Cui Wang Huang 2011 1

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/312768988
Basic Concepts of Fracture Mechanics
Chapter · January 2014

DOI: 10.1007/978-3-642-41831-0_4
CITATION READS
1 2,268
3 authors:
Weicheng Cui X.P. Huang

Westlake University Shanghai Jiao Tong University
384 PUBLICATIONS 5,754 CITATIONS 82 PUBLICATIONS 1,076 CITATIONS
SEE PROFILE SEE PROFILE
Fang Wang
Shanghai Ocean University
169 PUBLICATIONS 1,948 CITATIONS
SEE PROFILE
All content following this page was uploaded by Weicheng Cui on 21 February 2018.
The user has requested enhancement of the downloaded file.

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Marine Structures 24 (2011) 153–181
Contents lists available at ScienceDirect
Marine Structures
journal homepage: www.elsevier.com/locate/
marstruc
A unified fatigue life prediction method for marine

structures
Weicheng Cui a, *, Fang Wang a, Xiaoping Huang b
a
China Ship Scientific Research Center, P.O. Box 116, No. 222 East Shanshui Road, Wuxi, Jiangsu 214082, China
b
State Key Laboratory of Ocean Engineering, Shanghai Jiao Tong University, Shanghai 200030, China
a r t i c l e i n f o a b s t r a c t
Article history: Marine structures such as ships and offshore platforms are mostly
Received 10 February 2011 designed with damage tolerance and this design philosophy
Accepted 10 February 2011 requires accurate prediction of fatigue crack propagation process.
Now more and more people have realized that only a fatigue life
Keywords: prediction method based on fatigue crack propagation (FCP)
Unified fatigue life prediction (UFLP)
theory has the potential to satisfy the accuracy requirement and to
method
explain various fatigue phenomena observed. In the past several
Marine structures
Crack growth rate years, the authors’ group has made some efforts in developing
Fatigue strength assessment a unified fatigue life prediction (UFLP) method for marine struc-
tures. The key issue for this development is to establish a “correct”
crack growth rate relation. In this paper the improvement of the
crack growth rate model is dealt with first. A new crack growth
rate model based on the concept of partial crack closure is pre-
sented. The capability of the model is demonstrated. Secondly,
studies on the engineering approaches to determine the parame-
ters in the new crack growth rate model are carried out and vali-
dated by comparing with the experimental results on a wide range
of alloys. Thirdly, the preliminary studies on some significant
problems such as load sequence effect are presented. Finally,
further studies for the application of the UFLP method to the
fatigue strength assessment of marine structures are pointed out.
Ó 2011 Elsevier Ltd. All rights reserved.
* Corresponding author.
E-mail address: [email protected] (W. Cui).
0951-8339/$ – see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.marstruc.2011.02.007
154 W. Cui et al. / Marine Structures 24 (2011) 153–181
Nomenclature
3tot Total error of fatigue life prediction

