negative half cycles have
been reduced to zero height, whereas in Figure B.7(c) they have been re-
moved altogether. In Figure B.7(d) the negative peaks have been removed
altogether and the positive peaks joined to zero by sine waves. Usually, root
mean square (RMS) values are calculated for the original complete process.
Parameters can be calculated for a one sided process, but different results are
sometimes obtained for√the three methods. For example, mean values (Sm )
are σ/π , 2σ/π and σ/ 2 respectively, where σ is the RMS of the original
complete process The RMS of ranges, σr , is the same for all three methods
of removal, and √is equal to the RMS of peaks, σp , for the original complete
process, that is 2 σ , and would appear to be a good choice. The original
complete process is Gaussian, but instantaneous values of a one sided pro-
cess do not follow the Gaussian distribution (Equations (B.7) and (B.8)).
B.3.3.4 Finite Random Processes
Any practical random sinusoidal process must be of finite length and contain
a finite number of cycles, N. For a pseudo random process, N is the return
period after which it repeats exactly. One consequence is that the maximum
peak size, and hence the clipping ratio, are restricted (see Sections B.3.1 and
B.3.3.1). An intuitive approach is to set the exceedance, P (S/σ ), equal to
1/N and then take the corresponding value of S/σ from Equation (B.11)
@Seismicisolation
@Seismicisolation
Metal Fatigue 215
Figure B.7. One sided constant amplitude sinusoidal processes. (a) Original cycle. (b) Neg-
ative half cycles reduced to zero height. (c) Negative half cycles removed. (d) Negative half
cycles removed and positive half cycles reshaped (Pook 1987a). Reproduced under the terms
as the clipping ratio. However, in narrow band random loading, large cycles
occur in groups and the expected maximum value of S/σ , S̄/σ , which will
@Seismicisolation
@Seismicisolation
Table B.1. COLOS 7 level load history.

Level number Number of cycles in level RMS of level
7 1,000 4.07σ
6 4,000 3.46σ
5 40,000 2.90σ
4 180,000 2.27σ
3 575,000 1.68σ
2 1,250,000 1.10σ
1 2,950,000 0.426σ
be the expected clipping ratio, is somewhat less. It is given approximately by

(Pook 1978)

S̄ √ γ 1
≈ ln N + , (B.20)
σ 2 ln N
where γ is Euler’s constant = 0.5772. . . . For example, for N = 105 ,
P (S/σ ) = 10−5 and from Equation (B.11) S/σ = 4.80, whereas Equa-
tion (B.20) gives S̄/σ = 3.48.
B.3.4 N ON S TATIONARY NARROW BAND R ANDOM L OADING
In service, random loadings are usually statistically non stationary so that root
mean square (RMS) values, and perhaps other parameters, are a slowly vary-
ing function of time. In the short term they can usually be regarded as stat-
istically stationary. For a succession of narrow band random loadings whose
RMSs follow the positive half of a Gaussian distribution (Equation (B.7)) the
peaks sum to the Laplace distribution (Equation (B.14)) (Pook 1983b).
Corresponding load histories for fatigue testing also need to be non sta-
tionary. A procedure was developed (Pook 1984) which made it possible to
approximate a wide range of probability distributions as the sum of several
Rayleigh distributions and hence produce load histories which consist of a
sequence of narrow band random loadings. In one example a 7-level approx-
imation of the Laplace distribution was used as the basis of an agreed standard
load history known as the COmmon LOad Sequence (COLOS) (Anon. 1985).
The numbers of cycles and load levels are listed in Table B.1 in terms of the
of the overall RMS, σ .
The water surface elevations of ocean waves are an example of a process
which often approximates to a non stationary narrow band random process
(Pook and Dover 1989). In oceanography the primary parameter used in the
characterisation of sea state is the wave height, H , which is measured peak
to trough. Over a period of time short enough (conventionally 20 min) for
@Seismicisolation
@Seismicisolation
Metal Fatigue 217
Table B.2. Scatter diagram for M V Famita for full year.

H1/3 m Zero crossing wave period, s
4.5 5.5 6.5 7.5 8.5 9.5 10.5 11.5 12.5 13.5
0.3 14 40 34 8 0 0 0 0 0 0
0.91 64 159 135 40 4 0 0 0 0 0
1.52 18 103 164 78 24 2 0 0 0 0
2.13 6 53 126 95 33 7 1 0 0 0
2.74 0 19 103 72 41 8 2 0 0 0
3.35 0 9 46 71 31 3 1 0 0 0
3.96 0 1 23 63 38 6 1 0 0 0
4.47 0 0 6 20 31 10 2 1 0 0
5.18 0 0 5 13 15 12 1 0 0 0
5.79 0 0 1 9 4 6 3 0 0 0
6.40 0 0 0 2 2 4 2 0 0 2
7.01 0 0 0 0 1 3 7 0 0 0
7.62 0 0 0 1 1 0 4 0 2 0
8.23 0 0 0 1 2 0 0 0 0 0
8.84 0 0 0 0 0 2 2 0 0 0
9.45 0 0 0 0 0 0 0 0 0 2
Parts per 1924.
a sea state to be regarded as statistically stationary, the usual measure of its

severity is the significant wave height, H1/3 , which is the average height of
the highest one third portion of the waves, and is approximately equal to 4σ
(Sarpkaya and Isaacson 1981). One of the ways in which wave height data
can be usefully presented is by means of a scatter diagram that gives the
relative occurrences of sea states within specified small intervals of H1/3 and
wave period, T , which is the reciprocal of the wave passing frequency (Pook
1987b). Table B.2 is an example of a scatter diagram obtained by the M V
Famita (Holmes and Tickell 1975). The M V Famita is one of a number of
ships that have been stationed in the North Sea to collect oceanographic data.
Long term records show that variations in sea state have the appearance of
a random process (Anon. 1985), but examination of data from four different
sources showed that the distribution of H1/3 is not Gaussian (Pook 1987b).
A detailed examination of some data for five years (Pook and Dover 1989)
showed that the distribution of H1/3 was more accurately represented by the
Gumbel distribution (Gumbel 1958). In its simplest form the exceedance,
P (S), of the Gumbel distribution is given by
P (S) = 1 − exp{− exp(−S)}. (B.21)
@Seismicisolation
@Seismicisolation
For S > 4, P (S) ≈ exp(−S), and for S < 2, P (S) ≈ 1. Taking S as

the significant wave height, H1/3 , the sea state data were well fitted by the
expression

1.9 − H1/3
P (H1/3 ) = 1 − exp − exp , (B.22)
1.06
where P (H1/3) is the exceedance of H1/3.
B.4 Broad Band Random Loading
In a broad band random loading (Figure B.1) individual cycles cannot be dis-
tinguished (see Section 4.3.3). When such a process is encountered in metal
fatigue it is usually characterised by σ , that is the root mean square (RMS)
value of the whole process. A measure of bandwidth is also required; a com-
mon one in metal fatigue is the irregularity factor, I , which is the ratio of
mean crossings to peaks (see Section 4.3.3). It lies in the range 0 to 1. The
irregularity factor has the advantages that is easily understood, and is not
restricted to processes whose instantaneous values follow the Gaussian dis-
tribution.
B.4.1 S PECTRAL D ENSITY F UNCTION
In random process theory a process which is statistically stationary and er-

godic, and whose instantaneous values follow the Gaussian distribution, is
usually regarded as adequately described statistically if its root mean square
(RMS) value and spectral density function (SDF) are known. Mathematic-
ally, the SDF is obtained by first calculating the autocorrelation function,
R(τ ). This describes the relationship between the values of the random pro-
cess S(τ ) at times t and t + τ , and is given by (Papoulis 1965, Bendat and
Piersol 2000)

1 T
R(τ ) = lim(T → ∞) S(t)S(t + τ ) dt. (B.23)
T 0
The SDF, G(f ), where f is frequency, is the Fourier transform of R(τ ), and
is given by
∞
G(f ) = 2 R(τ ) exp(−iπf τ ) dτ
−∞
∞
=4 R(τ ) cos(2πf τ ) dτ. (B.24)
−∞
@Seismicisolation
@Seismicisolation
Metal Fatigue 219
Determining the SDF in this way is called transforming from the time domain
to the frequency domain. The SDF is sometimes plotted on a logarithmic scale
and sometimes on a linear scale. Figure B.1(b) shows the SDF for the broad
band random loading shown in Figure B.1(a). In practice it is calculated using
an algorithm known as the Fast Fourier Transform (FFT) (Bendat and Piersol
2000). Physically, the SDF gives the frequency content of the random process,
and in the narrow band random process case is sharply peaked at the centre
frequency (resonant frequency). An alternative method of determining the
SDF is to pass the process of interest through a bandpass filter of very narrow
bandwidth and plot the amplitude of the resulting signal against frequency.
The area under the PSD is equal to the mean square value of the process, as
given by Equation (B.2).
In electrical engineering the SDF is usually known as the power spectral
density (PSD) because it provides a measure of the electrical power, which
may be ascribed to the various frequency components.
Some useful results depend only on the spectral bandwidth, ε, which is a
measure of the RMS width of the SDF (Pook 1978, Bendat and Piersol 2000).
It lies in the range 0 to 1 and is given by

m22
ε = 1− , (B.25)
m0 m4
where m0 , m2 and m4 are the zeroth, second and fourth moments of the SDF
about the origin. It is related to the irregularity factor by
ε2 = 1 − I 2. (B.26)
As ε → 0 the distribution of peaks tends to the Rayleigh distribution (Equa-
tions (B.10) and (B.11)) and as ε → 1 to the Gaussian distribution (Equa-
tions (B.7) and (B.8)).
There is no generally accepted definition of what is meant by narrow band,
partly because of the physical difficulties of measuring bandwidth as ε → 0.
In metal fatigue a random process is usually called narrow band if the peak
distribution approximates to the Rayleigh distribution; this is generally so,
provided that I ≥ 0.99, corresponding to ε ≤ 0.14 (see Section 4.2.2).
Difficulties in determining the irregularity factor for narrow band random
process are illustrated by the example shown in Figure B.3. Inevitably, only
a finite length process can be examined, so decisions are needed on how to
deal with the beginning and end of the process. Also, the mean value of the
process has to be determined. The horizontal line in Figure B.3 is intended to
be the mean value. In the figure there are 44 upward going zero crossings of
this line and 45 positive peaks, giving an irregularity factor of 44/45 ≈ 0.98.
@Seismicisolation
@Seismicisolation
Figure B.8. Spectral density functions for a 0.76 m diameter horizontal member immersed
10.8 m, significant wave height 4.75 m. (a) Water surface elevation; (b) bending stress; (c) axial
stress (Pook 1989b). Reproduced under the terms of the Click-Use Licence.
@Seismicisolation
@Seismicisolation
Metal Fatigue 221
It could be argued that the mean crossing at the end of the process should
not be counted, giving an irregularity factor of 43/45 ≈ 0.96. However, if the
horizontal line were slightly lower there would be 45 upward going crossings,
and the irregularity factor would be 45/45 = 1. What should be taken as
the correct value is not easily resolved. In principle, the irregularity factor
could be determined by first finding the spectral bandwidth and then using
Equation (B.26) but there are corresponding difficulties in determining the
spectral bandwidth of a narrow band random process.
As an example of the sort of information that can be derived from spectral
density functions, Figure B.8 shows data for a tubular welded tall platform
in the North Sea (Pook 1989b). A 0.76 m diameter horizontal member, im-
mersed 10.8 m, was strain gauged so that bending and axial stresses could
be derived. In practice, although sea states have a dominant wave passing
frequency, they are not particularly narrow band so energy may be avail-
able to excite structural resonances (Pook 1987b). The SDF for the water
surface elevation (Figure B.8(a)) shows this. There is a clearly defined peak
corresponding to the dominant wave passing frequency, but there is signific-
ant energy at other frequencies. To avoid structural resonances, offshore plat-
forms are designed so that resonant frequencies are substantially greater than
the dominant wave passing frequency. This has been successful for the axial
stresses since, as might be expected, the SDF (Figure B.8(b)) is of similar
form, with no structural resonances exited. However, the SDF for the bend-
ing stress (Figure B.8(c)) does show two peaks corresponding to structural
resonances. The point of collecting data of this sort is to permit comparison
of actual structural behaviour with theoretical calculations.
@Seismicisolation
@Seismicisolation
C
Non Destructive Testing
C.1 Introduction
Non destructive testing (NDT) is not a clearly defined concept (Halmshaw

