Lecture 12.

1: Basic Introduction to Fatigue

OBJECTIVE/SCOPE:
To summarize the main factors affecting fatigue strength, as opposed to static strength, of
welded joints and to illustrate the method of carrying out a fatigue check.
PREREQUISITES
Lecture
11.1.2:
Introduction
Lecture 11.2.1: Generalities on Welded Connections

to

Connection

Design

RELATED LECTURES
Lecture
12.12:
Determination
of
Lecture 12.13: Fracture Mechanics Applied to Fatigue

Stress

Intensity

Factors

SUMMARY
This lecture gives an explanation of the mechanism of fatigue and the influence of welding on
that mechanism. It summaries the primary factors affecting fatigue strength and introduces S-N
Curves. The classification of fatigue details is presented and important details reviewed. The
calculation of stress range is summarised. The principal types of fatigue loading and the bases
for their calculation are presented with an introduction to cycle counting and damage
calculations for mixed amplitude loading.
NOTATION
a design weld strength parameter
DsR stress range
DsD non-propagating stress, i.e. the constant amplitude stress range below which cracks will not
grow
N endurance number of cycles.
1. INTRODUCTION
1.1 Nature of Fatigue
Fatigue is the mechanism whereby cracks grow in a structure. Growth only occurs under
fluctuating stress. Final failure generally occurs in regions of tensile stress when the reduced
cross-section becomes insufficient to carry the peak load without rupture. Whilst the loading on
the structure is stationary the crack does not grow under normal service temperatures. Many
structures, such as building frames, do not experience sufficient fluctuating stress to give rise to
fatigue problems. Others do, such as bridges, cranes, and offshore structures, where the live
loading is a higher proportion of the total load.
1.2 How Welds Fatigue
1

In welded steel structures, fatigue cracks will almost certainly start to grow from welds, rather
than other details, because:

Most welding processes leave minute metallurgical discontinuities from which cracks
may grow. As a result, the initiation period, which is normally needed to start a crack in
plain wrought material, is either very short or no-existent. Cracks therefore spend most
of their life propagating, i.e. getting longer.

Most structural welds have a rough profile. Sharp changes of direction generally occur
at the toes of butt welds and at the toes and roots of fillet welds, see Figure 1. These
points cause local stress concentrations of the type shown in Figure 2. Small
discontinuities close to these points will therefore react as though they are in a more
highly stressed member and grow faster.

1.3 Crack Growth History

The study of fracture mechanisms shows that the growth rate of a crack is proportional to the
square root of its length, given the same stress fluctuation and degree of stress concentration.
For this reason fatigue cracks spend most of their life as very small cracks which are hard to
detect. Only in the last stages of life does the crack start to cause a significant loss of crosssection area, as shown in Figure 3. This behaviour poses problems for in-service inspection of
structures.

2. FATIGUE STRENGTH
2.1 Definition of Fatigue Strength and Fatigue Life
The fatigue strength of a welded component is defined as the stress range (R) which
fluctuating at constant amplitude, causes failure of the component after a specified number of
cycles (N). The stress range is the difference between the maximum and minimum points in the
cycle, see Figure 4. The number of cycles to failure is known as the endurance or fatigue life.

2.2 Primary Factors Affecting Fatigue Life

For practical design purposes there are two main factors which affect the fatigue life of a detail,
namely:

The stress range (R) at the location of crack initiation. There are special rules for
calculating this range.

The fatigue strength of the detail. This strength is primarily a function of the geometry
and is defined by the parameter 'a', which varies from joint to joint.

The fatigue life (N), or endurance, in number of cycles to failure can be calculated from the
expression:
3

where m is a constant, which for most welded details is equal to 3. Predictions of life are
therefore particularly sensitive to accuracy of stress prediction.
2.3 S-N Curve
The expression linking N and Rm can be plotted on a logarithmic scale as a straight line,
Equation (2), and is referred to as an S-N curve. An example is shown in Figure 5. The
relationship holds for a wide range of endurance. It is limited at the low endurance end by static
failure when the ultimate material strength is exceeded. At endurances exceeding about 5-10
million cycles the stress ranges are generally too small to permit propagation under constant
amplitude loading. This limit is called the non-propagating stress (D). Below this stress
range cracks will not grow.

For design purposes it is usual to use design S-N curves which give fatigue strengths about 25%
below the mean failure values, as shown in Figure 5. 'a' is used to define these lines.
2.4 Effect of Mean Stress
In non-welded details the endurance is reduced as the mean stress becomes more tensile. In
welded details the endurance is not usually reduced in those circumstances. This behaviour
occurs because the weld shrinkage stresses (or residual stresses), which are locked into the weld
regions at fabrication, often attain tensile yield. The crack cannot distinguish between applied
and residual stress. Thus, for the purposes of design, the S-N curve always assumes the worst,
i.e. that the maximum stress in the cycle is at yield point in tension. It is particularly important
to appreciate this point as it means that fatigue cracks can grow in parts of members which are
nominally 'in compression'.
2.5 Effect of Mechanical Strength
4

The rate of crack growth is not significantly affected by variations in proof stress or ultimate
tensile strength within the range of low alloy steels used for general structural purposes. These
properties only affect the initiation period, which, being negligible in welds, results in little
influence on fatigue life. This behaviour contrasts with the fatigue of non-welded details where
increased mechanical strength generally results in improved fatigue strength, as shown in Figure
6.

3. CLASSIFICATION OF DETAILS
3.1 Detail Classes
The fatigue strength parameter (K2) of different welded details varies according to the severity
of the stress concentration effect. As there are a wide variety of detail in common use, details
with similar K2 values are grouped together into a single detail class and given a single K 2 value.
This data has been obtained from constant amplitude fatigue tests on simple specimens
containing different welded detail types. For the most commonly used details, it has been found
convenient to divide the results into fourteen main classes. The classes are:

As shown in Figure 7, these classes can be plotted as a family of S-N curves. The difference in
stress range between neighbouring curves is usually between 15 and 20%.

The above table has been taken from Eurocode 3 [1]. It does not include S-N data for
unstiffened hollow tubular joints.
3.2 Detail Types
There are usually a number of detail types within each class. Each type has a very specific
description which defines the geometry both microscopically and macroscopically. The main
features that affect the detail type, and hence its classification, are:
6

Form of the member:

e.g. plate, rolled section, reinforcing bar.

Location of anticipated crack initiation:

The location must be defined with respect to the direction of stress fluctuation. A given
structural joint may contain more than one potential initiation site, in which case the joint may
fall into two or more detail types.

Leading dimensions:

e.g. weld shape, size of component, proximity of edges, abruptness of change of cross-section.

Fabrication requirements:

e.g. type of weld process, any grinding smooth of particular parts of the joint.

Inspection requirements:

Special inspection procedures may be required on higher class details to ensure that detrimental
welding defects are not present.
It should be noted that if fatigue is critical in the design, the extra controls on fabrication
incurred by the last two requirements may increase the total cost significantly above that for
purely static strength.
Examples of different types of welded detail and their classes are shown in Eurocode 3: Part 1.1
[1].
3.3 Commonly Used Detail Types
Figure 8 shows some of the most important details to look out for in welded steelwork.

They are:

Load carrying fillet welds and partial penetration butt welds. These details are category
36 for failure starting at the root and propagating through the throat.

Welded attachments on edges. They are category 45. Note that the attachment weld may
not be transferring any stress. Failure is from the weld toe into the member.

Ends of long flat plates, e.g. cover plates are category 50.

Most short attachments in the stress direction are category 80 or 71 as long as they are
not at an edge.

Transverse full penetration butt welds can range from category 12,5 to 36 depending on
how they are made.

Long continuous welds on site welded structures are found to be category 100.

It should be borne in mind that most potential fatigue sites on welded structures are found to be
category 80 or below.
8

4. STRESS PARAMETERS FOR FATIGUE

4.1 Stress Area
The stress areas are essentially similar to those used for static design. For a crack starting at a
weld toe, the cross-section of the member through which propagation occurs is used. For a crack
starting at the root, and propagating through the weld throat, the minimum throat area is used, as
shown in Figure 8a.
4.2 Calculation of Stress Range
The force fluctuation in the structure must be calculated elastically. No plastic redistribution is
permitted.
The stress on the critical cross-section is the principal stress at the position of the weld toe (in
the case of weld toes cracks). Simple elastic theory is used assuming plane sections remain
plane, see Figure 9. The effect of the local stress concentration caused by the weld profile is
ignored as this is already catered for by the parameter 'd' which determines the weld class.

In the case of throat failures, the vector sum of the stresses on the weld throat at the position of
highest vector stress along the weld is used, as in static design.
Exceptions to these rules occur in the case of unstiffened joints between slender members such
as tubes. In this case the stress parameter is the Hot Spot Stress. This stress is calculated at the
point of expected crack initiation, taking into account the true elastic deformation in the joint,
i.e. not assuming plane sections to remain plane.
4.3 Effects of Geometrical Stress Concentrations and Other Effects
Where a member has large changes in cross-section, e.g. at access holes, there will be regions of
stress concentration due to the change of geometry. In static design the stresses are based on the
net area as plastic redistribution will normally reduce these peaks at ultimate load. With fatigue
this is not so, and if there is a welded detail in the area of the geometrical stress raiser the true
stress must be used, as shown in Figure 10.

4.4 Secondary Effects

Similarly any secondary effects, such as those due to joint fixity in latticed structures, and shear
lag and other distortional effects in slender beams, are allowed for in calculating the stresses.
5. LOADINGS FOR FATIGUE
5.1 Types of Loading
Examples of structures and the loads which can cause fatigue are:
Bridges: Commercial vehicles, goods trains
Cranes: Lifting, rolling and inertial loads
Offshore structures: Waves
Slender chimneys: Wind gusting
The designer's objective is to anticipate the sequence of service loading throughout the
structure's life. The magnitude of the peak load, which is vital for static design purposes, is
generally of little concern as it only represents one cycle in millions. For example, highway
bridge girders may experience 100 million significant cycles in their lifetime. The sequence is
important because it affects the stress range, particularly if the structure is loaded by more than
one independent load system.
For convenience, loadings are usually simplified into a load spectrum, which defines a series of
bands of constant load levels, and the number of times that each band is experienced, as shown
in Figure 11.

10

Slender structures, with natural frequencies low enough to respond to the loading frequency,
may suffer dynamic magnification of stress. This magnification can shorten the life
considerably.
A useful source of information on fatigue loading is Eurocode 1 [2].
5.2 Cycle Counting
In practice most stress histories in real structures are of the variable amplitude type, shown in
Figure 12, as opposed to the constant amplitude shown in Figure 4. Such histories pose a
problem in defining the number and amplitude of the cycles.

The first step is to break the sequence into a stress spectrum as shown in Figure 12 using a cycle
counting method. There are various methods in use. The two most used are the Rainflow
Method and the Reservoir Method. The latter, which is easy to use by hand for short stress
histories, is described in Lecture 12.2. The former is more convenient for analysing long stress
histories using a computer.
6. CALCULATION OF DAMAGE
Under variable amplitude loading the life is estimated by calculation of the total damage done
by each cycle in the stress spectrum. In practice the spectrum is simplified into a manageable
number of bands, as shown in Figure 13.

11

The damage done by each band in the spectrum is defined as

where n is the required number
of cycles in the band during the design life and N is the endurance under that stress range, see
Figure 14.

If failure is to be prevented before the end of the specified design life, the Palmgren-Miner's
Rule must be compiled with. This rule states that the damage done by all bands together must
not exceed unity, i.e.:

It should be noted that, when variable amplitude loading occurs, the bands in the spectrum
with values less than D may still cause damage. Damage occurs because the larger
amplitude cycles may start to propagate the crack. Once it starts to grow lower cycles become
effective. In this case, the horizontal constant amplitude fatigue limit D shown in Figure 5,
is replaced by a sloping line with a log gradient of

7. CONCLUDING SUMMARY

12

Fatigue and static failure (whether by rupture or buckling) are dependent on very
different factors, namely:

- Fatigue depends on the whole service loading sequence (not one extreme load event).
- Fatigue of welds is not improved by better mechanical properties.
- Fatigue is very sensitive to the geometry of details.
- Fatigue requires more accurate prediction of elastic stress.
- Fatigue makes more demands on workmanship and inspection.

It is therefore important to check early in the design whether fatigue is likely to be

critical. Acceptable margins of safety against static collapse cannot be relied upon to
give adequate safety against fatigue.

Areas with a high live/dead stress ratio and low category 36 details should be checked
first. The check must cover any welded attachment to a member, however insignificant,
and not just the main structural connections. Note that this check should include welded
additions to the structure in service.

If fatigue is critical, then the choice of details will be limited. Simplicity of detail and
smoothness of stress path should be sought.

Be prepared for fatigue critical structures to cost more.

8. REFERENCES
[1] Eurocode 3: "Design of Steel Structures": ENV1993-1-1: Part 1.1: General Rules and Rules
for Buildings, CEN 1992.
[2] Eurocode 1: "Basis of Design and Actions on Structures", CEN (in preparation).
9. ADDITIONAL READING
1. Maddox, S.J. "Fatigue Strength of Welded Structures", Cambridge, Abington
Publishing, 1991.
2. Gurney, T. R., "Fatigue of Welded Structures", 2nd ed., Cambridge University Press,
1991.
3. Narayanan, R. (ed), "Structures Subjected to Repeated Loading", London, Elsevier
Applied Science, 1991.

13

Lecture 12.2: Advanced Introduction to

Fatigue
OBJECTIVE/SCOPE:
To introduce the main concepts and definitions regarding the fatigue process and to identify the
main factors that influence the fatigue performance of materials, components and structures.
PREREQUISITES
Lecture 12.1: Basic Introduction to Fatigue
RELATED LECTURES
SUMMARY
The physical process of the initiation of fatigue cracks in smooth and notched test specimens
under the influence of repeated loads is described and the relevance of this process for the
fatigue of real structures is discussed.
The basis of different stress cycle counting procedures is explained for variable amplitude
loading. Exceedance diagram and frequency spectrum effects are described.
1. INTRODUCTION
Fatigue is commonly referred to as a process in which damage is accumulated in a material
undergoing fluctuating loading, eventually resulting in failure even if the maximum load is well
below the elastic limit of the material. Fatigue is a process of local strength reduction that
occurs in engineering materials such as metallic alloys, polymers and composites, eg. concrete
and fibre reinforced plastics. Although the phenomenological details of the process may differ
from one material to another the following definition given by ASTM [1] encompasses fatigue
failures in all materials:
Fatigue - the process of progressive localised permanent structural change occurring in a
material subjected to conditions that produce fluctuating stresses and strains at some point or
points and that may culminate in cracks or complete fracture after a sufficient number of
fluctuations.
The important features of the process relevant to fatigue in metallic materials are indicated by
the underlined words in the definition above. Fatigue is a progressive process in which the
damage develops slowly in the early stages and accelerates very quickly towards the end. Thus
the first stage consists of a crack initiation phase, which for smooth and mildly notched parts
that are subjected to small loads cycles may occupy more than 90 percent of the life. In most
case cases the initiation process is confined to a small area, usually of high local stress, where
the damage accumulates during stressing. In adjacent parts of the components, with only
slightly lower stresses, no fatigue damage may occur and these parts thus have an infinite
fatigue life. The initiation process usually results in a number of micro-cracks that may grow
more or less independently until one crack becomes dominant through a coalescence process at
the microcracks start to interact. Under steady fatigue loading this crack grows slowly, but starts
14

to accelerate when the reduction of the cross-section increases the local stress field near the
crack front. Final failure occur as an unstable fracture when the remaining area is too small to
support the load. These stages in the fatigue process can in many cases be related to distinctive
features of the fracture surface of components that have failed under fluctuating loads, the
presence of these features can therefore be used to identify fatigue as the probable cause of
failure.
2. CHARACTERISTICS OF FATIGUE FRACTURE SURFACES
Typical fracture surfaces in mechanical components that were subjected to fatigue loads are
shown in Slide 1. One characteristic feature of the surface morphology which is evident in both
macrographs is the flat, smooth region of the surface exhibiting beach marks (also called
clamshell marks). This part represents the portion of the fracture surface over which the crack
grew in a stable, slow mode. The rougher regions, showing evidence of large plastic
deformation, is the final fracture area through which the crack progressed in an unstable mode.
The beach marks may form concentric rings that point toward the areas of initiation. The origin
of the fatigue crack may be more or less distinct. In some cases a defect may be identified as the
origin of the crack, in other cases there is no apparent reason why the crack should start at a
particular point in a fracture surface. If the critical section is at a high stress concentration
fatigue initiation may occur at many points, in contrast to the case of unnotched parts where the
crack usually grows from one point only see Figure 1. While the presence of any defects at the
origin may indicate the cause of the fatigue failure, the crack propagation area may yield some
information regarding the magnitude of the fatigue loads and also about the variation in the
loading pattern. Firstly, the relative magnitude of the areas of slow-growth and final fracture
regions give an indication of the maximum stresses and the fracture toughness of the material.
Thus, a large final fracture area for a given material indicates a high maximum load, whereas a
small area indicates that the load was lower at fracture. Similarly, for a fixed maximum stress,
the relative area corresponding to slow crack growth increases with the fracture toughness of the
material (or with the tensile strength if the final fracture is a fully ductile overload fracture).

15

Slide 1 : Typical fatigue failures in steel components.

Beach marks are formed when the crack grows intermittently and at different rates during
random variations in the loading pattern under the influence of a changing corrosive
environment. Beach marks are therefore not observed in the surfaces of fatigue specimens tested
under constant amplitude loading conditions without any start-stop periods. The average crack
growth is of the order of a few millimetres per million cycles in high cycle fatigue, and it is
clear that the distance between bands in the beach marks are not a measure of the rate of crack
advance per load cycle. However, examination by electron microscope at magnifications
between 1,000x and 30,000x may reveal characteristic surface ripples called fatigue striations,
see Slide 2. Although somewhat similar in appearance, these lines are not the beach marks
described above as one beach mark may contain thousands of striations. During constant
amplitude fatigue loading at relatively high growth rates in ductile material such as stainless
steels and aluminium alloys the striation spacing represents the crack advancement per load
cycle. However, in low stress, high cycle fatigue where the striation spacing is less than one
16

atomic spacing (- 2.5 x 10-8m) per cycle. Under these conditions the crack does not advance
simultaneously along the crack front, growth occurring instead only along some portions during
a few cycles, then arrests while growth occurs along other segments. Striations as shown in
Figure 3 are not seen if the crack grows by other mechanisms such as microvoid coalescence or,
in brittle materials, microclevage. In structural steels the crack can propagate by all three
mechanism, and striations may be difficult to observe. Slide 3 shows an example of beach
marks and striations in the fracture originating at a large defect in a welded C-Mn steel with a
yield strength of about 360Mpa.

Slide 2 : Striations in an aluminium alloy.

Slide 3 : Fatigue failures in the Alexander L Kielland platform.

17

3. NATURE OF THE FATIGUE PROCESS

From the description of the characteristics of fatigue fracture surfaces, three stages in the fatigue
process may be identified:
Stage I: Crack initiation
Stage II: Propagation of one dominant crack
Stage III: Final fracture
Fatigue cracking in metals is always associated with the accumulation of irreversible plastic
strain. The crack process which is discussed in the following applies to smooth specimens made
of ductile materials.
In high cycle fatigue the maximum stress in cyclic loads that eventually cause fatigue failure
may be well below the elastic limit of the material, and large scale plastic deformation does not
occur. However, at a free surface plastic strains may accumulate as a result of dislocation
movements. Dislocations are line defects in the lattice structure which can move and multiply
under the action of shear stresses, leaving a permanent deformation. Dislocation mobility and
hence the amount of deformation (or slip) is greater at a free surface than in the interior of
crystalline materials due to lack of constraint from grain boundaries. Grains in polycrystalline
structural metals are individually oriented in a random manner. Each grain, however, has an
ordered atomic structure giving rise to directional properties. Deformation for example, takes
place on crystallographic planes of easy slip along which dislocations can move more easily
than other planes. Since slip is controlled primarily by shear stress, slip deformation takes place
along crystallographic planes that are orientated close to 45 to the tensile stress direction. The
results of such deformation is atomic planes sliding relative to each other, resulting in a
roughening of the surface in slip bands. During further cycling slip band deformation is
intensified at the surface and extending into the interior of the grain, resulting in so-called
persistent slip bands, (PSB's). The name originated from the observation in early studies of
fatigue that slip band would reappear - "persist" - at the same location after a thin surface layer
was removed by elastopolishing. The accumulation of local plastic flow result in surface ridges
and troughs called extrusions and intrusions, respectively, Figure 2. The cohesion between the
layers in slip band is weakened by oxidation of fresh surfaces and hardening of the strained
material. At some point in this process small cracks develop in the intrusions. These
microcracks grow along slip planes, ie. a shear stress driven process. Growth in the shear mode,
called stage I crack growth extends over a few grains. During continued cycling the microcracks
in different grains coalesce resulting in one or a few dominating cracks. The stress field
associated with the dominating crack cause further growth under the primary action of
maximum principal stress; this is called stage II growth. The crack path is now essentially
perpendicular to the tensile stress axis. Crack advancement is, however, still influenced by the
crystallographic orientation of the grains and the crack grows in a zigzag path along slip planes
and cleavage planes from grain to grain, see Figure 3. Most fatigue cracks advance across grain
boundaries as indicated in Figure 3, ie. in a transcrystalline mode. However, at high
temperatures or in a corrosive environment, grain boundaries may become weaker than the
18

grain matrix, resulting in intercrystalline crack growth. The fracture surface created by stage II
crack growth are in ductile metals characterised by striations whose density and width can be
related to the applied stress level.