3SCF Prediction error from stress concentration factor
3S–N Prediction error from the choice of S–N curve
3seqence Prediction error from load sequence
3initial crack Prediction error from initial crack
3final crack Prediction error from final crack
L Influence factors due to loading
M Influence factors due to material properties
S Influence factors due to structural geometry
E Influence factors due to environment
N Load cycles
da/dN Crack growth rate
Kmax Maximum stress intensity factor
Kmin Minimum stress intensity factor
Kcf The fracture toughness of the material under a fatigue condition
KIC Plane strain fracture toughness
KCmax Plane stress fracture toughness
KC Actual fracture toughness
Kop The stress intensity factor at the opening level
Kopmax The maximum stress intensity factor at the opening level for a macroscopic crack
Kopmax0 The maximum stress intensity factor at the opening level for a macroscopic crack under
zero load ratio
Kplate The stress intensity factor in the vicinity of crack tips in a plate
DK Stress intensity factor range
DKth The threshold stress intensity factor range
DKeff The effective stress intensity factor range
DKeffth The threshold effective stress intensity factor range
DKth0 The threshold stress intensity factor range under zero load ratio
DKeffth0 The effective threshold stress intensity factor range under zero load ratio
fop A crack opening function defined as the ratio Kop/DK
fth A function defined as the ratio of DKth/DKth0
feff A function defined as the ratio of DKeffth/DKth
fKc The fracture toughness coefficient
MK Stress concentration magnification factor for welded joints
smax The maximum stress level
smin The minimum stress level
sY The yield stress of the material
su The ultimate strength of the material
sfl The flow stress of the material usually defined as the average value between the
material yielding and ultimate strengths for convenience
sa The stress amplitude
sm Mean stress
sV The ‘virtual strength’ of the material representing the material strength at limit of
‘perfect’ condition (re ¼ 0)
Ds Stress range
s0.2 The reference yield strength of the material
W. Cui et al. / Marine Structures 24 (2011) 153–181 155
Y(a) A geometrical factor to calculate the stress intensity factors under crack length a
Y(re) A geometrical factor to calculate the stress intensity factors under crack length re
t Specimen thickness
w Plate width
E Elastic modulus of the material
n Poisson’s ratio
RA Area reduction of the cross-section of a specimen due to elongation
EL Elongation
CPO Crack plane orientation
T Test temperature
3f Fracture strain of the material
a The modified crack length which is equal to re plus the actual crack length
re An empirical material constant of the inherent flaw length of the order of 1 mm
ath The threshold value of crack length
c Surface crack depth
a0,c0 Initial values of surface crack length/depth
R The stress ratio defined by smin/smax
A A material- and environmentally sensitive constant of dimensions in the crack growth
rate model
m A constant representing the slope of the corresponding fatigue crack growth rate curve
in the crack growth rate model
k A material constant which reflects the rate of crack closure development with crack
advance
n The index indicating the unstable fracture in the crack growth rate model
a A parameter used to calculate the ‘virtual strength’ of the material
a0 The plane stress/strain constraint factor
B1,g The material constants in two equations of the threshold stress intensity factor range as
a function of load ratio, which should be determined by experiments
b,b1 The material constants in the segmented equation of the threshold stress intensity
factor range as a function of load ratio proposed by Huang (2006)
rps The crack tip plastic zone size under the plane stress state
rp3 The crack tip plastic zone size under the plane strain state
rp The crack tip plastic zone size
l The crack tip plastic zone coefficient
n0 The strain hardening exponent of the material
3a The strain amplitude
H0 ,m0 Cyclic stress–strain constants (Ramberg–Osgood Equation)
3max The maximum strain level
s0 f Fatigue strength coefficient
b,c Fatigue strength exponent/fatigue ductility exponent
1. Introduction
Fatigue is one of the most significant failure modes for marine structures such as ships and offshore
platforms which are mostly made of metals [1]. Accurate prediction of the fatigue life of marine
structures under service loading is very important for both safe and economic design and operation.
Metal fatigue has been studied for more than 170 years. Much progress has been made but is still not
completely satisfactory.
Many fatigue life prediction (FLP) methods for metals have been proposed and they can be divided
into: cumulative fatigue damage (CFD) theory and fatigue crack propagation (FCP) theory, both in
deterministic sense or probabilistic sense. Traditionally, the theoretical frameworks used for fatigue
strength assessment in most engineering rules are based on cumulative fatigue damage (CFD) theory.
Current Fatigue Strength Assessment (FSA) methods for marine structures are all stress-based using
nominal, hot-spot or notch stresses. Metal fatigue is an extremely complex phenomenon and many
factors will affect the fatigue life of metal structures. The total prediction error comes from the stress
concentration, the choice of stress versus cycles (S–N) curve, the load sequence, the initial crack, the
final crack, etc, which might be expressed as Eq. (1).
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
3tot ¼ u32SCF þ 32S-N þ 32sequence þ 32initialcrack þ 32finalcrack þ . (1)
t |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
ignored by CFD theory
Existing approaches based on CFD theory have only considered very limited number of influencing
factors such as the stress concentration and the choice of S–N curve, however, the influencing factors of
load sequence, the initial crack, the final crack, which have been experimentally proved of the similar
importance, have not been considered. This might explain a large part of the wide scatter of predicted
fatigue lives.
A recent comparative study on fatigue strength assessment procedures adopted by ship classifi-
cation societies indicated a large difference in the predicted fatigue lives for a very simple detail [2]. The
current approach adopted by international association of classification societies (IACS) for reducing the
scatter is the unification. It is believed that unification will not completely solve the actual problem of
scatter. The S–N curve method itself is also subjected to some theoretical flaws and is responsible for
some scatter. This is based on the following two arguments: (1) A scatter on fatigue lives as big as 10
times for practical structures is certainly not acceptable because a fatigue life of 3 years has
a completely different interpretation with a fatigue life of 30 years for a ship detail. However, if we
realize that this scatter also exists in well-controlled “identical” small specimens, the scatter can only
be attributed to the initial defects and the final state of failure. (2) Many experimental evidences have
showed that for variable amplitude loading, the load sequence effect is also very significant. It is well-
known that for the high-low loading, damage index is less than 1.0 due to the underload acceleration
while for the low-high loading, damage index is greater than 1.0 due to the overload retardation.
The scatter induced by initial defects and the load sequence is of the same order as that induced by
differences in definition of loading, calculation of the stress concentration factors (SCFs) and choice of
S–N curves, see Eq. (1). Therefore, currently used stress-based approaches for FLP of marine structures
are subjected to theoretical flaws. FCP theory could overcome these deficiencies. This is the funda-
mental philosophy for the development of a unified FLP method (UFLP) for marine structures.
It is true that crack-like defects always exist in welded structures. And the basic idea of our unified
method is that initial defects always exist in engineering structures and fatigue is purely a crack
propagation problem. This assumption has been confirmed due to two progresses: (1) fracture
mechanics (FM) has been successfully applied in microstructural level and (2) microstructural level
cracks can now be measured. However, for the purpose of easy application, we confine our model to the
macroscopic level. No microstructural parameters are involved. In FCP theory, the crack growth rate
curve is the most fundamental material property to be defined and many factors will influence the
crack growth rate curve. The most general one can be represented by the following expression:
da=dN ¼ f ð L; M; S; EÞ (2)
where L; M; S; E represent influence factors due to loading, material, structure and environment
respectively. Strictly speaking, if a method can be regarded as a unified FLP method, it must be able to
explain all the phenomena observed. In order to achieve this purpose, all the factors which affect the
fatigue behavior must be included in the crack growth rate model. This is an impossible task. We have
to approach it using empirical methods and including several most significant factors. The less the
parameters are used, the easier the method is applied. So every empirical crack growth rate curve is
a compromise of such an effort. Up to now hundreds of different empirical crack growth rate curves
have been proposed and each has its advantages and drawbacks. Since every empirical model only
includes a few parameters, it cannot be expected to explain every phenomenon. In that sense every
model is not a unified method but the unified approach can be regarded as the long-term target. So the
essence of developing a UFLP is to establish a “correct” crack growth rate relation.
The earliest crack growth rate based on linear elastic fracture mechanics is introduced by Paris et al.
who equated fatigue crack growth rate by using stress intensity factor range [3,4]. And later, people
found that crack growth rate curve is not linear for all ranges of stress intensity factor range so that
a threshold value of stress intensity factor range is introduced into the equation [5]. Recently, more
progress has been made to establish crack growth rate relations which are able to explain different
fatigue phenomena [6,7]. As one of the fatigue crack growth models in this sense, the modified
constitutive relation developed by McEvily and his co-workers [6] can not only account for the effects
of initial crack size and load sequence, but also explain various other phenomena of metal fatigue
observed in tests. But, the modified model is only for linear and near threshold regions and valid for
elastic-perfectly plastic materials. Consequently, this modified constitutive relation is further gener-
alized by Cui and Huang [8] in three main aspects: (1) introducing an unstable fracture condition; (2)
defining a virtual strength to replace the yield stress; and (3) defining an overload and underload
parameter. The performances of this general constitutive relation for fatigue crack growth have been
extensively studied and it is found that this general constitutive relation is able to explain various
phenomena observed with particular strong capability on load sequence effect. Based on the work of
Cui and Huang [8], a further extension is made by Wang et al. [9] concerning two aspects: (1) the slope
of the fatigue crack growth rate curve is regarded as a variable rather than a fixed value of 2 for different
materials; (2) both the maximum stress intensity factor at the crack opening level and the effective
stress intensity factor range at the threshold level are written as the functions of load ratio.
However, there are still some shortcomings in these empirical formulas, for example, the ability of
the model to explain some critical phenomena as overload retardation and underload acceleration,
specimen thickness effect, the unstable fracture condition, etc. In the past several years, the authors’
group has made some efforts in improving the ability of the crack growth rate model in UFLP for marine
structures. The extended McEvily model was further improved by Yi [10], Wang and Cui [11–14], Chen
and Cui [15,16]. And validation on the improved model has been made by comparing the predicted
results with test data for a wide range of materials. In this paper, the improvement of the crack growth
rate model based on the concept of partial crack closure will be introduced first. The ability of the
model is demonstrated. Secondly, studies on the engineering approaches to determine the parameters
in the improved model are carried out and validated by comparing with the experimental results on
a wide range of alloys. Thirdly, the preliminary studies on some significant problems such as load
sequence effect are presented. Finally, further issues related to the application of the UFLP method to
the fatigue strength assessment of marine structures are pointed out.
2. The improvement of the fatigue crack growth rate used in UFLP
2.1. General ideas of a unified fatigue life prediction (UFLP) method
The basic idea of our unified method is that initial defects always exist in engineering marine
structures and fatigue is purely a crack propagation problem. Therefore, the most important problem to
solve is to establish a “correct” crack growth rate relation as the most fundamental material behavior
under fatigue loading. For a given loading history, the accumulated crack length is calculated on
a cycle-by-cycle basis. This could accurately account for many loading effects such as mean stress effect,
load sequence effect and structural defects such as notch effect and size effect.
By unified approach in our terms the same method should be able to cover both low cycle fatigue
(LCF) and high cycle fatigue (HCF). The fatigue limit and the ultimate strength are the two extremes of
fatigue crack propagation. Various anomalous fatigue crack propagation phenomena [8] must be
explained by the method.
So in applying the unified procedure for FLP, the basic idea is to calculate the crack size on a cycle-
by-cycle basis and the fatigue loading must be given as a loading history rather than spectrum. Fig. 1
shows the general flow diagram of the unified fatigue life prediction method.
Fig. 1. The flow diagram of the UFLP method.
2.2. The improvement of the fatigue crack growth rate
Paris and co-workers [3,4] were the first to define the cyclic stress intensity factor range DKth as the
fatigue crack driving force and equated fatigue crack growth rate to DKth, which is the basis of the
improvement of general constitutive relation. Later, people found that the crack growth rate curve is
not linear for all ranges of DK but in sigmoidal shape, which can be subdivided into three regions,
namely, crack initiation and threshold region, stable crack growth region and unstable fracture region.
Furthermore, in order to explain the effects of load ratio on fatigue crack growth, Elber [17] introduced
the concepts of crack closure and the effective stress intensity factor as DKeff. This concept has been
highly appraised in 1980s and 1990s but now it is subjected to some challenges, e.g. [18]. Particularly
many people have agreed that the physical effects of crack closure have been greatly over-estimated in
the past, e.g. Ref. [19]. Partial crack closure model [6,20,21] was proposed to overcome the difficulty the
crack closure model met. Comparison with some experimental data showed that the constitutive
relationship proposed by McEvily et al. [6] is able to explain the six special phenomena they concerned.
Original McEvily model was written as:
8
>
>
da ¼ AM 2
>
> dN sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
>
> ps qffiffiffiffiffi
< max
M ¼ pre Sec þ 1 1 þ YðaÞ 2rae smax ð1 RÞ 1 eka Kopmax RKmax DKeffth
> 2 sY
>
> rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi qffiffiffiffiffi
>
> s
> p
: Kmax ¼ pre Sec 2 sY þ 1 1 þ YðaÞ 2rae smax
max
(3)
where re is the size of an inherent flaw, a parameter whose magnitude is of the order of several microns
in length [6]. Newly formed cracks are only significant when their lengths exceed re. a is the modified
crack length which is equal to re plus the actual crack length. Thus the minimum value of a is re which
represents a component contains no crack (plain specimen). Eq. (3) is derived by combining Irwin’s
concept of plastic zone with Dugdale’s result. This formula was valid for elastic-perfectly plastic
materials and will be modified by us to account for other materials. sY is the yield stress of the material,
MPa, smax is the maximum stress in a loading cycle, R is the stress ratio (¼smin/smax), Ds is the stress
range applied, MPa. k is a material constant which reflects the rate of crack closure development with
crack advance and is similar to that in other partial crack closure model. Y is a geometrical factor. Kopmax
pffiffiffiffiffi
is the maximum stress intensity factor at the opening level for a macroscopic crack, MPa m. Both
Kopmax and DKeffth are functions of R [6].
This relation was found to be able to account for various fatigue phenomena observed [6,21–23] and
valid from physically small crack to long crack regimes [24]. But it is only applicable for near threshold
region and valid for elastic-perfectly plastic materials. The limitations are overcome through the efforts
of Cui and Huang [8]. They further generalized McEvily model by: (1) introducing an unstable fracture
condition into the crack growth rate curve to cover the unstable growth region; (2) defining a ‘virtual
strength’ sV to replace the yield stress in Eq. (3). Thus the generalized constitutive relation for fatigue
crack growth of metals can be expressed as follows:
8
>
> da ¼ AM 2
>
> dN 1ðKmax =KC Þn
>
> sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

