You are on page 1of 11

Journal of Constructional Steel Research 62 (2006) 1199–1209

www.elsevier.com/locate/jcsr

Fatigue reliability of welded steel structures


M.K. Chryssanthopoulos ∗ , T.D. Righiniotis
School of Engineering, University of Surrey, Guildford, Surrey, UK

Abstract

In general, two different approaches to the formulation of the fatigue limit state are considered, the first based on S–N lines in combination
with Miner’s damage accumulation rule, and the second based on fracture mechanics crack growth models and failure criteria. Often, the two
approaches are used sequentially, with S–N being used at the design or preliminary assessment stage and fracture mechanics for more refined
remaining life or inspection and repair estimates. However, it is essential to link the results, and the decisions made, at the design and assessment
stages, and it is therefore important to develop compatible methodologies for using these two approaches in tandem. In doing so, it is essential
to understand and quantify different uncertainty sources and how they might affect the robustness of the results obtained, and the subsequent
decisions made about the structure. The objective of this paper is to highlight parts of recent research at the University of Surrey on the fatigue
assessment of steel bridges. The work includes the development of a probabilistic fracture mechanics methodology for the prediction of fatigue
reliability, using up-to-date crack growth and fracture assessment criteria and incorporating information on inspection and subsequent management
actions.
c 2006 Elsevier Ltd. All rights reserved.

Keywords: Crack growth; Fatigue; Fracture mechanics; Inspection; Steel bridges; Structural reliability

1. Introduction particular resistance and load modelling issues but also the
first papers dealing with a complete methodology for fatigue
Over the last twenty years or so, probabilistic methods for reliability evaluation were being produced, including updating
the assessment of fatigue reliability have attracted significant following inspection and repair [1–5]. In the following years,
attention. In the civil engineering field, much of the research, this approach was adopted by offshore operators for estimating
from the mid-80s onwards, was directed towards applications remaining fatigue life, and for determining inspection plans, of
in offshore structures, in particular tubular joints subject ageing structures [6–8].
to stochastic loading. This effort was aided considerably In the past decade, considerable interest has arisen in
by progress in experimental techniques associated with the adapting and implementing these techniques for applications
measurement of crack growth data under laboratory conditions, in metallic bridges, thus focusing on fatigue details found in
and the development of technology aimed at measuring cracks girders and plated structures subjected to traffic loading [9–13].
in actual structures. The characteristics of bridge live loading being substantially
At the same time, advances in probabilistic methods, different from wave loading on offshore structures, led to
especially the concerted effort in developing structural revised formulations for the reliability problem. In parallel with
reliability methods for damage accumulation problems under these developments, improved methods for fatigue and fracture
time-varying loads, made it possible to cast fatigue assessment assessment were actively being pursued for other structures,
problems in a reliability format. As a result, by the late 80s, not e.g. nuclear plants [14] and ships [15].
only had a large number of publications appeared addressing An additional factor, contributing to an increased interest
in fatigue design and assessment, has been the flurry of
activity associated with the development of a new generation
∗ Corresponding address: Department of Civil Engineering, School of
of structural codes, both at national and international level;
Engineering, University of Surrey, GU2 7XH, Guildford, Surrey, UK. Tel.: +44
1483 686 632; fax: +44 1483 450 984. for example, a new European standard for fatigue design [16]
E-mail address: mkchry@surrey.ac.uk (M.K. Chryssanthopoulos). has been prepared, whilst other documents have been updated

c 2006 Elsevier Ltd. All rights reserved.


0143-974X/$ - see front matter
doi:10.1016/j.jcsr.2006.06.007
1200 M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209

and extended in order to incorporate recent developments in to present a case study, pertaining to welded bridge details, in
fracture mechanics assessment methods [17]. The underlying which the proposed procedures are implemented and utilised
philosophy in these regulatory bodies, as well as amongst in support of decision making. As will become evident, much
owners and operators, is increasingly focusing on the need work has been, and is still being, carried out in this area
to introduce probabilistic concepts for fatigue life prediction. stemming from different industrial sectors and applications.
Thus, the development of probabilistic fatigue models, and their Thus, it is considered essential to sift through and process
testing and validation through examples and case studies, could information from experiments and field observations, as well
considerably enhance available guidance documents. In this as to integrate and consolidate the procedures to be followed in
respect, the effort of the Joint Committee of Structural Safety fatigue reliability analysis.
in developing the Probabilistic Model Code [18] is of particular
note. 2. S–N approach
In general, two different approaches to the formulation of
the fatigue limit state may be considered, the first based on An S–N curve is a relation between the stress range under
S–N curves in combination with Miner’s damage accumulation constant amplitude loading and the number of stress cycles to
rule, and the second based on fracture mechanics crack growth failure. The standard S–N curve can be expressed in the form
models and associated failure criteria. Often, as can be seen of:
in some of the references cited above, the two approaches are
N Sm = A (1)
used sequentially, with S–N being used at the ‘design’ stage
and fracture mechanics at the ‘assessment’ stage, in other words where N is the number of stress cycles to failure at a
for new and existing structures respectively. In the former case, constant amplitude stress range S, A and m are the material
the purpose of a fatigue analysis is to determine a design life, parameters. Sometimes a model with two segments is used,
associated with a target reliability, whereas in the latter the having parameters A1 and m 1 , A2 and m 2 . The stress range
objective is to determine inspection intervals or time to repair, level at which the two curves intersect is defined by A1 S0m 1 =
once more linked to target reliabilities. A2 S0m 2 .
It is often desirable to link the results, and hence the Many steels subjected to pure constant amplitude loading
decisions, at the ‘design’ and ‘assessment’ stages. Thus, it is in inert environments exhibit a fatigue limit, i.e. a stress level
important to develop compatible methodologies for using these below which fatigue failure appears to never occur. However, it
two approaches in tandem. Although the majority of engineers is generally accepted that even infrequent overloads (i.e. stress
working in the above mentioned industries are more familiar cycles that exceed the fatigue limit value) may lead to fatigue
with the S–N , rather than the fracture mechanics, approach for damage even though the vast majority of stress cycles are below
fatigue analysis, the latter is increasingly gaining ground as the fatigue limit. In essence, a fatigue limit no longer exists and
fitness-for-purpose criteria are becoming popular with owners every stress cycle is treated as damaging, as determined from
and regulators. the S–N curve(s) [19].
The fatigue process may be regarded as comprising three There are many sources of uncertainty in the fatigue process
stages; crack initiation, crack propagation and final failure. A and its analysis. Wirsching [20] has produced an itemised
fatigue analysis based on S–N curves, the latter being derived list, which includes the fatigue process itself, the extrapolation
from standard fatigue tests, typically lumps all three stages from laboratory test specimens and procedures to details in
into one, though the definition of failure within this context is real structures, the loading conditions, the local environment
not always clear. On the other hand, a fatigue analysis based (temperature, presence of water/humidity etc.), the dynamic
on fracture mechanics is concerned primarily with the second effects, as well as the stress analysis methods used to obtain
stage, though it can be extended to include final failure through estimates of local stress ranges from globally applied forces and
the introduction of appropriate limit state criteria related to displacements.
fracture resistance (which could be expressed in terms of a The uncertainty associated with S–N curves is typically as-
critical crack size). The extent to which a fracture mechanics sessed from laboratory tests on nominally identical specimens
approach can provide comparable information on fatigue life under constant amplitude loading. In a typical fatigue test, the
with that derived from S–N curves will depend, in part, on stress level (i.e. the independent variable) is specified and the
the number of load cycles expended during the initiation stage. cycles to failure (i.e. the dependent variable) are recorded. Un-
A common assumption in fatigue analysis of welded joints der these conditions, it has been observed that the distribution
is that the initiation stage is negligibly small compared to of ln N for a fixed value of S exhibits a variation which is in-
the propagation stage. This is because fatigue cracks develop dependent of S (at least until test results at very low stress lev-
from small defects introduced right from the outset in areas of els are considered), whereas the mean value of ln N varies lin-
stress concentration. The weld toe is considered as a critical early with ln S. In general, the lognormal distribution provides
area in which such cracks are often to be found. Clearly, this a better fit to N than other candidate distributions such as the
assumption needs to be evaluated for the specific prevailing Weibull distribution, though there is no apparent physical or
conditions. mathematical reason for this [20]. Assuming then that in Eq.
The objective of this paper is to highlight the key factors (1) the parameter m is deterministic and that the uncertainty is
that need to be considered in fatigue reliability analysis, and lumped into the second parameter A, it is easily shown that if
M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209 1201