Since crack nucleation is related to the magnitude of stress, any stress concentration in the form
of external or internal surface flaws can marked reduce fatigue life, in particular when the
initiation phase occupies a significant portion of the total life. Thus a part with a smooth,
polished surface generally has a higher fatigue strength than one with a rough surface. Crack
initiation can also be facilitated by inclusions, which act as internal stress raisers. In ductile
materials slip band deformations at inclusions are higher than elsewhere and fatigue cracks may
initiate here unless other stress raisers dominate.
In high strength materials, notably steels and aluminium alloys, a different initiation mechanism
is often observed. In such materials, which are highly resistant to slip deformation, the interface
between the matrix and inclusion may be relatively weak, and cracks will start here if
decohesion occurs at the inclusion surface, aided by the increased stress/strain field around the
inclusion. Slide 4 shows small fatigue cracks originating at inclusions in a high strength steel.
Alternatively, a hard brittle inclusion may break and a fatigue crack may initiate at the edges of
the cleavage fracture.

Slide 4 : Fatigue crack initiation at an inclusion in a high strength steel alloy.

From the discussion above it is evidently not possible to make a clear distinction between crack
nucleation and stage I growth. "Crack initiation" is thus a rather imprecise term used to describe
a series of events leading to stage II crack. Although the initiation stage includes some crack
growth, the small scale of the crack compared with microstructural dimensions such as grain
size invalidates a fracture mechanics based analysis of this growth phase. Instead, local stresses
and strains are commonly related to material constants in prediction models used to estimate the
19

length of stage I. The material constants are normally obtained from tests on smooth specimens
subjected to stress or strain controlled cycling.
4. FATIGUE LOADING
The simplest form of stress spectrum to which a structural element may be subjected is a
sinusoidal or constant amplitude stress-time history with a constant mean load, as illustrated in
Figure 4. Since this is a loading pattern which is easily defined and simple to reproduce in the
laboratory it forms the basis for most fatigue tests. The following six parameters are used to
define a constant amplitude stress cycle:

Smax = maximum stress in the cycle

Smin = minimum stress in the cycle
Sm = mean stress in the cycle = (Smax + Smin)/2
Sa = stress amplitude = (Smax Smin)/2
AS = stress range = Smax - Smin = 2Sa
R = stress ratio = Smin/Smax
The stress cycle is uniquely defined by any two of these quantities, except combinations of
stress range and stress amplitude. Various stress patterns are shown in Figure 5, with definitions
in accordance with ISO [2] terminology.

20

The stress range is the primary parameter influencing fatigue life, with mean stress as a
secondary parameter. The stress ratio is often used as an indication of the influence of mean
loads, but the effect of a constant mean load is not the same as for a constant mean stress. The
difference between S-N curves with constant mean stress or constant R-ratio is discussed in the
section on fatigue testing.
The test frequency is needed to define a stress history, but in the fatigue of metallic materials the
frequency is not an important parameter, except at high temperatures when creep interacts with
fatigue, or when corrosion influences fatigue life. In both cases a lower test frequency results in
a shorter life.
Typical stress-time histories obtained from real structures are one shown in Figure 6. The
sequence in Figure 6a has a constant mean stress, individual stress cycles are easily identifiable,
and it necessary to evaluate this stress history in terms of stress range only. The more "random"
stress variations in Figure 10b is called a broad band process because the power density function
(a plot of energy vs. frequency) spans a wide frequency range, in contrast to the one in Figure 6a
which contains essentially one frequency. The difference is illustrated in Figure 7. The load
history in Figure 6 can be interpreted as a variation of the main load with superimposed smaller
excursions that could be caused by eg. second order vibrations or by electronic noise in the load
acquisition system. In case of true mean load variations not only the range but also the mean of
each cycle needs to be recorded in order to estimate the influence of mean load on the damage
accumulation. In both cases it is necessary to eliminate the smaller cycles since they may be
below the fatigue limit and therefore cause no fatigue damage, or because they do not represent
real load cycles. Thus a more complicated evaluation procedure is required for identifying and
counting individual major stress cycles and their associated mean stresses. Counting methods
such as the range pair, rainflow and the reservoir methods are designed to achieve this. These
procedures are described in paragraph 7.

21

5. FATIGUE LIFE DATA

The total fatigue life in terms of cycles to failure can be expressed as:
Nt = Ni + Np (1)
where Ni and Np are number of cycles spent in the initiation and propagation stages,
respectively. As noted, the two stages are distinctly different in nature and different material
parameters control their length. The life of unnotched components, for example, is dominated
by crack initiation. In sharply notched parts, however, or in parts containing crack-life defects,
eg. welded joints, the crack growth stage dominates and crack propagation data may be used in
an assessment of fatigue life using fracture mechanics analysis. Therefore different test methods
are necessary to assess the fatigue properties of these types of components.
5.1 Fatigue Strength Curves
Fatigue data for components whose lives consist of an initiation phase followed by crack
propagation are usually presented in the form of S-N curves, where applied stress S is plotted
against total cycles to failure, N (= N t). As the stress decreases, the life in cycles to failure
increases, as illustrated in Figure 8. The S-N curves for ferrous and titanium alloys exhibit a
limiting stress below which failure does not occur; this is called the fatigue or the endurance
limit. The branch point or "knee" of the curve lies normally in the 10 5 to 107 cycle range. In
aluminium and other nonferrous alloys there is no stress asymptote and a finite fatigue life
22

exists at any stress level. All materials, however, exhibit a relatively flat curve in the high-cycle
region, ie. at lives longer than about 105 cycles.

A characteristic feature of fatigue tests is the large scatter in fatigue strength data, this is
particularly evident when a number of specimens are tested at the same stress level, as
illustrated in Figure 9. Plotting the data for a given stress level along a logarithmic endurance
axis gives a distribution which can be approximated by the Gaussian (or normal) distribution,
hence endurance data are said to have a log normal distribution. Alternatively the Weibull
distribution may be used, but the choice is not important since about 200 specimens, tested at
the same stress level, are required to make a statistically significant distinction between the two
distributions. This number is about one order of magnitude larger than the quantity of specimens
that typically are available for fatigue testing at one stress level.

Assuming the life distribution to be log normal, the associated mean life curve and the standard
deviation can be used to define a design S-N curve for any desired probability of failure.
When the crack propagation stage dominates fatigue life, design data may be obtained from
crack growth curves, an example of which is shown schematically in Figure 10. The stress
intensity factor K uniquely describes the stress field near the crack tip, and is therefore used in
the design against unstable fracture. Likewise, the range of the stress intensity factor, K, may
be expected to govern fatigue crack growth. The validity of this assumption was first proved by
Paris [3], and later verified by many other researchers. The crack growth curve, which has a
sigmoidal shape, spans three regions as indicated schematically in Figure 10. In Region I the
crack growth rate drops off asymptotically as K is reduced towards a limit or threshold, Kth,
below which no crack growth takes place. Life fatigue endurance data, crack growth data show

23

considerable scatter and test results must be evaluated by statistical methods in order to derive
useful design data.
5.2 Fatigue Testing
The basis for any design methodology aimed at preventing fatigue failures is data characterising
the fatigue strength of components and structures. Fatigue testing is therefore essential for the
fatigue design process. The ideal fatigue test may be defined as a test in which an actual
structure is subjected to the service load spectrum of that structure. However, life estimates are
required before the design is finalised or details of the loading history are known. Additionally,
each structure will experience a particular load history that is unique for that structure, so many
simplifications and assumptions need to be made regarding the test stress sequence which is
going to represent the many types of service histories that can occur in practice.
Fatigue testing is therefore performed in several ways, depending on the stage the design or
production of the structure has reached or the intended use of the data. The following four main
types of tests can be identified:
1. Stress-life testing of small specimens.
2. Strain-life testing of small specimens.
3. Crack growth testing.
4. S-N tests of components.
5. Prototype testing for design validation.
The first three tests are idealised tests that produce information on the material response. The
use of the results from these tests in life prediction of components and structures requires
additional knowledge of influencing factors related to the geometry, size, surface condition and
corrosive environment. S-N tests of components are also normally standardised tests that make
life predictions more accurate compared with the three other tests because the uncertainties
regarding the influence of notches and surface conditions are reduced. Service loading or
variable amplitude testing normally requires a knowledge of the response of the actual structure
to the loading environment, and is therefore normally used only for prototype or component
testing at a late stage in the production process.
Rotating bending machines were used in the past to generate large amounts of test data in a
relatively inexpensive way. Two types are shown schematically in Figure 11. The computercontrolled closed loop testing machines are widely used in all modern fatigue testing
laboratories. Most are equipped with hydraulic grips that facilitate the insertion and removal of
specimens. A schematic diagram of such a testing machine is shown in Figure 12. These
machines are capable of a precise control of almost any type of stress-time, strain-time or load
pattern and are therefore replacing other types of testing machines.

24

5.3 Presentation of Fatigue Test Data

Among the first systematic fatigue investigations reported in the literature are those set up and
conducted by the German railway engineer, August Whler, between 1852 and 1870. He
performed tests on full scale railway axles and also small scale bending, axial and torsion tests
on several types of materials. Typical examples of Whler's original data are shown in Figure
13. These data are presented in what is now well known as Whler or S-N diagrams. Such
diagrams are still commonly used in the presentation of fatigue data, although the stress axis is
often on a logarithmic scale in contrast to Whler's linear stress axis. Basquin's equation is often
fitted to test data, it has the form:
Sa Nb = constant (2)

25

where Sa is the stress amplitude, and b is the slope. When both axes have logarithmic scales,
Basquin's equation becomes a straight line.

Other types of diagrams are used, for instance to demonstrate the influence of mean stress;
examples are the Smith or Haigh diagrams which are shown in Figure 14. Low cycle fatigue
data are almost universally plotted in strain vs. life diagrams since strain is a more meaningful
and more easily measurable parameter than stress when the stress exceeds the elastic limit.

6. PRIMARY FACTORS AFFECTING FATIGUE LIFE

The difference in fatigue behaviour of full scale machine or structural components as compared
with small laboratory specimens of the same material is sometimes striking. In the majority of
cases the real life component exhibits a considerably poorer fatigue performance than the
laboratory specimen although the computed stresses are the same. This difference in fatigue
response can be examined in a systematic manner by evaluating the various factors that
influence fatigue strength. Qualitative and quantitative assessments of these effects are
presented in the following paragraphs.
6.1 Material Effects
Effect of static strength on basic S-N data

26

For small unnotched, polished specimens tested in rotating bending or fully reversed axial
loading there is a strong correlation between the high-cycle fatigue strengths at 10 6 to 107 cycles
(or fatigue limit) So, and the ultimate tensile strength Su. For many steel materials the fatigue
limit (amplitude) is approximately 50% of the tensile strength, ie. S o = 0.5 Su. The ratio of the
alternating fatigue strength So to the ultimate tensile strength Su is called the fatigue ratio. The
relationship between the fatigue limit and the ultimate tensile strength is shown in Figure 15 for
carbon and alloy steels. The majority of data are grouped between the lines corresponding to
fatigue ratios of 0.6 and 0.35. Another feature is that the fatigue strength does not increase
significantly for Su>1400 Mpa. Other relationships between fatigue strength and static strength
properties based on statistical analysis of test data may be found in the literature.

For real life components, the effects of notches, surface roughness and corrosion reduce the
fatigue strength, the effects being strongest for the higher strength materials. The variation in
fatigue strength with the tensile strength is illustrated in Figure 16. The data in Figure 16 are
consistent with the fact that cracks are quickly initiated in components that are sharply notched
or subjected to severe corrosion. The fatigue life then consists almost entirely of crack growth.
Crack growth is very little influenced by the static strength of the material, as illustrated in
Figure 16, and the fatigue lives of sharply notched parts are therefore almost independent of the
tensile strength. An important example is welded joints which always contain small crack-like
defects from which crack start growing after a very short initiation period. Consequently the
fatigue design stresses in current design rules for welded joints are independent of the ultimate
tensile strength.

27

Crack Growth Data

Fatigue crack growth rates seem to be much less dependent on static strength properties than
crack initiation, at least within a given alloy system. In a comparison of crack growth data for
many different types of steel, with yield strengths from 250 to about 2000 Mpa levels of steel,
Barsom [4] found that grouping the steels according to microstructure would minimise scatter.
His data for ferritic-pearlitic, matensitic and austenitic are shown in Figure 17. Also shown in
the same diagram is a common scatter band which indicates a relatively small difference in
crack growth behaviour between the three classes of steel. While data for aluminium alloys
show a larger scatter than for steels, it is still possible to define a common scatter band.
Recognising that different alloy systems seem to have their characteristic crack growth curves,
attempts have been made to correlate crack growth data on the basis of the following expression

=C

(3)

An implication of Equation 3 is that at equal crack growth rates, a crack in a steel plate can
sustain three times higher stress than the same crack in an aluminium plate. Thus, a rough
assessment of the fatigue strength of an aluminium component whose life is dominated by crack
growth can be obtained by dividing the fatigue strength of a similarly shaped steel component
by three.

28

6.2 Mean Stress Effects

In 1870 Whler identified the stress amplitude as the primary loading variable in fatigue testing;
however, the static or mean stress also affects fatigue life as shown schematically in Figure 10.
In general, a tensile mean stress reduces fatigue life while a compressive mean stress increases
life. Mean stress effects are presented either by the mean stress itself as a parameter or the stress
ratio, R. Although the two are interrelated through:

Sm = Sa

(4)

the effects on life are not the same, ie. testing with a constant value of R does not have the same
effect on life as a constant value of Sm, the difference is shown schematically in Figure 18.

As indicated in Figure 19a, testing at a constant R value means that the mean stress decreases
when the stress range is reduced, therefore testing at R = constant gives a better S-N curve than
29

the Sm = constant curve, as indicated in Figure 19b. It should also be noted that when the same
data set is plotted in an S-N diagram with R = constant or with S m = constant the two S-N curves
appear to be different, as shown in Figure 20.

The effect of mean stress on the fatigue strength is commonly presented in Haigh diagrams as
shown in Figure 21, where S a / So is plotted against S m / Su. So is the fatigue strength at a given
life under fully reversed (Sm = 0,R = -1) conditions. Su is the ultimate tensile strength. The data
points thus represent combinations of S a and Sm giving that life. The results were obtained for
small unnotched specimens, tested at various tensile mean stresses. The straight lines are the
modified Goodman and the Soderberg lines, and the curved line is the Gerber parabola. These
are empirical relationships that are represented by the following equations:
Modified Goodman Sa/So + Sm/Su = 1 (5)
Gerber Sa/ So + (Sm/ Su)2 = 1 (6)
Soderberg Sa/ So + Sm/ Sy = 1 (7)

30

The Gerber curves gives a reasonably good fit to the data, but some points fall below the line,
ie. on the unsafe side. The Goodman line represents a lower of the data, while the Soderberg
line is a relatively conservative lower bound that is sometimes used in design. These
expressions should be used with care in design of actual components since the effects of
notches, surface condition, size and environment are not accounted for. Also stress interaction
effect due to mean load variation during spectrum loading might modify the mean stress effects
given in the three equations.
6.3 Notch Effects
Fatigue is a weakest link process which depends on the local stress in a small area. While the
higher strain at a notch makes no significant contribution to the overall deformation, cracks may
start growing here and eventually result in fracture of the part. It is therefore necessary to
calculate the local stress and relate this to the fatigue behaviour of the notched component. A
first approximation is to use the S-N curve for unnotched specimens and reduce the stress by the
Kt factor. An example of this approach is shown in Figure 22 for a sharply notched steel
specimen. The predicted curve fits reasonably well in the high cycle region, but at shorter lives
the calculated curve is far too conservative. The tendency shown in Figure 22 is in fact a general
one, namely that the actual strength reduction in fatigues is less than that predicted by the stress
concentration factor. Instead the fatigue notch factor K f is used to evaluate the effect of notches
in fatigue. Kf is defined as the unnotched to notched fatigue strength, obtained in fatigue tests:

31

Kf =

(8)

From Figure 22 it is evident that K f varies with fatigue life, however, K f is commonly defined as
the ratio between the fatigue limits. With this definition K f is less than Kt, the stress increase due
to the notch is therefore not fully effective in fatigue. The difference between K f and Kt arise
from several sources. Firstly, the material in the notch may be subject to cyclic softening during
fatigue loading and the local stress is reduced. Secondly, the material in the small region at the
bottom of the notch experiences a support effect caused by the constraint from the surrounding
material so that the average strain in the critical region is less than that indicated by the elastic
stress concentration factor. Finally, there is a statistical variability effect arising from the fact
that the highly stressed region at the notch root is small, so there is a smaller probability of
finding a weak spot.
The notch sensitivity q is a measure of how the material in the notch responds to fatigue
cycling, ie. how Kf is related to Kt. q is defined as the ratio of effective stress increase in fatigue
due to the notch, to the theoretical stress increase given by the elastic stress concentration factor.
Thus, with reference to Figure 21

where max,eff is the effective maximum stress, see Figure 23. This definition of K f provides a
scale for q that ranges from zero to unity. When q = 0, K f = Kt = 1 and the material is fully
insensitive to notches, ie. a notch does not lower the fatigue strength. For extremely ductile, low
strength materials such as annealed copper, q approaches 0. Also materials with large defects,
eg. grey cast iron with graphite flakes have values of q close to 0. Hard brittle materials have
values of q close to unity. In general q is found to be a function of both material and the notch
root radius. The concept of notch sensitivity therefore also incorporates a notch size effect.

32

The fatigue notch factor applies to the high cycle range, at shorter lives K f approaches unity as
the S-N curves for notched and unnotched specimens converge and coincide at N = 1/4 (tensile
test). In experimental investigations involving ductile materials it was found that the fatigue
notch factor need to be applied only to the alternating part of the stress cycle and not to the
mean stress. For brittle materials, however, K f should be applied to the mean stress as well.
6.4 Size Effects
Although a size effect is implicit in the fatigue notch factor approach, a size reduction factor is
normally employed in when designing against fatigue. The need for this additional size
correlation arises from the fact that the notch size effect saturates at notch root radii larger than
about 3-4mm, ie. Kf Kt, while it is well known from tests on full scale components, also
unnotched ones, that the fatigue strength continues to drop off with increasing size, without any
apparent limit.
The size effect in fatigue is generally ascribed to the following sources:

A statistical size effect, which is an inherent feature of the fatigue process the nature of
fatigue crack initiation which is a weakest link process where a crack initiates when
variables such as internal and external stresses, geometry, defect size and number, and
material properties combine to give optimum conditions for crack nucleation and
growth. Increasing size therefore produces a higher probability of a weak location.

A technological size effect, which is due to the different material processing route and
different fabrication processes experienced by large and small parts. Different surface
conditions and residual stresses are important aspects of this type of size effect.

A geometrical size also called the stress gradient effect. This effect is due to the lower
stress gradient present in a thick section compared with a thin one, see Figure 24. If a
defect, in the form of a surface scratch or a weld defect, has the same depth in the thin
and thick parts, the defect in the thick part will experience a higher stress than the one
in the thin part, due to the difference in stress gradient, as indicated in Figure 24.

A stress increase effect, due to incomplete geometric scaling of the micro-geometry of

the notch. This takes place if the notch radius is not scaled up with other dimensions.

33

Examples of components for which the latter effect is important are welded joints and threaded
fasteners. The critical locations for crack initiation are the weld toe and the thread root,
respectively. In both cases the local stress is a function of the ratio of thickness (diameter) to the
notch radius. In welds the toe radius is determined by the welding process and is therefore
essentially constant for different size joints. The t/r ratio therefore increases and also the local
stress when the plate is made thicker, with r remaining constant. A similar situation exists for
bolts, due to the fact that the thread root radius is scaled to the thread pitch, rather than the
diameter for standard (eg. ISO) threads. Since the pitch increases much slower than the diameter
the result is an increase in the notch stress with bolt size. For bolts as well as welded joints the
increased notch acuity effect comes in addition to the notch size effect discussed earlier, the
result is that the experimentally determined size effects for these components are among the
strongest recorded. An example of size effects for welded joints is shown in Figure 25. The solid
line represents current design practice, according to eg. Eurocode 3 and the UK Department of
Energy Guidance Notes. The equation for this line is given by:

(10)

34

The exponent n, the slope of the lines in Figure 25a, is the size correction exponent.
The experimental data points indicate that the thickness correction with n = 1/4 is on the unsafe
side in some cases. As indicated in Figure 25a thickness correction exponent of n = 1/3 instead
of the current value of 1/4 gives a better fit to the data in Figure 25a. For unwelded plates and
low stress concentration joints in Figure 25b a value of n = 1/5 seems appropriate [7].
There is experimental evidence that indicate a relationship between the stress gradient and the
size effect. Based on an analysis of experimental data similar the following size reduction factor
has been proposed to account for the larger stress gradient found in notched specimens [8].
n = 0.10 + 0.15 log Kt (11)
6.5 Effects of Surface Finish
Almost all fatigue cracks nucleate at the surface since slip occurs easier here than in the interior.
Additionally, simple fracture mechanics considerations show that surface defects and notches
are much more damaging than internal defects of similar size. The physical condition and stress
situation at the surface is therefore of prime importance for the fatigue performance. One of the
important variables influencing the fatigue strength, the surface finish, commonly characterised
by Ru, the average surface roughness which is the mean distance between peaks and troughs
over a specified measuring distance. The effect of surface finish is determined by comparing the
fatigue limit of specimens with a given surface finish with the fatigue limit of highly polished
standard specimens. The surface reduction factor C r is the defined as the ratio between the two
fatigue limits. Since steels become increasingly more notch sensitive with higher strength, the
surface factor Cr decreases with increasing tensile strength, Su.