>
< p smax qffiffiffiffiffi
M ¼ pre Sec þ 1 1 þ YðaÞ 2rae smax ð1 RÞ 1 eka Kopmax RKmax DKeffth
> 2 sV
>
> rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
>
> qffiffiffiffiffi
> p s
: Kmax ¼ pre Sec 2 s þ 1 1 þ YðaÞ 2r smax
> max a
V e
(4)
Use of the last equation in Eq. (3) to define Kmax implies that the stress will never exceed the yield stress
sY even at static failure for “uncracked” plain specimens. In order to make Kmax ¼ Kc be also valid for the
uncracked plain specimens, an empirical approach is adopted by replacing the yield strength sY by
another “virtual strength” (sV). The “virtual strength” (sV) is defined by the condition that the
maximum stress of the uncracked (i.e. a ¼ re) plain specimen is equal to the ultimate strength (su) of
the material at unstable fracture condition. Based on this definition, the “virtual strength” (sV) can be
determined from the following equation:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Yðre Þ p su
1 þ pffiffiffi pre Sec þ 1 su ¼ Kc (5)
2 2 sV
And the solution of Eq. (5) is
sV p 1 Kc pffiffiffi
¼ $ ; a¼ pffiffiffiffiffiffiffiffi > 2 (6)
su 2 cos1 1 su pre 1 þ pffiffieffi
Yðr Þ
a 1
2
2
The virtual strength of the material may represent the material strength at limit of ‘perfect’ condition
(re ¼ 0) while the actual ultimate strength of the material (su) represents the strength under the
condition that the defect size is equal to the inherent flaw length (re > 0), a minimum crack size for
engineering metals.
However, it is shown in Li et al. [25] that McEvily model with a fixed slope of 2 is not in good
agreement with experimental results for some materials. Thus Wang et al. [9] proposed a further
extended model. In this model, the parameter m is adopted to represent the slope of the fatigue crack
growth rate curve for different materials. However, there are still some shortcomings in the formula.
The measurement method of the parameter Kopmax is still not explained very clearly in relevant
literature. It is even not a material constant, as the assumption of its independence on stress intensity
factor range DKeffth level may not always be proved valid [24] and it should be strongly dependent on
the stress range [26]. In order to reduce the uncertainty resulting from the measurement of this
parameter, the item DKop defined as the stress intensity factor range at the opening level and Kopmax
introduced in modified McEvily mode [24] will be replaced by Newman’s function [27,28] in the
improved crack growth rate model. Another shortcoming is the expression of the unstable fracture
condition. Strictly speaking, the unstable fracture condition in the fatigue crack growth model is
defined by Kmax ¼ Kcf, where, Kcf is the fracture toughness of the material under fatigue condition. In
view of the lack of this value from costly experiments, the static fracture toughness from a more easily
carried-out test is usually adopted as the unstable fracture condition in many models. But actually, it
should be strongly dependent on the load sequence. This problem has been specifically pointed out by
Schijve [29] and needs further experimental study. In our further improvement procedure, the
assumptions as follows will be added [11]: (1) The value of KC in the crack growth model varies with
crack length; (2) The unstable fracture occurs when the stress intensity factor with a certain crack
length reaches its corresponding KC; (3) The stress intensity factor at the opening level will also change
with crack length by modifying the crack tip stress/strain constraint factor. Then the general consti-
tutive relation in the improved crack growth rate model will be written in the following form,
8
>
> A½Kmax $ð1fop ÞDKeffth
m
>
> da=dN ¼
>
> 1ðKmax =KC Þ n
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
> ffi qffiffiffiffiffi
<K ¼ p r Sec p smax þ 1 1 þ YðaÞ a s
max e 2 sv 2re max
(7)
> rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffi
>
>
>
>
> Kmin ¼ pre Sec p2 ssmin þ 1 1 þ YðaÞ 2rae smin
>
> v
>
: DK ¼ Kmax K
min
where,
( n o
max R; A0 þ A1 R þ A2 R2 þ A3 R3 0R<1
fop ¼
A0 þ A1 R 2R<0
8 h i1=a0
>
> A ¼ 0:825 0:34 a 0 þ 0:05a02 $ cos ps =2s
>
> 0 max fl
>
>
>
> A1 ¼ ð0:415 0:071a0 Þ$smax =sfl
>
>
>
> A ¼ 1 A0 A1 A3 (8)
< 2
A3 ¼ 2A0 þ A1 1
>
> sfl ¼ ðsY þ su Þ=2
>
>
>
> 1 1
1 12
>
> a0 ¼ þ n
>
> 3:2251 0:75952
>
> 1 2n
: 1 þ 0:8861$ t=ðKmax =sY Þ2
pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi #
8 ð12nÞ2 1 n2 p$l 1 n2 1
>
> KC ¼ $ þ $KIC
>
> ð12nÞ2 1 ð12nÞ2 ð12nÞ2 1
>
> !
< 1
1 1 1 n0 ð11:65nÞ2 1 10
ð11:65nÞ2
n
>
>
> ð11:65nÞ 2
1 10 p 2:2n0 p 5 20n 0
>
>l ¼ ð11:65nÞ2 þ
n
>
: 2 1:6þ1=n
5 20n0 max =sY Þ
0
1 þ t=ðK
1þ1=n 0
(9)
It can be noticed that Kmax, fop and KC are all the functions of crack length a.
3. Engineering approaches to determine the parameters in the improved model
3.1. General methods to estimate the model parameters
Costly fatigue tests for determining so many model parameters will certainly be an obstacle for its
engineering application. As one may realize that fatigue failure is the consequence governed by these
model parameters if this model really reflects the correct fatigue crack growth behavior. Thus, we may
be able to estimate the model parameters from any set of fatigue test data. In this part, the approaches
to estimate the model parameters will be introduced for further research.
If the model parameters are regarded as deterministic, then determination of the model parameters
from available test data is basically an optimization problem. The general form of the model can be
expressed as follows,

yi ¼ f X ðiÞ ; q þ ei i ¼ 1; 2; .; N (10)
where the observed values of the yi constitute the responses or values of the dependent variables X; the
ðiÞ ðiÞ ðiÞ
known X ðiÞ ¼ ðx1 ; x2 ; .; xK Þare the vectors of values of the independent K variables; f is a known
function of an unknown parameter vector q ¼ ðq1 ; q2 ; .; qM Þ; and the values of ei are independently
distributed normal errors.
The residuals for the model are

ei ð qÞ ¼ yi f X ðiÞ ; q i ¼ 1; 2; .; N (11)
A value of q can be determined by minimizing the following error function,

N h
X i2
Error ¼ yi f X ðiÞ ; q (12)
i¼1
3.1.1. Estimation method from crack growth rate data

Theoretically speaking, all model parameters can be regarded as variables to be determined but this
involves very complicated mathematical manipulation. For simplicity, we generally treat some of the
parameters as constants and others as model parameters. If we know N testing points: [smaxi, Ri, ai, (da/
dN)i], then the error function is
X
N
AMim 2
Error ¼ ðda=dNÞi (13)
i¼1
1 ðKmaxi =Kc Þn
Based on these equations one can determine an optimal set of the parameters. Nonlinear curve fitting
theory can be used to solve this problem [11,30].
3.1.2. Estimation method from a–N curve

From the crack growth rate, one can determine the a–N curve
Z a
1 ðKmax =Kc Þn
N ¼ da (14)
a0 AM m
If we know N testing points [ai, Ni], namely the a–N curve is available, we can estimate the parameters
in the crack growth rate model by the error function as follows,
2 32
X
N Z
4Ni
ai
1 ðKmax =Kc Þn 5
Error ¼ da (15)
i¼1 a0 AM m
In these equations, one needs to calculate a lot of integrations which is a difficult task. However, we can
estimate some parameters by tensile test data of the material and left only A, m, n to be determined by
the method introduced here.
In order to estimate the model parameters (A, m, n), the Nonlinear Least-square Fitting method is
usually applied, which includes the establishment and the optimization calculation of the error
function. The particular explanation of mathematical manipulation of the Nonlinear Least-square
Fitting method will be introduced below by the flow diagram of calculation shown in Fig. 2. To
minimize the value of error function, Jacobian matrix must be calculated based on the explicit
Fig. 2. Flow diagram of the nonlinear least-square fitting method.
expression of integral function inside the right part of Eq. (15), which could be obtained by executing
the operation of numerical integration depending on the variable a first. Setting JacobianT equal to zero,
we can obtain the optimal value of the parameters (A, m, n) by solving the nonlinear equations group.
However, the operation of the numerical integration calculation and the nonlinear equation group
solution mentioned above will become very difficult when the error function is complex. In such cases,
Genetic Algorithm may be an effective way to avoid the difficulty of these mathematical manipulations.
The schematic diagram of genetic algorithm is illustrated in Fig. 3. The procedure will start with coding
in some way, then the initial population (A0, m0, n0) can be obtained. By substituting (A0, m0, n0) into Eq.
(15), we can calculate the fitness of every individual, based on which some individuals are selected as
next generation population and others are eliminated. Subsequently, crossover and mutation are
executed according to the genetic operators and the next generation population (A1, m1, n1) is
produced. By heredity of enough generations, the optimal individual (An, mn, nn) can be obtained when
the optimum criterion is satisfied.
3.1.3. Estimation method from S–N curve

From the crack growth rate, one can determine the S–N curve
Z af
1 ðKmax =Kc Þn
Nf ¼ da (16)
a0 AM m
If we know N testing points [smaxi, Ri, Ni], from smaxi, Kc, su, re, one can calculate afi, the error function is,
2 32
X
N Z
4Nfi
afi
1 ðKmax =Kc Þn 5
Error ¼ da (17)
a0 AMm
i¼1
Using the same method as explained in Fig. 3, one can determine a best set of model parameters. Cui
[31] comprehensively discussed the relation between crack growth rate curve and stress versus life
curve.
Fig. 3. Schematic diagram of genetic algorithm.