the variable N follows a lognormal distribution with a coeffi- factors. However, as can be evidenced from the literature, it is
cient of variation equal to C N then A is also lognormal with a difficult to establish rigid rules regarding these issues; it is thus
coefficient of variation C A = C N . More general treatment of essential in any analysis, whether deterministic or probabilistic,
the uncertainty in ln S– ln N lines has been undertaken [21] al- to be aware of the idealisations and simplifications adopted,
lowing for joint probabilistic modelling of the material param- and the implications for uncertainty modelling. Straub [25] has
eters m and A, but in many practical applications the simpler reviewed studies pertaining to marine structures under wave
approach outlined above has been adopted. loads and has tabulated uncertainty levels for a factor, which
It should be emphasised that statistical analysis of standard acts as a multiplier on calculated stress ranges.
fatigue tests is a challenging task, as it requires careful In general, structures experience variable loading during
consideration of many factors, such as the merging of data from their life. Thus, fatigue loading is characterised by the number
different sources, influence of different failure criteria adopted of stress cycles and the magnitude of stress range for each
by different laboratories (e.g. ‘visible’ or through-thickness cycle. Moreover, fatigue damage is quantified in terms of the
crack), influence of specimen preparation as well as the contrast Palmgren-Miner rule. According to this rule, each load cycle
between laboratory and field conditions, etc. In recent years, causes fatigue damage proportional to the inverse of the fatigue
effort has been directed towards the development of the new life at that stress range amplitude. Therefore, letting Si be
Eurocode for fatigue [16]. The above, and many other, factors the stress range of the ith cycle, damage may be defined in
were assessed in producing S–N lines for different fatigue accordance with the Palmgren-Miner rule as
details [22]. n
The stress ranges in a fatigue analysis using the S–N
X 1
Dn = (2)
approach should be obtained bearing in mind the location i=1
N (Si )
of the detail within the structure as well as the basis of the
S–N curves used. For example, chord to brace connections in where N (Si ) is the number of cycles to failure at stress level Si .
offshore tubular joints are assessed using the so-called T-curve, It can readily be shown that by introducing a simple linear S–N
for which the ‘hot-spot’ stress range has to be determined. curve, see Eq. (1), Eq. (2) can be written as
The ‘hot-spot’ stress is related to the nominal stress in the 1 Xn
1
brace through empirical stress concentration factors, usually Dn = Sim = ψ L (t) (3)
developed from limited parametric studies using experiments A i=1 A
or finite element analyses of typical geometries. On the other where ψ L (t) may be referred to as the ‘fatigue loading’. In a
hand, for many other fatigue details the corresponding S–N deterministic format the Palmgren-Miner rule simply states that
lines take into account the local stress concentrations created by damage increments, expressed as life fractions, are additive and
the joints themselves or by the weld profile [23]. The relevant independent of sequencing, and that fatigue failure is expected
stress ranges to be used could then be determined from the when such fractions sum to unity (i.e. Dn = 1). Experience
nominal stresses present in the vicinity of the joint. However, if shows that this sum at failure is subject to considerable
the detail is situated in a region of stress concentration resulting uncertainty; for example, analyses undertaken in relation to
from the overall shape of the structure, this must be accounted offshore structures have revealed median values between 0.8
for in estimating the nominal stress. The different approaches, and 1.2, and even wider variations are reported for bridge
adopted in fatigue assessment of welded joints, have been details. Wirsching [20] has suggested that, in the absence of
reviewed by Fricke [24]. specific data for the problem in hand, a lognormal distribution
As far as probabilistic modelling is concerned, the important with a median value of unity and a coefficient of variation of
message from the above is that modelling uncertainties are 0.30 may be adopted for the damage at failure.
involved both in the method for calculating nominal stresses On the other hand, ψ L (t) may be described through
in the vicinity of a fatigue sensitive detail and in the a probability function, whose nature depends on the
estimation of stress concentration factors due to gross and characteristics of the underlying loading process. For some
local discontinuities, or due to deviations from an ideal shape. models the expectation of the mth moment of the arbitrary
The mean value and coefficient of variation adopted for such point-in-time distribution of the stress range is sufficient. In
model uncertainty variables would vary depending on the general, the derivation of the appropriate probabilistic model
type of global analysis procedures (e.g. finite element vs. for ψ L (t) is application specific.
simplified theory), on the characteristics of the local geometry
and of the detail itself, as well as on the factors included in 3. Fracture mechanics approach
the derivation of the relevant S–N curve. Typically, effects
associated with residual stresses, weld profiling, environment 3.1. Crack growth relationships
(e.g. air vs. water), through thickness stress variation (for
a reference plate thickness) and fabrication tolerances are Crack growth is a process not fully understood at the atomic
included in the derivation of any particular S–N curve, whereas level but for which macroscopic observations have been made,
effects associated with post-weld treatment (e.g. grinding), and models have been developed. Crack growth can occur under
difference in plate thickness (from a reference value), global cyclic loading, however, in a hostile chemical environment, it
discontinuities etc. are accounted for through modification can even take place in the presence of statically applied loads.
1202 M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209