35

6.6 Residual Stress Effects

Residual stresses or internal stresses are produced when a region of a part is strained beyond the
elastic limit while other regions are elastically deformed. When the force or deformation
causing the deformation are removed, the elastically deformed material springs back and impose
residual stresses in the plastically deformed material. Yielding can be caused by thermal
expansion as well as by external force. The residual stresses are of the opposite sign to the
initially applied stress. Therefore, if a notched member is loaded in tension until yielding
occurs, the notch root will experience a compressive stress after unloading. Welding stresses
which are locked in when the weld metal contracts during cooling are an example of highly
damaging stresses that cannot be avoided during fabrication. These stresses are of yield stress
magnitude and tensile and compressive stresses must always balance each other, as indicated in
Figure 26. The high tensile welding stresses contribute to a large extent to the poor fatigue
performance of welded joints.

Stresses can be introduced by mechanical methods, for example by simply loading the part the
same way service loading acts until local plastic deformation occurs. Local surface deformation
a such as shot peening or rolling are other mechanical methods frequently used in industrial
applications. Cold rolling is the preferred method to improve the fatigue strength by axisymmetric parts such as axles and crankshafts. Bolt threads formed by rolling are much more
resistant to fatigue loading than cut threads. Shot peening and hammer peening have been
shown to be highly effective methods for increasing the fatigue strength of welded joints.
Thermal processes produce a hardened surface layer with a high compressive stress, often of
yield stress magnitude. The high hardness also produces a wear resistant surface; in many cases
this may be the primary reason for performing the hardness treatment. Surface hardening can be
accomplished by carburising, nitriding or induction hardening.
Since the magnitude of internal stresses is related to the yield stress their effect on fatigue
performance is stronger the higher strength of the material. Improving the fatigue life of
components or structures by introducing residual stresses is therefore normally only cost
effective for higher strength materials.
Residual stresses have a similar influence on fatigue life as externally imposed mean stresses,
ie. a tensile stress reduces fatigue life while a compressive stress increases life. There is,
36

however, an important difference which relates to the stability of residual stresses. While an
externally imposed mean stress, eg. stress caused by dead weight always acts (as long as the
load is present), residual stress may relax with time, especially if there are high peaks in the load
spectrum that cause local yielding at stress concentrations.
6.8 Effects of Corrosion
Corrosion in fresh or salt water can have a very detrimental effect on the fatigue strength of
engineering materials. Even distilled water may reduce the high-cycle fatigue strength to less
than two thirds of its value in dry air.
Figure 27 schematically shows typical S-N curves for the effect of corrosion on unnotched steel
specimens. Precorrosion, prior to fatigue testing introduces notch-like pits that act as stress
raisers. The synergistic nature of corrosion fatigue is illustrated in the figure by the drastic lower
fatigue strength which is obtained when corrosion and fatigue cycling act simultaneously. The
strongest effect of corrosion is observed for unnotched specimens, the fatigue strength reduction
is much less for notched specimens, as shown in Figure 28.

Protection against corrosion can successfully be achieved by surface coatings, either by paint
systems or through the use of metal coatings. Metal coating are deposited either by galvanic or
electrolytic deposition or by spraying. The preferred method for marine structures, however, is
cathodic protection which is obtained by the use of sacrificial anodes or, more infrequently, by
impressed current. The use of cathodic protection normally restores the high cycle fatigue
strength of welded structural steels to its in-air value, while at higher stresses hydrogen
embrittlement effects may reduce the fatigue life by a factor of 3 to 4 on life.
7. CYCLE COUNTING PROCEDURE FOR VARIABLE AMPLITUDE LOADING

37

In practice the pattern of the stress history with time at any particular detail is likely to be
irregular and may indeed be random. A more realistic pattern of loading would involve a
sequence of loads of different magnitude producing a stress history perhaps as shown in Figure
29. The problem now arises as to what is meant by a cycle and what is the corresponding stress
range. A number of alternative methods of stress cycle counting have been proposed to
overcome this difficulty. The methods most commonly adopted for use in connection with
Codes and Standards are the 'reservoir' or the 'rainflow' method.

7.1 The Reservoir Method

The basis of the reservoir method is shown in Figure 30 using the stress time history as Figure
29. it should be assumed that a stress time history of this kind has been obtained from strain
gauges attached to the structure at the detail under consideration or has been estimated by
computer simulation. It is important that the results analysed should be representative of long
term behaviour. To analyse these results, a representative period is chosen so that the peak stress
level repeats itself and a line is drawn to join the two peaks as shown in Figure 30a. The region
between these two peaks is then regarded as being filled with water to form a reservoir. The
procedure is then to take the lowest trough position and imaging that one opens a tap to drain
the reservoir. Water drains out from this trough T1 but remains tapped in adjacent troughs
separated by intermediate peaks as shown in Figure 30b. The draining of the first trough
T1 corresponds to one cycle of stress range St as shown, and the remaining level of water is now
lowered to the level of the next highest peak. A tap is now opened at the next lowest trough
T2 as shown in Figure 30c and the water allowed to drain out. The height of the water released
by this operation corresponds to one cycle of stress range S 2. This procedure is continued
sequentially through each next lowest trough, gradually building up a series of numbers of
cycles of different stress ranges. It is also essential to allow for the one cycle from zero to peak
stress. For the particular stress time history shown in Figure 29 the results obtained from the
sample time period taken would be:

38

1 cycle at 120N/mm2, 1 at 100N/mm2, 4 at 80N/mm2, 6 at 60N/mm2, 10 at 30N/mm2.

The important principle of the above procedure is the recognition that by taking the difference
between the lowest and highest stress levels (trough and peak) it is ensured that the greatest
possible stress range is counted first, and this procedure is repeated sequentially so that the
highest ranges are identified as the random fluctuations take place. In the assessment of the
effects of the different cycles the greatest damage is caused by the higher stress ranges since the
design curves follow a relationship of the kind S mN=constant. The reservoir method procedure
does ensure that practical combinations of minima and maxima are considered together whereas
this is not always the case in other stress cycle coating procedures.
An alternative way of carrying out the reservoir cycle counting method is to turn the diagram
upside down and use the complementary part of the diagram as shown in Figure 31. This
version of the reservoir method gives identical results to the normal method but has the
advantage of including the major cycle of stress from zero to maximum and back.

7.2 The 'Rainflow' Counting Method

The alternative 'rainflow' cycle counting procedure is illustrated in Figure 32a for the same
stress time history of Figure 29. This is essentially the same picture turned onto its side as
shown in Figure 32a. Water (rain) is allowed to fall from the top onto the pattern considered as a
roof structure and the paths followed by the rain are followed. However it is important that a
number of standard rules are followed and the procedure is rather more complex and subject to
error than the reservoir method. For each leg of the roof an imaginary flow of water is
introduced at its highest point as shown by the dots in Figure 32b. The flow of water is followed
for the outermost starting point first, allowing the water to drop onto any parts of the roof below
and continue to drain until it falls off the roof completely. The width from the stress level at
which the water started until it left the roof represents the magnitude of one cycle of stress. It is
necessary to follow the flow paths from each starting point sequentially, moving progressively
39

in from the points which are furthest out. If however the flow reaches a position where water
has drained from a previous flow, it is terminated at that point as shown in Figure 32c for the
flow starting from position 3 terminated by the previous flow position 1. The stress range for a
cycle terminated in this way is limited to the width between the starting point and the
termination point. The complete rainflow diagram for the stress pattern of Figure 29 is shown in
Figure 32d. This procedure when correctly applied also counts the highest stress range cycles
first and ensures that only practical combinations of minima and maxima within a sequence are
considered. The rainflow method is somewhat more difficult to apply correctly than the
reservoir method and it is recommended that both for teaching and for design purposes the
reservoir method should be used. The results for the stress ranges from the rainflow method
applied to the stress history from Figure 29 are identical to those from the reservoir method ie.
1 cycle at 120N/mm2, 1 at 100N/mm2, 4 at 80N/mm2, 6 at 60N/mm2, 10 at 30N/mm2.

40

There are two other cycle counting methods, the 'range pair counting' method and the 'mean
crossing level' which are sometimes used although they tend not to be specified in Codes.
Example 1 This design example is based on the stress cycle history of Figure 29 as analysed
above for stress cycle counting purposes. Firstly the stress history represents a relatively short
time period, and has to be extrapolated to represent the total required life. Obviously the first
requirement is to ascertain the required design life, and to multiply the numbers of cycles of
each stress range determined as above by the ratio of the design life to the period represented by
the sample time record taken. For example, if the design life was 20 years, and the sample time
period was 6 hours, the numbers of cycles should be multiplied by 20 x 365 x 4 = 29200.
Caution should be exercised with such an extrapolation however, as to whether such a short
length time sample is representative of long term behaviour. For example in the case of a bridge
structure the traffic flows are likely to vary at different times of day, peaking at rush hour times
and falling to low values in the middle of the night. Furthermore there is possibility that the
heaviest loads may not have occurred during the sampling time considered. Problems of
extrapolation from samples to full data are common in the statistical world and statistical
procedures may be necessary to ensure that potential differences in scaling up the data are
allowed for. To a large extent this depends on the absolute size of the sample taken.
To check whether the design is satisfactory for any particular detail, it is necessary to decide on
the appropriate design S-N design curve for the detail. The basis of doing this for Eurocode 3
will be explained in Lecture 12.9. For present purposes it will be assumed that the stress history
of Figure 29 analysed above applies to a detail for which the design S-N curve is S90, for which
the design life is 2 x 10 6 cycles at stress range 90N/mm2 with slope - 1/3 continued down to a
stress level of 66N/mm2 at design life 5 x 106 cycles, with a change in slope to -1/5 on down to a
stress range of 36N/mm2 which is the fatigue limit at 10 million cycles. For a twenty year design

41

life assuming the stress history of Figure 29 is representative of 6 hours typical loading the
following table can be constructed:

For these assumptions the loading is acceptable for the detail and life required. Indeed the
'Damage Sum' value of 0.1174 based on a 20 year design life indicates the available design life
is 20/0.1174 = 170 years. For this particular case the stress range of 60N/mm 2 fell in the
intermediate range between 36 and 66N/mm 2 and the available life N was calculated using the
changed slope of the S-N curve for this region. The stress range of 30N/mm 2 is below the cut off
for the S90 classification and does not contribute to the fatigue damage.
7.3 Exceedance Diagram Methods
A convenient way of summarising the fatigue loading applied to structures is by the use of
exceedance diagrams. These diagrams present a summary of the magnitude of a particular event
against the number of times this magnitude is exceeded. Whilst in principle this presentation can
be applied to a wide variety of phenomena for the purposes of fatigue analyses the appropriate
form is a graph of log (number of times exceeded) against the occurrence of different stress
levels. An example is shown in Figure 33. This might represent the stresses caused at a
particular location in a bridge by traffic passing over r by wave loading of an offshore structure.
A typical feature of natural phenomena of this kind is that the number of exceedances increases
as the stress level decreases. The form of the exceedance diagram for natural phenomena of this
kind is often close to linear as shown. It is important to note that the diagram represents
exceedances so that any particular point on the graph includes all of the numbers of cycles of
stress range above that value. For use in fatigue analysis using Miner's law the requirement is a
summary of the numbers of cycles of each stress level occurring. Thus the loading represented
by the exceedance diagram of Figure 33 can be treated as an equivalent histogram with cycles
as follows:

42

Some of the stress ranges will be found to be below the fatigue limit and hence will not
contribute to the Miners law damage sum. For example for the detail considered in Example 1
above, the cut off limit was 36N/mm2 and the stress ranges of 20N/mm2 would not contribute to
the fatigue damage. The stress ranges above this level will contribute however and their effects
must be included. This is done by finding the value of SmN separately for the remaining
stress levels above and below the change in slope of the S-N curve, and for the figures given
above this will be found to be 5.692 x 1010 for stress ranges of 80N/mm2 and above, and 1.621 x
1015 for the 40 and 60N/mm2 stress ranges. For an S90 detail with the spectrum of loading
shown above, the fatigue damage from each part of the S-N curve has to be calculated based on
the appropriate value of SmN=constant as follows:

= 0.298

From these figures the damage sum factor calculated as 0.298 is acceptable. Detailed
examination of the figures leading up to this result would indicate that the majority of the
damage calculated occurs at the lowest stress ranges of 40 and 60N/mm 2 contributing to the S5N
part of the design curve.
7.4 Block Loading
Block loading is a particular case of an exceedance diagram.
Consider the particular case of a one lane bridge structure on which the loading is idealised as
falling into three categories. Suppose that there are n 1 heavy lorries travelling across the bridge
during its lifetime, and that at a particular welded detail each lorry causes a stress range S 1. In
addition there are n2 medium lorries which cause a stress range S 2, and n3 cars which cause a
stress range S3 at the same welded detail as they cross the bridge. To assess the combined effect
of the different stress ranges all being applied in some form of sequence the procedure adopted
is to assume that the damage caused by each individual group of cycles of a given stress range is
the same as would be caused under constant amplitude loading at that stress range. It is
necessary first to decide on the appropriate classification for the geometric detail being
43

considered and to identify the appropriate S-N design curve. For present purposes, let us assume
that the design curve is as shown in Figure 34. If the only fatigue loading applied to the bridge
was the crossing of the heavy lorries with stress range S 1 at the detail concerned, the available
design life would be N1 cycles as shown in Figure 34. In fact the number of cycles applied at
this stress range is n1. It is assumed that the fatigue damage caused at stress range S t is n1/N1.
Similarly if the only fatigue loading applied to the bridge was the crossing of the medium lorries
with stress range S2 the available design life would be N2 and the fatigue damage caused would
be n2/N2. For the passage of the cars at stress range S 3 the available design life if this was the
only loading would be N3 and the fatigue damage caused would be n3/N3. When all three
loadings occur together the assumption for design purposes is that the total fatigue damage is
the sum of that occurring at each individual stress range independently. This is known as the
Palmgren-Miner law of linear damage, or more simply as Miner's law and is summarised as
follows:

+ .... + = 1 (11)

7.5 Frequency and Spectrum Aspects

It is not uncommon for loading to occur at more than one frequency. It is generally considered
that for non aggressive environmental conditions, eg. steel in air, there is little or no effect of
frequency on constant amplitude fatigue behaviour. In aggressive conditions however, eg. steel
in seawater, there may be significant effects of frequency on the crack growth mechanism
leading to increased crack growth rates, shorter lives and reduction or elimination of the fatigue
limit. In particular it is necessary in fatigue testing of materials where environmental conditions
may be important to carry out the testing at the same frequency as that of the service loading.
An example of this is the effect of wave loading on offshore structures where a typical
frequency of waves is about 0.16Hz. Clearly this has major implications on the time required for
testing since to accumulate one million cycles at 0.16Hz would take about 70 days whereas a
conventional test in air at say 16Hz would reach the same life in less than 1 day. With any
structure the response of the structure to dynamic loading depends on the frequency or rate of
the applied loading and on the vibration characteristics of the structure itself. It is most
important for the designer to ensure that the natural resonance frequencies of the structure are
44

well separated from the frequencies of applied loading which may occur. Even so the structure
may respond with frequencies of stress fluctuation which are a combination of the applied
loading frequency and its own natural vibration frequencies. Furthermore since the magnitude
of the loading may also vary with time it is necessary to consider both time domain and
frequency domain aspects. Figure 35 shows a typical frequency domain response for stress
fluctuations at a particular location in an offshore structure. This diagram gives information on
number of times different stress levels are exceeded as well as the frequency data. The peaks at
about 0.16Hz correspond to the applied loading whereas the higher frequency peaks are those
due to the vibration response of the structure.

With variable amplitude fatigue loading of this kind there are additional complexities with
regard to frequency effects to be considered. Where the stressing occurs close to or at a single
frequency the condition is known as 'narrow band' and when there are a range of different
frequencies involved it is known as 'broad band'. If the frequency domain response of Figure 35
is converted back into the time domain response in which the data was originally recorded the
result would look like Figure 36. Clearly some assumptions must have been in the conversion of
one diagram into the other and in this case it is that stress cycle counting has been carried out by
the reservoir method. In Figure 36 however, it is clear that because the higher frequency stress
cycles are superimposed on top of the lower frequency cycles, some of the higher frequency
cycles occur at higher mean stress or stress ratio.

8. CONCLUDING SUMMARY

In this lecture it has been shown that fatigue is a weakest link process of a statistical
nature in which a crack will initiate at a location where stress, local and global
45

geometry, defects and material properties combine to give a worst case situation. The
crack thus nucleates at a local peak spot, and may cause failure of the structure, even if
the rest of the structure has a high fatigue resistance. Good fatigue design practice is
therefore based on close attention to details that increases the stress locally and
therefore are potentially initiation sites for fatigue cracks.

A positive aspect of the local nature of the fatigue process is that only a relatively small
area of highly stressed material need to be improved in order to increase the load
carrying capacity of the structure when fatigue is the limiting design criterion.

Another general conclusion is that increasing the size of a structure generally leads to a
lower strength with respect to brittle fracture as well as fatigue. Size effects must
therefore be properly accounted for.

The larger number of factors influencing fatigue strength makes the combined effects of
these factors very difficult to predict. The safest way to obtain design data is therefore
still to perform fatigue tests on prototype components with realistic environmental
conditions.

A normal structural design analysis must be carried out for the maximum design loads
and for a series of intermediate loads with known number of occurrences in the design
life to give stress results at typical details. Alternatively if the application Code gives an
equivalent constant amplitude loading condition and associated number of cycles this
loading should be applied and stresses determined. The stresses should be analysed for
range of variation in principal stress or of direct stress aligned perpendicular or parallel
to the geometric detail as defined in Eurocode 3. Treatments for shear stresses are given
in Eurocode 3. The stress ranges should be multiplied by appropriate partial factors, and
for variable amplitude loading either combined together to give an equivalent constant
amplitude stress range and number of cycles or used to sum up fatigue damage.

The correct detail classification must be identified for typical critical details and the
applied fatigue damage for the design life checked against the design S-N curve for the
detail concerned. If the design is not satisfactory either the stress ranges must be
reduced or the detail changed until satisfactory results are obtained.

9. REFERENCES
1. Metals Handbook, ASM 1985.
2. ISO Standard, 373 - 1964.
3. P.C. Paris and F. Erdogan, "A Critical Analysis of Crack Propagation Laws", Trans,
ASME, Vol. 85, No. 4, 1963.
4. J.M. Barsom, "Fatigue Crack Propagation", Trans, ASME, SEr. B, No.4, 1971.
5. H. Neuber, "Kerbspannungslehre", Springer, 1958.
6. R.E. Peterson, "Stress Concentration Factors", John Wiley & Sons, 1974.
7. O. rjaster et al, "Effect of Plate Thickness on the Fatigue Properties of a Low Carbon
Micro-Alloyed Steel", Proc. 3rd Int. ECSC Conf. on Steel in Marine Structures
(SIMS'87), Delft, 15-18 June 1987.
46

8. P. J. Haagensen, "Size Effects in Fatigue of Non-Welded Components", Proc. 9th Int,

Conf. on Offshore Mechanics and Arctic Engineering, (OMAE), Houston, Texas, 18-23
February 1990.

47

Lecture 12.3: Effect of Workmanship on

Fatigue Strength of Longitudinal and
Transverse Welds
OBJECTIVE/SCOPE
Identification of factors influencing the fatigue strength of welded joints and of the
consequences for design, fabrication and inspection.
PREREQUISITES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.6: Fatigue Behaviour of Bolted Connections
RELATED LECTURES
Lecture 3.4: Welding Processes
Lecture 3.6: Inspection/QA Assurance
SUMMARY
The data on fatigue strength given in Eurocode 3 [1] are briefly reviewed. The strengths of
longitudinal and transverse welds are related to the quality of workmanship. The need for
inspection and the limitations of non-destructive testing are examined. The implications for
economic design, detailing and specification are set out.
1. INTRODUCTION
Any joint in a structure or in any part of it is a potential point of weakness, both in static
strength and in fatigue.
For fatigue the potential weakness is evident from the fatigue strength data given in Eurocode 3
[1] (Figure 1). There the perfect plate is in detail category 160, which is the fatigue strength at
2.106 cycles, whilst the joint detail with the worst geometry and hence stress concentration, is in
category 36.

48

In a welded joint potential sites for initiation of a fatigue crack are:

1. In the parent metal of either part joined, adjacent to:
(i) the end of the weld
(ii) a weld toe
(iii) a change of direction of the weld.
2. In the weld metal itself, starting from:
(i) the weld root
(ii) the weld surface
(iii) an internal flaw.
Even one type of joint, the longitudinal fillet or butt weld, can fall into any one of four
categories, from 140 to 100, depending on workmanship, see Figure 2.