3.1.4. Estimation method from 3–N curve

The strain-based approach differs significantly from the stress-based approach. Recall that the
stress-based approach emphasizes nominal (average) stress, rather than local stresses and strains, and
it employs elastic stress concentration factors and empirical modifications thereof. Employment of the
cyclic stress–strain curve is a unique feature of the strain-based approach, as is the use of a strain versus
life curve, instead of a nominal stress versus life curve [32]. In the following, the subscript ‘l’ stands for
the local value of the parameter.
The cyclic stress–strain curve can be expressed as,
( 1=m0
3a ¼ sa sa
E þ H’
1=m0 (18)
3max ¼ smax
E þ smax
H0
The Manson–Coffin equation considering mean stress effect is adopted here to express 3–N curve,
namely,
s0f sm
3a ¼ ð2NÞb þ30f ð2NÞc (19)
E
If the cyclic stress–strain curve and the strain versus life curve by tests for a material have been
obtained (the parameters E; H 0 ; m0 s0f ; e0f ; b; c are given), namely, we know N testing points [3lai,Ni], then
the value of slai and slmi can be calculated respectively according to Eqs. (18) and (19). Using the relation
slmi ¼ slmaxi slai and Eq. (18), the valued of slmaxiand 3lmaxi will be determined.
The concept of ‘crack tip damage zone’ proposed by Ma and Lu [33] will be used here for reference in
deducing the relation between 3–N data and S–N data. The basic assumptions include,
(a) A small crack tip damage zone with the length of x* exists inside the plastic zone, in which the
stress/strain gratitude is very small and can be regarded as constant. The stress/strain at the middle
point (0.5x*) can be taken as its mean value.
(b) The material in the crack tip damage zone will experience fatigue failure and crack will propagate
by an increment x* after fatigue stress cycles N.
(c) The material inside the damage zone is extremely damaged but the stress/strain gratitude outside
it decreases quickly and the material damage can be neglected.
For a center-cracked specimen under cyclic stress, the released energy when crack propagates can
R
2
be expressed as Kmaxi =E$2xi while the work inside the crack tip damage zone is 2 1=2slmaxi $2vðxÞdx;
where v(x) is the displacement normal to crack face in the damage zone. The two xi values should be
equal to each other, namely,

Z
1
2
Kmaxi =E$2xi ¼ 2 s $2vðxÞdx (20)
2 lmaxi
xi
Ma and Lu [33] deduced the expressions of stress and strain in the crack tip damage zone for power-
law material, which can be written as,
8 n0 =ðn0 þ1Þ
Ki2
>
< sl 0:5xi ¼ sY pð1þn0 Þs2 $0:5x

Y i
1=ðn0 þ1Þ (21)
>
: s K 2
3l 0:5xi ¼ E pð1þn0 Þs2 $0:5x
Y i
Y i
Then the relation between the stress and strain in crack tip damage zone can be expressed as,
2
Kmaxi
slmaxi elmaxi ¼ (22)
pð1 þ n0 ÞE$0:5xi
Substituting the calculated values of slmaxi and 3lmaxi into Eq. (22), the values of Kmaxi and xi can be
obtained by combining Eqs. (20) and (22). Furthermore, according to the following equation,
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi rffiffiffiffiffiffiffi
ps ai
Kmaxi ¼ pre Sec maxi þ 1 1 þ Yðai Þ s (23)
2 sV 2re maxi
P
whereai ¼ a0 þ xi , the value of smaxi can be obtained. By repeatedly employing the procedures
introduced above,i the 3–N curve can be transformed to S–N curve. And the crack growth rate model
parameters for this material can be determined using the method introduced in the above two parts.
At present, the fatigue crack growth rates are still mainly measured by experiments, especially the
threshold of stress intensity factor range, which are expensive and time consuming. Therefore, some
researchers attempt to find the methods for predicting the fatigue crack growth rates from the tensile
properties of the material, e.g. Refs. [34–37]. And the threshold effective stress intensity factor range
can be accordingly estimated.
3.2. Validation of the improved crack growth rate model with its model parameters on a wide range of alloys
under constant amplitude load
The approximate estimation approach of model parameters introduced above will be validated in
this part by a wide range of alloys, which include aluminum alloy [38–40], titanium alloy [41], and steel
[14,42,43]. Detailed information on test conditions can be referred to these references but engineering
estimation methods [34,35,44,45] are used for those test parameters not described in references. In all
examples, the model parameters are derived from experimental data under stress ratio R ¼ 0.1 and
then used to estimate the crack growth rate curves under other stress ratios. The comparisons between
test data and prediction curves are illustrated in Figs. 4–13, which show that the prediction curves
agree well with test data and the approximate estimation approach is accordingly validated. The model
parameters used in these figures are given in Table 1.
Another example is made to validate the improved model for welded structures. Fracture mechanics
based computations are extremely sensitive to the initial crack shape and size [46]. A realistic semi-
elliptical initial crack shape is usually assumed for calculation, which has been widely accepted. And
the surface crack under fatigue loading always changes its crack front as it grows. The changes in defect
aspect ratio depend principally upon initial configuration, relative crack depth and loading condition,
but also depend weakly upon stress ratio, loading frequency, growth rate exponent, mean stress and
the crack tip stress conditions which alter with changes in crack shape [47]. Many investigators have
0.1 R=0.1, experimental data

R=0.1, predicted curve
R=0.5, experimental data
1E-3 R=0.5, predicted curve
da / dN (m / cycle)

R=-1, experimental data
R=-1, predicted curve
1E-5
2324-T39
1E-6
1E-7
1E-8
0 .1 1 10 1 00
0.5
Δ K (MPa m )
Fig. 4. Comparison between predicted crack growth rate curves and test data of 2324-T39 aluminum alloy [38].

1E-3
R=0.5,predicted curve
da / dN (m / cycle)
R=0.7,experimental data
1E-4 R=0.7,predicted curve
1E-5 R=-1, predicted curve
6013-T651
1E-6
1E-7
1E-8
0. 1 1 10 100
0.5
Δ K (MPa m )
carried out the study on predicting crack shape changes [48–51]. It is necessary to pre-know the crack
shape changes before estimating the remaining fatigue life using fracture mechanics theory, as the
stress intensity factors along crack front are normally expressed by the function of crack aspect ratio.
Generally, the surface crack is assumed to keep its semi-elliptical profile when growing for simplicity,
so that the crack profile changes combined with semi-elliptical assumption are always investigated in
the past and proved to be reasonable. We will also adopt this assumption for the current study.
Then the basic assumptions of the methodology for welded joints include
(1) The semi-elliptical configuration is used in estimating the fatigue life of welded joints with surface
damage problems. It is assumed that the crack will propagate keeping the two-dimensional semi-
elliptical profile and every point along the crack front will advance following the same crack
growth model with different growth rates. Then the growth characteristics of two points at the
bottom and the end of the crack are enough to describe the growth of the crack front.
The change of the crack aspect ratio, or the growth rate relationship of the bottom and end of the
crack must be pre-known before estimation, because the stress intensity factors of the bottom and the

1E-3
da / dN (m / cycle)

1E-5 R=-1, predicted curve
7055-T7511
1E-6
1E-7
1E-8
0 .1 1 10 100
0.5
Δ K (MPa m )
1E-3
7075-T6
1E-4
da / dN (m / cycle)
1E-5
1E-6 R=0.1, experim ental data

R=0.5, experim ental data
R=-1, experim ental data
1E-8
1 10
0.5
Δ K (M Pa m )
end of crack are dependent on crack aspect ratio. Many studies have been carried out on this problem
and the empirical or numerical results in literature can be adopted currently for further analysis. In the
analysis of growing a single macro-crack, the empirical equation proposed by Lin and Smith [47] as
follows is recommended,
a a 2:5 a
a a 2:5
¼ g 0 g 0 0 (24)
c t a t c0
where
gðuÞ ¼ 1:331 þ 0:706u2 þ 0:155u4 þ 0:013u6 (25)

Eqs. (24) and (25) are used for growing a single semi-elliptical surface crack while the random coa-
lescence of cracks at the weld toe has not been taken into account. Typically, the experimental crack
aspect curve proposed by Engesvik and Moan [52] is usually adopted to consider coalescence effect,
and in practice, continuous edge crack with zero crack aspect ratio is always applied for simplicity.