The former is known as fatigue while the latter is described as


environmentally assisted crack growth. Under cyclic loading in
a hostile environment, both types may occur in combination.
Engineering analysis of crack growth can be undertaken
using relationships between the stress intensity factor (SIF),
which characterises the stress conditions at the crack tip, and
the associated crack growth rates. The underpinning (linear
elastic fracture mechanics—LEFM) theory, which is used to
determine SIFs for cracked bodies is based on stress analysis of
an isotropic and homogeneous solid containing an ideally sharp
crack. For cracks of the order of a few mm or greater, micro-
structural material (e.g. grain boundaries) and geometrical
(e.g. crack kinks) features are in effect assumed to occur on
such a small scale that only average behaviour needs to be
considered [23]. If, however, the crack is sufficiently small (of
the order of a few microns), it can interact with the micro-
structure and special treatment might be needed. Here, the
distinction between small and short cracks is relevant. For
the former, all the crack dimensions are similar or smaller
than the dimension of the greatest micro-structural feature, Fig. 1. Schematic representation of a typical crack growth curve and its
whereas for the latter, one crack dimension is large compared approximations.
to the microstructure. There is considerable research interest
in the behaviour and modelling of small and short cracks [23, summarise the treatment of uncertainty in relation to Paris
26], which, from an engineering point of view is focused on type models. This is believed to arise from inhomogeneities
determining conditions under which such cracks may grow in the metal, the type of crack measurement method, the
and reach the level at which LEFM becomes applicable. As measurement procedures (e.g. whether da/dn is determined
mentioned in the introduction, this can also be viewed in terms by averaging over predetermined intervals in n or a), and the
of the three stages of fatigue life of engineering components and statistical procedures adopted for parameter estimation. The
structures. The ensuing discussion focuses on the modelling of variability can also be viewed as having a component present
crack propagation stage, assuming that LEFM is applicable; within a particular test and a component manifested when
initial defect distributions are considered in the case studies results from several nominally identical tests are combined.
presented. The approaches that have been developed to deal with random
The most widely used fatigue crack growth model based fatigue crack propagation include: treatment of C and m as
on LEFM, commonly known as Paris’s Law, was proposed by random variables (individually or as a correlated pair) [2,21];
Paris and Erdogan [27] and can be written as: formulation as a first-passage problem focusing particularly on
the effect of averaging that is necessarily present in measuring
da
= C(1K )m (4) crack increments [28]; Markov process description [29]; and
dn the introduction of a lognormal random process as a multiplier
where a is the crack depth, da/dn is the instantaneous (median) of the right-hand side of the Paris law [30,31].
crack propagation rate and 1K is the alternating SIF at the Unfortunately, comparisons between the different ap-
crack tip, generated from an applied nominal stress range; C proaches on realistic structural details are virtually absent from
and m are material constants which can be determined from the literature. However, most of the engineering applications
experiments on simple specimens in which a small initial crack in offshore and bridge structures have been undertaken with the
is introduced and then propagated through the application of simplest of the approaches mentioned above, i.e. treatment of C
constant amplitude cyclic loading. Clearly, if the above model and m as random variables. One notable exception is the fatigue
is correct, then a plot of ln(da/dn) against ln(1K ) would result analysis of welded miter gates in river locks, which has been
in a straight line. As shown in many experimental studies on carried out using the lognormal random process approach [31].
a wide range of metals, this is the case for the middle range In the JCSS Probabilistic Model Code [18], the suggested
of growth rates, e.g. typically for a range between 10−5 and approach is to treat m as a deterministic parameter and to
10−3 mm/cycle. At higher rates, the model underestimates model C as a lognormal random variable. It is emphasised
actual rates, whereas for lower values it overestimates them. that this should be viewed as an operational model for which
Thus, fatigue crack growth over a wide range of crack growth a probabilistic fatigue analysis under different load processes
rates spanning several orders of magnitude invariably results in can be undertaken using standard reliability techniques, e.g.
the well-known ‘sigmoidal’ curve, the middle portion of which FORM/SORM or Monte Carlo simulation.
may be described accurately by Eq. (4) (see Fig. 1). As mentioned previously, the Paris model is only strictly
Before more detailed models that attempt to capture the valid in the central region of crack growth (Region II,
entire crack growth behaviour are presented, it is helpful to see Fig. 1), although in fatigue assessments of engineering
M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209 1203

components it is sometimes extended to the near-threshold 3.2. Stress intensity factor


region (Region I in Fig. 1). In many studies the effect of a cut-
off or threshold value, 1K thr , below which the crack is assumed The discussion has so far focused on crack growth rates.
to be non-propagating is also included. Thus, Eq. (4) is replaced However, the power law relationship between crack growth rate
by (da/dn = damage) and crack driving force (1K = loading)
implied by Eq. (4) or any other alternative crack growth model
da (see for example Eq. (6)) suggests that a closer examination of
= C(1K )m for 1K > 1K thr . (5)
dn K or 1K is warranted. Within the context of LEFM, the stress
intensity factor K for a given applied ‘far field’ stress Sa is
This equation forms the basis for the JCSS commonly given as:
adopted probabilistic crack growth model, with m assumed