49

Transverse butt welds can have an even wider range of strengths (Figure 2) - from category 125
to category 36, 7 categories in all. If one excludes butt welds made from one side only, with and
without backing strips, i.e. detail categories 71, 50 and 36, four categories are left for "good"
butt welds. Here the category depends on both weld geometry and workmanship.
Other welds (transverse fillets, welds to attachments, etc.) also show wide variations in strength
depending on geometry and workmanship.
It is important to note that a number of other (usually accidental) results of poor workmanship
can reduce the performance of a detail to below what its category would indicate:
(a) weld spatter
(b) accidental arc strikes
(c) unauthorised attachments
(d) corrosion pitting
(e) weld flaws, particularly in transverse butt welds
(f) poor fit-up
(g) eccentricity and misalignment.
Most of these are largely unquantifiable and must be controlled by adequate inspection and
repair.

50

It is the purpose of this lecture to describe in greater detail welded joints and the matters to be
considered by the designer before deciding the fatigue strength that will be used in calculations.
2. LONGITUDINAL WELDS
The highest category for longitudinal welds, 140, applies only where there are "no significant
flaws". This implies automatic welding, no stop/start positions, no slag inclusions or blow holes
- near perfection "demonstrated by specialist inspection".
The next category down, 125, requires automatic welding and expert repair, followed by
inspection, of any accidental stop/start positions. Leaving stop/start positions brings the
category of longitudinal fillet welds down to 112 and that of longitudinal butt welds down to
100.
Manual fillet or butt welds and one-sided butt welds are all in category 100, as are "repaired"
welds.
There is experimental evidence that small slag inclusions can bring the strength of a
longitudinal fillet weld down to category 90.
A lower limit to the strength of defective longitudinal fillet welds is probably that of an
intermittent weld, category 80, or even the end of such a weld at a cope hole, category 71.
3. TRANSVERSE BUTT WELDS
Transverse butt welds can reach category 125 when "high quality welding" is obtained and
proved to have been achieved by later inspection. Amongst other requirements, the proposed
welding standard limits solid inclusions in such welds to a width of 2mm and a length of 6mm,
thus acknowledging the importance of internal defects.
Lower quality welds fall into category 112 provided the welds are ground flush. Otherwise they
are in category 90, or 80 for splices in rolled sections or girders. Here the category depends on
the weld profile and the likely quality of workmanship; internal defects are not mentioned.
In fact, internal defects have at least as great an influence on the fatigue strength of transverse
butt welds as does the weld profile.
Another factor which affects the strength of splices in girders, and which is not mentioned
explicitly in the description of the detail categories, is the order in which the welds are made.
This can affect the level of residual stress.
The test results shown in Figure 3 illustrate these points. They are all results of tests on
transverse butt welds shown against the grid of lines representing the fatigue strengths given in
Eurocode 3 [1] for the various detail categories. The short thick lines represent test results on
small plate specimens, 40mm wide and 10mm thick. All other points represent results from tests
of complete beams.

51

The results range from category 112 for the plate specimens down to category 63 or so for a butt
joint in a rolled I beam.
The reasons for this spread of results are partly weld quality and partly residual stresses caused
by different welding procedures.
3.1 Effect of Internal Defects.
It is likely that the plate specimens were reasonably free from internal defects. The butt welds in
the plate girder flanges, shown as large circles, contained various small defects, in the range of
3mm2 to 30mm2 from which the cracks leading to failure originated. Allowing for the fact that
the plate girder flanges were 35mm thick, all results would fit into category 112. So would the
results from tests on small girders with 25mm thick flanges, shown as triangles.
The results shown as small dots were obtained for a butt weld between a rolled and a built-up I
section. The failure was due to a large "lack of fusion" defect in the 30mm thick flange directly
above the web to which it was joined by 24mm radii. The defect had an area of about
80mm2 and is sketched in Figure 4 and was attributed to faulty weld preparation. It must be
pointed out, however, that it was the work of experienced fabricators, who clearly had not
appreciated the difficulty of achieving full penetration at this point.

52

Even allowing for size effect, one would put this result into category 63. There is no information
on the strength of such welds in Eurocode 3 - they should not be used. A British Standard,
BS5400: Part 10 puts such welds in a class which corresponds, as regards strength, to category
63 [2]. This classification fits the test results.
These few test results suffice to indicate that internal weld defects occur and that they have a
decisive influence on the fatigue strength of a welded joint.
To determine this effect quantitatively, a fracture mechanics study has been undertaken, based
on fatigue test results from butt welds containing known defects. The results were used to obtain
basic fracture mechanics data. These results showed the scatter typical of all fatigue test results.
A lower bound of the values was then used to calculate the fatigue strength of butt welds of
various thicknesses containing defects of various sizes. The defect size was expressed as an
area; a reasonable approximation which avoided the need to give two dimensions for every
defect and to investigate various shapes.
The figures shown in Figure 5 are approximate and were obtained by interpolation, and some
extrapolation, from the results produced by the investigation.

The agreement between these figures and the few large beam results is quite good.

53

It will be noted that near surface defects cause a greater loss of fatigue strength than deep ones
and that a 12mm2 defect, such as suggested in the draft welding standard [1], would bring a butt
weld strength down to category 100, or even 90 if near the surface.
It is clear, therefore, that high fatigue strength in a butt weld requires nearly perfect welds.
3.2 Effect of Welding Procedure
The results in Figure 3 showing the effect of different welding procedures and, hence, residual
stress are the two groups of squares.
They were obtained from tests on butt joints through rolled I sections. The higher set, full
squares, were from specimens in which the flange butt welds were made before the web butt
welds. The lower set, open squares, were obtained from specimens in which the reverse
procedure had been used - web butt weld first, then flange butt welds so that their contraction
was resisted by the web.
One set fits category 100, the other category 80 - a considerable loss of strength through using
the wrong weld sequence.
These results do not stand alone; similar ones have been obtained in the United States and they
are confirmed by the results shown by the circles on the figure.
These results were obtained from plate girder specimens with 35mm thick flanges. The full
circles show results from specimens in which only the flange plates were butt welded, and that
before they were welded to the webs. Allowing for size effect, they fit category 112 or, possibly,
125. The open circles are results from butt welds right through similar plate girders. They are
little, if any, worse than those shown by full circles.
However, the welding procedure, shown in Figure 6, was designed to minimise residual stresses
in the flange welds. Initially, the webs were not welded to the flange for some 110mm either
side of the joint, the flange butt welds were made first, then the web butt weld and, finally, the
web was welded to the flanges. This weld also served to close the small slot which had been left
under the flange butt weld to allow radiography of these welds. Cope holes were neither needed
nor provided.

54

It is clear from these results that butt joints right through a girder can have the same fatigue
strength as a butt weld through a plate provided that the right welding procedure is specified by
the designer and followed in the fabrication shops. Otherwise there is a loss of fatigue strength
of the order of 25%.
The results were obtained from tests on plate girders, but the conclusions have a wider
application. They apply, for example, to joints in portal frames at or near the corners and any
situation where there is a risk of high restraint of butt welds.
There is some evidence that similar considerations apply to welding attachments to girders. In
one test, plates welded to the compression flange of a plate girder caused early cracks, as was
expected. When similar attachments were welded to the flange plate before it was welded to the
girder, no cracks were observed at about double the endurance; again an improvement of about
25% in the fatigue strength.
Given the effect welding procedure can have on the fatigue strength of a joint, it must be
considered at the design stage and be specified; it cannot be left to the fabricator.
4. OTHER WELDS
4.1 General
So far discussion has been limited to those types of weld (longitudinal butt and fillet welds, and
transverse butt welds) whose fatigue strength can be very significantly affected by embedded, or
usually more importantly, surface breaking defects. Many other types of weld can be and have
been used which, even if carried out perfectly, result in a considerable loss of fatigue strength.
This loss is normally a result of the welds forming geometrical discontinuities, or causing stress
raisers; frequently workmanship faults, unless gross, have little further degrading effect.
55

Three specific types of weld will be considered in this section:

a. Transverse fillet welds
b. Welds connecting load-carrying and non load-carrying attachments.
It is sometimes difficult to differentiate between the reduction of strength which results from
poor detailed design, and that which results from poor workmanship, since there can be a
correlation between the two; however, the main considerations are described below.
4.2 Transverse Fillet Welds
Transverse fillet welds may be used for connecting transverse stiffeners to a plate (e.g. web
stiffeners, or transverse stiffeners on a wide flange plate in compression, or diaphragms in box
girders); they always reduce the fatigue strength of the parent plate. Eurocode 3 Part 1 [1] shows
that the best that can normally be achieved is category 80 if the thickness of the stiffener or
diaphragm is 12 mm or less, or category 71 if it is more than 12 mm. As is discussed later, it
may be desirable or possible in some circumstances to dispense with such stiffeners by
thickening the parent plate.
As fatigue failure resulting from a transverse fillet weld is usually initiated by a crack growing
into the parent plate from the toe of the weld (Figure 7), faults in the weld itself (e.g. slag
inclusions or porosity) are unlikely to reduce the strength further. However, any faults in
workmanship which damage the parent plate can be serious; in particular any residual faults
which take the form of planar cracking in the heat affected zone of the plate must be strictly
forbidden, and inspection must be specified to ensure that any such faults can be detected and
rectified.

Undercut in the parent plate at the weld toe, if excessive, can also degrade the fatigue strength
although a small amount can generally be tolerated without reducing the category below the
figures given above. It is generally considered that undercut of not more than 0,5 mm is
permissible, provide that the total undercut does not reduce the cross-section by more than 5%.

56

One area where design and workmanship tend to overlap is where a transverse fillet weld ends
near the edge of the parent plate. For example, in the case of a web stiffener attached also to the
flange of a joist, what happens to that portion attached near to the edge of the flange? Where the
stiffener is terminated within 10 mm of the edge of the flange, the category is reduced to 50 at
that point; where it is terminated 10 mm or more from the edge, the "full" category of 71 or 80
as appropriate may be used.
Other applications of transverse fillet welds where the workmanship can affect the strength arise
where the weld is subject to bending about its longitudinal axis; such conditions can arise where
the traffic loads on a steel deck plate of a bridge are transferred to stiffeners or girders through
fillet welds between the plate and the stiffener web (Figure 8). In this case the quality of the
weld is all important; in particular the fit-up between the plate and the web should be very good
otherwise the flexural stresses will be increased and the root of the weld will be likely to be of
very rough profile. In such circumstances many instances have arisen of a crack initiating at the
root under the flexural stresses induced, and propagating through the throat of the weld. A
failure such as this is extremely difficult to detect until the whole flexural cross-section area of
the weld has been lost; furthermore attempts to use analytical methods to anticipate such an
occurrence are almost impossible because of the extreme sensitivity of the weld to such
problems of workmanship.

As a final example of transverse fillet welds, and as an introduction to load-carrying

attachments, consider the attachment of a doubling plate as shown in Figure 9. Such doubling
plates have frequently been used in the past to increase the static strength of (for example) the
flange plate of a steel girder. At the end of such a doubling plate there will be a transverse fillet
weld, but because it is having to transfer load into the plate its function is significantly different
from those discussed earlier.

57

Firstly, because of the load transfer the stress concentration factor is high and hence the "basic"
category of the weld is 50* when neither the flange plate nor the doubling plate is greater than
20 mm thick. If either exceeds 20 mm in thickness the category is reduced to 36*.
Secondly, not only will damage to the plate reduce the fatigue strength in a similar manner to
that described above, but faults in the weld itself will reduce the strength since it is actually
transferring load. An interesting development of this detail, which largely overcomes the
problems described, is to omit the welding at the ends of the doubling plate and instead connect
it to the flange plate in this region using high strength friction grip bolts, only beginning the
welding (as a longitudinal fillet) after most of the load has been transferred. This weld will then
behave as described for longitudinal fillets in Section 2.
4.3 Load-Carrying and Non Load-Carrying Attachments
As can be inferred from Section 4.2 above, a load-carrying attachment is likely to have a lower
category than a non load-carrying one. This statement must be treated with care, however, since
sometimes an attachment intended to be non load-carrying may carry load; a typical case is a
lug welded on, for example, to aid lifting during erection (see Figure 10). The lug is not
intended to carry load in service but will, of course, tend to attract some. This circumstance is
recognised in Eurocode 3 by varying the category depending on the length of the attachment in
the direction of stress in the parent plate (the longer it is, the more load it will attract and hence
the greater the stress concentration and the lower the category). Thus, even if it is nominally non
load-carrying, it comes into the following categories (even lower categories would apply if the
attachments were within 20mm of the edge of a plate):
length 50 mm: category 80
50 mm < length 100 mm: category 71
length > 100 mm: category 50*

58

If such attachments come in a fatigue sensitive area, a designer will normally require them to be
removed and ground flush after use for erection; however, sometimes they are required in the
final structure (e.g. batten plates, connection of bracing, etc.). In such cases care should be taken
to detail them in such a way as to minimise the load transfer and hence the stress concentration.
This is usually done by making them as short as possible and keeping them away from the
extreme edge of members.
True load-carrying attachments (such as the doubling plate of Section 4.2, or load-carrying
cruciform joints), are usually of very low detail category and are sensitive to weld defects in the
same way as transverse butt welds. Furthermore, cruciform and similar connections can be very
sensitive to lamellar faults in the plates, thus leading to lamellar tearing after welding. Such
faults would have to be repaired - they cannot be quantified or allowed for by simply reducing
the category of the weld.
5. INSPECTION
Since the fatigue strength of welded joints is greatly influenced by the quality of the welds, the
designer's first consideration in choosing a fatigue strength for design calculations must be the
quality which can be achieved at economically justifiable cost and which can be shown to have
been achieved by reasonably possible supervision of the welding process and inspection of the
finished product.
5.1 Inspection of Longitudinal Welds
Consider first longitudinal welds. The requirements for high fatigue strength are severe.
Any breakdown of an automatic welding process would drop a potential category 140 weld to
category 125, or to category 112 unless the accidental stop/start position is repaired, and seen to
be repaired, by a specialist.
59

To use a design strength of 140, therefore, requires faultless automatic welding under
continuous supervision and rejection of any component with an accidental stop/start position at
a highly stressed point.
Even category 125 requires a high degree of supervision to ensure identification and proper
repair of any stop/start. The choice of lower categories reduces the need for supervision.
Inspection of longitudinal welds is difficult. For a start, a 30m long plate girder has 120m of
longitudinal weld, some 30m of which would be stressed within about 10%of the maximum.
Secondly, practicable methods of inspection are limited to visual - just looking at the weld - and
magnetic particle methods to detect cracks.
Considerations of time and cost limit anything other than visual inspection to sample lengths,
say a number of 1m lengths, making up a total of some 5% to 10% of the total length of weld. In
addition, the ends of welds should be checked for cracks.
If such sample inspection reveals defects, the sampling rate must be increased to determine
whether there is something wrong with the welding process.
The suggested sampling rate applies where the only function of the longitudinal welds is to hold
web and flange together. In certain applications, such as crane girders or box girders,
longitudinal welds may be subject to bending about their longitudinal axis or to concentrated
vertical loads applied to the flange.
Eurocode 3 [1] gives categories for welds in such bending; category 71 for longitudinal butt
welds and 50 for such partial penetration butt welds and fillet welds.
There is experimental evidence that welds at box girder corners can reach strengths
corresponding to categories 80 or even 90, if butt welds, 125 if a pair of fillet welds, but only
50, as in Eurocode 3, if one-sided fillet welds. The stresses are taken as the greater of those in
the weld or the parent metal.
This suggests that the strengths required for design in accordance with Eurocode 3 can be
achieved with "ordinary" quality welds. Sample inspection, as for other longitudinal welds,
should be sufficient but the sample size should be increased, probably doubled.
In relating strength requirements to inspection and quality criteria it should be considered that at
least part of the partial factor in Eurocode 3 is intended to deal with variations in workmanship.
For example, a factor of about 1,12 would cover the difference between a "perfectly" repaired
stop/start position and a reasonably good one. If the repair led to slag inclusion, it would need a
factor of 1,40 to cover the resulting loss in strength.
5.2 Inspection of Transverse Butt Welds
The total length of transverse butt weld in a girder is much less than that of longitudinal welds,
say of the order of 4m. For this, and other reasons, they are more easily inspected than
longitudinal welds. Even so, it is not possible to be sure of finding small internal defects and to
size accurately those found.
60

In addition to visual inspection, two methods of non-destructive testing can be used radiography and ultrasonics.
Essentially radiography records the intensity with which X-rays are transmitted through the
weld onto photographic film. Porosity, for example, shows as dark grey spots on a grey
background.
Radiography is costly, requires safety precautions difficult to arrange where there is 24 hour use
of a workshop, and will not show thin defects such as lack of fusion, see Figure 11.

Ultrasonic testing can find such defects but is operator dependent. Even careful ultrasonic
examination will, in most cases, miss defects of the order of 2mm 2 in area - the sort of defect
which may bring fatigue strength down to category 112.
The effect of defect size on fatigue strength was discussed earlier. The use of a given strength in
design therefore implies a corresponding limit on defect size. Hence there is a need to measure
the size of detected defects by ultrasonics.
In ultrasonic examination (Figure 12) defects are detected by the ultrasonic beam they reflect.
The reflection is indicated on a display against a time base. The distance from the probe sending
the beam is determined by the location of the echo mark. The height of the mark above the base
measures the intensity of the reflected beam relative to that from some standard reflector.

61

The boundaries of a defect are established by moving the probe until the mark of the defect echo
has disappeared. The boundary is usually taken to be the centre of the probe at the position
where the defect echo height is halved. This is not a precise measurement. Errors are of the
order of 5mm on a defect dimension.
This method cannot be used when the dimension to be measured is less than the diameter of the
probe - some 10mm in diameter or 79mm 2 in area. The area of such small but significant defects
can be determined by the proportion of the ultrasonic beam they reflect, i.e. the echo height. The
strength of this reflection depends not only on the size of the defect but also on the efficiency of
its surface as a reflector. Again, this is not a precise measurement. The same measurement of
echo height can be produced by defects which differ in area by a factor of 2 if small, and a
factor of 4 if large.
Defects which are acceptable for reasonable fatigue strength are small. The errors in
determining their size by non-destructive testing are such that safe limits have to be set. This
means that welds containing only acceptable defects will have to be rejected to be reasonably
sure of not accepting welds with unacceptable defects.
Limits on defect size can be relaxed if part of the partial factor is taken to cover faults in
workmanship. A factor of 1,25, for example, applied to category 112 strength would cover the
sort of defects which can occur in good quality work.
5.3 Inspection of Other Welds
As was shown in Section 4, the performance of transverse fillet welds and welds to attachments
is less dependent on the quality of the welding than is the performance of butt welds or
longitudinal fillets. Hence inspection of such welds is frequently confined to:
(a) visual inspection to ensure that the weld is of the right size, of a good profile, and does not
suffer from excessive undercut.
(b) non-destructive testing (e.g. magnetic particle or dye penetrant) to ensure that there are no
cracks in the weld or in the parent plate at the weld toe.

62

In certain applications, such as cruciform joints, it may be necessary to check the welding,
probably by ultrasonic methods, to ensure that it has not caused lamellar tearing of the plate.
Where temporary gussets, etc. have been removed and ground, it is usual to check the surface of
the ground plate for cracks.
6. CHOICE OF QUALITY
A designer must appreciate that high quality is associated with considerable cost.
The costs arise from increased care in welding and supervision, inspection and repeated repair
of rejected work. Repeated repair occurs because the quality of a repaired weld is often worse
than that of the rejected weld. Added to this cost is the major cost of the delay caused to
fabrication.
It is, therefore, important not to specify a higher quality than is justified on economic grounds or
necessary on engineering grounds.
For economy must be balanced the cost of achieving a given quality against that of the extra
material which would have to be used to reduce stresses to a level for which a lower quality
would be acceptable.
Two considerations can limit the necessary quality.
Firstly, a part of a structure designed for the ultimate limit state may only be lightly stressed
under fatigue loading, or may not be subject to many cycles of stress so that a detail category
with a low fatigue strength is adequate. This situation applies, for example, to certain railway
bridge girders above 40m span.
Secondly, the welded joint considered may not determine the fatigue strength of the component.
For example, a stiffener or shear connector attached to the flange of a plate girder would bring
the fatigue strength down to category 80 or even 71; in such a case there is no need to specify a
quality of transverse butt weld in that flange which would give a fatigue strength corresponding
to category 112. The same applies to the longitudinal welds attaching that flange to the web.
7. DESIGN CONSIDERATIONS
The best way of avoiding the consequences of joints with low fatigue strength is not to have
such joints.
To show how to approach this desirable design, consider the plate girder shown in Figure 13,
with cope holes, stiffeners, transverse butt welds in the flanges and longitudinal fillet welds.