0.01 R=0.1, predicted curve
da / dN (mm / cycle)

1E-4
7075-T651
1E-5
1E-6
0 .1 1 10 100
0.5
Δ K (MPa m )
Fig. 8. Comparison between predicted crack growth rate curves and test data of 7075-T651 [40].
1E-4 R=0.1, experimental data

1E-5
1E-6
da / dN (m / cycle) 1E-7
1E-8
HTS-A steel
1E-9
1E-10
1E-11
1E-12
1 10 10 0
0.5
Δ K (MPa m )
Fig. 9. Comparison between predicted crack growth rate curves and test data of HTS-A steel [14].
(2) Plane strain condition can be assumed approximately when estimating the growth rate of bottom
point. And the corresponding crack opening level can be calculated under plane strain condition.
While the actual three-dimensional stress/strain constraint will be considered when calculating
the growth rate of end point, which is related to the plate thickness, yield stress and stress intensity
factor during propagation.
(3) As the welded structures usually have crack-like defects, the total fatigue life of the weld joints is
the summation of surface crack propagation life and through-thickness crack propagation life
when the deepest point of the crack reaches the plate bottom.
Stress concentration magnification factor of semi-elliptical cracks at the toe of fillet welded joints
can be expressed as follows,
K ¼ MK Kplate (26)
1E-4

da / dN (m / cycle)

1E-7 300 M steel
1E-8
1E-9
10 0.5
ΔK (MPa m )
Fig. 10. Comparison between predicted crack growth rate curves and test data of 300M steel [42].
1E-5
R=0.1, exprimental data
1E-6
da / dN (m / cycle) R=0.3, exprimental data
1E-7
350WT steel
1E-8
1E-9
1 10 1 00
0.5
ΔK (MPa m )
Fig. 11. Comparison between predicted crack growth rate curves and test data of 350WT steel [43].
where, Kplate is the stress intensity factor in the vicinity of crack tips in a plate and MK is a magnification
factor for welded joints. Bowness and Lee [53] recommended an empirical expression of MK of semi-
elliptical crack. And the calculation methods for stress intensity factors of surface cracks are introduced
in Ref. [54].
As an example, the surface crack in a T-joint is illustrated in Fig. 14. And the symbols for joint and
crack geometry parameters are labeled in the figure.
The test data given in NIMS [55] for some cruciform joints will be used for comparative study
here. Coalescence effect is considered. Table 2 gives parameters of cruciform joints. The comparison
(a) and correlation (b) between predicted results and test data [55] are shown in Fig. 15(a) and (b),
which can prove the suitability of the model used for welded structures. Details can be referred to
Wang and Cui [14].
1E-6
da / dN (m / cycle)
1E-8
CrMoV
1E-9
1E-10
1 10
0.5
ΔK (MPa m )
Fig. 12. Comparison between predicted crack growth rate curves and test data of CrMoV [45].
0.1
da / dN (m / cycle) R=0.8, predicted curve
1E-4
Ti-10V-Fe-3Al
1E-5
1E-6
1E-7
1 10 100
0.5
Δ K (MPa m )
Fig. 13. Comparison between predicted crack growth rate curves and test data of Ti–10V–Fe–3Al [41].
4. Capabilities of the unified crack growth rate model
In this part, the capability of the unified crack growth rate model will be exhibited. In our previous
work [8,9,11], the basic performances of the improved general constitutive relation have been exten-
sively studied and it is found that after improvement, various phenomena observed in tests can be
explained. Here, emphasis will be put on explaining some special phenomena such as specimen
thickness effect, compressive to compressive loading effect, and overload effect by the improved model
expressed in Eqs. (7)–(9).
4.1. Specimen thickness effect
Validation for the capability of our model to explain specimen thickness effect is made by the
comparison between the test results in Bao and McEvily [56]. The corresponding geometry and material
data can be referred to the original paper. However, the value of KIC is not given there. We will estimate
the value according to the following empirical equation proposed in Farahmand and Nikbin [34],

KIC ¼ 16:348 0:0685$ef $DKth0 (27)
where, the value of DKth0 under zero load equation should be calculated first by its corresponding value
at load ratio of 0.05 given in Ref. [57] according to the following equation proposed in Huang et al. [58],
Table 1
The derived model parameters from experimental data for different materials.
Material A m
2324-T39 Aluminum alloy 2.4357E06 2.2143
Ti–10V–Fe–3 Titanium alloy 3.5880E06 1.3069
HTS-A steel 1.3900E10 2.3668
300 M steel 1.5770E10 2.3031
350 WT steel 7.3784E10 2.0628
CrMoV 3.2097E10 1.9858
Fig. 14. Illustration of surface crack in T-joint: (a) surface crack sketch (b) 3D T-joint model.
8 0:5
< ð1 RÞ0:3
> 5 R < 0
DKth =DKth0 ¼ fth ¼ ð1

RÞ
0:3
0 R < 0:5 (28)
>
:
1:05 1:4R þ 0:6R2 0:5 R < 1
During the simulation, we used multivariable nonlinear regression method to determine the param-
eters in the model first according to the test data for the thick plate and obtain the values of A, m, n, re
which are equal to 1.5144 106 (MPa)m m1m/2, 1.36204, 6 and 3.15 107m respectively. The
detailed approaches to estimate parameters will have been introduced in Section 3 of this paper. These
values for thick plate will be used for the calculation of the crack growth rate for the thin plate. The
simulation results and their comparison with the test data [56] are illustrated in Fig. 16. It can be seen
that they agree with each other very well. The comparison proves the validity of the method proposed
in this paper. Details can be referred to Wang et al. [12].
4.2. Compressive to compressive loading effect
There are some components subjected to compressive and bending loading in ship and offshore
structures and fatigue cracks were found in compressive side of these components caused by fluctu-
ation loadings, during their service [59]. Generally, the negative portion of the applied stress intensity
factor range is always ignored in fatigue life prediction. Till now, only a few studies have been done to
examine the details of crack initiation and growth under compressive cycling [60–64]. Actually, the
crack can be initiated and propagated to a certain length and the residual stresses play a very important
role in the crack growth under compressive loading. But in order to explain the compressive to
compressive loading effect, the calculation method of some parameters must be further specified in
our model. Residual stresses which cause crack propagation must be included in the model.
Typically residual stress distribution along the crack surface must be determined first. However, it is
very difficult to determine it by theoretical analysis. Finite element method can be employed to do that.
Table 2
Parameters of cruciform joints used for comparative study between predicted results and test data [55].
sY (MPa) sb (MPa) R T (mm) q ( ) l (mm) a0 (mm)

485 570 0 9 39.2 11.8 0.2
485 570 0 20 50.9 19.5 0.2
485 570 0 40 37.4 33.9 0.2
Fig. 15. Comparison (a) and correlation (b) between predicted results and test data [55].
From the residual stress distribution, the stress intensity factor calculation may also be difficult. Tan
and his co-workers [65] proposed an analytical equation to determine the stress intensity factor (SIF) of
arbitrary weld residual stresses. For a stress distribution sres(x) along the crack surface, the stress
intensity factor can be determined by the following equation [65],
pffiffiffiffiffiffiffiffiffi Z a sres ðxÞ
Kres ðlinearÞ ¼ 2 a=p pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (29)
0 a2 x2
Here, we call this value calculated by Eq. (29) the linear one of SIF to distinguish it with the nonlinear
one expressed in our model. An average value of residual stress will be determined by,
pffiffiffiffiffiffi
sres ¼ Kres ðlinearÞ= YðaÞ pa (30)
And the nonlinear value of SIF induced by residual stress can be obtained,
1E-4
t/w=0.111014 (test)
t/w=0.111014 (simulation)
t/w=0.011583 (test)
1E-5 t/w=0.011583 (simulation)
da/dN (m/cycle)
1E-6
1E-7
9Cr-1Mo steel
1E-8
0.1 1 10 100
0.5
Δ K (MPa m )
Fig. 16. Comparison between test data (Bao, 1998) and simulation results of the effect of plate thickness on the curve of crack growth
rate versus stress intensity factor range [12].
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi rffiffiffiffiffiffiffi
ps a
Kres ¼ pre Sec res þ 1 1 þ YðaÞ sres (31)
2 sv 2re
Then we can define the equivalent values of parameters in our model as follows,
8s
> max;eff ¼ smax þ sres
>
< min;eff ¼ smin þ sres
s
>
>
Reff ¼ smin;eff =smax;eff (32)
>
> Kmax;eff ¼ Kmax þ Kres
>
>
: 0
aeff ¼ a0 Kmax;eff ; fop;eff ¼ fop Kmax;eff ; leff ¼ l Kmax;eff ; KC;eff ¼ f leff
The final expression of crack growth model considering compressive to compressive loading effect will
be written as,
h im
A Kmax;eff $ 1 fop;eff DKeffth Reff
da=dN ¼ n (33)
1 Kmax;eff =KC;eff
In order to testify the method proposed in this part, prediction results will be compared with the
results of fatigue test performed by us on specimens made of HTS-A material with double edge crack
(DECP) under compressive to compressive loading. Detailed description on the specimen sizes and
loading parameters is given in Fig. 17 and Table 3. And Fig. 18 gives the comparison of test data with
predicted result. It can be seen that the predicted values are in good agreement with the test data.
4.3. Overload effect
Crack growth retardation due to overload is the most commonly observed load interaction effects
and has the most notable influence on fatigue life [66,67]. Many fatigue crack growth retardation
models have been proposed up to now but none of them has been universally accepted. In considering
overload effect, in our improved constitutive model it is assumed that the load sequence effect is due to
variations in crack closure stress in metals and the large plastic zone as a result of overloading can raise
the stress intensity factor at the opening level.
Taking single overload as an example, a large plastic zone is created as a result of an overload, and
the crack closure level instantaneously rises to the maximum value, and then gradually recovers to the
initial level under constant amplitude loading when the crack penetrates the large plastic zone in the
subsequent cycles. A modified coefficient, F, on the basis of Wheeler model will be introduced as
a magnification factor to correct the amount of the stress intensity factor at crack opening level during
Fig. 17. Geometry of the specimen DECP.