deterministic and C and 1K thr treated as random variables [18]. K = Sa Y Mk πa (7)
Eq. (5) is conservative and often quite acceptable, particularly
in benign environments, such as inert gases or vacuum. where Y is the “geometry factor”. The Y -factor depends on the
generic geometry of the cracked joint (i.e. excluding any local
However, more refined descriptions may be necessary when a
stress concentration effects), the nature of load distribution
large part of the fatigue damage is due to a very large number
(e.g. bending or membrane) and the crack size normalised with
of low-amplitude stress cycles, or the environmental response
respect to other relevant dimensions (e.g. thickness). The factor
of the material deviates significantly from linear behaviour.
Mk accounts for the effect of a crack being in the immediate
As highlighted by King et al. [32], both these factors vicinity of a stress concentration, such as a weld toe, a notch
become important in the fatigue assessment of offshore details. or a hole. It therefore depends on the local geometry and is a
Furthermore, the presence of a very large number of low- function of crack size and loading. Solutions for Y -factors are
amplitude stress cycles is a key feature of the traffic load available for a number of standard load cases (tension, bending,
on highway bridges [12,33,34]. A re-appraisal of available torsion) and generic geometries in handbooks (for example
data for fatigue crack growth rates of steels in air and Refs. [37–39]). Solutions for Mk may be found in the literature
seawater environments [32,35] has led to the development and [40,41] or can be developed through FE analysis.
characterisation of a bi-linear relationship to represent crack The stress Sa should be determined from the applied loading
growth in the near threshold region (Region I) as well as in the in the vicinity of the crack (but not influenced by it). The
traditional intermediate (Paris) region (Region II). Eq. (5) may through-thickness distribution of this stress is often assumed
thus be replaced by to consist of ‘membrane’ and ‘bending’ components and is
characterised in terms of the ratio of the bending stress to the
da total stress, called ‘degree-of-bending’ [42–44]. Eq. (7), may be
= C1 (1K )m 1 for 1K thr < 1K ≤ 1K tr
dN (6) re-cast as follows
da √
= C2 (1K )m 2 for 1K tr < 1K . K = (Sam Ym Mkm + Sab Yb Mkb ) πa (8)
dN
The above bi-linear crack growth model is also included in the where the subscripts ‘m’ and ‘b’ refer to membrane and bending
JCSS document [18], with the information on variability of C1 components respectively.
and C2 taken from the work by King et al. [32]. It is worth In addition to the stresses due to the applied loading, residual
noting that the coefficient of variation of C1 is significantly stresses Sres may also be present, e.g. in the case of welded
higher than the corresponding value for C2 (or C in the case of joints. The distribution of residual stresses is usually obtained
the simpler linear model). This reflects the higher uncertainty from experimental measurements on typical joint and weld
involved in both the actual crack growth process, as well as its details; the uncertainty associated with such distributions is
measurement, when the rates are small. Finally, it should be large both in terms of physical and statistical components.
Note that the presence of residual stresses should also be taken
emphasised that the mean values adopted for the C parameters
into account in determining the stress ratio R, which in turn
should properly reflect the influence of the environmental
determines the appropriate values of C and m parameters in
conditions (e.g. air vs. seawater), and the stress ratio R (defined
Paris type models.
as the ratio between minimum and maximum applied stresses),
as there are important differences in mean values depending on
3.3. Crack growth modelling
the prevailing conditions.
For either Eq. (5) or (6), the variability associated with the For welded joints it is often observed that micro-cracks
threshold value for the stress intensity factor range, 1K thr , initiate from surface-breaking defects at the toe of the weld.
should be included. As reported by King et al. [32] the These micro-cracks tend to coalesce to form a single, dominant
coefficient of variation for 1K thr is quite high; furthermore, its fatigue crack of roughly semi-elliptical shape. Hence, semi-
mean value appears to be dependent on the R ratio. A lognormal elliptical cracks in plated structures are of interest in many
distribution with a CoV of 0.4 [36] has been proposed. Note that practical applications (see Fig. 2). In this case, two crack
in the case of freely corroding steels 1K thr tends to zero. dimensions, the depth a and the half-length at the surface c,
1204 M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209

where G(a) and G(c) are ‘threshold correction factors’, given


by [46]
E[S m ]∞
Sthr (a)
G(a) = (13)
E[S m ]∞
0
and
E[S m ]∞
Sthr (c)
G(c) = (14)
Fig. 2. Butt welded plate containing a semi-elliptical toe crack. E[S m ]∞
0
where Sthr is the stress range for either crack dimension (a or c)
become relevant both of which are functions of the fatigue corresponding to 1K thr , which can be obtained from Eq. (7)
loading process. or (8). The values of G(a) and G(c) are in the range from
For each principal direction of crack growth, a Paris type zero to unity, with the latter corresponding to the case where
expression may be formulated thus all cycles contribute to fatigue damage. 1K effA and 1K effB
are ‘effective stress intensity factor ranges’, which account for
da residual compressive stresses and stress ratio effects [25,45].
= C(1K A )m for 1K A > 1K thr
dn They can be defined through ‘effective stress intensity factor
(9)
dc ratios’ U (a) and U (c), according to [4]
= C(1K B )m for 1K B > 1K thr
dn 1K effA = U (a)1K A (15)
where the first expression relates to point A and growth in the and
depth direction, whereas the second expression relates to point
B and growth in the length direction (Fig. 2). For each of these 1K effB = U (c)1K B . (16)
two points, the stress intensity factor range, 1K A and 1K B , is However, as pointed out by Straub [25], although more
given by sophisticated crack growth relationships may lead to a better
comparison between predicted and measured values, they
1K A = SY A Mk A πa/Q
p
(10) also lead to difficulties with regard to calibration and model
1K B = SY B Mk B π c/Q
p
uncertainty characterisation, due to the additional parameters
that are introduced. There is considerable debate in the fatigue
where S is the applied stress range, Y A , Y B are geometry and fracture community regarding the significance of crack
factors, Mk A , Mk B are stress magnification factors for points closure effects for different applications.
A and B respectively, and Q is the elliptic shape factor which Integration of the first differential equation (12) from
may be approximated by [44] an initial crack size a0 to a crack size a(t) after time t
corresponding to a number of stress cycles n results in
 a 1.65
Q = 1 + 1.464 . (11) 1 a(t)
Z
dx n
c
X
m (π x/Q )m/2 = Sim = ψ L (t) (17)
C a0 G a Uam Yam Mka a i=1
Once Eqs. (10) and (11) are substituted into (9), a pair of
coupled differential equations is obtained. With the exception of For brevity, the subscript ‘a’ is used here to denote
the material parameters (C and m) and the applied stress range the functional dependence of the variable on crack size.
S, all other terms are a function of the crack size (a, c), which Comparison of Eq. (17) with Eq. (3) reveals the similarity of
clearly changes during the fatigue loading process. the two approaches adopted for fatigue damage accumulation
Under variable loading and considering the uncertainties problems, i.e. S–N and fracture mechanics.
In general, the above integral is evaluated through an
associated with it, some further modifications may be
incremental numerical procedure, which involves sub-division
introduced to Eq. (9). The first pertains to the effect of
of the interval between a(t) and a0 . At each step, following
the threshold value, 1K thr , which essentially acts as a filter
the evaluation of Eq. (17), the second crack dimension c is
distinguishing between damaging and non-damaging cycles
computed as well, using the relationship that can be obtained
in the loading process. A second factor is related to crack
by combining the two differential equations, i.e.
closure [45], which deals with the possibility that a crack may
G(c) 1K effB m
 
not propagate (i.e. remain ‘closed’) due to local compressive dc
= . (18)
stresses at the crack tip, even though the applied stresses are da G(a) 1K effA
tensile. As a result, Eq. (9) can be re-written as
When the life of a fatigue sensitive detail is required, the
da upper bound of the integral in Eq. (17) may be taken as
= C(1K effA )m G(a) the critical value ac determined through the application of a
dn (12)
dc fracture criterion or some other requirement, e.g. a maximum
= C(1K effB )m G(c) permissible crack size.
dn
M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209 1205

work of Lassen [50], in order to distinguish between crack


initiation and crack propagation under LEFM conditions. As
mentioned earlier, in fatigue analysis of welded joints it is
commonly assumed that the initiation stage is negligibly small
compared to the propagation stage, and hence the methods
presented in the previous section can be adopted for an
evaluation of the entire fatigue life.