63

The cope holes limit the fatigue strength to category 71. Figure 6 showed that they are
unnecessary and therefore they should not be there. Eliminating them is the first improvement.
The stiffeners, if more than 12mm thick, are in the same low category. They may be required to
strengthen the web to resist shear forces or concentrated loads applied to the top flange, to resist
lateral loads, e.g. wind forces, or to stabilise the top flange against lateral buckling.
If the stiffeners are only required to strengthen the web, they can be avoided if the web is made
thicker. This improves the fatigue strength of the girder and will often be more economical than
providing stiffeners, except in deep girders.
If the stiffeners are deleted, the fatigue strength of the girder is now limited by the transverse
butt welds and the longitudinal fillet welds.
The transverse butt welds can and should be avoided if plates for the length of the girder can be
obtained in one piece. This is on the basis that the butt welds are likely to cost more than the
material which can be saved by using them. If butt welds must be used, they can and should be
placed at a point where the fatigue strength of a reasonable quality weld does not limit the
fatigue strength of the component. In short, they should not be placed at points of maximum
stress.
If butt welds are made "harmless" as suggested, the longitudinal fillet welds remain. Here the
choice of quality may be a matter of economics if the fatigue loading is such that it determines
the design.
In short built-up girders where severe fatigue loading is the main design criterion, even
longitudinal welds can be moved from highly stressed areas by using T sections for the flanges.
This is not cheap, but may, in some cases, be economical.
Finally, if the required girder is short enough, it should be possible to find a rolled section which
can be used without any welded joints, at least in highly stressed areas. This would achieve the
objective of "no joints".
This short example does not, of course, represent all design problems. It does, however, show
that it is often possible to reduce or eliminate the effect of joints on the fatigue strength of a
component of a structure.

64

It must be remembered, however, that eliminating joints is not the primary objective in design.
If joints are required they must be provided. Their design must be based on realistic assumptions
as to their quality and fatigue strength.
8. CONCLUDING SUMMARY

All joints are potential points of weakness in a structure.

Faults in workmanship reduce the fatigue strength of joints.

Assuming a fatigue strength in design implies ensuring a corresponding quality of

workmanship.

Inspection and non-destructive tests have limitations in proving quality.

Choice of quality is affected by the likely available quality of workmanship, practicable

inspection and cost.

Good design can reduce the effect of joints on the fatigue strength of a structure.

9. REFERENCES AND FURTHER READING

[1] Eurocode 3 "Design of Steel Structures" ENV 1993-1-1:Part 1.1 General Rules and Rules
for Buildings, CEN, 1992.
[2] British Standard BS5400: Part 10: 1980, Code of Practice for Fatigue, British Standards
Institution, London
10. ADDITIONAL READING
1. ORE/D 130/RP 1. 1974 and ORE/D 86/RP 3, 1971
published by the European Rail Research Institute (ERRI), formerly ORE, Oudenoord
500, NL-3513 EX, Utrecht, The Netherlands.
2. Gurney, T.R., "Fatigue of welded structures" Cambridge University Press, 1968.
3. Haibach, E., "Betriebsfestigkeit - Verfahren und Daten zur Bauteilberechnung." VDI
Verlag, Dsseldorf, 1989.

65

Lecture 12.4.1: Fatigue Behaviour of

Hollow Section Joints (I)
OBJECTIVE/SCOPE:
To gain an understanding of the fatigue behaviour of hollow section joints and the available
methods of design.
PRE-REQUISITES:
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.3: Effect of Workmanship on Fatigue Strength of Longitudinal and Transverse Welds
Lecture 13.1: Application of Hollow Sections in Steel Structures
Lecture 13.2: The Behaviour and Design of Welded Connections between Circular Hollow
Sections under Predominantly Static Loading
Lecture 13.3: The Behaviour and Design of Welded Connection between Rectangular Hollow
Sections under Predominantly Static Loading
RELATED LECTURES
Lecture 12.4.2: The Fatigue Behaviour of Hollow Section Joints (II)
SUMMARY
66

In hollow section joints the stiffness along the intersection of the connected members is
generally rather non-uniform, which may result in large peak stresses. The peak stress ranges
determine the fatigue behaviour to a large extent.
This lecture describes the basic behaviour and introduces the methods of analysis. Detailed
design is presented in Lecture 12.4.2.
NOTATION
The notation of Eurocode 3 [1] has been adopted.
1. INTRODUCTION
Hollow sections (circular, square and rectangular) are used in many applications subjected to
fatigue loading, e.g. cranes, bridges, offshore jacket structures and several applications in
mechanical engineering. The phenomenon of fatigue, the factors influencing it, definitions and
loading are described in the Lectures 12.1 to 12.3. In these lectures it is shown that the peak
stress ranges determine the fatigue life of a particular connection to a large extent.
In Lecture 13.1 it is shown that the most economical construction of hollow section structures is
obtained by the direct welded connection of hollow section members avoiding stiffeners or
gusset plates. In such a connection the stiffness around the intersection is not uniform, resulting
in a geometrical non-uniform stress distribution, as shown in Figure 1, for an X-joint of circular
hollow sections. This non-uniform stress distribution depends on the type of loading (axial,
bending in-plane, bending out-of-plane) and the connection (type and geometry). Thus, many
cases exist. For this reason the fatigue behaviour of hollow section joints is generally treated in
a different way to that, for example, for welded connections between plates.

The fatigue behaviour can be determined either by -N methods or with a fracture

mechanics (F.M.) approach.
67

The various -N methods are based on experiments resulting in -N graphs with a

defined stress range on the vertical axis and the number of cycles N to a specified failure
criterion on the horizontal axis. The F.M. approach is based on a fatigue crack growth model.
The material crack growth parameters of the model can be determined from standardized small
specimens and the influence of the connection geometry is incorporated in the stress intensity
factor K, see Lectures 12.10 to 12.15. This lecture describes the -N methods.
2. GEOMETRIC STRESS OR HOT SPOT STRESS APPROACH
2.1 Definition of Geometric Stress and Stress Concentration Factors
In the -N concepts the stress range and failure criterion have to be defined. Considering the
X-joint in Figure 1, "nominal" stresses are shown in the brace members and the peak stresses at
the connection, i.e. at the intersection of the brace and chord members.
For axial loading, the nominal stresses in the members are defined. However, for bending
moments a certain cross-section has to be defined. Considering the peak stress (Figure 2) at the
intersection, the peak stress considered has to be defined, since for a certain loading the actual
peak stress is determined by:

the global geometry of the joint (type of joint and joint parameters)

the overall configuration of the weld (fillet weld, butt weld; flat, convex, concave)

the local condition at the weld toe (radius of the weld toe, undercut, etc.).

The local condition at the weld toe depends on the fabrication (welder, welding conditions and
welding process). This effect is generally indirectly incorporated in the scatter of the test results.
The same applies to the shape of the weld (flat, convex or concave). The fatigue behaviour of
fillet welded specimens is sometimes related by factors to that of butt welded specimens.
Initially it was thought that if a geometric (or so-called hot spot) peak stress range is used which
takes the global geometry and the loading into account, the fatigue behaviour of all types of
68

joints could be related to one basic S r-N line. However, the crack propagation not only depends
on the actual peak stress range, but on the whole stress pattern. Consequently, the stress
gradients also have an influence. At present these stress gradients cannot be incorporated in a
proper manner in a concept based on the geometric or hot spot stress range. Furthermore, as will
be shown later the thickness also has an influence which is independently taken into account.
In the geometric stress or hot spot stress approach the geometrical stress range is used as the
basis for analysis. The geometric hot spot stress (range) is defined as the maximum extrapolated
stress (range) to the weld toe, taking the global geometrical effects into account.
The extrapolation is defined in such a way that the effects of the global geometry of the weld
(flat, concave, convex) and the condition at the weld toe (angle, undercut) are not included in
the geometric stress. Therefore, the first point of extrapolation should be outside the influence
area of the weld, see Figure 2.
For linear extrapolation, two points are defined at the crown and saddle position of chord and
brace [2]. Based on the work of Gurney [3] and Van Delft [4] this first point can be taken at 0,4
(to or t1) with a minimum of 4 mm from the weld toe. The second point is defined depending on
the type of hollow sections used (circular or rectangular).
In those cases where the geometric stress distribution is not linear, a quadratic extrapolation is
defined with well defined measuring points, see Lecture 12.5.
In some codes it is stated that the principal stress should be extrapolated to the weld toe.
However, this has several disadvantages [5]:

The stress component perpendicular to the weld toe governs the crack growth along the
weld toe.

The direction of the geometrical principal stress close to the weld toe is (nearly)
perpendicular to the weld toe, but changes at locations further away from the weld toe.
This would favour an extrapolation along a curved line, which is difficult.

If the extrapolation is carried out along a line perpendicular to the weld toe, an
extrapolation of the principal stresses might result in a lower geometric principal stress
than by an extrapolation of the stresses perpendicular to the weld toe.

For different load cases the direction of principal stress might differ, prohibiting
superposition of load cases.

Another aspect is that the exact direction of the principal stress due to residual stresses
is not known.

In view of the above arguments, extrapolation of stresses perpendicular to the weld toe is
favoured.
For a particular loading case and a particular type of joint with a defined geometry, the
extrapolated geometric or hot spot stress can be determined from measurements on actual steel
specimens or acrylic models or with finite element calculations. However, this is not suitable for
design. For this reason the geometric or hot spot peak stresses are related by stress concentration
69

factors to the nominal stress in the member (in most cases the brace) which causes the
geometric stress at the intersection of the brace with the chord. For example, for an X-joint
without chord loading the stress concentration factor for a particular location (chord, brace;
crown or saddle) is defined as:

SCFi.j.k =

(1)

where
i is the chord or brace
j is the location, e.g. crown or saddle for CHS joints
k is the type of loading
In this way, stress concentration factors can be determined for various load conditions (axial
loading, bending in-plane and bending out-of-plane) at various locations, e.g. crown and saddle
positions of chord and brace.
Based on parametric finite element studies, parametric formulae have been developed which
give the stress concentration factors for various locations and loading. For a combined loading,
the nominal stress ranges of the brace have to be multiplied by the relevant stress concentration
factor for that particular location i.j. and the relevant loading case k, e.g:
i.j.k = axbrace . SCFaxi.j + bipbrace . SCFbipi.j + bopbrace . SCFbopi.j
(2)
This has to be done for the chord and brace (i) at various locations (j), see Figure 3, for a T-joint.

The geometric stresses considered until now are caused by the forces or moments in the brace.
But the forces in the chord will also cause stress concentrations at the intersection, although
they are considerably smaller. These effects also have to be incorporated. But here the stress
70

concentration factors are related to the nominal stress in the chord. The effect of chord loading
on the geometric peak stress in the brace is generally small and can be neglected. For the chord,
however, the stress concentration factor can reach values up to 2,5 (see Lecture 12.4.2).
With the method described above, the geometric stress range can be determined for various
locations of the chord and the brace considering the relevant loading.
Since the fatigue behaviour depends on the thickness, the maximum geometric or hot spot stress
range has to be determined for the chord and the brace considering different thicknesses. Using
the -N curve for geometric or hot spot stress, the number of cycles to failure can be
determined.
2.2 Definition of Fatigue Life
The fatigue life is generally specified as the number of cycles N for stress or strain of a
specified character that a given joint sustains before failure of a specified nature occurs.
Various modes of failure can be considered [6], e.g.

first visible crack

crack through the wall

a certain crack length

end of test (complete loss of strength)

Figure 4 shows the relationship between extrapolated, measured geometric stress at a critical
location versus the number of cycles. The fatigue life of welded hollow section joints is related
to both crack initiation and crack propagation. Their importance depends on the size and type of
joint, e.g. the initiation period may cover 10 to 80% of the total fatigue life.

71

Usually, a crack through the wall is adopted as the failure criterion for hollow section joints,
which corresponds to about 80% of the total fatigue life of a joint.
2.3 Thickness Effect
The reason that a lower fatigue strength is found for specimens with larger thicknesses, where
specimens have the same geometry and loading and the same geometric stress range but a
different size, is attributed to the following [5, 7]:

Geometrical effects

Although the geometry might be the same, the stress gradient at the notch is less steep for larger
thicknesses. As a result the stresses at the crack tip are larger, thus increasing the crack growth.
The geometry is not completely scaled, e.g. the radius of the weld toe is not increased as much
as the wall thickness, resulting in a larger thickness effect.

Statistical effects:

Statistically, in a larger volume, the probability of a larger defect increases and the fatigue
strength decreases with increasing defect size.

Technological effects:

In larger thicknesses, the grain size is coarser, the yield strength is lower, the residual stresses
are higher, the toughness is lower and the probability of hydrogen cracking increases, all
resulting in a lower fatigue strength for thicker specimens.

Another factor contributing to the influence of thickness is the stress state, i.e. plane
strain vs plain stress.

Early work by Gurney based on plated specimens gave the following thickness correction for
the fatigue strength for a particular number of cycles:
t = t reference . [t/treference]-0,25

(3)

This influence has also been adopted in Eurocode 3 for thicknesses exceeding 25mm. For
smaller thicknesses, no correction is given in Eurocode 3 at present, although the thickness
effect is even larger especially for hollow section joints, since the effect increases with the stress
or strain gradient.
Further work in France and the U.K. (within the framework of the offshore research
programmes) on thicknesses of 16 mm and more resulted in:
t = t reference . [t/treference]-0,30

(4)

This relation is now proposed for new standards [8]. For smaller thicknesses there is not only a
larger thickness effect, but also the slope of the -N curve changes from m = - 3 for higher
thicknesses to m = - 4 to - 5 for very small thicknesses [5].
72

Based on the results of ECSC and CIDECT sponsored research programmes, the following
thickness corrections for hollow section joints have now been proposed for Eurocode 3 [1], see
Figures 5a and 5b:

For thicknesses of 4 to 16 mm:

t = t = 16 mm .

(5)

For thicknesses of 16 mm and more:

t = t = 16 mm .

(6)
73

For thicknesses below 4 mm no guidance is given, since the fatigue behaviour may be adversely
affected by the welding imperfections at the root of the weld.
2.4 Fatigue Limit
Various investigations have shown that the fatigue limit, i.e. where the -N line changes to a
horizontal line, depends on the notch effect. For example, for basic steel, the fatigue limit for
constant amplitude loading might be of the order of 2 106 cycles, whereas for welded
connections with high peak stresses it will be about 10 7 cycles. Many offshore codes and also
the IIW recommendations adopt 107 for tubular connections. In Eurocode 3 [1], one general
limit of N = 5 106 is given.
For random or variable amplitude loading with stresses exceeding the , at 5 106 certain
interaction effects may appear with the result that smaller stress ranges can have an influence on
the fatigue life. This is incorporated by changing the slope after 5 106 cycles to m = - 5. The
fatigue limit for variable amplitude loading is given for all welded connections including hollow
section joints at 108 cycles. In certain offshore codes 2 108 is given. However, no tests are
available to check whether this is correct.
2.5 Fatigue Class and -N Curves
Various re-analyses of test results [5, 8, 9] have shown that for the geometric stress or hot spot
stress concept, the following classes (stress range at 2 106 cycles) can be adopted for hollow
section joints with 16 mm thickness:

Circular hollow section joints

Square or rectangular hollow section joints
For other thicknesses, the thickness corrections according to Equations (5) and (6) have to be
adopted for N 5 106. For N > 5 106 the -N curve remains parallel to the -N
line for 16 mm thickness (thus, the same thickness effect is used as for N = 5 106).
As stated above, the curves for thicknesses of 16 mm and larger have a slope m = -3,
whereas below 16 mm the curves change to a smaller slope due to the thickness correction.
Figure 6 shows the resulting basic curves for 16 mm thickness. In Figure 7 this is worked out
for square hollow section joints including the thickness correction. ECSC and CIDECT [5, 9]
research has proved that for thicknesses up to 8 mm, these curves can be used for butt welded
and fillet welded connections or for combinations of both welds. For thicknesses larger than 8
mm, butt welds should be used.

74

In Eurocode 3 [1], Class 90 is given for butt welded joints with controlled weld profile and
lower classes are given for butt welds (Class 71) and fillet welds (Class 36). However, it is
stated that higher values may be used if sufficient data is available for justification. The
outcome of the research with basic classes of 112 and 90 for 16 mm thickness will be the basis
for the next revision of Eurocode 3.
2.6 Low Cycle Fatigue
In Eurocode 3, -N curves are given for N = 104 cycles and higher [1]. It is further stated
that the stress range should not exceed 1,5 times the yield stress in order to avoid alternating
yielding. In general, in the case of low cycle fatigue, the stress range concept is not valid and the
fatigue strength is determined more by strain range. This limitation is true for concepts based on

75

nominal stress; but with a geometric stress concept the geometric peak stress ranges are
considered, which are only locally present.
As shown in Figure 5, the fatigue test results of hollow section joints for N = 10 3 are still in line
with the -N curves given. However, this basis might result in very high theoretical stress
ranges (in some cases 5 times the yield stress). These extended -N curves can be used; but
a brittle failure check should be carried out to determine the critical crack depth.
2.7 Design Procedure
For each potential crack location the long term distribution of relevant stress ranges should be
established and the probable fatigue life should satisfy the Palmgren-Miner linear cumulative
damage rule: ni/Ni 1,0.
An arbitrary joint could be checked by following the steps given below, also shown in a flow
chart in Figure 8, for a T- and X-joint of square hollow sections.

1. Load and geometry of the joint should be determined first.

2. From simple formulae the nominal stresses in the members can be determined.

76

3. Determine the joint parameters = b1/bo, or

2 = b0/t0 or

and = t1/t0.

4. From SCF formulae or graphs, the SCFs for the various load cases can be determined,
e.g. for lines A to E for square hollow sections, see Figure 3.
5. Determine the geometric stress range for the various lines by multiplication of the
nominal stress ranges with the relevant stress concentration factors. Only the highest
stress range in the chord and the highest stress range in the brace need to be considered.
6. The number of cycles to failure N for both brace and chord is obtained from the -N
line for the appropriate wall thickness. The lowest number of cycles in brace or chord
determines the fatigue strength of the joint.
3. CLASSIFICATION METHOD
In the previous section it has been shown that the geometric (peak) stress range largely
determines the fatigue life. The stress range depends on the type of joint, geometry and loading.
If the fatigue strength were to be based on nominal stress ranges, then on the one hand it would
be simpler for the designer, but on the other hand, an atlas would be required to cover all cases
[10].
The classification method now offers a compromise between both concepts. The test results are
analyzed based on nominal stress range in the brace and then grouped together in such a way
that the main influencing geometrical parameters are taken into account. For connections in
which the geometrical stress concentration factor varies to a large extent, e.g. X- or T-joints, see
Figure 9, such grouping is not possible. However, for K- and N-joints, a classification can be
adopted if the thickness parameter t o/ti is included and certain parameters are kept nearly
constant, e.g. gap, overlap, angle. Further, stringent limitations should be made for the range of
validity. This approach is currently included in Eurocode 3 [1] for K and N joints with
thicknesses up to 12,5 mm. A more detailed description is given in Lecture 12.4.2.

77

4. OTHER METHODS
Other methods [7, 10] are also given in the literature and guides, e.g.

failure criterion method

punching shear method

relation to static strength.

4.1 Failure Criterion Method

This method, developed at the University of Karlsruhe, is also based on the fact that the stress
concentration factor is indirectly taken into account by giving nominal stress ranges at
2 106 (all classes) in relation to the joint geometry parameters and loading. The data is
presented in diagrams. The -N curve to be used is then fixed by the design class. This
method is not used at present.
4.2 Punching Shear Method
This method has a certain relation with the classification method. However, here the punching
shear stress range is taken as the basis instead of the nominal stress range in the brace. This
method is used in American codes, such as the API and AWS standards.
4.3 Relation to Static Strength
In many Japanese publications the stress range is related to the static strength for the same
hollow section joints, but with a yield stress f y = 235 N/mm2. As fatigue is a quite different
phenomenon from static strength, one could expect poor correlation between the test results and
the -N line. However, the correlation is not worse than for the other simplified methods
discussed above.
This funding may be explained by the fact that the static strength depends on the same
geometrical parameters which influence the geometric stress and also that the incremental load
capacity of the joint between initial localized yielding and failure has some relation to the stress
gradient. Thus, within certain parameter ranges, a reasonable relationship can be obtained. This
method has of course no general validity since various factors are not included, e.g. tension vs
compression loading, effect of secondary bending moments, etc.
5. EFFECT OF SECONDARY BENDING MOMENTS
In lattice girder joints (e.g. K- and N-joints) secondary bending moments exist. For static design
these moments are not important if the critical members or joints have sufficient rotation
capacity. However, for fatigue design the peak stress range is the governing parameter and
secondary bending moments influence the peak stress (range). As a consequence, secondary
bending moments have to be considered in fatigue design.
Secondary bending moments are caused by various influences, such as:

the overall bending stiffness of the joint

78

the stiffness distribution in the joint along the intersection perimeter

eccentricities in noding of the members

Figure 10 shows, as an example, a K-joint in which three sides at the intersection are stiffer than
the side at the heel, resulting in a secondary bending moment since the reaction force is not in
line with the force in the brace member.

The secondary bending moment can only be accurately taken into account if the joints are
modelled as substructures with finite elements as shown in Figure 11. However, such modelling
is still the future for engineering offices.