1991). NDT has a wide range of applications in the detection and evaluation
of flaws in materials. Many different methods are used, and a wide range
of commercially available instruments has been developed. These are often
automated under computer control. The key feature of NDT is that it has no
deleterious effect on the item tested.
In the context of metal fatigue the usual meaning of non destructive test-
ing is the detection and sizing of cracks and crack like flaws in compon-
ents, structures and laboratory specimens. This includes monitoring of fatigue
crack propagation in service and in laboratory specimens; the advantages and
disadvantages of various methods are summarised by Richards (1980). The
accuracy of crack sizing that can be achieved varies widely.
Some of the non destructive testing techniques used in metal fatigue work
are described briefly in this appendix, together with the important statistical
concepts of probability of detection and probability of sizing. In order to make
the appendix more self contained some figures in the main text are repeated
here.
C.2 Visual Inspection
Visual inspection is the simplest method of detecting surface cracks, usually

called surface breaking cracks in the non destructive testing literature. This
term is used in this appendix. The utility and importance of visual inspection
are often underestimated. Under good conditions fatigue cracks with a surface
length of 3 mm can be detected by the naked eye, but in general 25 mm is a
@Seismicisolation
@Seismicisolation
224 Appendix C: Non Destructive Testing
Figure C.1. Fatigue cracks in an aircraft engine nacelle.
more realistic detection limit. A low power lens (say ×3) and additional port-
able lighting are useful. For a permanent record photographs may be taken or
replicas of the surface made. If the surface is irregular, as in welds, surface
breaking cracks are difficult to detect by visual methods.
Regulatory authorities often call for periodic visual inspection of struc-
tures for defects, including cracks. For example, Figure C.1 shows unexpec-
ted fatigue cracking found in an aircraft engine nacelle during a routine in-
spection (Pook 2004). Another example is the cracking in a burner from a
domestic central heating boiler shown in Figure 8.13. Routine visual inspec-
tion is tedious, and fatigue cracks are sometimes missed. For example, one
of the concerns at the official inquiry into the catastrophic fatigue failure of
a fairground ride was why fatigue cracks, which should have been detected,
were missed during routine visual inspections (Pook 1998).
When fatigue crack propagation is being monitored visually, crack length
measurement is often aided by markings etched or scribed onto the speci-
men surface, for example the grid shown in Figure C.2 (see Section 8.2.2).
Visual methods have been widely used to collect data during fatigue crack
propagation rate tests, scribed marks were used to collect the data shown in
Figure C.3. The use of guide markings does not meet resolution accuracy re-
quirements in modern fatigue crack propagation rate testing standards such
as Anon. (2003b). A common technique, which does meet the requirements,
is to use a micrometer thread travelling microscope with a magnification of
×20 to ×50.
Visual methods of inspection have the advantage that the equipment
needed is relatively inexpensive, but they are labour intensive when used to
monitor fatigue crack propagation, and are not amenable to automation. The
@Seismicisolation
@Seismicisolation
Metal Fatigue 225
Figure C.2. Fatigue crack path in a Waspaloy sheet under biaxial fatigue load. The grid is
0.1 inch (2.54 mm). National Engineering Laboratory photograph. Reproduced under the
Figure C.3. Fatigue crack propagation curve for a central crack, length 2a, in a 0.76 m wide
× 2.5 mm thick mild steel specimen. Nominal stress 108 ± 31 MPa (Frost et al. 1974).
major disadvantage is that only crack surface lengths can be measured. In a

plate of constant thickness a fatigue crack front of a through the thickness
crack is often curved (Figure C.4). This curvature can affect the calculation
of stress intensity factors (see Section A.3.2). With a surface breaking crack
@Seismicisolation
@Seismicisolation
Figure C.4. Fracture surface of 19 mm thick aluminium alloy fracture toughness test specimen
(Pook 1968). Reproduced under the terms of the Click-Use Licence.
Figure C.5. Uniform alternating current on the surface of a plate containing a surface breaking
crack.
(Figure C.5) it would not be possible to calculate stress intensity factors be-
cause these are largely dependent on crack depth (see Section A.3.1.2).
In practice, the major use of visual inspection is to detect surface breaking
cracks. Any cracks found are then sized using an appropriate technique such
as ultrasonics (see Section C.6) or alternating current potential drop (see
Section C.7.2).
C.3 Magnetic Particle Inspection
Magnetic particle inspection (MPI) is a well established technique for the

detection of surface breaking cracks. The method can be used only on fer-
romagnetic materials which can be strongly magnetised. These include irons
@Seismicisolation
@Seismicisolation
Metal Fatigue 227
Figure C.6. Principle of magnetic particle inspection.
and ferritic steels, but not all steels. Magnetic effects arise through electro-
magnetic fields. These can be represented as lines of magnetic force through
space which form a magnetic flux.
The principal of magnetic particle inspection is shown in Figure C.6. A
magnetic flux is established in the material by placing the poles of a magnet
(usually an electromagnet) in contact with the material. If the magnetic flux
encounters a transverse surface breaking crack the flux becomes distorted.
Some of the magnetic flux passes through the crack, some passes around
the crack tip, and some leakage flux passes around the crack at the surface.
This leakage flux attracts ferromagnetic particles to the crack mouth, and the
resulting visible concentration of particles marks the crack.
The magnetic particles are applied as a suspension in a carrier liquid, such
as light oil or water, at a concentration by volume of about 2 per cent. If water
is used, a wetting agent and a corrosion inhibitor are incorporated. The sus-
pension is normally supplied in an aerosol, and is sometimes called a mag-
netic ink. The particles are usually black iron oxide of around 1–25 µm in
size, and may be dyed to improve visibility. Florescent dyes are sometimes
used, and the particles are then viewed under ultra violet light
Magnetic particle inspection is the most widely used non destructing test-
ing method for detecting surface breaking cracks in welded joints. MPI is
easily carried out using portable equipment, but expertise is needed for satis-
factory results, and it can be a messy procedure. The major disadvantages are
that only the surface length of a crack can be determined, and the accuracy of
crack sizing is low. An advantage is that, with special equipment, magnetic
particle inspection can be used under water.
@Seismicisolation
@Seismicisolation
Figure C.7. Schematic view of dye penetrant in crack after removal of excess penetrant from
the surface.
C.4 Dye Penetrant
The dye penetrant method is used to detect surface breaking cracks. The
method can be applied to any material that has a non absorbent surface. Most
of the cracks found by dye penetrants can be seen visually in good condi-
tions, but dye penetrants make them much easier to detect. The principle of
the method is shown in Figure C.7). After the surface has been cleaned a pen-
etrant, which contains a dye in solution, is applied. The penetrant is chosen
so that it wets the material being inspected, and it is drawn into cracks by ca-
pillary action. Excess penetrant is then removed from the surface, and a thin
layer of a porous developer is applied. Penetrant is drawn out of cracks by the
developer, thus making cracks visible.
The dye and developer colours are chosen to provide good contrast. Pen-
etrant dyes and developers are usually supplied in aerosols. A wide range
of techniques is available, and for good results the technique chosen must
be carefully matched to the intended application (Halmshaw 1991). Unfortu-
nately, much published information on the results of dye penetrant non de-
structive testing is of little value because full details of techniques used are
not included.
The dye penetrant method is widely used for aluminium alloys and other
metallic materials which cannot be magnetised so that magnetic particle in-
spection is impossible (see previous section). The main advantage of dye pen-
etrant is that it is simple to use, and particularly suitable for field work. The
main disadvantage, as with visual inspection and magnetic particle, is that
only the surface length of a crack can be determined (see Sections C.2 and
C.3). If fatigue crack propagation is being monitored, a potential disadvantage
is that dye penetrant remaining in a fatigue crack could affect its subsequent
propagation behaviour.
@Seismicisolation
@Seismicisolation
Metal Fatigue 229
Figure C.8. Fatigue cracking from shrinkage cavity in 30 × 35 mm cast steel bar. National
Engineering Laboratory photograph. Reproduced under the terms of the Click-Use Licence.
Figure C.9. Schematic view of shrinkage cavity in 30 × 35 mm cast steel bar.
C.5 Radiography
The main use of radiography in metal fatigue is the detection of voluminous

defects including internal cavities, porosity and inclusions. An example is
the shrinkage cavity in a cast steel bar shown in Figure C.8. This has an
irregular boundary and can be regarded as a crack like flaw. The general shape
of the cavity was obtained by radiography, and it is shown schematically in
Figure C.9 (Pook et al. 1981).
The principle of radiography in its original form is shown in Figure C.10.
X-rays are emitted from a small source, and travel in straight lines towards
a sheet of photographic film, which acts as a detector. The X-rays are partly
absorbed as they pass through the specimen, and then strike the photographic
film to produce a radiograph. There is less absorption when the X-rays pass
through a cavity, and a two dimensional image of the three dimensional cav-
ity is formed on the film. To get three dimensional information on the cavity,
@Seismicisolation
@Seismicisolation
Figure C.10. Principle of radiography in its original form.
including its location within the specimen, radiographs have to be taken from
more than one direction. This was done to produce the sketches shown in Fig-
ure C.9. As an alternative to using film, X-rays can be captured electronically,
and images displayed in real time on a monitor.
X-rays are a form of electromagnetic radiation, similar to light, but with
very much shorter wavelengths. They are produced when a beam of high en-
ergy electrons strikes a metal anode in a vacuum. Wavelengths of X-rays used
in radiography range from about 10−4 nm to about 10 nm. Long wavelength
X-rays are sometimes called soft X-rays and will penetrate only small dis-
tances. Short wavelength X-rays are sometimes called hard X-rays, and can
penetrate up to about 50 cm thick steel. Gamma rays are sometimes used in
radiography. They are also a form of electromagnetic radiation and are pro-
duced by the decay of a radioactive isotope such as cobalt-60.
Radiography is a very versatile and easily used technique. It is probably
the oldest non destructive testing technique used for the quality control of
welded joints. Radiographs provide a convenient, permanent record. The ma-
jor disadvantage of radiography is that stringent safety precautions have to
be taken to protect operators, and the public at large, from radiation. From a
metal fatigue viewpoint its major disadvantage is that it is difficult to detect
and size tight cracks, that is cracks whose opposite surfaces are either close
together or touching.
@Seismicisolation
@Seismicisolation
Metal Fatigue 231
C.6 Ultrasonics
In metal fatigue the main use of ultrasonics is crack sizing, and a wide range
of techniques is available. The method is based on the propagation of sound
waves through the material at frequencies above the audible range, hence the
term ultrasonics. Sound waves are mechanical vibrations. Hence the velocity
of propagation is different in different materials, and also depends on the
type of wave. Frequencies used are of the order of a MHz, and the resulting
wavelengths are of the order of a mm. The essential feature of the waves used
in ultrasonics is that they propagate through the material in the same way that
ocean waves move across the water surface. This contrasts with the standing
waves observed in the vibrations of a tuning fork. Inadvertent standing waves
can be a problem in the use of ultrasonics.
Two main types of wave are used in ultrasonics. One is compressional
waves, also known as longitudinal waves, where particles vibrate in the dir-
ection of wave propagation. The other is shear waves, also known as trans-
verse waves, where vibration is at right angles to the propagation direction.
Several other types of wave are used for special purposes (Halmshaw 1991).
Ultrasonic waves used to interrogate the specimen under test are generated in
pulses, not continuously.
Ultrasonic wave pulses are generated by applying short electric pulses to
a suitable probe. One type of probe uses a piezoelectric disc, which resonates
at a selected frequency (Figure C.11). A couplant, such as thin layer of oil,
is used to ensure good transmission of ultrasonic waves from the probe into
the specimen under test. Ultrasonic waves emerging from the specimen are
received by a suitably positioned probe, and processed to give information
about defects within the specimen. The same probe is sometimes used for
both transmission and reception.
The arrangement for ultrasonic testing of a cracked specimen, using a
compressional probe, is shown schematically in Figure C.11. Ultrasonic wave
pulses are reflected from the crack, and also from the top and bottom sur-
faces of the specimen. These echoes are displayed on an oscilloscope using
what is known as an A-scan, shown schematically in Figure C.12. The time
base of the oscilloscope is triggered as each ultrasonic pulse is transmitted.
Hence, the positions of the echoes on the time axis provide information on
the crack location. In practice, dispersion effects within the specimen mean
that subsidiary echoes also appear. These subsidiary echoes can complicate
interpretation of the display.
Other methods of display may be used when a probe is scanned across
a specimen surface. A B-scan is obtained from a scan along a line on the
@Seismicisolation
@Seismicisolation
Figure C.11. Arrangement for ultrasonic testing of a cracked specimen using a compressional
probe.
Figure C.12. A-scan showing crack.
surface. The display is arranged so that it shows the sizes and positions of
flaws on a cross section perpendicular to the surface of the specimen. A C-
scan produces a radiograph like display (see previous section). It is obtained
from a series of scans along parallel lines, and the display shows a plan view
of the specimen in which defects appear in their correct positions, but with no
information on their through the thickness locations. A D-scan is similar to a
B-scan, but is obtained from a series of scans along parallel lines. The terms
A-scan, etc., are often used in the ultrasonics literature without explanation.
Various ultrasonic techniques are used to size cracks and crack like flaws.
Two of these, which are used for surface breaking cracks, are shown schem-
atically in Figures C.13 and C.14. In the end on technique a high intensity
compressional probe, located at the opposite surface, is used to find the tip
@Seismicisolation
@Seismicisolation
Metal Fatigue 233
Figure C.13. End on technique for measuring the depth of a surface crack.
of the crack (Figure C.13). The depth of the crack can then be determined
directly from an A-scan. A different approach is used in the time of flight
diffraction (TOFD) technique, shown schematically in Figure A.14. In this
technique there are separate transmitter and receiver compressional probes.
The principle is that when a compressional ultrasonic beam meets a crack
tip some of the energy is diffracted. The diffracted waves spread over a large
angular range, and may be detected by a suitably placed receiver probe. If
the transmitter and receive probes are symmetrically placed about the crack
tip, then a simple calculation gives the position of the crack tip, and hence
the crack depth. To ensure symmetry about the crack tip the two probes may
be linked mechanically, and traversed across the crack. The probes are sym-
metrically positioned when the time of flight is minimised. Shear waves are
sometimes used in the TOFD technique; these are more suitable for deeper
cracks. Both the end on technique and the TOFD technique are compatible
with methods of approximation of stress intensity factors since, in effect, an
oblique crack is projected onto a plane perpendicular to the surface (see Sec-
tion A.5.2).
Simple theory suggests that, for a flaw of a given size, the height of the
echo on an A-scan is inversely proportional to the square of the distance of the
flaw from the probe. A distance amplitude correction curve, usually known
as a DAC curve, is used to correct for this effect. The term DAC level refers
to the heights of echoes, relative to background noise, that are regarded as
significant. (Background noise is known as grass, because of its appearance
on an oscilloscope screen.) Thus, 50 per cent DAC (level) means that only
@Seismicisolation
@Seismicisolation
Figure C.14. Time of flight diffraction technique for measuring the depth of a surface crack.
echo heights that are at least 50 per cent greater than the height of the grass
are considered significant. The choice of DAC level is important when the
probability of detection and probability of sizing are being determined (Visser
2002) (see Section C.8).
Ultrasonics is a very versatile and well established method of crack sizing,
and a wide range of techniques is available. Technique details are readily
adaptable to specific applications. The major disadvantages of ultrasonics are
cost and that it is not suitable for small, thin specimens. It is also difficult to
use on austenitic steels.
C.7 Electromagnetic Fields
In metal fatigue electromagnetic field methods are used both for crack detec-
tion and for crack sizing. They are based on the injection of a uniform al-
ternating current field into the surface of a specimen, such as the plate shown
in Figure C.5. Eddy current and alternating current potential drop (ACPD)
methods are usually regarded as distinct methods of non destructive testing,
but they are actually limiting cases of general electromagnetic field methods .
Due to the skin effect the alternating current density is greatest at the surface,
and decreases exponentially with depth below the surface. The skin depth, δ,
is usually defined as the depth at which the alternating current density is 1/e
(36.8%) of its surface value, and this depth is given by (Lewis et al. 1988)
1
δ=√ , (C.1)
π µσf
@Seismicisolation
@Seismicisolation
Metal Fatigue 235
where µ is magnetic permeability, σ is electrical conductivity, and f is