Fig. 3. Buried elliptical crack located in a plate of thickness t (The direction of 4. Limit state functions
the applied loading is perpendicular to the indicated plane).
4.1. S–N approach
3.4. Initial defect size
For the S–N approach, the limit state function may be given
As can be seen from Eq. (17), the fracture mechanics by
approach considers the growth of a crack from an initial size to ξ
ψ L (t)

a critical size. The initial crack size distribution is one of the key Tp
g( X̃ , t) = 1 − D = 1 − m
Bscf m
Bglob (19)
inputs in determining fatigue lives through the LEFM methods A TB
presented in the preceding section. It should be emphasised
that the application of LEFM implies that the stress intensity where X̃ is the vector of random variables, t is any point in
factor concept is meaningful in describing crack growth [47]. time, ξ is the thickness correction factor [51], 1 is Miner’s
As reported by Straub [25], recent studies seem to indicate damage sum at failure, T p is the actual thickness of the plate
that 0.1 mm is a reasonable lower bound for the application and TB is the reference plate thickness. In Eq. (19), Bscf and
of LEFM in common metallic structures. Bglob are model uncertainties associated with global and local
In engineering structures, many possible mechanisms lead stress analysis. In the work of Shetty and Baker [3–5] pertaining
to the presence of cracks, arising from material processing or to welded tubular joints for offshore applications, a lognormal
manufacturing factors. Cracks may be classified as buried or distribution is proposed with a mean of unity and a coefficient
surface, and for analysis purposes such cracks are generally of variation of 0.20 and 0.10 for Bscf and Bglob respectively.
idealised as elliptical, semi-elliptical or quarter-elliptical. Fig. 3 Straub [25] reports higher CoV values from published literature
shows schematically an elliptical crack that lies below the pertaining to ships, offshore structures and FPSOs, and mean
surface, and is characterised through the dimensions, a, c and values in the range from 0.7 to 1.0.
s. Surface defects (i.e. s = 0) are usually more dangerous than
4.2. Fracture mechanics approach
embedded defects, particularly in welded structures, because
(a) they are associated with higher Y factors, (b) they are often Using the fracture mechanics approach, the limit state
located at stress concentrations, which further increase Y and function can be formulated as
(c) they are oriented normal to the principal stress. Moreover,
their presence is caused by limitations in workmanship and g( X̃ , t) = ac − a(t) (20)
their initial size tends to be larger than that of embedded
where ac is a limiting crack depth (for example plate thickness)
cracks for the normal range of component thickness. Intentional
and a(t) is the crack depth after a service exposure of time
or unintentional lack of penetration in butt welds form an
t. Starting from an initial crack depth of a0 , the crack depth
exception to this last observation, however, experience supports
a(t) after time t can be calculated using the crack growth
the view that many fatigue cracks in welded structures tend to
propagation models highlighted above. In this case, a(t) is a
grow from an initial surface defect [2,48].
function of random variables such as initial defect size, fatigue
Meaningful statistics on crack frequency, size and location
material properties, uncertainties in service loading, model
are difficult to obtain, due to measuring and sampling issues.
uncertainties, etc. (see Eq. (17)).
That said, in the past twenty-five years there have been
Alternatively, in terms of a fatigue resistance function and
several studies aimed at improving our understanding and
a fatigue loading function, the limit state function can also be
providing quantitative information on relevant crack parameters
expressed as
for different detail geometries used in a range of structural
1 ac
Z
applications (ships, offshore platforms, nuclear etc.). Crack dx
g( X̃ , t) = m Y m M m (π x/Q )m/2
size distributions are typically modelled by exponential or C a0 G a Uam Bsif a ka a
lognormal distributions, whereas the occurrence rate is assumed m
− Bglob m
Bscf ψ L (t) (21)
to be Poisson distributed. Information on crack aspect ratios
(a/c) is very limited, though there have been some attempts where Bsif is a model uncertainty associated with the estimation
to determine the sensitivity of fatigue reliability towards this of the stress intensity factor; in general, this will be influenced
parameter under simplified conditions (e.g. c and a are either by the methods used for determining both Y and Mk factors.
independent or fully correlated). A review of published data on At present, it is suggested to model Bsif as a lognormal variable
initial defect distributions can be found in [2,25,49]. Straub [25] with a mean of unity and coefficient of variation in the range
also addresses the issue of small crack growth, based on the from 0.07 to 0.20. Lower variability may be appropriate if the
1206 M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209

factors are computed using finite element models and/or weight


function techniques.
4.3. Fatigue crack growth approach with fracture resistance
model
The maximum crack size, ac , that can be sustained by the
component will, in general, depend on the material’s resistance
to fracture i.e. its fracture toughness K mat . The limit state
function with respect to fracture can be defined as