Table 3
The specimen size and loading parameters.
DECP Pmax/KN Pmin/KN f/HZ a0/mm W/mm t/mm

10.1 252.7 10 20 100 5.85
the recovering period after an overload in the improved constitutive model illustrated in Fig. 19, and
the crack closure level is described in Eq. (34).
(
aOL þrOL ry g
0 Kop aOL a aOL þ rOL ry
Kop ¼ Kop $F ¼ a (34)
Kop a aOL þ rOL ry
Considering the effect of load sequence interactions, it is important to estimate the plastic zone size in
front of the crack tip. The following continuous function is recommended to calculate plastic zone size
to consider effects of material Poisson’s ratio and strain hardening effects [11,68],
2
Kmax
rp ¼ l (35)
sy
where, l is calculated according to Eq. (9).
The plastic zone size mentioned above is also called monotonic plastic zone. As the minimum cyclic
load in a cycle is approached, yielding in compression occurs in a region of a smaller size, called the
cyclic plastic zone, as expressed in Eq. (36) [69].
2
DK
rp ¼ l (36)
2sy
When the sequence effect becomes significant, the loading history must be defined in such a way that
the real order of the loading cycles can be reproduced. Under these conditions, cycle-by-cycle calcu-
lation is the most widely accepted procedure for simulating the real crack propagation [70]. For each
loading cycle, a crack growth increment Dai is separately calculated by integration and the results are
accumulated to predict the fatigue crack length for whole loading history. According to the improved
constitutive model, the total crack length can be mathematically represented in Eq. (37) using a cycle-
by-cycle integration procedure.
1E-7
da/dN (cycles)
HTS-A steel
Test
Predicted
1E-8
1E-9
0.02 0.021 0.022 0.023 0.024 0.025 0.026
a (m)
Fig. 18. Comparison of test data with predicted result (DECP).
Fig. 19. Schematic representation of the assumed change of Kop under load sequence with single overloading.
X
n X
n
a ¼ a0 þ Dai ¼ a0 þ f ðDK; R; F; .Þ (37)
i¼1 i¼1
For the numerical implementation of the improved constitutive model, a cycle-by-cycle integration
program starts with inputting the variables concerning material characteristics and geometric
dimensions and the model parameters determined by fitting those data of crack growth rate under
constant amplitude loading. Another important input file is loading spectrum which lists peak and
valley values in the real order of applied sequence. The flow diagram for a cycle-by-cycle integration
procedure of the improved constitutive model is illustrated in Fig. 20.
The improved constitutive model can be validated by comparing the predictions with fatigue crack
growth test data including D12Cz aluminum alloy [32], 7075-T651 aluminum alloy [23], 18G3A steel
[71], 350WT steel [43], HTS-A steel [72]. The comparisons are illustrated in Figs. 21–25, which show
that the prediction curves agree well with test data both in the case of single overload and multiple
overloads.
5. Issues to be solved in developing a UFLP method for marine structures
The unified fatigue life prediction method is currently under development. In order to apply the
UFLP method to the fatigue assessment of marine structures, more relevant issues related to code
development should be investigated further such as the problems related to fatigue loading, fatigue
resistance of material, the calculation method of the fatigue life for complicated structures etc. Most of
them are still unsolved though illustrations to some of them are given above. In this section, the
problems to be solved for the development of a unified fatigue life prediction method for marine
structures will be briefly identified as the possible research areas, which can be listed as follows,
Fig. 20. Flow diagram for the cycle-by-cycle integration procedure of the improved constitutive model.
1E-6
Baseline D16Cz Aluminium Alloy
Experimental data
Predicted curve
R=0 OLR=2
1E-7
da/dN (m/cycle)
-m 1-m/2
A=1.42E-8 MPa m
m=1.4903
1E-8 n=6
re=1E-6 m
KIC=26.72 MPa
ΔKeffth=3.6 MPa
γ =4.18
1E-9
0.005 0.01 0.015 0.02 0.025 0.03
0.5
ΔK (MPa m )
Fig. 21. Comparison of prediction with test data of D12Cz Al alloy under single overload [32].
5.1. Issues related to fatigue loading part
(1) The long-term measurement of fatigue loading histories;

(2) The extrapolation of short-term measurement to long-term time history;
(3) Sequential flow cycle counting;
(4) Definition of a standardized loading histories for the design of marine structures;
(5) Multiaxial fatigue loading.
5.2. Issues related to fatigue resistance of material
(1) The actual driving forces for a crack development;

(2) The threshold of these driving forces;
(3) The unstable fracture condition for these driving forces;
1E-6
Baseline
Experimental data
Predicted curve
R=0.05 OLR=2
1E-7 18G3A steel

da/dN (m/cycle)
-m 1-m/2
A=9.78E-12 MPa m
m=3.8665
1E-8 n=6
re=1E-6 m
KIC=18.64 MPa
ΔKeffth=5.62 MPa
γ =2.12
1E-9
20 40
0.5
ΔK (MPa m )
Fig. 22. Comparison of prediction with test data of 18G3A steel under single overload [71].
1E-5
D Predicted curve
B Experimental data
R=0 OLR=2
1E-6
da/dN (m/cycle)
1E-7
-m 1-m/2
A=9.001E-9 MPa m
1E-8 m=2.281
n=6
re=1E-6 m
KIC=29 MPa
1E-9
ΔKeffth=4.72 MPa
γ=18.32
1E-10
10 12 14 16 18 20
0.5
ΔK (MPa m )
Fig. 23. Comparison of prediction with test data of 7075-T651Al alloy under single overload [23].
(4) The expression of a unified fatigue crack growth rate (FCGR);

(5) Criterion for crack growth path.
5.3. Issues related to the calculation method of the fatigue life
(1) Determination of model parameters using nonlinear regression analysis;

(2) Process of fatigue life calculation under random load sequence;
(3) Efficient and accurate integration of FCGR in a cycle-by-cycle basis.
5.4. Issues related to the application of UFLP method
(1) Target probability of failure for fatigue mode;

(2) Efficient and accurate calculation of fatigue driving forces for complex structures;
1E-6
Baseline
Experimental data
Predicted curve
R=0.1 OLR=1.75
350 WT steel
da/dN (m/cycle)
1E-7
-m 1-m/2
A=4.0311E-8 MPa m
m=0.9987
n=6
re=1E-6 m
KIC=25.85 MPa
ΔKeffth=13.24MPa
γ=0.52
1E-8
20 40 60
0.5
ΔK (MPa m )
Fig. 24. Comparison of prediction with test data of 350WT steel under single two overloads [43].
1E-5
experimental data
predicted curve
R=0.3 OLR=2
HTS-A steel
1E-6
da/dN (m/cycle)
-m 1-m/2
A=2.737E-9 MPa m
m=1.349
1E-7
n=6
re=1E-6 m
KIC=150 MPa
ΔKth0=13.2 MPa
1E-8 γ=14.05
10 100
0.5
ΔK (MPa m )
Fig. 25. Comparison of prediction with test data of HTS-A steel under multiple overloads [72].
(3) Calculation of residual stresses;