the frequency of the alternating current. Some typical skin depths shown in
Table C.1. Magnetic permeability is sensitive to precise material composition,
and it is also a function of current density, Hence data in the table should only
be used as a guide to the selection of an appropriate frequency.
Table C.1. Typical skin depths.

Material Frequency
1 kHz 10 kHz 100 kHz 1 MHz
18/8 stainless steel 13.1 mm 4.14 mm 1.31 mm 0.414 mm
Brass 3.90 mm 1.23 mm 0.390 mm 0.123 mm
Aluminium 2.65 mm 0.838 mm 0.265 mm 0.084 mm
Copper 2.00 mm 0.632 mm 0.200 mm 0.063 mm
Mild steel 0.148 mm 0.047 mm 0.015 mm 0.005 mm
For satisfactory results the skin depth must be small compared with the
crack depth. Typically δ is about 0.1 mm so, in general, electromagnetic field
methods cannot be used for cracks less than about 1 mm deep.
The response of a crack being interrogated by an alternating current de-
pends on the value of the dimensionless parameter m, which is given by
(Lewis et al. 1988)
µ0 a
m= , (C.2)
µδ
where µ0 is the magnetic permeability of a vacuum (= 4π ×10−7 H m−1 ). For
non magnetic materials of high electrical conductivity, such as aluminium,
µ ≈ µ0 , m is large because δ/a is small, and eddy current testing is appropri-
ate. For magnetic materials, such as ferritic steels, µ µ0 , m is small, and
ACPD testing is appropriate. Typical values of m are 12 for aluminium and
0.6 for mild steel (Lewis et al. 1988).
C.7.1 E DDY C URRENT
In metal fatigue the main uses of eddy current testing are the detection and
sizing of cracks in non magnetic materials, especially aluminium alloys. Fre-
quencies of the order of one MHz are used in order to ensure a small skin
depth (Table C.1).
The principle of eddy current testing is shown schematically in Fig-
ure C.15. A probe with a current carrying coil is scanned across the specimen
at a small fixed lift off distance. The alternating current in the coil produces
@Seismicisolation
@Seismicisolation
Figure C.15. Principle of eddy current testing of a cracked specimen.
an alternating magnetic flux, which induces eddy currents in the specimen.

These eddy currents are sometimes called Foucault currents. The induced
eddy currents in turn produce an alternating magnetic flux. This opposes the
flux produced by the current carrying coil, and changes the impedance of the
coil. As the probe passes over a crack the eddy currents are distorted and the
presence of a crack is detected, using suitable instrumentation, by changes
in the coil impedance. In some systems the effect of the magnetic flux pro-
duced by the eddy currents is monitored by voltages induced in a second coil,
similar to the current carrying coil. An important practical advantage of eddy
current testing is that physical contact between the probe and the specimen is
not necessary.
Eddy current testing is a very versatile method of crack detection and siz-
ing, and a wide range of techniques is available. It is possible to detect cracks
as small as 10 µm deep. However, for good results the technique and instru-
mentation chosen must be carefully matched to the intended application. Its
major disadvantages are cost and, at times, interpretational difficulties. It is
possible to monitor fatigue crack propagation by using a system in which the
probe is traversed by a motor, and locks onto the crack tip.
When used on magnetic materials, eddy current testing is sometimes
called alternating current field measurement (Lewis et al. 1988). Lower fre-
quencies are used, and methods of data reduction are different.
C.7.2 A LTERNATING C URRENT P OTENTIAL D ROP
In metal fatigue the main uses of alternating current potential drop (ACPD)
techniques are the detection and sizing of cracks in magnetic materials, espe-
@Seismicisolation
@Seismicisolation
Metal Fatigue 237
Figure C.16. Principle of alternating current potential drop testing of a cracked specimen.
cially steel. Frequencies of the order of around 1–10 MHz are used in order
to ensure a small skin depth (Table C.1).
The principle of ACPD testing is shown in Figure C.16. The current is im-
pressed through two contacts some distance apart. It flows along the specimen
skin from one contact to the other, passing down one side of the crack and up
the other. The voltage (potential drop) is measured using a probe with known
contact spacing, , placed across the crack. In automatic systems for monit-
oring fatigue crack propagation, current impression and potential drop meas-
urement measuring wires are spot welded to the specimen (Austin 1999).
Calibration is straightforward. A voltage, V1 , is first measured by placing
the probe near the crack, as shown in Figure C.17 (left). The voltage, V2 ,
across the crack is then measured (Figure C.17 (centre)). Assuming that the
current is constant for the two measurements, then the crack depth, a, is given
by

V2
a= −1 . (C.3)
V1 2
This one dimensional solution has to be modified for a two dimensional crack,
such as that shown in Figure C.5. Modifiers are available for various config-
urations.
Because the technique is self calibrating, the impressed current field does
not have to be completely uniform. This makes the technique suitable for
irregularly shaped specimens such as welded joints. The technique is par-
ticularly suitable for automatic collection of fatigue crack propagation data
during structural fatigue tests. For fatigue crack propagation rate testing on
@Seismicisolation
@Seismicisolation
Figure C.17. Alternating current potential drop calibration.
plate specimens, in which a crack front may be curved (Figure C.4), contact
points can be arranged so as to measure an average crack length through the
thickness.
The main advantages of the ACPD method are its versatility and its ease
of calibration. Voltages are low, so no safety precautions are needed, and the
method can be used under water. Care has to be taken in arranging the im-
pressed current and probe leads so as to avoid spurious interactions, and also
interference in electrically noisy environments. Crack bridging by metallic
particles is not usually a serious problem. Various systems are available com-
mercially. Techniques are still evolving. It is moderately expensive.
A disadvantage is that for an oblique crack the method gives the crack
length, not the depth of the crack tip below the surface (Figure C.17 right).
The ACPD method is therefore not compatible with methods of approxima-
tion of stress intensity factors where an oblique crack is projected onto a plane
(see Section A.5.2). In effect, the projected crack length is overestimated, as
is the concomitant approximated stress intensity factor.
C.8 Probability of Detection
Traditionally, it has been assumed that a particular non destructive testing

technique is capable of detecting all cracks larger than a critical size, and
that no cracks smaller than the critical size will be detected. What is meant
by crack size depends on the application; for a surface breaking crack (Fig-
ure C.5) this is usually either the surface length or the maximum depth. In
practice a critical crack size is not clearly defined, and the probability of de-
tection (POD) increases with the crack size as shown schematically in Fig-
ure C.18, which also shows the ideal situation where a critical crack size is
clearly defined.
@Seismicisolation
@Seismicisolation
Metal Fatigue 239
Figure C.18. Actual and ideal probability of detection (POD) curves.
Figure C.19. Schematic relative operating characteristic (ROC) curve.
Another possible outcome of an inspection is a false call where no crack

is present but one is apparently detected. In a relative operating characteristic
(ROC) curve the probability of detection is plotted against the false call prob-
ability (FCP) as shown schematically in Figure C.19. A good performance,
shown by dashed lines on the figure, is defined by Visser (2002) as a probab-
ility of detection of ay least 80 per cent combined with a false call probability
of at most 20 per cent.
@Seismicisolation
@Seismicisolation
Table C.2. Typical data obtained from a probability of detection trial.

Crack size Whether Crack size Whether Crack size Whether
(mm) detected (mm) detected (mm) detected
2.2 No 8.8 Yes 21.5 Yes
2.5 No 9.5 No 21.7 Yes
3.0 No 10.7 Yes 21.8 Yes
3.4 Yes 11.5 Yes 22.2 No
3.8 No 13.4 Yes 22.4 Yes
4.2 Yes 13.9 No 22.9 Yes

5.1 Yes 16.0 Yes 23.9 Yes
5.8 No 17.6 Yes 24.4 Yes
6.5 Yes 17.9 No 26.0 Yes
7.1 Yes 19.4 Yes 26.2 Yes
7.3 Yes 19.7 Yes 28.2 Yes

8.0 No 19.9 Yes 28.9 Yes
C.8.1 D ETERMINATION OF P ROBABILITY OF D ETECTION
Probability of detection (POD) curves are determined experimentally by blind

trials on a set of specimens containing cracks of known sizes. The objective
is to see what can be achieved by a skilled inspector using a particular non
destructive testing technique, not to test the inspector. It is good practice to
include some uncracked specimens so that data on false calls can be obtained.
Sets of specimens kept for blind trials are sometimes called a library. Inspect-
ors carrying out blind trials are not given any information on the cracks in the
specimens, or on the results of the trials.
There are two particular difficulties in carrying out POD trials. The first is
producing specimens with cracks of the desired size range. Cracks are usually
produced by fatigue loading but fatigue cracks are, in general, difficult to
control. The second is ensuring that crack sizes, especially crack depths, are
accurately known. The most satisfactory way of determining crack depths is
to section specimens, but this destroys them so that they cannot be used for
further trials. What is sometimes done is to size the cracks using an accurate
method, such as time of flight diffraction (see Section C.6) and, from time
to time, section a few specimens in a library as a cross check. The sectioned
specimens are replaced by new specimens containing similar cracks.
Typical data obtained from a POD trial are shown in Table C.2. In order
to obtain point estimates of POD, specimens are grouped according to crack
size, as shown in Table C.3. It is good practice to have about the same number
of groups as there are specimens in each group. The number of successful
@Seismicisolation
@Seismicisolation
Metal Fatigue 241
Table C.3. Point estimates of probability of detection (POD).

Crack size Number in POD
range (mm) group (per cent)
0–5 6 33.3
5–10 8 62.5
10–15 4 75
15–20 6 83.3
20–25 8 87.5
25–30 4 100
Figure C.20. Probability of detection point estimates.
detections in each group divided by the number of specimens in the group is

a point estimate of POD, and is usually expressed as a percentage. The point
estimates are plotted at the upper limit of each size range, and are usually
joined by straight lines, as shown in Figure C.20.
The probabilities of detection shown in Table C.3 are point estimates
based on small samples of 4 to 8 cracks. However, point estimates obtained
from small samples may not accurately reflect probabilities of detection ob-
tained from large samples. Point estimates obtained from small samples are
sometimes analysed statistically to provide lower bound estimates of probab-
ilities of detection at, say, the 95 per cent confidence level. However, the small
sample sizes typically used in practice for cost reasons make rigorous statist-
ical analysis difficult, and lower bound estimates of probabilities of detection
are not in general use (Visser 2002).
@Seismicisolation
@Seismicisolation
Table C.4. Typical data obtained from a probability of sizing trial.