79

To avoid these complicated analyses, Eurocode 3 [1] gives factors to account for the secondary
bending moment effects (Tables 1a and 1b). The stress ranges obtained for axial loading should
be multiplied by these factors if the secondary bending moments are not included in the
analysis. The values given in these tables are based on measurements in actual girders and tests
as well as finite element calculations. Comparison of the various values shows that the
secondary bending moments in girders with N-joints are larger than those in girders with Kjoints. The secondary bending moments in girders of rectangular or square hollow section joints
are larger than those in girders with circular hollow section joints. These effects are caused by
the stiffness along the intersection perimeter and the stiffness of the brace members (vertical
brace for N-joints). In general, overlap joints of rectangular hollow sections give lower
secondary bending moments in the braces than in the case of gap joints. For girders of circular
hollow sections a similar effect is given in the tables, although current research does not confirm
this [11]. More evidence will be obtained when current research projects are finished.
6. SIMPLE CONNECTIONS AND ATTACHMENTS
For simple connections, such as butt welded end-to-end connections or hollow sections
connected to plates, a simple classification is used as for other sections. The same procedure is
followed for attachments, cover plates, etc. The classification is in line with the classifications
given for similar welded details of I-sections, see Lecture 12.4.2.
7. PARTIAL SAFETY FACTORS
For hollow section joints, the same partial safety factors apply for the stress range as for other
structures loaded in fatigue. Eurocode 3 recommends factors (Table 2) which depend on the
type of structure (fail safe and non fail safe) and the possibility of inspection and maintenance.
For more detailed information, see Lecture 12.8.
8. CUMULATIVE DAMAGE
Eurocode 3 has adopted the Palmgren-Miner rule to determine cumulative fatigue damage, i.e.:

D=

(7)

Although the real damage also depends on the spectrum of the loading and the sequence of
stress ranges, the Miner rule is the simplest available for determining damage. Its use gives
results which are not worse than those given by other rules. For more detailed information,
see Lecture 12.2.
9. CONCLUDING SUMMARY

The fatigue behaviour of hollow section joints is largely influenced by the geometric
stress range (also called the hot spot stress range)

Various analyses and design methods are used, of which the geometric or hot spot stress
method is the most generally valid one

80

The determination of the geometric or hot spot stress should be carried out by
extrapolation to the weld toe excluding the effects of the weld geometry and the local
effects at the weld toe

Hollow section joints show a considerable thickness effect, especially for small
thicknesses

Stress concentration factors should be used with care (only within the range of validity
of the parametric formulae)

The classification method can only be used for K- and N-joints within a limited range of
validity

The secondary bending moments in lattice girders have to be incorporated in the fatigue
analysis.

10. REFERENCES
[1] Eurocode 3: "Design of Steel Structures": ENV 1993-1-1: Part 1, General Rules and Rules
for Buildings, CEN, 1992.
[2] Steel in marine structures, Proceedings of the 2nd International ECCS Offshore Conference,
Institut de Recherches de la Siderurgie Franaise, Paris, October 1981.
[3] Gurney, T.R.: "Fatigue of welded structures", Cambridge, 2nd ed.
[4] Van Delft, D.R.V.: "A two dimensional analysis of the stresses at the vicinity of the weld
toes of welded tubular joints", 1981, Report 6-81-8, Stevin Laboratory, Delft University of
Technology.
[5] Van Wingerde, A.M.: "The fatigue behaviour of T- and X-joints made of square hollow
sections", Ph.D. thesis, Delft University of Technology, 1992.
[6] Noordhoek, C. and De Back, J., Eds.: Steel in Marine Structures, Proceedings of the 3rd
International ECSC Offshore Conference on Steel in Marine Structures (SIMS '87), Delft, The
Netherlands, June 15-18, 1987.
[7] Marshall, P.W.: "Design of welded tubular connections: Basis and use of AWS code
provisions", Ph.D. thesis, Elsevier Applied Science Publishers Ltd., Amsterdam/London/New
York/Tokio.
[8] Reynolds, A.G., Sharp J.V.: "The fatigue performance of tubular joints - An overview of
recent work to revise Department of Energy guidance", 4th International Symposium of
Integrity of Offshore Structures, p. 261-277, Elsevier Applied Science Publishers Ltd.,
Amsterdam/ London/New York/Tokio. Glasgow, U.K., July 1990.
[9] Wardenier, J., Mang, F., Dutta D.: "Fatigue strength of welded unstiffened RHS joints in
lattice structures and Vierendeel girders". ECSC Final Report, ECSC 7210-SA/111.
[10] Wardenier, J.: "Hollow Section Joints", Delft University Press, 1982, ISBN 90.6275.084.2.
81

[11] Romeijn, A., Wardenier, J., de Komming, C. H. M., Puthli, R. S,, Dutta,D.: "Fatigue
Behaviour and Influence of Repair on Multiplanar K-Joints Made of Circular Hollow Sections".
Proceedings ISOPE '93, Singapore.
Table 1a Coefficients to account for secondary bending moments in joints of lattice girders
made from circular hollow sections
Type of joint
Gap joints

Overlap joints

Chords

Verticals

Diagonals

1,5

1,3

1,5

1,8

1,4

1,5

1,2

1,5

1,65

1,25

Table 1b Coefficients to account for secondary bending moments in joints of lattice girders
made from rectangular hollow sections
Type of joint
Gap joints

Overlap joints

Chords

Verticals

Diagonals

1,5

1,5

1,5

2,2

1,6

1,5

1,3

1,5

2,0

1,4

Table 2 Partial safety factors M according to Eurocode 3

Inspection and access
Periodic inspection and maintenance. Accessible joint detail.
Periodic inspection and maintenance.

82

Poor accessibility.

Lecture 12.4.2: Fatigue Behaviour of

Hollow Section Joints II
OBJECTIVE/SCOPE:
To give information for the design of hollow section joints with guidelines for an optimal
design.
PRE-REQUISITES
Lecture 12.1: Basic Introduction to Fatigue

83

Lecture 12.3: Effect of Workmanship on Fatigue Strength of Longitudinal and Transverse Welds
Lecture 13.1: Application of Hollow Sections in Steel Structures
Lecture 13.2: The Behaviour and Design of Welded Connections Between Circular Hollow
Sections under Predominantly Static Loading
Lecture 13.3: The Behaviour and Design of Welded Connections Between Rectangular Hollow
Sections under Predominantly Static Loading
RELATED LECTURES
Lecture 12.4.1:Fatigue Behaviour of Hollow Section Joints I
SUMMARY
The geometric stress concept and the classification method used for the fatigue design of hollow
section joints are discussed in more detail and illustrated by design examples. Guidelines and
simplified design graphs are given to give insight into the stress concentration factors to
stimulate optimal design.
The limitations and the validity of various approaches are also discussed.
NOTATION
The notation of Eurocode 3 has been adopted.
1. INTRODUCTION
In Lecture 12.4.1, the fatigue behaviour of hollow section joints and methods of analyses are
discussed. In Eurocode 3, two methods are adopted for the design of hollow section joints [1].
The geometric stress method (sometimes called the hot spot stress method) can generally be
used for all types of connections between hollow sections. However, sufficient data regarding
stress concentration factors for various types of joints and loadings are required.
The classification method also given in Eurocode 3 for K- and N-type joints with wall
thicknesses up to 12,5 mm is a simplified method, only valid for a particular range of
geometrical parameters.
In this lecture these two methods are discussed in more detail and the method of design is given.
2. MODELLING OF THE STRUCTURE
For fatigue design, the stress ranges in the members have to be known. The stresses determined
depend directly on the idealization of the structure.
Trusses made of hollow sections generally have welded connections with continuous chords.
For ease of fabrication the diagonals are sometimes connected with a certain noding eccentricity
e, see Figure 1.
84

To obtain the correct load and stress distribution in the members, there are various methods of
modelling [2].
a. The best method is to model the connections as substructures, as shown in Fig. 11 of Lecture
12.4.1. In this way, the influence of the geometry of the connections is directly taken into
account. Even the governing extrapolated geometric stresses at the weld toes can be determined
directly. However, this method requires excellent pre- and post-processors for the finite element
modelling and sufficient computer capacity. Only specialized research institutes and companies

85

can handle this method at present. Furthermore, the designer who is modelling should be a
specialist in the use of the correct elements and the correct meshes.
b. Another method is to use parametric formulae to determine the connection stiffness. The truss
is then modelled with continuous chords and the diagonals are connected by springs
representing the connection stiffness to the chord. Eccentricities should be incorporated in the
model. In this way the proper bending moments in the members can be determined. The axial
load distribution in the members is hardly influenced by the modelling, e.g. pin, spring or
rigidly connected diagonals, see Figure 2. For bending moments, however, considerable
differences may occur, since the actual bending stiffness is influenced by the axial loads present
(interaction). In offshore engineering it is common practice to assume rigidly connected
members. However, the value of the bending moments is reduced by taking not the moments at
the noding points but those at the chord face. Due to a lack of evidence of the stiffness
characteristics and for simplicity this simplification is used.

c. The simplest method of modelling is to use continuous chords with the diagonals connected
with pins to the chord. Eccentricities exceeding 0,55 (d o or ho) e 0,25 (do or ho) should be
incorporated as shown in Figure 3.

86

The effect of secondary bending moments should be included by multiplying the stress ranges
due to axial forces by the factors given in Tables 1a and 1b of Lecture 12.4.1.
This simplified modelling assumes indirectly that the stress concentration factors for bending
are similar as those for axial forces, which is not generally valid. The stress concentration
factors for bending in-plane are generally lower than those for axial force.
There are of course other methods of modelling, but those mentioned above represent the most
commonly used ones.
3. END-TO-END CONNECTIONS AND ATTACHMENTS
End-to-end connections and connections with plates and hollow sections with attachments are
classified in the normal way, i.e. a fatigue class is given for 2 106 cycles as shown in Table 1.
Here a slope constant of m = 3 is adopted for N 5 106. For spectrum loading, a slope
constant of m = 5 is adopted for 5 106 to the cut off limit 10 8. It should be noticed that for
end-to-end connections an opposite "thickness effect" is included. In these connections no or
only small geometrical effects are present, causing the crack to initiate from the root of the
weld. Since these welds are made from one side (not counter welded) the root effect is more
severe for smaller thicknesses. Here the limit is taken at 8 mm. For thicknesses larger than 25
mm the normal thickness effect should be applied, i.e.
t = t=25[t/25]-0,25

(1)

4. GEOMETRIC STRESS METHOD

4.1 Reference Curves
The basic reference curves for 16 mm wall thickness, according to Eurocode 3 [1] are given in
Figure 4. The equations for the reference curves and the thickness effects to be adopted are also
given.

87

As discussed in Lecture 12.4.1, for hollow section joints with partial penetration welds or fillet
welds, sufficient evidence is now available to use the same curves as for full penetration butt
welds.
4.2 Determination of Geometric Stresses by F.E. Modelling
The geometric stress should be determined by extrapolation of the geometric stress outside the
influence region of the weld to the weld toe. For most connections of circular hollow sections
the geometric stress has a linear part and a linear extrapolation can be used as shown in Figure
5a.

88

For square or rectangular hollow section joints the geometric stress is non-linear and a quadratic
extrapolation gives a better accuracy and lower scatter in test results, Figure 5b. This
extrapolation is more consistent with the mode of failure (location of the crack).

89

Determination of the geometric stresses by F.E. modelling requires proper modelling (type of
elements, mesh), preferably by methods which have been calibrated by tests [2]. Due to the
difference between actual dimensions and nominal dimensions, the calibration should be done
with care using the actual dimensions, see Figure 6.

4.3 Stress Concentration Factors

The geometric stress can also be obtained by multiplying the nominal stresses by the relevant
stress concentration factors. Many parametric formulae are now available for circular hollow
section joints. The most extensive set which gives good agreement with measurements is given
by Efthymiou [3]. For square hollow section joints, a set of parametric formulae for T- and Xjoints is given by Van Wingerde [4]. For K- and N-joints of square hollow sections, formulae are
available, however only for a particular combination of axial loads and bending moments [5].
Thus, they are not generally applicable. Currently, new parametric formulae are being
developed in the framework of a CIDECT programme. These formulae will become available in
1994.
To give formulae for all types of joints, for various locations and for various loadings is beyond
the scope of this lecture. However, some graphical presentations are given in Figures 7 - 15 for
circular hollow section joints and in Figures 16 - 18 for T- and X-joints of square hollow
sections. If the effect of the stiffness distribution along the intersection perimeter is clearly
understood, then the tendency of the graphs will be clear. All graphs show that the SCF for the
saddle location achieves a maximum for medium ratios. For T- and X-joints minimum values
are obtained for = 1,0. Here, the SCF changes very rapidly for small variations of . For this
reason, for X-joints of circular hollow sections, lower values than those for = 0,95 are not
taken, because for = 1,0 the weld might cause an eccentricity.

90

91

92

93

94

95

96

97

98

Note that the effect of the bending moment in the chord in a T-joint is included in the SCF's for
T-joints of circular hollow sections, whereas for T-joints of square hollow sections it has been
excluded, allowing the same presentation as for X-joints.
It is recommended that a minimum stress concentration factor of 2 is adopted to cover cracking
from the root of the weld. Furthermore, the SCF's should be multiplied by the following
correction factors for square hollow section joints if fillet welded connections are used:
Brace: 1,4 (for lines A and E)
Chord: 1,0 (no correction factors)
These factors cover the effect of brace wall bending. For the chord, the weld toe is located
further away from the brace, resulting in a lower SCF. However, insufficient evidence is
available yet to quantify this in general. Therefore, a factor 1,0 is adopted for the time being.
For an optimal design the geometric stress concentration factors should be low, i.e.
low and values and low or high values.
As indicated in Lecture 12.4.1, the geometric stress range can be obtained by multiplying the
relevant stress concentration factor (location and loading) by the nominal stress range, which
causes the stress peak.
Care has to be taken that the stress concentration factors are only used within the range of
validity. Figure 19 shows, for example, that the effect of multiplanar loading can be
considerable.

99

4.4 Fatigue Life

The fatigue life is determined by checking the highest geometric stress for the chord as well as
for the brace multiplied by the appropriate M factor with the basic -N curve using the
appropriate thickness.
5. CLASSIFICATION METHOD
The classification method is only used for axially loaded K- and N-type joints of circular and
rectangular hollow sections, since the variation in SCF is too large for other types of joints.
In principle, the classes should be a lower bound, defined by the basic reference curve for the
geometric stress range divided by the SCF's which can be obtained within the range of validity.
The classification is based on an analysis of relevant test results, taking account of the
parameter and using a lower bound. In this approach, the effects of other influencing
parameters and the thickness effects are combined to some extent.
Multiplication of the class by a minimum SCF of 2,0 gives values far exceeding those of the
basic reference curve for geometric stress, even if the thickness effect is incorporated. On the
other hand, the two approaches cannot be compared directly, since the slopes are different, i.e.
for the classification method a fixed slope constant of m = 5 is adopted.
Although Eurocode 3 [1] gives the range of validity shown in Table 1, it would be better to
change some of the limits, i.e.

100

4 to or t1 8 mm instead of to or t1 12,5 mm
bo/to t1 or do/to t1 100 instead of bo/to or do/to 25
These limits are in better agreement with the test results on which this classification method has
been based.
In view of these comments, one could ask: "Why has the classification method been adopted?"
At the time of drafting Eurocode 3, insufficient evidence was available for SCF's of hollow
section joints. Further, many designers are not familiar with the geometric stress concept.
Therefore the classification method was favoured by many countries.
The design is very simple, since the designer determines the nominal stress ranges in the brace
with one of the methods described in Section 2 considering the factors for secondary bending
moments and the relevant M factor. The design class gives the -N curve and the fatigue
life can be determined.
6. GENERAL REQUIREMENTS FOR WELDING
The connections at the welded joints should be made over the entire perimeter of the hollow
sections by means of a full penetration weld, partial penetration weld, fillet weld, or a
combination. Full penetration welds should be used if:

The brace has a wall thickness larger than 8 mm.

The angle at the toe of the brace is larger than 120 .

For joints between square or rectangular hollow sections with = 1,0, fillet welds can
only be used at the side of the brace perpendicular to the axis of chord. Therefore, the
sides parallel to the chord axis should be welded by a full penetration butt weld.

Attention should be paid to the proper selection of materials and the welding procedure. To
avoid failure of the weld under static loading, the throat thickness of the fillet weld for steel up
to S355 is equal to or greater than the wall thickness of the brace (a t1).
Welding should start at the middle of the sides. If welding of joints commences at the corners of
the brace, the fatigue strength deteriorates. This may result in a decrease by a factor 2 on stress
range.
7. CONCLUDING SUMMARY
For the geometric stress method and the classification method:

The modelling of the structure should be done with care, considering secondary bending
moment effects.

In using F.E. modelling, proper consideration of elements and meshes is required, as

well as calibration with test results. The geometric stresses should be extrapolated to the
weld toe in a standardized manner.

101

Stress concentration factor formulae should be used within their range of validity. For
example the formulae for uniplanar joints cannot be used for multiplanar loading.

Fillet welds exhibit higher SCF's in the brace than butt welds.

For axially loaded T- and X-joints the highest SCF's occur for medium ratios (0,5 to
0,6)

The classification method and the geometric stress approach are not yet consistent.

Welding of hollow section joints should not start at the locations of high stress
concentrations, e.g. not at the corners of rectangular or square hollow sections, but at
the middle of the sides.

8. REFERENCES
[1] Eurocode 3: "Design of Steel Structures": ENV 1993-1-1: Part 1, General rules and rules for
buildings, CEN, 1992.
[2] Romeijn, A., Puthli, R.S., Wardenier, J.: "Finite element modelling of multiplanar joint
flexibility in tubular structures", Proceedings ISOPE '92, San Francisco, U.S.A.
[3] Efthymiou, M.: "Development of SCF formulae and generalised influence functions for use
in fatigue analysis", Offshore Tubular Joints Conference (OTJ '88), UEG Offshore Research,
Englefield Green Near Egham, U.K., 1988.
[4] Van Wingerde, A.M.: "The fatigue behaviour of T- and X-joints made of square hollow
sections", Ph.D. thesis, Delft University of Technology.
[5] Mang, F., Herion, S., Bucak, ., Dutta, D.: "Fatigue behaviour of K-joints with gap and with
overlap made of rectangular hollow sections", p. 297-310 of Proceedings "Tubular Structures",
edited by E. Niemi and P. Mkelinen.
9. ADDITIONAL READING
1. Dutta, D., Mang, F., Wardenier, J.: "Fatigue behaviour of welded hollow section joints",
CIDECT Monograph No. 7.
2. "Design of tubular joints for offshore structures", vol. 1, 2 and 3, UEG publication
UR33.
3. Romeijn, A., Puthli, R.S., Wardenier, J.: "The flexibility of uniplanar and multiplanar
joints made of circular hollow sections", Proceedings ISOPE '91, Edinburgh, U.K.
Table 1 Detail category: Hollow sections and simple connections
Details loaded by nominal normal stresses
Detail
Constructional detail
category

Description

102

m=3
Rolled and extruded products
160

See Fig T1-1

Non-welded elements.
Sharp edges and surface flaws to be improved by grinding
Continuous longitudinal welds

140

See Fig T1-2

Automatic longitudinal welds with no stop-start positions; proven

free of detectable discontinuities

Transverse butt welds

Butt welded end-to-end connection of circular hollow sections
71

Requirements
See Fig T1-3

- Height of the weld reinforcement less than 10% of weld

width; smooth transitions to the flat surface
- Welds made in flat position and proven free of detectable
discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified two detail categories higher, i.e. > 90
Transverse butt welds
Butt welded end-to-end connection of rectangular hollow sections

56

Requirements
See Fig T1-4

- Height of the weld reinforcement less than 10% of weld

width; smooth transitions to the flat surface
- Welds made in flat position and proven free of detectable
discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified two detail categories higher, i.e. > 71
Welded attachments (non load-carrying welds)

103

71

See Fig T1-5

Circular or rectangular section, fillet welded to another section.