g( X̃ , t) = f (K r , L r ) (22)
where K r is the fracture ratio, L r is a measure of the proximity
to plastic collapse and f (.) is an appropriate interaction
criterion, as for example given in BS7910 [17]. In this approach
failure occurs when a stress cycle occurs that causes the residual
load bearing capacity of the cross section to be exceeded. The Fig. 4. Comparison between the BS 5400 [58] mean and 97.7% probability of
residual load bearing capacity depends on the actual crack size survival lines for class E and their FM for a butt weld counterparts. Also shown
and the interaction between plastic resistance and the brittle in the figure are the results presented in [56] and [57].
fracture. The quantities K r and L r are defined as follows [17]:
5.1. S–N based FM model calibration
K s + K res Sref
Kr = + ρ; Lr = (23) As was previously discussed, one of the greatest uncertain-
K mat Sy
ties related to the fatigue process arises from the initial crack
where K mat is the material fracture toughness, K s is the stress size distribution. This distribution will be a function of the man-
intensity factor for maximum applied stress, K res is the stress ufacturing process and the type of welded detail in question.
intensity factor for residual stresses, Sref (=S Bglob Bscf ) is the Since models describing the fatigue crack growth process (dis-
net section stress (function of the crack size), S y is the yield tributions and their parameters for C, m, K mat etc.) are fairly
stress and ρ is a factor, which accounts for the interaction well established (for example [17,32,52]), it is expedient to
between primary (loading) and secondary (residual) stresses. use the appropriate S–N curve, or alternatively published S–N
The crack size a may be determined from the fatigue data, to calibrate the FM model in terms of the initial defect
crack growth models described above. Different residual stress size. This approach has been used in the past in the offshore
profiles for various cracked geometries and restraint conditions industry [6]. Note that for FM modelling as discussed in Sec-
may be assumed but their use implies that K res would need tion 3.3, whereby a crack possesses both depth and length, this
to be evaluated using either finite element techniques or calibration would have to be carried out in terms of both a0 and
the weight function method. A simplified (and conservative) c0 , or, alternatively, for a0 and (a/c)0 .
method would be to approximate the residual stress field via The limit state function (Eq. (21)) may be readily modified
a linear stress field subtended from the surface and crack tip to cater for the constant amplitude load case and the statistics of
stress values. The corresponding K res can then be obtained the fatigue life at different stress ranges obtained from Monte
by superposition of the tensile and bending solutions for the Carlo simulation. Stress intensity factors for the butt-welded
geometry in question. Further details of this approximation are detail are presented in [53]. Crack growth was modelled based
given by Tada et al. [37]. If the bending solution is not known, on the bi-linear model discussed in Section 3.1 (see Eq. (6))
in which case the previously mentioned procedure cannot be coupled with a fracture resistance model (see Section 4.3).
applied, the residual stress field may be assumed to be uniform. Distribution types and their parameters used for C, K mat etc.
This approach will, in general, yield very conservative results as well as the relevant detail dimensions may be found in [12,
for deep cracks and less conservative results for shallow cracks. 53]. Fig. 4 depicts the results of this type of analysis for a butt-
Clearly, if this approach is followed, the randomness in the welded plate. Also shown in this figure are the results reported
material fracture toughness needs to be considered. Several in [56] and the scatter band reported in [57].
models for fracture toughness have been developed. For The FM results (mean and mean − 2 std.dev.) presented
example, two [11,52,53], three [54] and four [55]-parameter in Fig. 4 were generated by assuming a0 and (a/c)0 log-
Weibull distributions have been used to describe fracture normally distributed with means 0.2, 0.01 and coefficients of
toughness. variation (CoV) of 0.2 and 0.2, respectively. Although these
5. Case studies values are not unique and other distribution types with different
mean–CoV combinations for a0 and (a/c)0 could have equally
To highlight some of the issues discussed previously, a been used to yield similar results, the proposed values compare
number of case studies are presented in this section. The fatigue well with published results in the literature (see [53] for further
reliability problem is here considered in terms of the FM discussion).
approach as this is applied to a butt-welded bridge detail (see It has to be noted that in the absence of specific data on
Fig. 2). initial defect sizes pertinent to the welded detail in question,
M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209 1207

Fig. 5. Basic reliability curve for a butt-welded detail. Fig. 6. Reliability curve for a butt-welded detail following multiple inspections
(no detection).
S–N based calibration of the FM model discussed here forms
a crucial step in a fatigue reliability analysis. Naturally, good
where the bar indicates a conditional event, Ti is the time
correlation between the experimentally based S–N data and
the FM results increases confidence when considering variable at which the inspection takes place and ai is the previously
amplitude loading. mentioned subspace, which is defined as a(Ti ) ≥ ad in the case
of crack detection and a(Ti ) < ad otherwise. The parameter ad
5.2. Fatigue reliability based on inspections is the detectable crack size, which is randomised through the
inspection method’s probability of detection (PoD).
Calibration of the FM model allows the treatment of the Using the modified limit state function of Eq. (24) and
variable amplitude case to proceed. Here the butt-welded detail assuming inspections using the ACFM method whose PoD
is assumed to be located at the mid-span of a 10 m bridge curve is given by [59]
girder. The loading characteristics were obtained by traversing
the BS 5400 Pt 10 Table 11 of vehicles [58] over the span 1
PoD(ad ) = 1 − (25)
and obtaining the relevant bending moment history assuming 1 + e2.2 (a d − 0.7)1.9
an annual vehicle flow of 106 . Bending moment ranges were
derived using the rainflow method and converted into stress leads to the results shown in Fig. 6. The scenario considered
ranges on the assumption that the detail’s S–N design fatigue in Fig. 6 involves multiple inspections at the times at which
life is 120 years. The associated basic reliability curve, obtained the reliability is violated (years 36, 47, 62 and 80) under the
using the limit state function of Eq. (20) is shown in Fig. 5. As assumption of no detection during these four inspections. Note
can be seen in this figure, the detail’s failure probability P f that the first branch of the ‘saw-tooth’ curve is the initial part
at year 120 of continuous operation is approximately 1.5 × of the curve shown in Fig. 5 and that each segment of the curve
10−2 . This compares well with the S–N anticipated failure represents a separate analysis. Fig. 6 demonstrates that in order
probability of 2.3 × 10−2 , which is not surprising in view of to maintain a P f = 10−3 for this detail, using ACFM and
the results presented in Fig. 4. A similar curve could have been assuming that at no stage cracks are detected, a minimum of
obtained using an S–N based probabilistic methodology (see four inspections would need to be carried out throughout the
Section 4.1). However, the great utility of the FM approach is detail’s life.
demonstrated when inspections and/or other types of invasive However, it has to be emphasised that different inspection
action are considered. To illustrate this point it is here assumed scenarios could have been considered if for example the bridge
that following the design and construction of the bridge, the owner had decided that inspections at those specific times were
owner has decided to impose a more stringent requirement on
unsuitable. It also has to be noted that the results of this type of
the detail’s reliability, namely P f = 10−3 . As can be seen
reliability analysis are rather sensitive to the inspection method
in Fig. 5, this requirement is violated at year 36 of operation.
used as well as the inspection outcome. This last point is
Clearly, an inspection is warranted at that time, which may or
highlighted in Fig. 7, whereby the fourth inspection at year
may not lead to crack detection.
80 results in crack detection. Since the failure probabilities
Modifications to the basic reliability problem to cater for a
become unacceptably high leading to a failure probability of
series of n inspections are introduced by considering a subset
of the fatigue crack sizes a(t) to reflect the outcome of these 0.26 at year 120, a decision has to be made regarding the
inspections. Accordingly, the limit state function (Eq. (20)) detail’s operation. Methods that can be used by the bridge
becomes owner in order to maintain the desirable reliability include
" # repair and/or imposition of load restrictions. The way some of
n
\ these techniques may be quantified within an FM-based fatigue
g( X̃ , t) = ac − a(t) a (T ) (24)

i=1 i i reliability analysis is discussed in the next case study.
1208 M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209

Fig. 8. Reliability curve for a butt-welded detail following an initial crack


Fig. 7. Reliability curve for a butt-welded detail following multiple inspections
detection and repair.
(no detection–detection).