Actual crack Measured crack Actual crack Measured crack
size (mm) size (mm) size (mm) size (mm)
1.5 Not detected 13.1 13.7
1.9 1.9 13.5 15.1
2.7 3.7 16.1 16.2
3.3 5.5 16.5 17.3

3.9 5.0 17.7 19.7
5.2 4.7 18.1 19.9
5.5 7.9 18.3 Not detected
5.9 5.4 18.5 20.2

7.1 7.4 19.7 21.2
7.7 9.6 20.1 19.4
9.4 9.4 21.6 22.0

11.1 10.7 23.3 23.6
Table C.5. Point estimates of probability of sizing (POS) using 90 per cent accuracy criterion
compared with probability of detection (POD).
Size range (mm) Number in group POS (per cent) POD (per cent)
0–5 8 12.5 50
5–10 8 50 75
10–15 5 60 80
15–20 10 60 90
20–25 5 80 100
C.8.2 P ROBABILITY OF S IZING
The probability of sizing (POS) is a refinement of probability of detection in

which a crack is counted as detected only if its size is measured to within
an accuracy criterion. An accuracy criterion of x per cent means that the
measured crack size is within ±(100 − x) per cent of the actual crack size.
Table C.4 shows typical data obtained from a probability of sizing trial, and
Table C.5 point estimates of probability of sizing, using a 90 per cent accuracy
criterion. Point estimates of probability of detection are also shown in the
@Seismicisolation
@Seismicisolation
Metal Fatigue 243
Figure C.21. Comparison of probability of sizing (90 per cent accuracy criterion) with prob-
ability of detection.
table. These point estimates are plotted in Figure C.21; probabilities of sizing
are significantly lower than probabilities of detection.
@Seismicisolation
@Seismicisolation
References
Almen JO, Black PH (1963) Residual Stresses and Fatigue in Metals. McGraw-Hill, New
York.
Anon. (1871) Wöhler’s experiments on the ‘fatigue’ of metals. Engineering (London), 11:
199–200, 221, 244–245, 261, 299–300, 326–327, 349–350, 397, 439–441.
Anon. (1954) Report on Comet accident investigations. Accident Note No. 263. Part 3. Fatigue
tests on the pressure cabin and wings. Royal Aircraft Establishment, Farnborough.
Anon. (1958) Proc. Int. Conf. on Fatigue of Metals, London and New York, 1956. Institution
of Mechanical Engineers, London.
Anon. (Ed) (1965a) Fracture toughness testing and its applications. ASTM STP 381. American
Society for Testing and Materials, Philadelphia, PA.
Anon. (1965b) BS 1515 Welded pressure vessels (advanced design and construction) for use
in the chemical, petroleum and allied industries. Part I: Carbon and ferritic alloy steels.
British Standards Institution, London
Anon. (1974a) Airplane damage tolerance requirements. Military specification MIL-A-8344.
United States Air Force,
Anon. (1974b). Description of UK Offshore Steel Research Project. Objective and current
progress. Reference MAP 015G(1)/20-2. Department of Energy, London.
Anon. (1976) BS 5500: Specification for unfired fusion welded pressure vessels. British Stand-
ards Institution, London.
Anon. (1977) Rules for the design, construction and inspection of offshore structures. Ap-
pendix C. Steel structures. Det Norske Veritas, Hovik, Norway.
Anon. (1979) Flow measurement facilities at NEL on 1 October 1979. NEL Report No. 665.
National Engineering Laboratory, East Kilbride, Glasgow.
Anon. (1980a) BS 5400 Steel, concrete and composite bridges. Part 10: 1980. Code of practice
for fatigue. British Standards Institution, London.
Anon. (1980b) PD 6493: 1980. Guidance on some methods for the derivation of acceptance
levels for defects in fusion welded joints. British Standards Institution, London.
Anon. (1981a) ASTM E647 – 78T. Tentative test method for constant-load-amplitude fatigue
crack growth rates above 10−8 m/cycle. In: Hudak SJ, Bucci RJ (Eds) Fatigue Crack
Growth Measurement and Data Analysis. STP 738. American Society for Testing and
Materials, Philadelphia, PA, pp. 321–339.
@Seismicisolation
@Seismicisolation
246 References
Anon. (1981b) Kielland report ignores vulnerability findings. Offshore Eng., April: 13–16.
Anon. (1985) The common load sequence for fatigue evaluation of offshore structures.
Background and generation. IABG report TF-1892. Industrieanlagen-Betriebsgesellschaft
GmbH, Ottobrunn.
Anon. (1986) Subcritical crack growth. Royal Society, London.
Anon. (1996) BS AU 50-2c: 1996. Tyres and wheels. Wheels and rims. Specification for road
wheels manufactured wholly or partly of cast light alloy for passenger cars. British Stand-
ards Institution, London.
Anon. (1997) BS EN 818-4: 1997 Short link chain for lifting purposes. Safety. Chain sling.
Grade B. British Standards Institution, London.
Anon. (2000a) BS EN 60669-1: 2000, BS 3676-1: 2000. Switches for household and similar
fixed electrical installations. General requirements. British Standards Institution, London.
Anon. (2000b) BS EN 12983-1: 2000 Cookware. Domestic cookware for use on top of a stove,
cooker or hob. British Standards Institution, London.
Anon. (2001a) Guide to the Consumer Protection Act 1987. Department of Trade and Industry,
London.
Anon. (2001b) SAE J 442 Test strip, holder and gage for shot peening. Society of Automotive
Engineers, Warrendale, PA.
Anon. (2002) BS ISO 10771-1: 2002. Hydraulic fluid power. Fatigue pressure testing of metal
pressure containing envelopes. Test method. British Standards Institution, London.
Anon. (2003a) BS ISO 12107: 2002. Metallic materials – fatigue testing – statistical planning
and analysis of data. British Standards Institution, London.
Anon. (2003b) BS ISO 12108: 2002. Metallic materials – fatigue testing – fatigue crack growth
method. British Standards Institution, London.
Anon. (2004) The proposed architecture for NII’s revised safety assessment principles for
nuclear safety and radioactive waste management. A discussion document. Nuclear Safety
Directorate, London.
Anon. (2005a) Safety of nuclear power reactors. Nuclear Issues Briefing Paper 14. Uranium
Information Centre Limited, Melbourne.
Anon. (2005b) BS 0-1 A standard for standards. Development of standards. Specification.
British Standards Institution, London.
Anon. (2005c) BS 2A 241: 2005. General requirements for steel protruding-head bolts of
tensile strength 1250 MPa (180 000 lbf/in2 ) or greater. Specification. British Standards
Institution, London.
Anon. (2005d) The general product safety regulations 2005. Guidance for businesses, con-
sumers and enforcement authorities. Guidance notes. Department of Trade and Industry,
London.
Anon. (2005e) BS EN ISO 9000: 2005 Quality management systems. Fundamentals and
vocabulary. British Standards Institution, London.
Anon. (2005f) BS 7910: 2005: Guide to methods for assessing flaws in metallic structures.
British Standards Institution, London.
Anon. (2005g) BS EN ISO 12737: 2005. Metallic materials – determination of plane-strain
fracture toughness. British Standards Institution, London.
Anzai H (1992) Algorithm of the rainflow method. In: Murakami Y (Ed) The Rainflow Method
in Fatigue. Butterworth-Heinemann, Oxford, pp. 11–20.
Austin JA (1994) The rôle of corrosion fatigue crack growth mechanisms in predicting the
fatigue life of offshore tubular joints. PhD Thesis. University of London, London.
Austin JA (1999) Measurement of fatigue crack growth in large compact tension specimens
using an AC magnetic field method. Fatigue Fract. Engng. Mater, Struct., 22: 1–9.
@Seismicisolation
@Seismicisolation
Metal Fatigue 247
Ayer AJ (1956) The Problem of Knowledge. Penguin Books, Harmondsworth.