Section width parallel to stress direction 100 mm
Welded connections (load-carrying welds)
Circular hollow sections, end-to-end butt welded with an
intermediate plate

50

See Fig T1-6

Requirements
- Welds proven free of detectable discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified one detail category higher, i.e > 56
Welded connections (load-carrying welds)
Rectangular hollow sections, end-to-end butt welded with an
intermediate plate

45

See Fig T1-7

Requirements
- Welds proven free of detectable discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified one detail category higher, i.e. > 50

Fig T1-1

Fig T1-2

Fig T1-3

Fig T1-4

104

Fig T1-5

Fig T1-6

Fig T1-7

Table 1 (continued) Hollow sections and simple connections

Details loaded by nominal normal stress (continued)
Detail
Constructional detail
category

Description

m=3
Welded connections (load-carrying welds)
40

See Fig T1-8

Circular hollow sections, end-to-end fillet welded with

an intermediate plate
Requirements
- Wall thickness less than 8 mm
Welded connections (load-carrying welds)

36

See Fig T1-9

Rectangular hollow sections, end-to-end fillet welded

with an intermediate plate
Requirements

105

- Wall thickness less than 8 mm

80

l 50 mm

71

50 < l 100 mm

50

l > 100 mm

80

t 12 mm

71

t > 12 mm

80

t 12 mm

71

t > 12 mm

Longitudinal attachments (non-load-carrying welds)

See Fig T1-10

The detail category varies according to the length of the

attachment l

Transverse attachments
See Fig T1-11

The end of the weld more than 10 mm from the edge of

the plate

Transverse attachments
See Fig T1-12

Diaphragms of rectangular girders welded to the flange

or web

Transverse attachments
80

See Fig T1-13 The effect of welded shear connectors on base material
Cruciform joints (load-carrying welds)
See Fig T1-14

Full penetration weld, inspected free of detectable

discontinuities

71
Requirements
- The maximum misalignment of the loadcarrying plates should be less than 15% of the
thickness of the intermediate plate
Cruciform joints (load-carrying welds)
See Fig T1-15

36

Fillet welded connection. Two fatigue assessments are

required
Firstly root cracking is evaluated by determining the
stress range in the weld throat area, Category 36
Secondly toe cracking is evaluated by determining the

106

stress range in the load-carrying plates, Category 71

Requirements
- The maximum misalignment of the loadcarrying plates should be less than 15% of the
thickness of the intermediate plate
50

t and tc 20 mm

36

t and tc 20 mm

Cover plates (load-carrying welds)

See Fig T1-16

End zones of single or multiple welded cover plates, with

or without frontal weld.
When the reinforcing plate is wider than the flange a
frontal weld, carefully ground to remove undercut, is
necessary

Fig T1-8

Fig T1-9

Fig T1-10

Fig T1-11

107

Fig T1-12

Fig T1-13

Fig T1-14

Fig T1-15

Fig T1-16

Table 2: Classes for classification method

108

Detail categories for lattice girder joints

Detail
category
m=5

Construction Details

Description

Joints with gap

90

to/ti 2,0

45

to/ti 1,0

See Fig T2-1

Circular hollow sections, K and

N joints

Joints with gap

71

to/ti 2,0

See Fig T2-2

Rectangular hollow sections, K

and N joints
Requirements

36

to/ti 1,0

0,5(bo-bi) g 1,1(bo-bi)
g 2to

Joints with overlap

71

to/ti 1,4

See Fig T2-3

K joints
Requirements

56

overlap between 30% and

100%

to/ti 1,0

Joints with overlap

71

to/ti 1,4

See Fig T2-4

N joints

109

to/ti 1,0

50

General Requirements
to, ti 12,5 mm 35 50 bo/to 25
0,4 bi/bo 1,0 0,25 di/do 1,0 bo/to 25
bo 200 mm do 300 mm -0,5do e 0,25do
-0,5ho e 0,25ho
Out-of-plane eccentricity: 0,02bo or 0,02do
Fillet welds are permitted in braces with wall thicknesses 8 mm
Table 2: Classes for classification method

For intermediate to/ti values, use linear interpolation between nearest detail categories

Note that the braces and the chords require separate fatigue assessments

Note regarding General Requirements:

Preferably use 4 to or t1 8 mm instead of 12,5 mm
Preferably use bo/to t1 or do/to t1 100 mm instead of bo/to or do/to 25

Fig T2-1

110

Fig T2-2

Fig T2-3

Fig T2-4

111

Lecture 12.4.2: Fatigue Behaviour of

Hollow Section Joints II
OBJECTIVE/SCOPE:
To give information for the design of hollow section joints with guidelines for an optimal
design.
PRE-REQUISITES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.3: Effect of Workmanship on Fatigue Strength of Longitudinal and Transverse Welds
Lecture 13.1: Application of Hollow Sections in Steel Structures
Lecture 13.2: The Behaviour and Design of Welded Connections Between Circular Hollow
Sections under Predominantly Static Loading
Lecture 13.3: The Behaviour and Design of Welded Connections Between Rectangular Hollow
Sections under Predominantly Static Loading
RELATED LECTURES
Lecture 12.4.1:Fatigue Behaviour of Hollow Section Joints I
SUMMARY
112

The geometric stress concept and the classification method used for the fatigue design of hollow
section joints are discussed in more detail and illustrated by design examples. Guidelines and
simplified design graphs are given to give insight into the stress concentration factors to
stimulate optimal design.
The limitations and the validity of various approaches are also discussed.
NOTATION
The notation of Eurocode 3 has been adopted.
1. INTRODUCTION
In Lecture 12.4.1, the fatigue behaviour of hollow section joints and methods of analyses are
discussed. In Eurocode 3, two methods are adopted for the design of hollow section joints [1].
The geometric stress method (sometimes called the hot spot stress method) can generally be
used for all types of connections between hollow sections. However, sufficient data regarding
stress concentration factors for various types of joints and loadings are required.
The classification method also given in Eurocode 3 for K- and N-type joints with wall
thicknesses up to 12,5 mm is a simplified method, only valid for a particular range of
geometrical parameters.
In this lecture these two methods are discussed in more detail and the method of design is given.
2. MODELLING OF THE STRUCTURE
For fatigue design, the stress ranges in the members have to be known. The stresses determined
depend directly on the idealization of the structure.
Trusses made of hollow sections generally have welded connections with continuous chords.
For ease of fabrication the diagonals are sometimes connected with a certain noding eccentricity
e, see Figure 1.

113

To obtain the correct load and stress distribution in the members, there are various methods of
modelling [2].
a. The best method is to model the connections as substructures, as shown in Fig. 11 of Lecture
12.4.1. In this way, the influence of the geometry of the connections is directly taken into
account. Even the governing extrapolated geometric stresses at the weld toes can be determined
directly. However, this method requires excellent pre- and post-processors for the finite element
modelling and sufficient computer capacity. Only specialized research institutes and companies

114

can handle this method at present. Furthermore, the designer who is modelling should be a
specialist in the use of the correct elements and the correct meshes.
b. Another method is to use parametric formulae to determine the connection stiffness. The truss
is then modelled with continuous chords and the diagonals are connected by springs
representing the connection stiffness to the chord. Eccentricities should be incorporated in the
model. In this way the proper bending moments in the members can be determined. The axial
load distribution in the members is hardly influenced by the modelling, e.g. pin, spring or
rigidly connected diagonals, see Figure 2. For bending moments, however, considerable
differences may occur, since the actual bending stiffness is influenced by the axial loads present
(interaction). In offshore engineering it is common practice to assume rigidly connected
members. However, the value of the bending moments is reduced by taking not the moments at
the noding points but those at the chord face. Due to a lack of evidence of the stiffness
characteristics and for simplicity this simplification is used.

115

c. The simplest method of modelling is to use continuous chords with the diagonals connected
with pins to the chord. Eccentricities exceeding 0,55 (d o or ho) e 0,25 (do or ho) should be
incorporated as shown in Figure 3.

The effect of secondary bending moments should be included by multiplying the stress ranges
due to axial forces by the factors given in Tables 1a and 1b of Lecture 12.4.1.
This simplified modelling assumes indirectly that the stress concentration factors for bending
are similar as those for axial forces, which is not generally valid. The stress concentration
factors for bending in-plane are generally lower than those for axial force.
There are of course other methods of modelling, but those mentioned above represent the most
commonly used ones.
3. END-TO-END CONNECTIONS AND ATTACHMENTS
End-to-end connections and connections with plates and hollow sections with attachments are
classified in the normal way, i.e. a fatigue class is given for 2 106 cycles as shown in Table 1.
Here a slope constant of m = 3 is adopted for N 5 106. For spectrum loading, a slope
constant of m = 5 is adopted for 5 106 to the cut off limit 10 8. It should be noticed that for
end-to-end connections an opposite "thickness effect" is included. In these connections no or
only small geometrical effects are present, causing the crack to initiate from the root of the
weld. Since these welds are made from one side (not counter welded) the root effect is more
severe for smaller thicknesses. Here the limit is taken at 8 mm. For thicknesses larger than 25
mm the normal thickness effect should be applied, i.e.
t = t=25[t/25]-0,25

(1)

4. GEOMETRIC STRESS METHOD

4.1 Reference Curves

116

The basic reference curves for 16 mm wall thickness, according to Eurocode 3 [1] are given in
Figure 4. The equations for the reference curves and the thickness effects to be adopted are also
given.

As discussed in Lecture 12.4.1, for hollow section joints with partial penetration welds or fillet
welds, sufficient evidence is now available to use the same curves as for full penetration butt
welds.
4.2 Determination of Geometric Stresses by F.E. Modelling
The geometric stress should be determined by extrapolation of the geometric stress outside the
influence region of the weld to the weld toe. For most connections of circular hollow sections
the geometric stress has a linear part and a linear extrapolation can be used as shown in Figure
5a.

117

For square or rectangular hollow section joints the geometric stress is non-linear and a quadratic
extrapolation gives a better accuracy and lower scatter in test results, Figure 5b. This
extrapolation is more consistent with the mode of failure (location of the crack).

Determination of the geometric stresses by F.E. modelling requires proper modelling (type of
elements, mesh), preferably by methods which have been calibrated by tests [2]. Due to the
difference between actual dimensions and nominal dimensions, the calibration should be done
with care using the actual dimensions, see Figure 6.
118

4.3 Stress Concentration Factors

The geometric stress can also be obtained by multiplying the nominal stresses by the relevant
stress concentration factors. Many parametric formulae are now available for circular hollow
section joints. The most extensive set which gives good agreement with measurements is given
by Efthymiou [3]. For square hollow section joints, a set of parametric formulae for T- and Xjoints is given by Van Wingerde [4]. For K- and N-joints of square hollow sections, formulae are
available, however only for a particular combination of axial loads and bending moments [5].
Thus, they are not generally applicable. Currently, new parametric formulae are being
developed in the framework of a CIDECT programme. These formulae will become available in
1994.
To give formulae for all types of joints, for various locations and for various loadings is beyond
the scope of this lecture. However, some graphical presentations are given in Figures 7 - 15 for
circular hollow section joints and in Figures 16 - 18 for T- and X-joints of square hollow
sections. If the effect of the stiffness distribution along the intersection perimeter is clearly
understood, then the tendency of the graphs will be clear. All graphs show that the SCF for the
saddle location achieves a maximum for medium ratios. For T- and X-joints minimum values
are obtained for = 1,0. Here, the SCF changes very rapidly for small variations of . For this
reason, for X-joints of circular hollow sections, lower values than those for = 0,95 are not
taken, because for = 1,0 the weld might cause an eccentricity.

119

120

121

122

123

124

125

126

127

Note that the effect of the bending moment in the chord in a T-joint is included in the SCF's for
T-joints of circular hollow sections, whereas for T-joints of square hollow sections it has been
excluded, allowing the same presentation as for X-joints.
It is recommended that a minimum stress concentration factor of 2 is adopted to cover cracking
from the root of the weld. Furthermore, the SCF's should be multiplied by the following
correction factors for square hollow section joints if fillet welded connections are used:
Brace: 1,4 (for lines A and E)
Chord: 1,0 (no correction factors)
These factors cover the effect of brace wall bending. For the chord, the weld toe is located
further away from the brace, resulting in a lower SCF. However, insufficient evidence is
available yet to quantify this in general. Therefore, a factor 1,0 is adopted for the time being.
For an optimal design the geometric stress concentration factors should be low, i.e.
low and values and low or high values.
As indicated in Lecture 12.4.1, the geometric stress range can be obtained by multiplying the
relevant stress concentration factor (location and loading) by the nominal stress range, which
causes the stress peak.
Care has to be taken that the stress concentration factors are only used within the range of
validity. Figure 19 shows, for example, that the effect of multiplanar loading can be
considerable.

128

4.4 Fatigue Life

The fatigue life is determined by checking the highest geometric stress for the chord as well as
for the brace multiplied by the appropriate M factor with the basic -N curve using the
appropriate thickness.
5. CLASSIFICATION METHOD
The classification method is only used for axially loaded K- and N-type joints of circular and
rectangular hollow sections, since the variation in SCF is too large for other types of joints.
In principle, the classes should be a lower bound, defined by the basic reference curve for the
geometric stress range divided by the SCF's which can be obtained within the range of validity.
The classification is based on an analysis of relevant test results, taking account of the
parameter and using a lower bound. In this approach, the effects of other influencing
parameters and the thickness effects are combined to some extent.
Multiplication of the class by a minimum SCF of 2,0 gives values far exceeding those of the
basic reference curve for geometric stress, even if the thickness effect is incorporated. On the
other hand, the two approaches cannot be compared directly, since the slopes are different, i.e.
for the classification method a fixed slope constant of m = 5 is adopted.
Although Eurocode 3 [1] gives the range of validity shown in Table 1, it would be better to
change some of the limits, i.e.
4 to or t1 8 mm instead of to or t1 12,5 mm

129

bo/to t1 or do/to t1 100 instead of bo/to or do/to 25

These limits are in better agreement with the test results on which this classification method has
been based.
In view of these comments, one could ask: "Why has the classification method been adopted?"
At the time of drafting Eurocode 3, insufficient evidence was available for SCF's of hollow
section joints. Further, many designers are not familiar with the geometric stress concept.
Therefore the classification method was favoured by many countries.
The design is very simple, since the designer determines the nominal stress ranges in the brace
with one of the methods described in Section 2 considering the factors for secondary bending
moments and the relevant M factor. The design class gives the -N curve and the fatigue
life can be determined.
6. GENERAL REQUIREMENTS FOR WELDING
The connections at the welded joints should be made over the entire perimeter of the hollow
sections by means of a full penetration weld, partial penetration weld, fillet weld, or a
combination. Full penetration welds should be used if:

The brace has a wall thickness larger than 8 mm.

The angle at the toe of the brace is larger than 120 .

For joints between square or rectangular hollow sections with = 1,0, fillet welds can
only be used at the side of the brace perpendicular to the axis of chord. Therefore, the
sides parallel to the chord axis should be welded by a full penetration butt weld.

Attention should be paid to the proper selection of materials and the welding procedure. To
avoid failure of the weld under static loading, the throat thickness of the fillet weld for steel up
to S355 is equal to or greater than the wall thickness of the brace (a t1).
Welding should start at the middle of the sides. If welding of joints commences at the corners of
the brace, the fatigue strength deteriorates. This may result in a decrease by a factor 2 on stress
range.
7. CONCLUDING SUMMARY
For the geometric stress method and the classification method:

The modelling of the structure should be done with care, considering secondary bending
moment effects.

In using F.E. modelling, proper consideration of elements and meshes is required, as

well as calibration with test results. The geometric stresses should be extrapolated to the
weld toe in a standardized manner.

Stress concentration factor formulae should be used within their range of validity. For
example the formulae for uniplanar joints cannot be used for multiplanar loading.
130

Fillet welds exhibit higher SCF's in the brace than butt welds.

For axially loaded T- and X-joints the highest SCF's occur for medium ratios (0,5 to
0,6)

The classification method and the geometric stress approach are not yet consistent.

Welding of hollow section joints should not start at the locations of high stress
concentrations, e.g. not at the corners of rectangular or square hollow sections, but at
the middle of the sides.

8. REFERENCES
[1] Eurocode 3: "Design of Steel Structures": ENV 1993-1-1: Part 1, General rules and rules for
buildings, CEN, 1992.
[2] Romeijn, A., Puthli, R.S., Wardenier, J.: "Finite element modelling of multiplanar joint
flexibility in tubular structures", Proceedings ISOPE '92, San Francisco, U.S.A.
[3] Efthymiou, M.: "Development of SCF formulae and generalised influence functions for use
in fatigue analysis", Offshore Tubular Joints Conference (OTJ '88), UEG Offshore Research,
Englefield Green Near Egham, U.K., 1988.
[4] Van Wingerde, A.M.: "The fatigue behaviour of T- and X-joints made of square hollow
sections", Ph.D. thesis, Delft University of Technology.
[5] Mang, F., Herion, S., Bucak, ., Dutta, D.: "Fatigue behaviour of K-joints with gap and with
overlap made of rectangular hollow sections", p. 297-310 of Proceedings "Tubular Structures",
edited by E. Niemi and P. Mkelinen.
9. ADDITIONAL READING
1. Dutta, D., Mang, F., Wardenier, J.: "Fatigue behaviour of welded hollow section joints",
CIDECT Monograph No. 7.
2. "Design of tubular joints for offshore structures", vol. 1, 2 and 3, UEG publication
UR33.
3. Romeijn, A., Puthli, R.S., Wardenier, J.: "The flexibility of uniplanar and multiplanar
joints made of circular hollow sections", Proceedings ISOPE '91, Edinburgh, U.K.
Table 1 Detail category: Hollow sections and simple connections
Details loaded by nominal normal stresses
Detail
Constructional detail
category

Description

m=3

131

Rolled and extruded products

160

See Fig T1-1

Non-welded elements.
Sharp edges and surface flaws to be improved by grinding
Continuous longitudinal welds

140

See Fig T1-2

Automatic longitudinal welds with no stop-start positions; proven

free of detectable discontinuities

Transverse butt welds

Butt welded end-to-end connection of circular hollow sections
71

Requirements
See Fig T1-3

- Height of the weld reinforcement less than 10% of weld

width; smooth transitions to the flat surface
- Welds made in flat position and proven free of detectable
discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified two detail categories higher, i.e. > 90
Transverse butt welds
Butt welded end-to-end connection of rectangular hollow sections

56

Requirements
See Fig T1-4

- Height of the weld reinforcement less than 10% of weld

width; smooth transitions to the flat surface
- Welds made in flat position and proven free of detectable
discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified two detail categories higher, i.e. > 71
Welded attachments (non load-carrying welds)

71

See Fig T1-5

Circular or rectangular section, fillet welded to another section.

Section width parallel to stress direction 100 mm
132

Welded connections (load-carrying welds)

Circular hollow sections, end-to-end butt welded with an
intermediate plate
50

See Fig T1-6

Requirements
- Welds proven free of detectable discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified one detail category higher, i.e > 56
Welded connections (load-carrying welds)
Rectangular hollow sections, end-to-end butt welded with an
intermediate plate

45

See Fig T1-7

Requirements
- Welds proven free of detectable discontinuities
- Details with wall thicknesses greater than 8 mm may be
classified one detail category higher, i.e. > 50

Fig T1-1

Fig T1-2

Fig T1-3

Fig T1-4

133

Fig T1-5

Fig T1-6

Fig T1-7

Table 1 (continued) Hollow sections and simple connections

Details loaded by nominal normal stress (continued)
Detail
Constructional detail
category

Description

m=3
Welded connections (load-carrying welds)
40

See Fig T1-8

Circular hollow sections, end-to-end fillet welded with

an intermediate plate
Requirements
- Wall thickness less than 8 mm
Welded connections (load-carrying welds)

36

See Fig T1-9

Rectangular hollow sections, end-to-end fillet welded

with an intermediate plate
Requirements
- Wall thickness less than 8 mm

80

l 50 mm

71

50 < l 100 mm

Longitudinal attachments (non-load-carrying welds)

See Fig T1-10

134

50

l > 100 mm

The detail category varies according to the length of the

attachment l

80

t 12 mm

Transverse attachments

71

t > 12 mm

80

t 12 mm

71

t > 12 mm

See Fig T1-11

The end of the weld more than 10 mm from the edge of

the plate

Transverse attachments
See Fig T1-12

Diaphragms of rectangular girders welded to the flange

or web

Transverse attachments
80

See Fig T1-13 The effect of welded shear connectors on base material
Cruciform joints (load-carrying welds)
See Fig T1-14

Full penetration weld, inspected free of detectable

discontinuities

71
Requirements
- The maximum misalignment of the loadcarrying plates should be less than 15% of the
thickness of the intermediate plate
Cruciform joints (load-carrying welds)
See Fig T1-15

36

Fillet welded connection. Two fatigue assessments are

required
Firstly root cracking is evaluated by determining the
stress range in the weld throat area, Category 36
Secondly toe cracking is evaluated by determining the
stress range in the load-carrying plates, Category 71
Requirements
- The maximum misalignment of the load-

135

carrying plates should be less than 15% of the

thickness of the intermediate plate
50

t and tc 20 mm

36

t and tc 20 mm

Cover plates (load-carrying welds)

See Fig T1-16

End zones of single or multiple welded cover plates, with

or without frontal weld.
When the reinforcing plate is wider than the flange a
frontal weld, carefully ground to remove undercut, is
necessary

Fig T1-8

Fig T1-9

Fig T1-10

Fig T1-11

Fig T1-12
136

Fig T1-13

Fig T1-14

Fig T1-15

Fig T1-16

Table 2: Classes for classification method

Detail categories for lattice girder joints

Detail
category
m=5

Construction Details

Description

137

Joints with gap

90

to/ti 2,0

45

to/ti 1,0

See Fig T2-1

Circular hollow sections, K and

N joints

Joints with gap

71

to/ti 2,0

See Fig T2-2

Rectangular hollow sections, K

and N joints
Requirements

36

to/ti 1,0

0,5(bo-bi) g 1,1(bo-bi)
g 2to

Joints with overlap

71

to/ti 1,4

See Fig T2-3

K joints
Requirements

56

overlap between 30% and

100%

to/ti 1,0

Joints with overlap

71

to/ti 1,4

50

to/ti 1,0

See Fig T2-4

N joints

General Requirements
to, ti 12,5 mm 35 50 bo/to 25
0,4 bi/bo 1,0 0,25 di/do 1,0 bo/to 25

138

bo 200 mm do 300 mm -0,5do e 0,25do

-0,5ho e 0,25ho
Out-of-plane eccentricity: 0,02bo or 0,02do
Fillet welds are permitted in braces with wall thicknesses 8 mm
Table 2: Classes for classification method

For intermediate to/ti values, use linear interpolation between nearest detail categories

Note that the braces and the chords require separate fatigue assessments

Note regarding General Requirements:

Preferably use 4 to or t1 8 mm instead of 12,5 mm
Preferably use bo/to t1 or do/to t1 100 mm instead of bo/to or do/to 25

Fig T2-1

Fig T2-2

139

Fig T2-3

Fig T2-4
Lecture 12.5: Improvement Techniques in
Welded Joints
OBJECTIVE/SCOPE
To introduce the more commonly used weld improvement techniques and their effects on the
fatigue performance of welded joints.
PRE-REQUISITES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.2: Advanced Introduction to Fatigue
RELATED LECTURES
Lecture 12.3: Effect of Workmanship on Fatigue Strength of Longitudinal and Transverse Welds
SUMMARY

140

This lecture introduces improvement techniques primarily as remedial measures for welded
structures. Initial analyses of the reasons for the poor fatigue performance of welded joints leads
to a classification system for improvement methods. The various methods used in practice are
then described and evaluated. The methods described are:
The AWS improved profile
The use of special electrodes
Grinding
Weld toe remelting
TIG dressing
Plasma dressing
Hammer peening
Shot peening
Guidance on current design rules is summarised; the need for improvement to design standards
and guidance is highlighted.
1. INTRODUCTION
1.1 General
Any weld in a structure usually represents a weakness both with regard to brittle fracture and
fatigue strength. The low fatigue strength of welded joints is a limiting factor for the design of
more efficient structures, in particular since the fatigue strength normally does not increase with
static strength. Upgrading the fatigue performance of a welded structure can be achieved in
several ways such as:

Good detail design, e.g. by substituting a lower class joint with one having a higher
fatigue strength.