5.3. Fatigue reliability based on inspections and invasive play an important role in fatigue and fracture behaviour and to
actions focus attention on those which are of particular relevance to
the case in hand. Fatigue problems, and decisions, are multi-
Once again, for this case study the butt-welded detail is faceted and it is unwise to opt for a ‘one-size fits all’ approach.
selected. Here, however, it is assumed that, following the initial Procedures have been identified, tools exist, required data are
inspection at year 36 a crack is detected. The results of this slowly becoming available, and case studies can be consulted,
analysis are shown in Fig. 8 in the form of triangles. As can be but the need for well thought out benchmarking of analytical
seen, the detail’s fatigue reliability becomes unacceptably low results against experimental databases, careful probabilistic
almost immediately following crack detection (Pf = 2.1×10−2 modelling and clear understanding of the limitations associated
at year 36). It is here assumed that in the light of these results with any approach should not be underestimated.
the bridge owner decides to repair the crack by re-welding
the toe. This type of operation is assumed here to restore the Acknowledgments
crack depth to its as-welded state, in other words, a1 (T1 =
36) = a0 . Subsequent analysis results in the curve denoted in The first author would like to express his sincere thanks
Fig. 8 as ‘Detection & Re-welding’, which leads to a reliability to Prof. Ton Vrouwenvelder, TNO/TU Delft with whom he
violation at year 64. Thus, an inspection would have to be has had the opportunity to collaborate on probabilistic fatigue
undertaken at that time. The remaining two curves reflect the models within the Joint Committee of Structural Safety (JCSS).
two possible outcomes of this inspection i.e. detection and no The case studies presented in this paper have been produced
detection at year 64. Note that in this case, even the latter in the course of a project sponsored by the Highways Agency,
curve leads to violation of the reliability criterion at around year in co-operation with Flint and Neill and TWI. The opinions
104 of operation. At that point the bridge owner may consider expressed in the paper are those of the authors and do not
the option of imposing load restrictions on the bridge and/or necessarily reflect those of the sponsors.
preventative repair even though a crack is not detected. The
imposition of load restrictions and a different type of repair, References
namely weld toe grinding, were considered for a different type
of welded detail in [13]. [1] Madsen HO, Skjong RK, Tallin AG, Kirkemo A. Probabilistic fatigue
crack growth analysis of offshore structures, with reliability updating
through inspection. In: Proc. SNAME marine structural reliability
6. Concluding remarks
symposium. 1987.
[2] Kirkemo F. Applications of probabilistic fracture mechanics to offshore
The application of structural reliability techniques to structures. Appl Mec Rev 1988;41(2):61–84.
fatigue related problems in welded steel structures has [3] Shetty NK, Baker MJ. Fatigue reliability of tubular joints in offshore
occupied intensively the engineering community for the past structures: Fatigue loading. In: Proc. 9th offshore mechanics and arctic
twenty years. An appropriate probabilistic framework, and engineering conf. ASME; 1990. p. 33–40.
[4] Shetty NK, Baker MJ. Fatigue reliability of tubular joints in offshore
associated methodology, has by now been established and
structures: Crack propagation model. In: Proc. 9th offshore mechanics and
used in the context of assessment, inspection and repair. arctic engineering conf. ASME; 1990. p. 223–30.
However, it is clear, from reviewing the literature, that [5] Shetty NK, Baker MJ. Fatigue reliability of tubular joints in offshore
considerable differences can still be found on key assumptions structures: Reliability analysis. In: Proc. 9th offshore mechanics and arctic
regarding fatigue and fracture behaviour, the parameters of the engineering conf. ASME; 1990. p. 231–9.
[6] Pedersen C, Nielsen JA, Riber JP, Madsen HO, Krenk S. Reliability based
probabilistic models and the idealised processes adopted for
inspection planning for the Tyra field. In: Proc. 11th OMAE offshore
representing random loading. It is of paramount importance mechanics and arctic engineering conf. ASME; 1992. p. 255–63.
to examine carefully for each application, be it an offshore [7] Hovde GO, Moan T. Fatigue reliability of TLP tether systems considering
node, a ship’s deck or a bridge girder detail, the factors that In: effect of inspection and repair. In: 7th BOSS conference. 1994.
M.K. Chryssanthopoulos, T.D. Righiniotis / Journal of Constructional Steel Research 62 (2006) 1199–1209 1209