Bacon F (1930) Fatigue Stresses with Special Reference to the Breakage of Rolls. The South
Wales Institute of Engineers, Cardiff.
Basquin OH (1910) The exponential law of endurance tests. Proc. Am. Soc. for Test. Mat. 10:
625–630.
Bastenaire F (1963) A study of the scatter of fatigue test results by statistical and physical
methods. In: Barrois W, Ripley EL (Eds) Fatigue of Aircraft Structures. Pergamon Press,
Oxford, pp. 53–85.
Bathias C (2001) An understanding of the gigacycle fatigue of metals. In: Materials Science
for the 21st Century. The Society of Materials Science, Kyoto, Japan, Vol. A, pp. 12–18.
Bažant P, Estenssoro LF (1979) Surface singularity and crack propagation. Int. J. Solids
Struct., 15: 405–426.
Bendat JS, Piersol AG (1971) Random Data: Analysis and Measurement Procedures. Wiley-
Interscience, New York.
Bendat JS, Piersol AG (2000) Random Data: Analysis and Measurement Procedures. Third
Edition. John Wiley and Sons, New York.
Bignall VF, Peters G, Pym C (1977) Catastrophic Failures. Open University Press, Milton
Keynes.
Bishop NWM, Sherratt F (1990) A theoretical solution for the estimation of rainflow ranges
from power spectral density data. Fatigue Fract. Engng. Mater. Struct., 13: 311–326.
Blakey W (1780) The art of making watch springs. In: Wayman ML (Ed) (2000) The Fer-
rous Metallurgy of Early Clocks and Watches. Studies of Post Medieval Steels. Occasional
Paper No. 136. British Museum Press, London, pp. 113–122.
Braithwaite F (1854) On the fatigue and consequent fracture of metals. Institution of Civil
Engineers. Minutes of Proceedings, 13: 463–474.
Broberg KB (1987) On crack paths. Eng. fract. Mech., 28: 663–679.
Broek D (1988) Elementary Engineering Fracture Mechanics. Fourth edition. Kluwer Aca-
demic Publishers, Dordrecht.
Bruder T, Heuler P, Klätschke H, Störzel K. (2004) Analysis and synthesis of standardized
multi-axial load-time histories for structural durability assessment. In: Sonsino CM, Zen-
ner H, Portella PD (Eds) Proceedings of the Seventh International Conference on Bi-
axial/Multiaxial Fatigue and Fracture, Berlin, 28 June–1 July 2004. Deutscher Verband
für Materialforschung und -prüfung E. V, Berlin, pp. 63–77.
Camm FJ (1941) Watches. Adjustment and Repair. Second edition. George Newnes, London.
Carpinteri A, Pook LP (Eds) (2005) Fatigue crack paths. Special issue. Fatigue Fract. Engng.
Mater. Struct., 28: 1–264.
Carpinteri A, Spagnoli A, Vantadori S (2004) Application of a criterion for lifetime prediction
under multiaxial random loading. In: Sonsino CM, Zenner H, Portella PD (Eds) Proceed-
ings of the Seventh International Conference on Biaxial/Multiaxial Fatigue and Fracture,
Berlin, 28 June–1 July 2004. Deutscher Verband für Materialforschung und -prüfung E.
V, Berlin, pp. 369–374.
Cartwright DE, Longuet-Higgins MS (1965) The statistical distribution of the maxima of a
random function. Proc. Roy. Soc. A, 237: 212–232.
Cazaud R (1953) Fatigue of Metals. Chapman & Hall, London.
Chang R (1982) Fracture mechanics of two dimensional noncircular flaws – The equivalent
area approximation. Eng. Fract. Mech., 16: 675–681.
Chaudhury GK, Dover WD (1985) Fatigue analysis of offshore platforms subject to sea wave
loadings. Int. J. Fatigue, 7: 13–19.
@Seismicisolation
@Seismicisolation
248 References
Connolley T, Mchugh PE, Bruzzi (2005) A review of deformation and fatigue of metals at
small size scales. Fatigue Fract. Engng. Mater. Struct., 28: 1119–1152.
Cotterell B (1965) On brittle fracture paths. Int. J. Fract. Mech., 1: 96–103.
Cotterell B (1966) Notes on the paths and stability of cracks. Int. J. Fract. Mech., 2: 526–533.
Cottrell AH (1964) Theory of Crystal Dislocations. Blackie and Son, London.
Crisp HG (1974) Description of UK Offshore Steels Research Project: Objectives and current
progress. MAP 01SG(1)/20-2. London: Department of Energy.
Darlaston BJL, Harrison RP (1977) The concept of leak-before-break and associated safety
arguments for pressure vessels. In: Stanley P (Ed) Fracture Mechanics in Engineering
Practice. Applied Science Publishers, Barking, Essex, pp. 165–172.
Davis EH, Ellison EG (1989) Hydrodynamic pressure effects of viscous fluid flow in a fatigue
crack. Fatigue Fract. Engng. Mater. Struct., 12: 327–342.
Derby B, Hills DA, Ruiz C (1992) Materials for Engineering. A Fundamental Design Ap-
proach. Longman Scientific and Technical, Harlow, Essex.
Dhondt G (2005) Cyclic crack propagation at corners and holes. Fatigue Fract. Engng. Mater.
Struct., 28: 25–30.
Dover WD (1979) Variable amplitude fatigue of welded structures. In: Smith RA (Ed) Frac-
ture Mechanics. Current Status, Future Prospects. Pergamon Press, Oxford, pp. 125–147.
Dowling NE (1993) Mechanical Behavior of Materials. Engineering Methods for Deforma-
tion, Fracture and Fatigue. Prentice Hall, Englewood Cliffs, NJ.
Eastabrook JN (1981) On the validity of the Palmgren-Miner rule in fatigue crack growth.
RAE TR 81029. Procurement Executive, Ministry of Defence, Farnborough.
Edwards PR (1988) Full-scale fatigue testing of aircraft structures. In: Marsh KJ (Ed) Full-
Scale Testing of Components and Structures. Butterworth Scientific, Guildford, pp. 16–43.
Edwards PR, Darts J (1984) Standardized fatigue loading sequences for helicopter rotors
(Helix and Felix). RAE TR 84084. Royal Aircraft Establishment, Farnborough.
Elber W (1970) Fatigue crack closure under cyclic tension. Eng. Fract. Mech., 2: 37–45.
Etube LS (2001) Fatigue and Fracture Mechanics of Offshore Structures. Professional Engin-
eering Publishing, London.
Ewing DK (1986) Implementation of wave loading histories at N.E.L. National Engineering
Laboratory, East Kilbride, Glasgow.
Ewing JA, Humphrey JC (1903) The fracture of metals under rapid alternations of stress. Phil.
Trans. Roy. Soc. London, A200: 241–250.
Fairbairn W (1864) Experiments to determine the effect of impact, vibratory action and long-
continued changes of load on wrought iron girders. Phil. Trans. Roy. Soc., 154: 311.
Ford DG, Graff DG, Payne AO (1963) Some statistical aspects of fatigue life variation. In:
Barrois W, Ripley EL (Eds) Fatigue of Aircraft Structures. Pergamon Press, Oxford,
pp. 179–208.
Forsyth PJE (1961) A two stage process of fatigue crack growth. In: Proc. Crack Propagation
Symposium, Cranfield. The College of Aeronautics, Cranfield, Vol. 1, pp. 76–94.
Forsyth PJE (1969) The Physical Basis of Metal Fatigue. Blackie and Son, London.
Forsyth PJE, Ryder DA, Smale AC, Wilson RN (1959) Some further results obtained from the
microscopic examination of fatigue, tensile and stress corrosion fracture surfaces. RAE
TN MET 312. Royal Aircraft Establishment, Farnborough.
Frost NE (1975) The current state of the art of fatigue: Its development and interaction with
design. J. Soc. Env. Eng., 14-2: 21–24, 27–28.
Frost NE, Pook LP, Denton K (1971) A fracture mechanics analysis of fatigue crack growth
in various materials. Eng. Fract. Mech., 3: 109–126.
@Seismicisolation
@Seismicisolation
Metal Fatigue 249
Frost NE, Marsh KJ, Pook LP (1974) Metal Fatigue. Clarendon Press, Oxford. Reprinted with
minor corrections (1999), Dover Publications, Mineola, NY.
Fuchs HO (1980) Guest editorial. Strategies in design. J. Mech. Design, 102: 1.
Gassner E (1954) A basis for the design of structural parts subjected to statistically alternating
loads in service. Konstruction, 6: 97–104 [in German]. English translation. W & T Avery,
Birmingham.
Gerber W (1874) Bestimmung der zulässigen Spannungen in Eisen-konstructionen. Zeitschrift
des Bayerischen Architeckten und Ingenieur-Vereins, 6: 101–110.
Gere JM, Timoshenko SP (1991) Mechanics of Materials. Third SI Edition. Chapman and
Hall, London.
Gerlach HD (1980) The German pressure vessel code – Philosophy and safety aspects. Int. J.
Pres. Ves. & Piping, 8: 283–302.
Goldstein RV, Salganik RL (1974) Brittle fracture of solids with arbitrary cracks. Int. J. Fract.,
10: 507–523.
Goodman J (1899) Mechanics Applied to Engineering. Longmans-Green, London.
Gough HJ (1926) The Fatigue of Metals. Ernest Benn, London.
Grandt AF (2004) Fundamentals of Structural Integrity. John Wiley & Sons, Hoboken, NJ.
Gumbel EJ (1958) Statistics of Extremes. Columbia University Press, New York.
Gurney TR (1976) Fatigue design rules for welded steel joints. Weld. Res. Bull., 17: 115–124.
Gurney TR (1979) Fatigue of Welded Structures. Second edition. Cambridge University Press,
Cambridge.
Hall N (Ed) (1992) The New Scientist Guide to Chaos. Penguin Books, London.
Halmshaw R (1991) Non-Destructive Testing. Second edition. Edward Arnold, London.
Head AK (1953) The growth of fatigue cracks. Phil. Mag. Seventh series, 44: 925–938.
Head AK (1956) The propagation of fatigue cracks. J. Appl. Mech., 23: 407–410.
Heuler P, Klätschke H (2005) Generation and use of standardised load spectra and load-time
histories. Int. J. Fatigue, 27: 974–990.
Heuler P, Seeger T (1986) A criterion for omission of variable amplitude loading histories. Int.
J. Fatigue, 8: 225–230.
Heywood RB (1962) Designing against Fatigue. Chapman and Hall, London.
Hibberd RD, Dover WD (1977) The analysis of random load fatigue crack propagation. In:
Taplin DMR (Ed) Fracture. University of Waterloo Press, Waterloo, Vol. 2, pp. 1187–
1194.
Hill R (1950) The Mathematical Theory of Plasticity. Clarendon Press, Oxford.
Holmes P, Tickell RG (1975) The long term probability distribution of peak wave induced
loads on structural sections. Supplementary Report MCE/APR/75. Department of Civil
Engineering, The University, Liverpool.
Holmes R (1980a) Fatigue and corrosion fatigue of welded joints under random load condi-
tions. In: Proc. Select Seminar on European Offshore Steels Research, 27–29 November
1978. Welding Institute, Cambridge, pp. 11.1–11.26.
Holmes R (1980b) The fatigue behaviour of welded joints under North Sea environmental and
random loading conditions. OTC Paper No 3700. In: Proc. 12th Ann. Offshore Technology
Conf., Houston, pp. 219–230.
Holmes R, Kerr J (1982) The fatigue strength of welded connections subjected to North Sea
environmental and random loading conditions. In: Proc. BOSS 82. Third Int. Conf. on
Behaviour of Offshore Structures, Vol. 2, Paper S2. Massachusetts Institute of Technology,
Cambridge, MA.
Johnson HH, Paris PC (1968) Sub-critical crack growth. Eng. Fract. Mech., 1: 1–45.
@Seismicisolation
@Seismicisolation
250 References
Jones DP, James LA (1996) Integrable K-solutions for common fatigue crack growth speci-
mens. Int. J. Fract., 81: 89–97.
Jones DRH (2003) Fatigue: practical applications and failure analysis. In: Milne I, Ritchie RO,
Karihaloo B (Eds) Comprehensive Structural Integrity. Elsevier, London, Vol. 4, pp. 489–
512.
Kallaby J, Price JB (1978). Evaluation of fatigue considerations in the design of framed off-
shore structures. J. Pet. Technol., 30: 357–366.
Kam JCP, Dover WD (1989) Mathematical background for applying multiple axes random
stress histories in the fatigue testing of offshore tubular joints. Int. J. Fatigue, 11: 319–
326.
Kirkby WT, Forsyth PJE, Maxwell RJ (1980) Design against fatigue – Current trends. Aero-
nautical J., 84: 1–12.
Kitagawa H, Takahashi S (1976) Applicability of fracture mechanics to very small cracks or
the cracks in the early stage. In: Proc. 2nd Int. Conf. on Mechanical Behavior of Materials.
American Society for Metals, Metals Park, OH, pp. 627–631.
Knauss WG (1970) An observation of crack propagation in anti-plane shear. Int. J. Fract.
Mech., 6: 183–187.
Kobayashi M, Matsui T, Murakami Y (1998) Mechanism of creation of compressive residual
stress by shot peening. Int. J. Fatigue, 20: 359–364.
Kolk K, Kuhn G (2005) A predictor-corrector scheme for the optimization of 3D crack front
shapes. Fatigue Fract. Engng. Mater. Struct., 28: 117–126.
Kreyszig E (1983) Advanced Engineering Mathematics. Fifth Edition. John Wiley & Sons,
New York.
Kueppers M, Sonsino CM (2004) Critical plane approach for the assessment of welded alu-
minium under multiaxial spectrum loading. In: Sonsino CM, Zenner H, Portella PD (Eds)
Proceedings of the Seventh International Conference on Biaxial/Multiaxial Fatigue and
Fracture, Berlin, 28 June–1 July 2004. Deutscher Verband für Materialforschung und -
prüfung E. V, Berlin, pp. 361–368.
Lam YC (1993) Mixed mode fatigue crack growth with a sudden change in loading direction.
Theor. Appl. Fract. Mech., 19: 69–74.
Langer BF (1937) Fatigue failure from stress cycles of varying amplitude. J. Appl. Mech., 59:
A160–A162.
Lawn BR, Wilshaw TR (1975) Fracture of Solids. Cambridge University Press, Cambridge.
Lee HY, Nikbin KM, O’Dowd NP (2005) A generic approach for a linear elastic fracture
mechanics analysis of components containing residual stress. Int. J. Pres. Ves. Piping, 82:
797–806.
Lewis AM, Michael DH, Lugg MC, Collins R (1988) Thin-skin electromagnetic fields around
surface-breaking cracks in metals. J. Appl. Phys., 64: 3777–3784.
Lin XB, Smith RA (1999) Finite element modelling of fatigue crack growth of surface cracks.
Part II: Crack shape change. Eng. Fract. Mech., 63: 523–540.
Lipson C, Sheth NJ (1975) Statistical Design and Analysis of Engineering Experiments.
McGraw-Hill, New York.
Maddox SJ (1991) Fatigue Strength of Welded Structures. Second edition. Abington Publish-
ing, Cambridge.
Mandelbrot BM (1977) Fractals, Form, Chance and Dimension. WH Freeman and Company,
San Francisco.
Mandelbrot BM (1983) The Fractal Geometry of Nature. WH Freeman and Company, New
York.
@Seismicisolation
@Seismicisolation
Metal Fatigue 251
Marsh KJ (Ed) (1988) Full-Scale Testing of Components and Structures. Butterworth Sci-
entific, Guildford.
Matthys RJ (2004) Accurate Clock Pendulums. Oxford University Press, Oxford.
McClintock FA, Irwin GR (1965) Plasticity aspects of fracture mechanics. In: Fracture Tough-
ness Testing and Its Applications. ASTM STP 381. American Society for Testing and Ma-
terials, Philadelphia, PA, pp. 84–113.
Miller AG (1988) Review of limit loads of structures containing defects. Int. J. Pres. Ves.
Piping, 32: 197–327.
Miller KJ (1982) The short crack problem. Fat. Eng. Mat. Struct., 5: 223–232.
Miller KJ, de los Rios ER (Eds) (1992) The Behaviour of Short Fatigue Cracks. Mechanical
Engineering Publications, London.
Miller KJ, McDowell DL (1999a) Overview. In: Miller KJ, McDowell DL (Eds) Mixed-Mode
Crack Behavior. ASTM STP 1359. American Society for Testing and Materials, West
Conshohocken, PA, pp. vii–ix.
Miller KJ, McDowell DL (Eds) (1999b) Mixed-Mode Crack Behavior. ASTM STP 1359.
American Society for Testing and Materials, West Conshohocken, PA.
Milne I, Ritchie RO, Karihaloo B (Eds) (2003) Comprehensive Structural Integrity. Elsevier,
London.
Miner MA (1945) Cumulative damage in fatigue. J. Appl. Mech., 12: A159–A164.
Minoshima K, Suezaki, Komai K (2000) Generic algorithms for high-precision reconstruc-
tions of three-dimensional topographies using stereo fractographs. Fatigue Fract. Engng.
Mater. Struct., 23: 435–443.
Mounsey J (1958) An Introduction to Statistical Calculations. English Universities Press, Lon-
don.
Murakami Y (Ed) (1987) Stress Intensity Factors Handbook. Vols 1 and 2. Pergamon Press,
Oxford.
Murakami Y (Ed) (1992a) The Rainflow Method in Fatigue. Butterworth-Heinemann, Oxford.
Murakami Y (Ed) (1992b) Stress Intensity Factors Handbook. Vol. 3. Pergamon Press, Oxford.
Murakami Y (Ed) (2001) Stress Intensity Factors Handbook. Vols 4 and 5. Elsevier Science,
Oxford.
Murakami Y (2002) Metal Fatigue: Effects of Small Defects and Nonmetallic Inclusions. El-
sevier Science, Oxford.
Murray JAH (Ed) (1901) A New English Dictionary on Historical Principles. Vol. 4. Claren-
don Press, Oxford.
Nabarro FRN (1967) Theory of Crystal Dislocations. Clarendon Press, Oxford.
Nakagaki M, Atluri SN (1979) Fatigue crack closure and delay effects under Mode I spectrum
loading: An efficient elastic-plastic analysis of behaviour. Fatigue Engng. Mater. Struct.,
1: 421–429.
Newhall DH (1999) The Effect of Machining after Full Autofrettage. Harwood Engineering
Company, Walpole, MA.
Nixon F, Frost NE, Marsh KJ (1975) Choosing a factor of safety. In: Whyte RR (Ed) Engineer-
ing Progress through Trouble. Institution of Mechanical Engineers, London, pp. 136–139.
Palmgren A (1924) Die Lebensdauer von Kugellagern (The durability of ball bearings). Verein
Deutscher Ingenieure, 68: 339–341.
Papoulis A (1965) Probability, Random Variables, and Stochastic Processes. McGraw-Hill
Book Company, New York.
Paris PC (1962) The growth of cracks due to variations in load. PhD Thesis. Lehigh University.
@Seismicisolation
@Seismicisolation
252 References
Paris PC, Sih GC (1965) Stress analysis of cracks. In: Fracture Toughness Testing and Its
Applications. ASTM STP 381. American Society for Testing and Materials, Philadelphia,
PA, pp. 30–81.
Paris PC, Bucci RJ, Wessel ET, Clark WG, Mager TR (1972) Extensive study of low fatigue
crack growth rates in A533 and A508 steels. In: Stress Analysis and Growth of Cracks.
ASTM STP 513. American Society for Testing and Materials, Philadelphia, PA, pp. 141–
176.
Paris PC, Lados D, Tada H (2006) Progress in identifying the real Keffective in the threshold
region and beyond. In: Carpinteri A, Pook LP (Eds) Proceedings of the International Con-
ference on Crack Paths (CP 2006), Parma (Italy), 14–16 September 2006. University of
Parma, Parma, Proceedings on CD.
Parton ZV (1992) Fracture Mechanics. From Theory to Practice. Gordon and Breach Science
Publishers, Philadelphia, PA.
Parsons RH (1947) History of the Institution of Mechanical Engineers. Institution of Mechan-
ical Engineers, London.
Peterson RE (1953) Stress Concentration Design Factors. Wiley, New York.
Peterson RE (1974) Stress Concentration Factors. John Wiley & Sons, New York.
Petrone N, Susmel L (2003) Biaxial testing and analysis of bicycle-welded components for
the definition of a safety standard. Fatigue Fract. Engng. Mater. Struct., 26: 491–505.
Pilkey WD (1997) Peterson’s Stress Concentration Factors. Second edition. John Wiley &
Sons, New York.
Pippan R, Stüwe HP, Golos K (1994) A comparison of different methods to determine the
threshold of fatigue crack growth. Int. J. Fatigue, 16: 579–582.
Pook LP (1960) Microscopic examination of fatigue fracture surfaces. Laboratory Test Note
LTN 135. Hawker Siddeley Aviation, Coventry.
Pook LP (1962) Quantitative fractography. Examples showing information derived from air-
craft components after fatigue testing. Laboratory Test Note LTN 212. Hawker Siddeley
Aviation, Coventry.
Pook LP (1968) Brittle fracture of structural materials having a high strength weight ratio.
PhD thesis, University of Strathclyde.
Pook LP (1971) The effect of crack angle on fracture toughness. Eng. Fract. Mech., 3: 205–
218.
Pook LP (1976a) Basic statistics of fatigue crack growth. J. Soc. Env. Eng., 15-4: 3–8.
Pook LP (1976b) Proposed standard load histories for fatigue testing relevant to offshore struc-
ture. NEL Report 624. National Engineering Laboratory, East Kilbride, Glasgow.
Pook LP (1978) An approach to practical load histories for fatigue testing relevant to offshore
structures. J. Soc. Env. Eng., 17–1: 22–23, 25–28, 31–35.
Pook LP (1979) Fatigue crack propagation. In: Chell GG (Ed) Developments in Fracture
Mechanics – I. Applied Science Publishers, London, pp. 183–220.
Pook LP (1982a) Mixed mode threshold behaviour of mild steel. In: Bäcklund J, Blom AF,
Beevers CJ (Eds) Fatigue Thresholds. Fundamentals and Engineering Applications. En-
gineering Materials Advisory Services, Warley, West Midlands, Vol. 2, pp. 1007–1032.
Pook LP (1982b) Fatigue crack growth in cruciform-welded joints under non-stationary
narrow-band random loading. In: Residual Stress Effects in Fatigue. ASTM STP 776.
American Society for Testing and Materials, Philadelphia, PA, pp. 97–114.
Pook LP (1983a) The Role of Crack Growth in Metal Fatigue. Metals Society, London.
Pook LP (1983b) The effect of mean stress on fatigue-crack growth in cruciform-welded joints
under non-stationary narrow-band random loading. NEL Report 690. National Engineer-
ing Laboratory, East Kilbride, Glasgow.
@Seismicisolation
@Seismicisolation
Metal Fatigue 253
Pook LP (1984) Approximation of two parameter Weibull distributions by Rayleigh distribu-