(4) Treatment of contact problem of fracture surfaces;
(5) Code development for design and evaluation purposes.
6. Summary and conclusions
In this paper, some results in developing a unified fatigue life prediction (UFLP) method for marine
structures are presented. The key issue for the development of a UFLP method is to establish a “correct”
crack growth rate relation. The improved crack growth rate model presented in this paper could take
most of the important influencing factors into account. By using this model, most of the observed
macroscopic fatigue phenomena can be explained satisfactorily. The capability of the model is
demonstrated firstly in this paper. Then studies on the engineering approaches to determine the model
parameters are carried out and validated by comparing with the experimental results on a wide range
of alloys. Furthermore, the preliminary studies on some significant problems such as load sequence
effect are presented. Finally, further studies for the application of the UFLP method to the fatigue
strength assessment of marine structures are identified. Through these analyses the following
conclusions can be drawn:
(1) The improved crack growth rate model developed by the authors based on the concept of partial
crack closure can reflect many influential factors observed in fatigue tests by including model
parameters with specialized physical meaning;
(2) The improved model shows strong capability in simulating the crack growth curves under different
load ratio of various metal materials;
(3) A variable fracture toughness during crack propagation should be used in crack growth rate model;
and the crack tip stress/strain constraint factor used in calculating the stress intensity factor at
crack opening level should be changed continuously during crack propagation; So the improved
model could be successfully to explain the three-dimensional stress state on fatigue crack growth
rate;
(4) The model parameters can be quite accurately estimated by using the tensile test data of the
materials and limited fatigue test data; the estimation approach saves costly fatigue tests and can
be used when quickly obtaining the fatigue crack growth characteristics is demanded;
(5) The improved model can be used for welded joints. Although only constant amplitude loading is
considered in the validation example, the variable amplitude loading can be treated in the same
manner as examples for the overload effect;
(6) By reasonable treatment, the improved crack growth rate model can be successfully applied for
simulating some special fatigue phenomena such as compressive to compressive loading effect and
overload retardation effect;
(7) In order to apply the UFLP method to the fatigue assessment of marine structures, many further
issues have been identified related to code development.
Acknowledgement
This study was supported by the Innovative Scholars Support Program of Jiangsu Province, Project
No. BK2008004, 2008–2010.
References
[1] Cui WC. A state-of-the-art review on fatigue life prediction methods for metal structures. Journal of Marine Science and
Technology 2002;7:43–56.
[2] Fricke W, Cui WC, Kierkegaard H, Kihl D, Koval M, Lee HL, et al. Comparative fatigue strength assessment of a structural
detail in a containership using various approaches of classification societies. Marine Structures 2002;15(1):1–13.
[3] Paris PC, Gomez MP, Anderson WP. A rational analytical theory of fatigue. The Trends in Engineering 1961;13:9–14.
[4] Paris PC, Erdogan F. A critical analysis of crack propagation laws. Journal of Basic Engineering 1963;85:528–34.
[5] Donahue RJ, Clark HM, Atanmo P, Kumble R, McEvily AJ. Crack opening displacement and the rate of fatigue crack growth.
International Journal of Fracture Mechanics 1972;8:209–19.
[6] McEvily AJ, Bao H, Ishihara S. A modified constitutive relation for fatigue crack growth. In: Wu XR, Wang ZG, editors.
Fatigue’99: proceedings of the seventh international fatigue congress. Beijing, China: Higher Education Press. p. 329–336.
[7] Sadananda K, Vasudevan AK, Holtz R. Extension of the unified approach to fatigue crack growth to environmental
interactions. International Journal of Fatigue 2001;23:S277–86.
[8] Cui WC, Huang XP. A general constitutive relation for fatigue crack growth analysis of metal structures. Acta Metallurgica
Sinica 2003;16:342–54.
[9] Wang YF, Cui WC, Wu XY, Wang F, Huang XP. The extended McEvily model for fatigue crack growth analysis of metal
structures. International Journal of Fatigue 2008;30:1851–60.
[10] Yi GJ, Wang YF, Cui WC, Huang XP. Application of the general fatigue crack growth model on the single overload fatigue
problem. Journal of Ship Mechanics 2008;12(6):947–54.
[11] Wang F, Cui WC. Approximate method to determine the model parameters in a new crack growth rate model. Marine
Structure 2009;22(4):744–57.
[12] Wang F, Chen FL, Cui WC. Applicability of the improved crack growth rate model and its parameters estimation method.
Journal of Ship Mechanics 2010;14(3):252–62.
[13] Wang F, Cui WC. Effect of three dimensional stress state on unstable fracture condition and crack opening level in a new
crack growth model. Acta Metallurgica Sinica 2010;1:41–9.
[14] Wang F, Cui WC. On the engineering approach to estimate the parameters in an improved crack growth rate model for
fatigue life prediction. Ship and Offshore Structures 2010;3(8):227–41.
[15] Chen FL, Wang F, Cui WC. Applicability of the improved crack growth rate model for a wide range of alloys under constant
amplitude load. Journal of Ship Mechanics, in press.
[16] Chen FL, Wang F, Cui WC. An improved constitutive model to predict fatigue crack growth rate under variable amplitude
loading with single and multiple overload. Journal of Engineering for the Maritime Environment, in press.
[17] Elber W. Fatigue crack closure under cyclic tension. Engineering Fracture Mechanics 1970;2:37–45.
[18] Hertzberg RW, Newton CH, Jaccard R. Crack closure: correlation and confusion. In: Mechanics of fatigue crack closure,
ASTM STP 982. Philadelphia, PA: American Society for Testing and Materials; 1988. p. 139–148.
[19] Vasudevan AK, Sadananda K, Louat N. A review of crack closure, fatigue crack threshold and related phenomena. Materials
Science and Engineering 1994;A188:1–22.
[20] Kujawski D. Enhanced model of partial crack closure for correlation of R-ratio effects in aluminum alloys. International
Journal of Fatigue 2001;23:95–102.
00
[21] McEvily AJ, Ishihara S. On the dependence of the rate of fatigue crack growth on the sa (2a) parameter. International
[22] Ishihara S, McEvily AJ. A coaxing effect in the small fatigue crack growth regime. Scripta Materialia 1999;40:617–22.
[23] McEvily AJ, Ishihara S, Mutoh Y. On the number of overload induced delay cycles as a function of thickness. International
[24] Ishihara S, McEvily AJ. Analysis of short fatigue crack growth in cast aluminum alloys. International Journal of Fatigue
2002;24:1169–74.
[25] Li XY, Cui WC, Zhang WM. A modified constitutive relation fatigue crack growth. Journal of Ship Mechanics 2006;10:54–61
[in Chinese].
[26] Matos de PFP, McEvily AJ, Moreira PMGP, de Castro PMST. Analysis of the effect of cold-working of rivet holes on the
fatigue life of an aluminum alloy. International Journal of Fatigue 2007;29:575–86.
[27] Newman Jr JC. Crack opening stress equation for fatigue crack growth. International Journal of Fatigue 1984;24(3):131–5.
[28] Newman Jr JC, Crews JH, Bigelow CA, Dawicke DS. Variations of a global constraint factor in cracked bodies under tension
and bending loads. ASTM STP. 1995;1224:21–42.
[29] Schijve J. Fatigue of structures and materials in the 20th century and the state of the art. International Journal of Fatigue
2003;25:679–702.
[30] Zhou C, Cui WC. Determination of fatigue crack growth rate using existing data. Shipbuilding of China 2003;44:74–9
[in Chinese].
[31] Cui WC. Relation between crack growth rate curve and S-N curve for metal fatigue. Journal of Ship Mechanics 2002;6(6):93–106.
[32] Dowling NE. Mechanical behavior of materials-engineering methods for deformation, fracture, and fatigue. 3rd ed. United
States of America: Pearson Prentice Hall; 2007.
[33] Ma JF, Lu GZ. An analytical model for determining DKeff. Mechanical Science and Technology 1999;18(3):383–9.
[34] Farahmand B, Nikbin K. Predicting fracture and fatigue crack growth properties using tensile properties. Engineering
Fracture Mechanics 2008;75:2144–55.
[35] Kang JY, Choi BI, Lee HJ. Application of artificial neural network for predicting plain strain fracture toughness using tensile
test results. Fatigue and Fracture of Engineering Materials and Structures 2006;29:321–9.
[36] Tiryakioglu M, Hudak D. On estimating the fracture toughness of Al–7Si–Mg alloys by sharp-notch tensile test results.
Materials Science and Engineering 2008;A498:501–3.
[37] Firrao D, Matteis P, Scavino G, Ubertalli G, Ienco MG, Pinasco MR, et al. Relationships between tensile and fracture
mechanics properties and fatigue properties of large plastic mould steel blocks. Materials Science and Engineering 2007;
A468–A470:193–200.
[38] Paris PC, Tada H, Donald JK. Service load fatigue damage – a historical perspective. International Journal of Fatigue 1999;
21:S35–46.
[39] Newman Jr JC, Phillips EP, Everett RA. Fatigue analyses under constant and variable amplitude loading using small-crack
theory. NASA/TM-1999–209329, ARL-TR-2001.
[40] Zhao W, Zhang JX, Jiang YY. A study of fatigue crack growth of 7075-T651 aluminum alloy. International Journal of Fatigue
2008;30:1169–80.
[41] Jha SK, Ravichandran KS. Effect of mean stress (stress ratio) and aging on fatigue-crack growth in a metastable beta
titanium alloy, Ti–10V–2Fe–3Al. Metallurgical and Materials Transactions 2000;3:703–14.
[42] Dinda S, Kujawski D. Correlation and prediction of fatigue crack growth for different R-ratios using Kmax and DKþ
parameters. Engineering Fracture Mechanics 2004;71:1779–90.
[43] Taheri F, Trask D, Pegg N. Experimental and analytical investigation of fatigue characteristics of 350WT steel under
constant and variable amplitude loadings. Marine Structures 2003;16:69–91.
[44] Oh HK. Determination of fracture toughness by means of the uniaxial tensile test. Journal of Materials Processing Tech-
nology 1995;54:372–4.
[45] Bulloch JH. Near threshold fatigue crack propagation behavior of CrMoV turbine steel. Theoretical and Applied Fracture
Mechanics 1995;23:89–101.
[46] Nykänen T, Marquis G, Björk T. A simplified fatigue assessment method for high quality welded cruciform joints. Inter-
national Journal of Fatigue 2009;31(1):79–87.
[47] Lin XB, Smith RA. Finite element modelling of fatigue crack growth of surface cracked plates part II: crack shape change.
Engineering Fracture Mechanics 1999;63:523–40.
[48] Newman Jr JC, Raju IS. Analyses of surface cracks in finite plates under tension or bending loads. NASA Technical Paper
1578; 1979.
[49] Mahmoud MA. Quantitative prediction of growth patterns of surface fatigue cracks in tension analysis. Engineering
[50] Mahmoud MA. Growth patterns of surface fatigue cracks under cyclic bending: a quantitative analysis. Engineering
[51] Wu SX. Shape change of surface during fatigue growth. Engineering Fracture Mechanics 1985;22:897–913.
[52] Engesvik KM, Moan T. Probabilistic analysis of the uncertainty in the fatigue capacity of welded joints. Engineering
Fracture Mechanics 1983;18(4):743–62.
[53] Bowness D, Lee MMK. Prediction of weld toe magnification factors for semi-elliptical cracks in T-butt joint. International
[54] Han Y, Huang XP, Zhang Y, Cui WC. A comparative study of simplified SIF calculations of surface cracks at weld toe. Journal
of Ship Mechanics 2005;9(3):87–96.
[55] NIMS. Data sheets on fatigue properties of non-load-carrying cruciform welded joints of SM50B rolled steel for welded
structure. Effect of specimen size. Available from: http://tsuge.nims.go.jp/.1979.
[56] Bao H, McEvily AJ. On plane stress–plane strain interactions in fatigue crack growth. International Journal of Fatigue 1998;
20(6):441–8.
[57] Ray A, Patanker P. Fatigue crack growth under variable amplitude loading: part I – model formulation in state space
setting. Applied Mathematical Modelling 2001;25:979–94.
[58] Huang XP, Moan T, Cui WC. An engineering model of fatigue crack growth under variable amplitude loading. International
[59] Huang XP, Wang AQ, Cui WC, Bian RG. The fatigue crack growth under compressive to compressive fluctuating loading. In:
Proceedings of the ASME 2010 29th international conference on ocean, offshore and arctic engineering, OMAE 2010, June
6–11, Shanghai, China.
[60] Fleck NA, Shin CS, Smith RA. Fatigue crack growth under compressive loading. Engineering Fracture Mechanics 1985;
21(1):173–85.
[61] Suresh S. Crack initiation in cyclic compression and its applications. Engineering Fracture Mechanics 1985;21(3):453–63.
[62] Pippan R. The growth of short cracks under cyclic compression. Fatigue and Fracture of Engineering Materials and
Structures 1987;9(5):319–28.
[63] Hermann R. Fatigue crack growth in ductile materials under cyclic compressive loading. Fatigue and Fracture of Engi-
neering Materials and Structures 1994;17(1):93–103.
[64] Kasaba K, Sano T, Kudo S, Shoji T, Katagiri K, Sato T. Fatigue crack growth under compressive loading. Journal of Nuclear
Materials 1998;258–263:2059–63.
[65] Tan JML, Fitzpatrick ME, Edwards L. Stress intensity factors for through-thickness cracks in a wide plate: derivation and
application to arbitrary weld residual stress fields. Engineering Fracture Mechanics 2007;74:2030–54.
[66] Khan SU, Alderliesten RC, Rans CD, Benedictus R. Application of a modified wheeler model to predict fatigue crack growth
in fibre metal laminates under variable amplitude loading. Engineering Fracture Mechanics 2010;77(9):1400–16.
[67] Borrego LP, Ferreira JM, Costa JM. Partial crack closure under block loading. International Journal of Fatigue 2008;30:
1787–96.
[68] Liu Q, Wang F, Huang XP, Cui WC. Three dimensional FE analysis of the plastic zone size near the crack tip. Journal of Ship
Mechanics 2006;10(5):90–9 [in Chinese].
[69] Voorwald HJC, Torres MAS. Modeling of fatigue crack growth following overloads. International Journal of Fatigue 1991;
13(5):423–7.
[70] Zhang J, He XD, Du SY. A simple engineering approach in the prediction of the effect of stress ratio on fatigue threshold.
International Journal of Fatigue 2003;25:935–8.
[71] Schijve J, Skorupa M, Skorupa A, Machniewicz T, Gruszczynski P. Fatigue crack growth in the aluminium alloy D16 under
constant and variable amplitude loading. International Journal of Fatigue 2004;26:1–15.
[72] Huang XP. Numerical simulation and spectrum experimental study of fatigue behavior of deep-sea structures. Shanghai
technical report of Shanghai Jiangtong University; 2007 [in Chinese].
View publication stats