[8] Goyet J, Paygnard JC, Maroini A, Faber MH. Optimal inspection and [35] King RN. A review of fatigue crack growth rates in air and seawater.
repair planning: Case studies using IMREL software. In: Proc. 13th Offshore technology report OTR 511. London: Health and Safety
OMAE offshore mechanics and arctic engineering conf. ASME; 1994. Executive; 1998.
[9] Zhao Z, Haldar A, Breen FL. Fatigue reliability evaluation of steel [36] Austen I. Measurement of fatigue crack threshold values for use in design.
bridges. J Struct Eng (ASCE) 1994;120(5):1624–42. BSC Report SH/EN/9708/2/83/B. British Steel Corporation. 1983.
[10] Cremona C. Reliability updating of welded joints damaged by fatigue. Int [37] Tada H, Paris PC, Irwin GR. The stress analysis of cracks handbook. New
J Fatigue 1996;18(8):567–75. York: ASME; 2000.
[11] Lukić M, Cremona C. Probabilistic assessment of welded joints versus [38] Murakami Y. Stress intensity factors handbook. Oxford: Pergamon Press;
fatigue and fracture. J Struct Eng (ASCE) 2001;127(2):211–8. 1987.
[12] Righiniotis TD, Chryssanthopoulos MK. Fatigue and fracture simulation [39] Rooke DP, Cartwright DJ. Compedium of stress intensity factors. London:
of welded bridge details through a bi-linear crack growth law. Struct Saf HMSO; 1976.
2004;2:141–58. [40] Hobbacher A. Stress intensity factors of welded joints. Eng Fract Mech
[13] Righiniotis TD. Influence of management actions on fatigue reliability of 1993;46(2):173–82.
a welded joint. Int J Fatigue 2004;26(3):231–9. [41] Bowness D, Lee MMK. Weld toe magnification factors for semi-elliptical
[14] Bullough R, Green VR, Tomkins B, Wilson R, Wintle JB. A review of cracks in T-butt joints. Offshore technology report. OTO 1999 014. HSE.
methods and applications of reliability analysis for structural integrity London. 1999.
assessment of UK nuclear plant. Int J Press Vessel Pip 1999;76: [42] Newman JC, Raju IS. An empirical stress intensity factor equation for the
909–19. surface crack. Eng Fract Mech 1981;15.
[15] Guedes Soares C, Garbatov Y. Reliability of maintained ship hull girders [43] Mattheck C, Morawietz P, Munz D. Stress intensity factor at the surface
subject to corrosion and fatigue. Struct Saf 1998;20(3):201–19. and at the deepest point of semi-elliptical surface crack in plates under
[16] EN 1993-1-9. Eurocode 3: Design of Steel Structures, Part 1-9: Fatigue. stress gradients. Int J Fract 1983;23:201–12.
Brussels: CEN/TC250; 2005. [44] Newman JC, Raju IS. Stress-intensity factor equations for cracks in
[17] British Standards Institution. BS7910: Guide on methods for assessing the three-dimensional finite bodies subjected to tension and bending loads.
acceptability of flaws in fusion welded structures. London. 2000. In: Atluri SN, editor. Computational methods in the mechanics of fracture.
[18] Joint Committee on Structural Safety. The probabilistic model code. New York: Elsevier North Holland; 1986.
Internet Publication. http://www.jcss.ethz.ch, 2001. [45] Elber W. The significance of fatigue crack closure. In: Damage tolerance
[19] Dowling NE. Mechanical behavior of materials. 2nd edition. New Jersey: in aircraft structures. American Society for Testing and Materials. ASTM
Prentice Hall; 1999.
STP 486. 1971.
[20] Wirsching PH. Probabilistic fatigue analysis. In: Sundararajan C, editor.
[46] Wirsching PH, Ortiz K, Chen YN. Fracture mechanics fatigue model in a
Probabilistic structural mechanics handbook. New York: Chapman and
reliability format. In: Proc. 6th offshore mechanics and arctic engineering
Hall; 1995.
conf. ASME; 1987.
[21] Madsen HO, Krenk S, Lind NC. Methods of structural safety. Englewood
[47] Schijve J. Fatigue of structures and materials. Kluwer Academic
Cliffs (NJ): Prentice-Hall Inc; 1986.
Publishers; 2001.
[22] Background Document for prEN 1993-1-9. RWTH. Institute of Steel
[48] Maddox S. Fatigue strength of welded structures. Cambridge: Abington
Construction. 1st Draft. 2002.
Publishers; 1991.
[23] Suresh S. Fatigue of materials. Cambridge: Cambridge University Press;
[49] Harris DO. Probabilistic fracture mechanics. In: Sundararajan C, editor.
1991.
Probabilistic structural mechanics handbook. New York: Chapman and
[24] Fricke W. Fatigue analysis of welded joints: State of development. Mar
Hall; 1995.
Struct 2003;16(3):185–200.
[50] Lassen T. Experimental investigation and stochastic modelling of
[25] Straub D. Generic approaches to risk based inspection planning for steel
the fatigue behaviour of welded steel joints. Ph.D. thesis. Denmark:
structures. In: IBK-Bericht Nr. 284. Institute of Structural Engineering.
University of Aalborg; 1997.
ETH Zurich. 2004.
[26] Taylor D. Geometrical effects in fatigue: A unifying theoretical treatment. [51] Guerney TR. Fatigue of welded structures. Cambridge: Cambridge
Int J Fatigue 1999;21:413–20. University Press; 1979.
[27] Paris PC, Erdogan F. A critical analysis of crack propagation laws. J Basic [52] Burdekin FM, Hamour W. Partial safety factors for SINTAP procedure.
Eng (ASME) 1963;85:528–34. Offshore technology report. OTO 2000 020. HSE. London. 2000.
[28] Ditlevsen O. Random fatigue crack growth—a first passage problem. Eng [53] Righiniotis TD, Chryssanthopoulos MK. Probabilistic fatigue analysis
Fract Mech 1986;23(2):467–77. under constant amplitude loading. J Constr Steel Res 2003;59(5):867–86.
[29] Lin YK, Yang JN. A stochastic theory of fatigue crack propagation. AIAA [54] Wallin K. The scatter in KIC results. Eng Fract Mech 1984;19(6):
J 1985;23(1). 1085–93.
[30] Zheng R, Ellingwood BR. Stochastic fatigue crack growth in steel [55] Kunin B. A new type of extreme value distributions. Eng Fract Mechs
structures subject to random loading. Struct Saf 1998;20:303–23. 1997;58(5–6):557–70.
[31] McAllister TP, Ellingwood BR. Evaluation of crack growth in miter [56] Keating PB, Fisher JW. Evaluation of fatigue tests and design criteria
gate weldments using stochastic fracture mechanics. Struct Saf 2001;23: on welded details. Washington: Transportation Research Board; NCHRP
445–65. 286; 1986.
[32] King RN, Stacey A, Sharp JV. A review of fatigue crack growth rates for [57] Gurney TR. The basis of the new fatigue design rules for welded joints.
offshore steels in air and seawater environments. In: Proc. 15th OMAE In: Rockey KC, Evans HR, editors. The design of steel bridges. 1982.
offshore mechanics and arctic engineering conf. ASME; 1996. p. 475–85.
[33] Fisher JW, Mertz DR, Zhong A. Steel bridge members under variable [58] British Standards Institution. BS 5400: Part 10. Steel, concrete and
amplitude long life fatigue loading. Washington: Transportation Research composite bridges. Code of practice for fatigue. London. 1980.
Board; NCHRP 267. 1983. [59] TWI Ltd. NDT capability for the detection and sizing of surface-breaking
[34] Fisher JW, Nussbaumer A, Keating PB, Yen BT. Resistance of welded fatigue cracks in welded steel bridges. Rep no 12539/1/01. In: Fatigue
details under variable amplitude long life fatigue loading. Washington: assessment of steel bridge components with existing cracks. Highways
Transportation Research Board; NCHRP 354. 1993. Agency Contract 3/284. 2001.

You might also like