tions for fatigue testing. NEL Report 694. National Engineering Laboratory, East Kilbride,
Glasgow.
Pook LP (1985a) A failure mechanism map for mixed Mode I and II fatigue thresholds. Int. J.
Fract., 28: R21–R23.
Pook LP (1985b) The fatigue crack direction and threshold behaviour of mild steel under
mixed Mode I and III loading. Int. J. Fatigue, 7: 21–30.
Pook LP (1986). Keyword scheme for a computer based bibliography of stress intensity factor
solutions. NEL Report 704. National Engineering Laboratory, East Kilbride, Glasgow.
Pook LP (1987a) Random load fatigue and R.M.S. NEL Report 711. National Engineering
Laboratory, East Kilbride, Glasgow.
Pook LP (1987b) Some factors affecting wave loading of tubular members. NEL Report 712.
National Engineering Laboratory, East Kilbride, Glasgow.
Pook LP (1989a) Current Fatigue Standards. Engineering Integrity Society, Daventry.
Pook LP (1989b) Spectral density functions and the development of Wave Action Standard
History (WASH) load histories. Int. J. Fatigue, 11: 221–232.
Pook LP (1989c) The significance of Mode I branch cracks for mixed mode fatigue crack
growth threshold behaviour. In: Brown MW, Miller KJ (Eds) Biaxial and Multiaxial Fa-
tigue. EGF 3. Mechanical Engineering Publications, London, pp. 247–263.
Pook LP (1992) A brief introduction to fracture mechanics and the role of NDT. British
Journal of Non-Destructive Testing, 34: 595–600.
Pook LP (1993) A finite element analysis of the angle crack specimen. In: Rossmanith HP,
Miller KJ (Eds) Mixed-Mode Fracture and Fatigue. ESIS 14. Mechanical Engineering
Publications, London, pp. 285–302.
Pook LP (1994a) Some implications of corner point singularities. Eng. Fract. Mech., 48: 367–
378.
Pook LP (1994b) Mixed mode fatigue crack propagation. In: Carpinteri A (Ed) Handbook of
Fatigue Crack Propagation in Metallic Structures. Elsevier Science, Amsterdam, Vol. 2,
pp. 1027–1071.
Pook LP (1995) On fatigue crack paths. Int. J. Fatigue, 17: 5–13.
Pook LP (1997) Standard fatigue tests for components. In: Marquis G, Solin J (Eds) Fatigue
Design of Components. Elsevier Science, Amsterdam, pp. 203–214.
Pook LP (1998) A fatal fairground accident. Int. J. Fatigue, 20: 169–173.
Pook LP (2000a) Linear Elastic Fracture Mechanics for Engineers. Theory and Applications.
WIT Press, Southampton.
Pook LP (2000b) Crack profiles and corner point singularities. Fatigue Fract. Engng. Mater.
Struct., 23: 141–150.
Pook LP (2000c) Finite element analysis of corner point displacements and stress intens-
ity factors for narrow notches in square sheets and plates. Fatigue Fract. Engng. Mater.
Struct., 23: 979–992.
Pook LP (2002a) Crack Paths. WIT Press, Southampton.
Pook LP (2002b) Linear and nonlinear finite element analysis of cracked square plates under
Mode I and Mode II loading. In: Neimitz A, Rokach, V, Kocańda D, Gołoś K (Eds) Frac-
ture Mechanics beyond 2000. ECF 14. EMAS Publishing, Sheffield, Vol. II, pp. 695–702.
Pook LP (2004) An unusual fatigue failure. In: Sonsino CM, Zenner H, Portella PD (Eds)
Proceedings of the Seventh International Conference on Biaxial/Multiaxial Fatigue and
Fracture, Berlin, 28 June–1 July 2004. Deutscher Verband für Materialforschung und -
prüfung E. V, Berlin, pp. 617–621.
@Seismicisolation
@Seismicisolation
254 References
Pook LP, Crawford DG (1990) Stress intensity factors for twist cracks. Int. J. Fract., 42: R27–
R32.
Pook LP, Dover WD (1989) Progress in the development of a Wave Action Standard History
(WASH) for fatigue testing relevant to tubular structures in the North Sea. In: Watanabe
RT, Potter JM (Eds) Development of Fatigue Loading Spectra. ASTM STP 1006. Amer-
ican Society for Testing and Materials, Philadelphia, PA, pp. 99–120.
Pook LP, Frost NE (1973) A fatigue crack growth theory. Int. J. Fract., 9: 53–61.
Pook LP, Greenan AF (1976) Various aspects of the fatigue crack growth threshold in mild
steel. In: Proc. Fatigue Testing and Design Conf. Society of Environmental Engineers Fa-
tigue Group, Buntingford, Herts, Vol. 2, pp. 30.1–30.33.
Pook LP, Holmes R (1976) Biaxial fatigue tests. In: Proc. Fatigue Testing and Design Conf.
Society of Environmental Engineers Fatigue Group, Buntingford, Herts, Vol. 2, pp. 36.1–
36.33.
Pook LP, Short AM (1988) Fracture mechanics analysis of fatigue tests on thick walled steel
cylinders containing sharp notches. In: Heron R (Ed) Fluid Power 8. Elsevier Applied
Science Publishers, London, pp. 323–338.
Pook LP, Smith RA (1979) Theoretical background to elastic fracture mechanics. In: Smith
RA (Ed) Fracture Mechanics. Current Status, Future Prospects. Pergamon Press, Oxford,
pp. 26–67.
Pook LP, Greenan AF, Found MS, Jackson WJ (1981) Tests to determine the fatigue strength
of steel castings containing shrinkage. Int. J. Fatigue, 3: 149–156.
Portela A (1993) Dual Boundary Element Analysis of Crack Paths. Computational Mechanics
Publications, Southampton.
Pugsley AG (1966) The Safety of Structures. Edward Arnold, London.
Radaj D, Sonsino CM (1999) Fatigue Assessment of Welded Joints by Local Approaches.
Abington Publishing, Cambridge.
Rice JR (1967) Mechanics of crack tip deformation by fatigue. In: Fatigue Crack Propagation,
ASTM STP 415. American Society for Testing and Materials, Philadelphia, PA, pp. 247–
309.
Richard HA, Fulland M, Sander M (2005) Theoretical crack path prediction. Fatigue Fract.
Engng. Mater. Struct., 28: 3–12.
Richard HA, Fulland M, Sander M, Kullmer G (2006) Examples of fatigue crack growth in real
structures. In: Carpinteri A, Pook LP (Eds) Proceedings of the International Conference
on Crack Paths (CP 2006), Parma (Italy), 14–16 September 2006. University of Parma,
Parma, Proceedings on CD.
Richards CE (1980) Some guidelines on the selection of techniques. In: Beevers CJ (Ed)
The Measurement of Crack Length and Shape during Fatigue and Fracture. Engineering
Materials Advisory Services, Warley, West Midlands, pp. 461–468.
Romanovskaya AV, Botvina LR (2003) Effect of crack path on statistical distribution of the
fatigue lifetime. In: Carpinteri A, Pook LP (Eds) Proceedings of the International Confer-
ence on Fatigue Crack Paths (FCP 2003), Parma (Italy), 18–20 September 2003. Univer-
sity of Parma, Parma, Proceedings on CD (6 pp.).
Rossmanith HP (Ed) (1997) Fracture Research in Retrospect. AA Balkema, Rotterdam.
Ryman RJ (1962) Programme load fatigue tests. Aircraft Engng., 34: 34–42.
Sarpkaya T, Isaacson M (1981) Mechanics of Wave Forces on Offshore Structures. Van Nos-
trand Rheinhold Company, New York.
Schijve J (1974) Fatigue damage accumulation and incompatible crack front orientation. Eng.
Fract. Mech., 6: 245–252.
Schijve J (1980) Stress gradients around notches. Fatigue Engng. Mater. Struct., 4: 325–338.
@Seismicisolation
@Seismicisolation
Metal Fatigue 255
Schijve J (1985) The significance of flight simulation fatigue tests. In: Salvetti A, Cavallini
G. (Eds) Durability and Damage Tolerance in Aircraft Structures. Engineering Materials
Advisory Services, Warley, West Midlands, pp. 71–170.
Schijve J (2001) Fatigue of Structures and Materials. Kluwer Academic Publishers,
Dordrecht.
Schijve J (2005) Statistical distribution functions and fatigue of structures. Int. J. Fatigue, 27:
1031–1039.
Schöllmann M, Fulland M Richard HA (2003) Development of a new software for adaptive
crack growth simulations in 3D structures. Eng. Fract. Mech., 70: 249–268.
Schütz W (1979) The prediction of fatigue life in the crack initiation and propagation stages
– A state of the art survey. Eng. Fract. Mech., 11: 405–421.
Schütz W (1989) Standardized stress-time histories – An overview. In: Watanabe RT, Potter
JM (Eds) Development of Fatigue Loading Spectra. ASTM STP 1006. American Society
for Testing and Materials, Philadelphia, PA, pp. 3–16.
Schütz W (1996) A history of fatigue. Eng. Fract. Mech. 54: 263–300.
Schütz W, Pook LP (1987) WASH (Wave Action Standard History). A standardized stress-
time history for offshore structures. In: Noordhoek C, de Back J (Eds) Developments in
Marine Technology, 3. Steel in Marine Structures. Elsevier, Amsterdam, pp. 161–178.
Schütz W, Klätschke H, Hück M, Sonsino CM (1990) Standardised load sequence for offshore
structures – WASH 1. Fatigue Fract. Engng. Mater. Struct., 13: 15–29.
Sherratt F, Edwards PR (1974) The use of small on-line computers for random loading fatigue
testing and analysis. J. Soc. Env. Engrs., 14-4: 3–14.
Sherry AH, France CC, Goldthorpe MR (1995) Compendium of T-stress solutions for two and
three dimensional geometries. Fatigue Fract. Engng. Mater. Struct., 18: 141–155.
Shiozawa K, Sakai K (Eds) (1996) Databook on Fatigue Strength of Metallic Materials. El-
sevier Science, Amsterdam.
Sih GC, Lee YD (1989) Review of triaxial crack border stress and energy behaviour. Theor.
Appl. Fract. Mech., 12: 1–17.
Smith MC (1984.) Some aspects of Mode II fatigue crack growth. PhD thesis. University of
Cambridge.
Smith RA (Ed) (1979) Fracture Mechanics. Current Status, Future Prospects. Pergamon
Press, Oxford.
Smith RA (1990) The Versailles railway accident of 1842 and the first research into metal
fatigue. In: Kitagawa H, Tanaka T (Eds) Fatigue ’90. MCE Publications, Birmingham,
Vol. 4, pp. 2033–2041.
Smith SH (1965) Fatigue crack growth under axial narrow and broad band axial loading.
In: Trapp JW, Fourney PM (Eds) Acoustical Fatigue in Aerospace Structures. Syracuse
University Press, New York, pp. 331–360.
Socie D (1992) Rainflow cycle counting: a historical perspective. In: Murakami Y (Ed) The
Rainflow Method in Fatigue. Butterworth-Heinemann, Oxford, pp. 3–10.
Sonsino CM, Zenner H, Portella PD (Eds) (2004) 7th ICBMFF. Proc. Seventh International
Conference on Biaxial/Multiaxial Fatigue and Fracture, Deutscher Verband für Materi-
alforschung und -prüfung E. V., Berlin, pp. 17–25.
Speidel MO (1982) Influence of environment on fracture. In: François D (Ed) Advances in
Fracture Research. Pergamon Press, Oxford, Vol. 6, pp. 2685–2704.
Srawley JE, Jones MH, Brown WF (1967) Determination of plane strain fracture toughness.
Mat. Res. & Std., 7: 262–266.
Stanzl-Tschegg S (Ed) (2002) Fatigue in the very high cycle regime (Vienna Conference).
Fatigue Fract. Engng. Mater. Struct. 25: 725–896.
@Seismicisolation
@Seismicisolation
256 References
Steele LE, Stahlkopf KE (Eds) (1980) Assuring Structural Integrity of Steel Reactor Pressure
Vessels. Applied Science Publishers, London.
Stephens RI, Fatemi A, Stephens RR, Fuchs HO (2001) Metal Fatigue in Engineering. Second
edition. John Wiley & Sons, New York.
Sumi Y (2005) Simulation-based fatigue crack management in welded structural details. Ab-
stract. In: Abstract Book. 11th International Conference on Fracture, Turin, 20–25 March
2005, p. 5457. Extended abstract. In: CD Ibid.
Suresh S (1998) Fatigue of Materials. Second edition. Cambridge University Press, Cam-
bridge.
Taylor D (1989) Fatigue Thresholds. Butterworth-Heinemann, London.
Thompson N, Wadsworth NJ (1958) Metal fatigue. Phil. Mag. Suppl. 7: 62–169.
Timoshenko SP (1953) History of the Strength of Materials. McGraw-Hill, London.
Troughton AJ (1960) Relationship between theory and practice in aircraft structural problems.
J. Roy. Aero. Soc., 64: 653–667.
Troughton AJ, Ryman RJ, McLennan G (1963) Fatigue lessons learnt from the Argosy. In:
Barrois W, Ripley EL (Eds) Fatigue of Aircraft Structures. Pergamon Press, Oxford,
pp. 263–291.
Vinas-Pich J, Kam JCP, Dover WD (1996) Variable amplitude corrosion fatigue of medium
strength steel tubular welded joints. In: Dover WD, Dharmavasan S, Brennan FP, Marsh
KJ (Eds) Fatigue Crack Growth in Offshore Structures. Engineering Materials Advisory
Services, Solihull, pp. 147–178.
Visser P (2002) POD/POS Curves for Non-Destructive Examination. HSE Books, Sudbury.
Wanhill RJH (2003) Milestone case histories in aircraft structural integrity. In: Milne I, Ritchie
RO, Karihaloo B (Eds) Comprehensive Structural Integrity. Elsevier, London, Vol. 1,
pp. 61–72.
Watanabe RT, Potter JM (Eds) (1989) Development of Fatigue Loading Spectra. ASTM STP
1006. American Society for Testing and Materials, Philadelphia, PA.
Watson P, Dabell BJ (1976) Cycle counting and fatigue damage. J. Soc. Env. Engrs., 15-3:
3–9.
Wayman ML, Lang J, Leopold JH, Evans J (2000). Clock, watch and chronometer springs. In:
Wayman ML (Ed) The Ferrous Metallurgy of Early Clocks and Watches. Studies of Post
Medieval Steels. Occasional Paper No 136. British Museum Press, London, pp. 29–52.
Westergaard HM (1939) Bearing pressures and cracks. J. Appl. Mech., 6: 49–53.
Wheeler OE (1972) Spectrum loading and crack growth. J. Bas. Engng., 94D: 181–186.
Whyte RR (Ed) (1975) Engineering Progress through Trouble. Institution of Mechanical En-
gineers, London.
Zapffe CA, Clogg M, (1945) Fractography – A new tool for metallurgical research. Trans.
ASM, B: 71–107.
Zapffe CA, Worden CO (1951) Fractographic registrations of fatigue. Trans. Amer. Soc.
Metals, 43: 958–969.
Zastrow U (1985) Basic geometrical singularities in plane-elasticity and plate-bending prob-
lems. Int. J. Solids Structures, 21: 1047–1067.
Zenner H (2004) Multiaxial fatigue – Methods, hypotheses and applications. An overview.
In: Sonsino CM, Zenner H, Portella PD (Eds) 7th ICBMFF. Proc. Seventh International
Conference on Biaxial/Multiaxial Fatigue and Fracture, Deutscher Verband für Material-
forschung und -prüfung E.V., Berlin, pp. 3–16.
@Seismicisolation
@Seismicisolation
Metal Fatigue 257
Zerbst U, Schwalbe K-H, Ainsworth RA. (2003) An overview of failure assessment methods
in codes and standards. In: Milne I, Ritchie RO, Karihaloo (Eds) Comprehensive Structural
Integrity. Elsevier, London, Vol. 7, pp. 1–48.
Zuidema J (1995) Square and Slant Fatigue Crack Growth in Al 2024. Delft University Press,
Delft.
@Seismicisolation
@Seismicisolation
Index
K-calibration, 174 blind trial, 240