Cui Wang Huang 2011 1

Uploaded by

Copyright:

Available Formats

Cui Wang Huang 2011 1

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

Cui Wang Huang 2011 1

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Basic Concepts of Fracture Mechanics

Chapter · January 2014

Weicheng Cui X.P. Huang

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

Marine Structures 24 (2011) 153–181

Contents lists available at ScienceDirect

A uniﬁed fatigue life prediction method for marine

154 W. Cui et al. / Marine Structures 24 (2011) 153–181

3tot Total error of fatigue life prediction

W. Cui et al. / Marine Structures 24 (2011) 153–181 155

156 W. Cui et al. / Marine Structures 24 (2011) 153–181

W. Cui et al. / Marine Structures 24 (2011) 153–181 157

2. The improvement of the fatigue crack growth rate used in UFLP

2.1. General ideas of a uniﬁed fatigue life prediction (UFLP) method

158 W. Cui et al. / Marine Structures 24 (2011) 153–181

Fig. 1. The ﬂow diagram of the UFLP method.

2.2. The improvement of the fatigue crack growth rate

W. Cui et al. / Marine Structures 24 (2011) 153–181 159

And the solution of Eq. (5) is

160 W. Cui et al. / Marine Structures 24 (2011) 153–181

3. Engineering approaches to determine the parameters in the improved model

3.1. General methods to estimate the model parameters

W. Cui et al. / Marine Structures 24 (2011) 153–181 161

A value of q can be determined by minimizing the following error function,

3.1.1. Estimation method from crack growth rate data

3.1.2. Estimation method from a–N curve

162 W. Cui et al. / Marine Structures 24 (2011) 153–181

Fig. 2. Flow diagram of the nonlinear least-square ﬁtting method.

3.1.3. Estimation method from S–N curve

Fig. 3. Schematic diagram of genetic algorithm.

W. Cui et al. / Marine Structures 24 (2011) 153–181 163

3.1.4. Estimation method from 3–N curve

equal to each other, namely,

164 W. Cui et al. / Marine Structures 24 (2011) 153–181

0.1 R=0.1, experimental data

1E-4 R=0.7, predicted curve

W. Cui et al. / Marine Structures 24 (2011) 153–181 165

0.1 R=0.1, experimental data

0.1 R=0.1, experimental data

R=0.7, experimental data

166 W. Cui et al. / Marine Structures 24 (2011) 153–181

1E-6 R=0.1, experim ental data

gðuÞ ¼ 1:331 þ 0:706u2 þ 0:155u4 þ 0:013u6 (25)

R=0.1, experim ental data

R=-1, experim ental data

W. Cui et al. / Marine Structures 24 (2011) 153–181 167

1E-4 R=0.1, experimental data

R=0.05, experimental data

1E-6 R=0.5, predicted curve

168 W. Cui et al. / Marine Structures 24 (2011) 153–181

R=0.75, predicted curve

W. Cui et al. / Marine Structures 24 (2011) 153–181 169

4. Capabilities of the uniﬁed crack growth rate model

4.1. Specimen thickness effect

170 W. Cui et al. / Marine Structures 24 (2011) 153–181

4.2. Compressive to compressive loading effect

sY (MPa) sb (MPa) R T (mm) q ( ) l (mm) a0 (mm)

W. Cui et al. / Marine Structures 24 (2011) 153–181 171

sY (MPa) sb (MPa) R T (mm) q ( ) l (mm) a0 (mm)