K-dominated region, 141, 144, 180, 192 block fatigue loading, 42, 127
S/N curve, 16 block loading, 42
T -stress, 143 block programme, 53
T -stress criterion, 143 boundary layer, 188, 193
T -stress ratio, 144 branch crack, 138, 155, 198
branch crack formation, 140, 149
A-scan, 231 branch crack propagation, 140
absolute liability, 78 branch crack propagation angle, 141, 148
acceptance testing, 76 branch point, 93
accuracy criterion, 242 brittle fracture, 103, 110, 136, 138, 154
additive distribution, 31 broad band random loading, 35, 50, 218, 219
Almen strip, 99
alternating current field measurement, 236
C-scan, 232
alternating current potential drop, 226, 234,
carburising, 98
236
centre cracked tension specimen, 106
alternating stress, 15, 16
analytical approach, 72, 75, 162 centre frequency, 208, 219
antisymmetric mode, 187, 190 centre line average roughness, 84
aspect ratio, 153, 177 chaos theory, 141
attractor, 143, 156 chaotic event, 141
autocorrelation function, 218 characteristic crack dimension, 175, 199
autofrettage, 99 civil liability, 78
cliff, 156
B-scan, 231 clipping, 209, 212, 213
bandpass filter, 219 clipping ratio, 42, 59, 209, 212, 214
bandwidth, 40, 208, 218 code, 70, 73
basic situation, 69 code design, 70
Basquin equation, 18 COLOS, 216
bending mode, 174 compliance function, 174
biaxial fatigue loading, 37 compressional probe, 232
biaxial loading, 37, 59 compressional wave, 231
biaxiality ratio, 143 conditional distribution, 32
bimodal distribution, 31 constant amplitude loading, 15
@Seismicisolation
@Seismicisolation
260 Index
constant amplitude fatigue loading, 15, 39, D-scan, 232

42, 96, 105, 106, 117, 120, 122, 124, DAC curve, 233
127 DAC level, 233
Consumer Protection Act, 78, 79 damage density curve, 49
core region, 141, 180, 189 damage density function, 49
corner point, 151, 182, 187, 190, 191 damage tolerance, 71, 103
corner point singularity, 152, 187 defective, 69, 77, 78, 80, 164
corner point singularity dominated region, design defect, 69, 78
189, 192 Det Norske Veritas, 132
corner region, 188 detector, 229
couplant, 231 developer, 228
crack area, 199 differential geometry, 11, 155
crack bridging, 238 diffracted wave, 233
crack closure, 114, 127 directionally stable crack, 142
crack front, 168 directionally unstable crack, 142
crack front curvature, 138 disclination, 171, 173
crack front inclination angle, 191 dislocation, 152, 171, 173, 199
crack front intersection angle, 152, 188, 189 dispersion effect, 231
crack front shape, 129 distance amplitude correction curve, 233
crack initiation dominated, 83, 88, 92, 98 dye penetrant, 228
crack mouth, 227
Eastabrook’s theorem, 125
crack opening, 113
eddy current, 234, 235
crack path prediction, 149, 151, 158
edge sliding mode, 169
crack path stability, 142
effective crack length, 182, 184
crack propagation, 138, 172, 173, 186
effective value, 113
crack propagation dominated, 92, 101
effects of thickness, 110
crack sizing, 223, 227, 231
electromagnetic field method, 234
crack surface displacement, 168, 170, 192 electromagnetic radiation, 230
crack surface intersection angle, 187, 190, end on technique, 232
191 endurance, 17
crack tip, 168 endurance limit, 20
crack tip plastic zone, 105, 112, 113, 125, enforcement authority, 79, 80
137, 150, 182 enforcement officer, 79
crack tip surface displacement, 169, 170, equivalent Mode I crack, 197
183, 187, 190 equivalent constant amplitude stress range,
cracked situation, 101 47
criminal liability, 77, 79 equivalent cycle, 50, 124
critical crack front intersection angle, 190, equivalent single crack, 201
191 equivalent stress, 63, 64, 89, 96
critical length, 27 equivalent stress intensity factor, 159
critical plane, 63 equivalent tensile stress, 62
critical plane approach, 64 ergodic, 204, 218
crystal dislocation, 170, 199 escape clause, 75
cumulative damage, 44 exceedance, 40, 206, 208, 210, 217
cumulative probability, 40, 206 exponential distribution, 210
cycle, 15
cycle counting, 50, 125 facet, 156
cyclic load, 1 fail safe, 70, 154
@Seismicisolation
@Seismicisolation
Metal Fatigue 261
failure, 2, 4 Gaussian distribution, 21, 39, 206, 208, 211,

failure analysis, 68, 83, 97, 102, 109, 122, 213, 214, 216, 218
130, 162 geometric correction factor, 175
failure mechanism map, 139 Gerber diagram, 86
false call, 239 Gerber parabola, 85
false call probability, 239 gigacycle fatigue, 19, 58
FALSTAFF, 54 Goodman diagram, 86, 96
Fast Fourier Transform, 219 good practice, 73
fatigue, 7 grass, 233
fatigue assessment, 67, 69, 71–74, 83, 93, Gumbel distribution, 217
97, 109, 117
fatigue crack growth, 26 hard X-rays, 230
fatigue crack initiation, 25, 26, 31, 83, 98, high cycle fatigue, 17
132
fatigue crack path, 10, 75, 135, 138 ideal crack path, 142
fatigue crack propagation, 11, 26, 105, 109, in house tool, 70
114, 117, 118, 122, 124, 128, 129, induction hardening, 98
135, 138, 155, 159, 164, 165, 167, initial crack, 139, 154, 155, 158, 198
223, 224, 228, 236, 237 initial crack size, 101, 102, 122
fatigue crack propagation life, 102, 122, 124 integral approach, 64
fatigue crack propagation rate, 105, 106, interaction effect, 125, 127, 196, 200
internal crack, 142, 196, 200
109, 110, 114, 116, 119, 122, 124,
127, 133, 224, 237 internal defect, 32
intrinsic fatigue strength, 84
fatigue cycle, 15, 105
irregular crack, 196
fatigue design, 67, 109
irregularity factor, 40, 218
fatigue fracture toughness, 104
fatigue life, 83, 122, 124, 130 JOSH, 54
fatigue limit, 18, 120
fatigue load, 1, 55 Kirchoff plate bending theory, 174
fatigue loading, 15, 38, 61, 94, 98, 138, 149, Kitagawa diagram, 120
154, 156, 199, 204, 240 knee, 19
fatigue strength, 90, 162 KoNoS hypothesis, 65
fatigue strength reduction factor, 90
fatigue testing, 8, 9 Laplace distribution, 49, 210, 213, 216
final crack size, 101, 103, 122 LBF Normal distribution, 53
fish eye, 29 leak before break, 5, 154
flame hardening, 98 leakage flux, 227
Forsyth’s notation, 26, 137, 148 library, 240
Foucault current, 236 life, 17
fractals, 11 lift off, 235
fractography, 11, 129 limit load, 103
fracture criterion, 64 line tension, 152, 199
fracture mechanics, 12, 164, 167 linear damage rule, 45
fracture toughness, 104, 110, 155, 172, 183 linear elastic fracture mechanics, 167
frequency dependence, 116 load cycle, 16
frequency domain, 219 load history, 12, 35, 44, 49, 53, 56, 58, 65,
frequency independent, 18, 76, 116 127, 130, 135
load spectrum, 13, 44
gamma rays, 230 long crack, 118
@Seismicisolation
@Seismicisolation
262 Index
longitudinal wave, 231 narrow band random process, 41, 59, 208,
low cycle fatigue, 17, 65 219
negative peak, 42
macrocrack, 27, 136 negligence, 78, 79
magnetic flux, 227, 236 nitriding, 98
magnetic ink, 227 no fault liability, 78
magnetic particle inspection, 226 non destructive testing, 103, 164, 196, 223,
magnetic permeability, 235 238
main crack, 139 non metallic material, 4
major project, 70 non propagating crack, 93, 116, 120, 142
manufacturing defect, 69, 78 non proportional fatigue loading, 59, 64
mass product, 70 non proportional loading, 59
mathematical description, 37 non proportional random loading, 62
maximum normal stress criterion, 64 non stationary random processes, 42
maximum principal stress dominated crack nonlinear dynamics, 143
propagation, 137 Normal distribution, 21, 39, 206
maximum stress, 15, 20 notch, 83, 89, 180
mean stress, 10, 15, 85, 88 notch insensitive, 92
mean stress insensitive, 110 notch sensitive, 92
mean stress sensitive, 110, 114 notch sensitivity index, 92
mechanical description, 1, 7, 161, 165 number of cycles, 16
metal fatigue, 1, 7, 37, 161–163, 165, 204,
207, 210, 212, 218, 223, 229, 231, ocean wave, 216, 231
235, 236 omission dilemma, 58
metal fatigue damage, 10 omission level, 58, 213
metal fatigue mechanism, 24 one sided process, 214
metallurgical description, 1, 161, 165 open crack, 181
microcrack, 25, 27 opening mode, 169
micromechanisms, 165 overload, 127
Miner’s law, 45 oxide induced crack closure, 115
Miner’s rule, 13, 45, 49, 50, 97, 126
minimum stress, 15 P-S-N curves, 23
Mises criterion, 63 Palmgren–Miner law, 45
mixed mode, 26, 138, 155, 156, 158, 169, Palmgren–Miner rule, 45
198 Paris equation, 105
mixed mode threshold for fatigue crack Paris law, 105
propagation, 139 Paris region, 117, 122
Mode I, 169, 170, 172, 183, 184, 186, part through crack, 152, 153, 177
190–192, 196, 197, 199 peak counting, 51
Mode II, 169, 171, 172, 186, 191, 192 penny shaped crack, 176
Mode III, 169, 171, 172, 186, 191, 192 periodic processes, 206
modified Goodman diagram, 86 philosophies of design, 70, 154
multiaxial failure criterion, 62 physical crack length, 184
multiaxial fatigue loading, 35, 39, 60, 63, 65, plane strain, 109, 182, 183, 185
88 plane strain fracture toughness, 104, 154,
multiaxial loading, 35, 59 184
plane stress, 109, 154, 183, 184, 188
narrow band random loading, 35, 56, 58, plane stress fracture toughness, 154, 184
210, 214–216 plastic collapse, 103, 154
@Seismicisolation
@Seismicisolation
Metal Fatigue 263
plastic wake, 113, 119, 125, 127 safety factor, 72

plastic zone, 128, 182, 184 safety regulation, 78, 79, 80
plastic zone correction, 185 scalar criterion, 63
plastic zone radius, 185 scatter, 20, 31, 72, 96, 97, 107, 122, 133
point estimate, 240, 242 scatter band, 106
positive peak, 41 scatter diagram, 217
postulated cracks, 103 sea state, 216, 221
power spectral density, 219 service loading testing, 72, 75, 76, 163
probability density, 21, 40, 206, 208, 210 shakedown, 31
probability of detection, 223, 234, 238, 240, shear crack propagation, 138, 186
242 shear dominated crack propagation, 137
probability of failure, 23, 96, 97 shear lip, 111
probability of sizing, 223, 234, 242 shear mode, 169
product liability, 67, 77, 78 shear wave, 231, 233
programme loading, 42, 129 short crack, 118
programme marking, 129 shot peening, 88, 99
proof loading, 99 shrinkage, 101
proportional fatigue loading, 59, 64 sigmoidal, 117
proportional loading, 59 significant wave height, 217
pseudo random, 35, 214 skin depth, 234, 235
skin effect, 234
quality system, 80 slant crack propagation, 137, 186, 198
slant fatigue crack propagation, 110
radiograph, 229 slip, 25
radiography, 229 slip line, 10
rainflow counting, 51 small scale argument, 180, 186
random process, 204, 207 Soderberg line, 88
random process theory, 35, 39, 204, 207, 218 soft X-rays, 230
random walk, 142 spectral bandwidth, 219
range counting, 51 spectral density function, 40, 56, 218
range of scales, 24 stable state, 114
ratchetting, 30 Stage I crack, 26, 137, 140, 158
Rayleigh distribution, 41, 208, 210, 213, 216 Stage II crack, 26, 101, 138, 148
re-assessment, 73 Stage III, 27
redundant, 71 standard deviation, 21, 206, 207
reference stress, 38 standard load history, 13, 53–56, 62, 216
regulatory authority, 75, 80, 224 standard procedure, 13, 70, 73, 162
relative operating characteristic, 239 standard test method, 12, 107
residual stress, 88, 98, 99, 124, 133, 180 standing wave, 231
resonant frequency, 219 static failure, 101
return period, 35, 42, 58, 214 static failure region, 118, 122
root mean square, 39, 131, 206, 207, 210, static strength, 1
212–214, 216, 218 statistically non stationary, 75, 216
rotating bending, 16 statistically stationary, 39, 204, 216, 218
rotation, 172 stochastic process, 38
roughness induced crack closure, 115 stress concentration factor, 91, 150
stress criterion, 28, 117
safe life, 70 stress cycle, 15, 124
safe product, 77, 79, 80 stress history, 15
@Seismicisolation
@Seismicisolation
264 Index
stress intensity factor, 12, 104, 113, 116, truncation dilemma, 58

119, 122, 125, 132, 138, 143, 150, truncation level, 56
155, 158, 159, 164, 167, 169, 172, twist crack, 156, 158
180, 186, 190–192, 195, 196, 199,
200, 225, 233, 238 ultrasonics, 226, 231
stress intensity factor range, 105 uncracked situation, 83, 96
stress intensity measure, 187, 189 underload, 127
stress range, 10, 15, 105 uniaxial fatigue loading, 59
stress ratio, 15, 45, 110 uniaxial loading, 35, 59, 90
stress relief, 124 unsafe product, 80
stress state, 154
striation, 27, 129 validity corridor, 142
strict liability, 78, 79 variable amplitude fatigue loading, 35, 37,
subsidiary echo, 231 96, 124
surface breaking crack, 223, 226, 228, 232, variable amplitude loading, 35
238 variance, 206
surface crack, 25, 151, 196, 200, 223 viscous fluid induced crack closure, 115
surface factor, 85 visual inspection, 223
surface finish, 84 Volterra distorsioni, 170
surface hardening, 98 von Mises criterion, 63, 88, 92, 184
surface irregularity, 84
symmetric mode, 187, 190 Wöhler curves, 17
Wöhler’s laws, 9
thermal loading, 31, 150 WASH, 54
thermodynamic criterion, 28, 117, 137 water surface elevation, 216, 221
threshold for fatigue crack propagation, 11, wave height, 216
29, 116, 117, 119, 122, 132, 138, 158 wave loading, 55, 56, 58
threshold region, 117, 122 wave passing frequency, 56, 59, 217, 221
through the thickness crack, 154, 196, 225 wave period, 217
tight crack, 230 waveform, 18, 76, 211
time domain, 219 Weibull distribution, 210, 213
time history, 40, 52 weighted average stress range, 47, 124
time of flight diffraction, 233, 240 welded joint, 124, 130, 227, 230, 237
toughness, 172 Wheeler’s model, 128
trading standards officer, 79 white noise, 42
transformation induced crack closure, 115
transition region, 111, 186 X-rays, 229
translation, 172
transverse wave, 231 yield criterion, 63
Tresca criterion, 63 Young’s modulus, 109
trough, 42, 209
truncation, 42, 58, 210, 212, 213 zero crossing, 40
@Seismicisolation
@Seismicisolation

Metal Fatigue What It Is, Why It Matters Pook 2007

Uploaded by

Copyright:

Available Formats

Metal Fatigue What It Is, Why It Matters Pook 2007

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Metal Fatigue What It Is, Why It Matters Pook 2007

Uploaded by

Copyright:

Available Formats

What is metal fatigue and why does it matter?

What is metal fatigue and why does it matter?

What is the author's background and experience with metal fatigue?

What is the author's background and experience with metal fatigue?

@Seismicisolation

Series Editor: G.M.L. GLADWELL

Aims and Scope of the Series

For a list of related mechanics titles, see final pages.

ISBN 978-1-4020-5596-6 (HB)

Printed on acid-free paper

All Rights Reserved

3 Constant Amplitude Fatigue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.5 Effect of Mechanisms on S/N Curves . . . . . . . . . . . . . . . . . . . . 31

4 Variable Amplitude and Multiaxial Fatigue . . . . . . . . . . . . . . . . . . 35

6 The Uncracked Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

6.7 Scatter under Variable Amplitude Fatigue Loading . . . . . . . . . 96

7 The Cracked Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

8 Fatigue Crack Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

9 Why Metal Fatigue Matters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

A Fracture Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

B Random Load Theory and RMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 203

C Non Destructive Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

C.7.1 Eddy Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

brief description of the mechanisms of metal fatigue, to describe the basis of

I, II, III modes of crack surface displacement

KλS symmetric mode stress intensity measure

From an engineering viewpoint, metal fatigue matters because it is a ma-

Figure 1.2. Cracking in aircraft jet engine nacelle.

under unusually severe operating conditions, could be regarded as fair wear

Figure 1.5. Plastic domestic tap.

2.2 Experimental Work

2.2.1 FATIGUE T ESTING OF C OMPONENTS AND S TRUCTURES

2.2.2 FATIGUE T ESTING OF L ABORATORY S PECIMENS

Fatigue testing of specially designed laboratory specimens started in the

2.2.3 I NVESTIGATION OF THE M ECHANISMS OF M ETAL FATIGUE

Investigation of the mechanisms of metal fatigue started in earnest in the

2.2.4 I NVESTIGATION OF FATIGUE C RACK PATHS

The complete solution of a fatigue crack propagation problem includes de-

2.2.5 FATIGUE C RACK P ROPAGATION R ATE T ESTING

Fatigue crack propagation rate testing, in both laboratory specimens and

2.3 The Modern Era

2.3.1 F RACTURE M ECHANICS

2.3.2 S ERVOHYDRAULIC T ESTING E QUIPMENT

The introduction of closed loop servohydraulic testing equipment (Marsh

2.3.3 I NFLUENCE OF C OMPUTERS

Standard procedures of various degrees of formality are the most satisfactory

assessments are based on simpliﬁed standard procedures, which despite their

A constant amplitude fatigue loading (or constant amplitude loading) is a fa-

3.2 S/N (Wöhler) Curves

Figure 3.1. Notation for constant amplitude fatigue loading.

Figure 3.4. Plain fatigue specimen for testing in direct stress.

deﬁnitions are sometimes used. For example, loss of a speciﬁed amount of

where A and m are empirically determined constants. The intersection

3.3 Scatter in Fatigue Data

may be made sufﬁciently narrow for the histogram to be represented by a

Table 3.1. Normalised stresses for various probabilities of failure.

Table 3.2. Metallurgical feature scales.

3.4 Mechanisms of Metal Fatigue

Mechanisms responsible for the failure of metals, and associated metallur-

Table 3.3. Stress analysis feature scales.