Improving the fatigue strength of the joint using an improvement method.

Improvement methods are usually employed as remedial measures to extend the fatigue life of
welds that have failed prematurely and have been repaired. They are also used to extend the life
of welds which, through service load monitoring, have been shown to be more severely loaded
than assumed during the design phase.
The incentive for applying improvement methods is to make an improved welded joint behave
like a mildly notched component, as shown in Figure 1.

141

The use of higher allowable stresses for welded joints in higher strength steels entails other
benefits as well: the thickness effect in fatigue is reduced, bringing about a further reduction in
weight as compared with a lower strength steel joint with the same load bearing resistance. A
reduced size of section in general also improves the brittle fracture properties of the joint. The
lower welding, handling and erection costs may partially offset the higher fabrication expenses
incurred by the improvement methods.
In this lecture the emphasis is on the degree of improvement in fatigue strength which it is
possible to obtain using the various improvement techniques. However, from a practical point of
view, other considerations such as cost and reliability of the treatment (quality assurance) may
be important. Various aspects of quality control and cost are discussed briefly at the end of the
lecture.
1.2 The Potential for Improving Fatigue Strength
To understand the full potential of improvement methods for fatigue life, it is useful to look at
the reasons for the poor fatigue performance of welded joints. The low fatigue strength of
welded joints as compared with other notched components is illustrated in Figure 2. Welded
joints differ from other notched components in several ways even if the elastic stress
concentration factors Kt are similar. It is important to identify the main factors that tend to
reduce fatigue life in order to choose efficacious methods for improving the fatigue
performance. The main differences between welded and unwelded, notched components are:

142

(a) Notch shape and defects: The geometrical notch of the weld toe region, which normally is
the most fatigue critical area, is generally less uniform than notches in a machined component,
see Figure 3. Moreover, welded joints contain an assortment of defects, most of which are so
sharp that they start growing as fatigue cracks when the structure is subjected to dynamic loads,
thus reducing or eliminating the crack initiation stage of the fatigue life.

(b) Metallurgical changes in the base material: The material in the heat affected zone (HAZ), in
which the fatigue crack is likely to initiate and propagate, undergoes metallurgical changes that
may affect the local fatigue properties. Thus the softened material in the HAZ of a higher grade
steel, whose high strength has been obtained by thermo-mechanical treatment, may limit the
fatigue strength that is possible to obtain by improvement techniques.
(c) Residual stresses are set up in and near the weld due to the contraction of the weld metal as
it cools to ambient temperature. These local residual stresses due to welding, which may reach
the yield stress in magnitude, affect the fatigue properties in a similar manner to externally
143

imposed mean loads, i.e. a tensile residual stress reduces fatigue life while a compressive stress
increases life. The distribution of transverse residual stresses in a welded plate of simple shape
is shown in Figure 3(b).
Residual stresses do not arise only from the thermal strains associated with the welding process
and subsequent cooling. Global or long range residual stresses are introduced in a structure
whenever members are forced together, due to misfit, uneven thermal expansion or when
restraint is being used. Long range stresses act over large areas and are therefore not relaxed by
peak loads at stress concentration or by local treatment. They are generally of smaller
magnitude than welding stresses.
(d) Environmental effects: A corrosive environment may have a strong adverse effect on fatigue
life. The fatigue lives of common welded joints are typically reduced by a factor of two or four
under free corrosion in seawater. However, the prevention of corrosion by either cathodic
protection or protective coatings, which may restore the air fatigue properties of a welded joint,
are not regarded as improvement methods per se because corrosion protection is part of normal
practice for the construction and operation of offshore structures.
2. IMPROVEMENT METHODS - OPERATING PRINCIPLES
The low fatigue strength of welded connections is generally attributed to the very short crack
initiation period which is generally found to be in the range of about 10 to 30% of the total life,
depending on the method of observation and definition of the initial crack. Comparing this with
a crack initiation period of more than 90% typically observed for smooth specimens tested at
low stresses, there is obviously scope for a substantial life increase by delaying crack initiation.
The principal ways of achieving this increase are by:
(a) Reducing the stress concentration factor of the weld.
(b) Removing the crack-like defects at the weld toe.
(c) Removing the harmful tensile welding residual stresses or introducing compressive stresses.
Since both (a) and (b) both involve altering the local geometry, weld improvement methods can
be placed in two broad categories:
(1) Weld geometry modification methods
(2) Residual stress methods.
Various methods which have been investigated [1,2] are listed in Figure 4. Table 1 presents an
evaluation of the improvement methods that are currently used in practise, together with some
information on relative costs where available. In addition, two other methods have been tried in
the last few years; water gouging and laser remelting. However, the two methods are still very
much experimental in nature and have, as far as is known, not been employed in industry. The
following methods have reached a more mature stage in the sense that they are used in industrial
applications:

144

(1) Weld toe grinding, using either a disk grinder or a rotary burr tool.
(2) Tungsten inert gas (TIG) remelting of the weld toe region.
(3) Weld profile control, i.e. performing the welding such that the overall weld shape gives a
low stress concentration and the weld metal blends smoothly with the plate.
(4) Special electrodes with good wetting characteristics to give a favourable weld toe geometry.
(5) Hammer peening of the weld toe region.
(6) Shot peening.
145

Grinding and TIG remelting methods may be classed as weld toe geometry modification
methods, whereas hammer preening and shot preening are residual stress techniques.
Particularly large improvements may be obtained when techniques from the two groups are
combined.
3. SOME IMPROVEMENT METHODS AND THEIR EFFECT ON FATIGUE
STRENGTH
3.1 Improved Welding Techniques
Weld profiling and the use of special electrodes are methods that are integral parts of the
welding process itself. These methods are attractive from a production point of view since there
is no need to come back with a different type of equipment for a final treatment of the weld,
which would increase costs and make quality control more difficult.
3.1.1 The AWS improved profile
In the Structural Welding Code [3] of the American Welding Society (AWS), a low stress
concentration factor is sought by controlling the overall shape of the weld to obtain a concave
profile and requiring a gradual transition at the weld toe. The "disc test" or "dime test" shown in
Figure 5, as specified by AWS, is used to ensure an acceptable weld. If the weld does not pass
the disc test, remedial grinding at the weld toe or at the interbead notches has to be carried out.
If profile control is carried out the designer can use the X curve in Figure 6, if not the lower
X curve must be used.

146

3.1.2 Special electrodes

Specially developed electrodes with coatings that have good wetting and flow characteristics
have been used in Japanese test programs aimed at improving the fatigue performance of high
strength steels of 500 to 800 MPa yield strength [4]. These electrodes are understood to have
been widely used in the construction of high strength steel bridges. The electrodes give a
smooth transition at the weld toe and a reduction of the calculated stress concentration factors,
typically from around 3 to 1,2-1,5 for fillet welds. The improvements reported in the Japanese
tests were from 50 to 85%, the largest increases in fatigue life being reported for the highest
strength steels. However, other tests on T-joints made recently in Norway gave improvements of
approximately 25%. The main doubts about special electrodes concern their use in positional
welding where the easy flow of the filler material may be a disadvantage.
3.2 Grinding
Grinding (Figure 7) can be carried out with a rotary burr grinder or disc grinder, the former
requiring much more time and therefore incurring higher costs. To ensure the removal of slag
intrusions, grinding has to be extended to a minimum depth of 0,5mm below the bottom of any
visible undercut [5]. The lower stress concentration factor and the removal of crack-like defects
at the weld toe generally give large increases in fatigue life, typically from 25 to 100% at long
lives (N > 1 million cycles), see Figure 8 [6]. However, the scatter is large, particularly for disc
grinding which may be difficult to perform in confined areas; also an inexperienced operator
may inadvertently remove too much material.

147

Grinding is currently the only improvement method allowed in European codes for offshore
structures [5,7]. However, the higher fatigue strength is not intended for use in initial design,
instead grinding may be used as a remedial measure if the design life is shown to be inadequate
at a late stage during design or construction.
3.3 Weld Toe Remelting
Remelting of the weld toe using either TIG or plasma welding equipment generally results in
large gains in fatigue strength, for several reasons. Firstly, the smoother weld toe transition
reduces the stress concentration factor; secondly, slag inclusions and undercuts are removed;
and thirdly, according to some Japanese publications, the higher hardness in the heat affected
zone is claimed to contribute to the higher fatigue strength.
Plasma dressing generally tends to give better results than TIG dressing. This with plasma
dressing.
3.3.1 TIG dressing
148

Standard TIG dressing equipment is used, usually without any filler material. For the older type
C-Mn steels (e.g. St 52) with a relatively high carbon content, a second TIG round was
necessary to temper the first run at the toe [8], see Figure 9. The second run also contributes to a
better weld toe geometry. The hardness problem associated with TIG dressing of C-Mn steel is
eliminated with the use of modern low carbon steels. TIG dressing is somewhat sensitive to
operator skill, the weld and plate must be clean to avoid pores.

The magnitude of the improvement depends as for most improvement techniques, primarily on
the joint severity and base material strength. Improvements ranging from about 10% for butt
welds in mild steel plates to about 100% for fillet welded high strength steels have been
reported. Figure 10 shows typical results in the latter case [8].

149

3.3.2 Plasma dressing

Plasma dressing is similar to TIG dressing, the main difference being the higher heat input
(about twice that used in TIG dressing), and a wider weld pool. The latter tends to make plasma
dressing less sensitive to electrode position relative to the weld toe, and the resulting
improvements in fatigue strength are generally larger than for TIG dressing.
3.4 Residual Stress Methods
Some improvement in fatigue behaviour is obtained by removing residual welding stresses by
postweld heat treatment, especially if the applied load cycle is wholly or partly in compression.
However, the largest benefits are obtained if compressive residual stresses are introduced. The
more commonly used residual stress methods are hammer peening and shot peening.
3.4.1 Hammer peening
Hammer peening is carried out with a solid tool with a rounded tip of 6-14mm radius. A similar
technique consists of using a wire bundle instead of a solid tool. Both types of tool are normally
pneumatically operated. The solid tool gives a far more severe deformation and gives better
improvements than either wire bundle or shot peening [6].
Optimum results for hammer peening are obtained after four passes, giving a severely deformed
weld toe, with an indentation depth of about 0,6mm, providing a simple inspection criterion [6].
Like burr toe grinding, hammer peening is a noisy and tedious operation and has perhaps, for
this reason, not attained widespread use. The improvements are among the highest reported, see
Figure 11. Most test results show larger improvements for higher strength steels [6].

150

3.4.2 Shot peening

In the shot peening process the surface is blasted with small steel or cast iron shots in a high
velocity air stream, producing compressive residual surface stresses of about 70 to 80% of the
yield stress. Assessing the quality of the treatment entails time consuming residual stress
measurements. In practice, the intensity or the degree of surface plastic deformation is
determined by Almen strips, which are small steel strips attached to the surface of the
component. The curvature developed in the strip is a measure of the peening intensity. A second
parameter is area coverage. 100% coverage is obtained when visual examination at 10X
magnification of the surface shows that all dimples just overlap. The time required to obtain
100% is doubled to obtain 200% as normally specified. A major advantage of shot peening is
that it covers large areas at low cost.
Results from fatigue tests on shot peened welded joints show substantial improvements for all
types of joints, the magnitude of the improvements varying with type of joint and static strength
of the steel. Typical results are 30 to 100% increase in fatigue in fatigue lives in the long life
region; however, at shorter lives (N < 105 cycles) the improvements tend to disappear. Tests in
sea water show that the improvements are retained even under freely corroding conditions [9].
High peak loads in variable amplitude loads sequences may be assumed to relax the residual
stresses and reduce the efficacy of such methods, but German results have shown no such
adverse effects [9].
3.5 Compounding
The combination of two improvement methods, particularly a weld geometry method and a
residual stress method, are likely to give large improvements. One example is full profile
grinding and hammer peening which resulted in the fatigue strength of fillet welds in mild steel
being restored to that of the base material [10]. More common combinations are grinding and
151

shot peening and AWS weld profile control and shot peening [11]. In such cases the resulting
improvement may be double that of a single method.
4. APPLYING IMPROVEMENT METHODS TO REAL STRUCTURES
Most current knowledge on improvement methods has been gained from tests on small scale
planar specimens. When considering the application of weld improvement methods to actual
structures, the differences in fatigue behaviour has to be evaluated. One important factor is size.
In a large structure long range residual stresses due to forcing the members together are present
and influence fatigue life. Another consideration is the existence of alternative failure sites.
Obviously no improvement can be expected for a joint with load-carrying fillet welds whose toe
regions are ground or TIG dressed if the untreated joint is as likely to fail from the root as from
the toe; the failure would only be shifted to the root.
In contrast to small joints where the peak stress is limited to the weld toe, the peak stress region
in a large multi-pass joint may include several weld beads. Cracks may initiate anywhere in this
highly stressed area.
In some welds, e.g. in tubular joints with low beta ratios, there is a very steep stress gradient at
the weld toe which is caused partly by the global geometry. If the weld leg length is reduced,
e.g. by grinding as indicated in Figure 12, the resulting peak stress may well be higher and the
resulting improvement could be marginal or non-existent.

5. IMPROVEMENT METHODS AND DESIGN RULES

5.1 Current design rules incorporating improvement techniques
As noted in Section 3.1 the weld profile improvement method is included in the AWS/API
design rules in terms of the X curve that may be used generally if profile control is carried out,
otherwise the lower X must be used. The two curves intersect at a life that is somewhat less than
104 cycles, i.e. the improvement is lost at this life.
152

In the UK Department of Energy rules, S-N the curves for all types of joints can be moved by a
factor of 1,3 on strength (2,2 on life) if grinding is carried out [5]. Thus the two curves are
parallel, and the improvement applies also in the low life/high stress region, contradicting most
test data which tend to show very small or no improvements at all in this region, i.e. giving
intersecting as-welded and improved S-N curves, as exemplified by Figure 11.
The Swedish design code [12] for welded structures consists of 10 S-N curves, each of which is
identified by its Kx factor, see Figure 13. The code also includes a weld quality system
containing four basic classes plus an additional class designated U for improved fatigue
strength. Use of the improved class requires that:

undercuts, weld reinforcements, penetration beads, un-filled grooves and root

concavities must blend smoothly with the base material.

incomplete root penetration is not permitted.

arc strikes must be avoided or removed.

The use of improvement techniques such as grinding, TIG dressing and hammer peening is
permitted to obtain the highest quality class. The combination of weld geometry, probability of
survival level, and weld quality thus determines the S-N curve to be used. The Swedish system
of S-N curves is similar to the British rules insofar that employing an improvement method
leads to a parallel shift of the S-N curve.
5.2 Improved Welds and Size Effects
Size effects in notched components are generally attributed to three origins [13], i.e. a
technological size effect, a statistical size effect or a geometrical or stress gradient size effect.

153

Technological size effects result from differences in production parameters, generally leading to
lower mechanical strength for the thicker parts. Also residual stresses and surface quality may
vary with thickness.
Statistical size effects arise from the higher probability of encountering a large defect in a large
volume of material compared with a smaller volume.
Geometric size effects arise from the stress gradient at the notch root. Even if geometric scaling
is maintained the stress gradient is steeper for the thicker part and a crack will grow in a higher
stress field. If geometric scaling is not maintained which is usually the case for welded joints,
the stress magnification factor increases with thickness, see Figure 14.

Fracture mechanics calculations [14] have shown that the influence of thickness increases with
the SCF of the joint. A statistical analysis of published data on size effects in welded joints gave
a size exponent of n=0,33 for as-welded joints and n=0,20 for improved joints, where n is the
size exponent in the thickness correction equation
S/So=(to/t)n (1)
This tendency to get a smaller influence of size for unnotched or mildly notched parts has been
shown to exist for mechanical components and the following relation between n and the SCF
has been proposed [13]
n=0,1 + 0,14logKt (2)
5.3 Future Modification to Design Rules
The current status of improvement methods is not satisfactory as several methods with proven
ability to improve the fatigue strength of a large variety of small scale specimens as well as
large structural components are not included in design rules. Moreover, the European rules [5, 7,
13] which give the same improvement at all lives, are not consistent with test data which
indicate that the largest improvements are obtained in the high-cycle region, and very small or
no improvements occur in the low-cycle region (N < 104 cycles).
154

Secondly, both theoretical and experimental results indicate that size effects are less severe for
mildly notched parts than for the more severe joints with very short crack initiation lives. Thus a
size exponent of 0,2 would probably be adequate for low SCF joints like simple butt welds or Tjoints with small attachment thicknesses. For the higher SCF joints, e.g. Class F and lower, a
size exponent of n=0,33 would be more suitable [13]. For improved welds an exponent n=0,2
for all weld classes would probably be adequate.
Life predictions that include a crack initiation stage using local stress strain concepts plus
fracture mechanics methods for the crack growth stage have given reasonably accurate life
estimates for improved welds [6], and support the experimental observation that the fatigue
lives of improved welds generally increase with base material strength. Thus a third, and
perhaps more controversial modification to the design rules, would be to allow higher fatigue
strength for higher strength steels. However more data has to be collected before specific
recommendations regarding the degree of improvement can be made.
An effort is now being made within the International Institute of Welding's Commission:
Fatigue Behaviour of Welded Components and Structure, to collect data on improvement
methods with the aim of developing recommended shop practices and design guidance for
improvement methods.
6. CONCLUDING SUMMARY

The low fatigue life of welded connections is generally attributed to the very short crack
initiation period.

Weld improvement methods are primarily aimed at extending the resistance to crack
initiation.

Substantial increases in fatigue strength can be obtained consistently when

improvement methods are used.

However, the full potential of weld improvement method can only be obtained if
premature failures from other locations, e.g. the weld root, can be avoided.

The degree of improvement is generally large for higher strength steels than for mild
steels.

Size effects are lower for low severity joints, implying that size effects are mitigated by
weld improvement methods which reduce the local stress concentration.

The problems of quality control are similar to those involved in the welding process
itself.

The question of employing an improvement method is related to cost and the benefit
allowed in design rules.

Further work is needed to improve the quality of design codes and design guidance.

7. REFERENCES

155

[1] Haagensen, P.J.: "Improving the Fatigue Strength of Welded Joints", Fatigue Handbook.
Offshore Steel Structures. Ed. A. Almar Naess, Tapir 1985.
[2] Bignonnet, A.: "Improving the Fatigue Strength of Welded Steel Structures", PS4, Steel in
Marine Structures, Int. Conf. Delft, Elsevier, June 1987.
[3] Structural Welding Code - Steel, ANSI/AWS D1.1-86, American Welding Society, Feb.
1986.
[4] Kobyashi, K. et al.: "Improvements in the Fatigue Strength of Fillet Welded Joint by Use of
the New Welding Electrode", IIW doc. XIII-828-77.
[5] Department of Energy, "Offshore Installations: Guidance on Design and Construction".
HMSO, London 1984.
[6] Knight, J.W.: "Improving the Fatigue Strength of Fillet Welded Joints by Grinding and
Peening", Welded Res. Int. Vol. 8(6), 1978.
[7] "Fatigue Strength Analysis for Mobile Offshore Units", Det Norske Veritas Classification
Note 30.2, Aug. 1984.
[8] Haagensen, P.J.: "Effect of TIG Dressing on Fatigue Performance and Hardness of Steel
Weldments", ASTM STP 648, 1978.
[9] Grimme D. et al.: "Untersuchungen zur Betriebsfestigkeit von geschweissten OffshoreKonstruktionen in Seewasser", ECSC Agreement 7210 KG/101 Final Report 1984.
[10] Gurney, T.R.: "Effect of Grinding and Peening on the Fatigue Strength of Fillet Welded
Joints", British Welding Journal, December 1968.
[11] Haagensen et al.: "Prediction of the Improvement in Fatigue Life of Welded Joints due to
Grinding, TIG Dressing, Weld Shape Control and Shot Peening", TS35, Steel in Marine
Structures, Int Conf. Delft, Elsevier, June 1987.
[12] Swedish Regulations for Welded Steel Structures 74 StBk-N2, National Swedish
Committee on Regulations for Steel Structures, 1974.
[13] Haagensen, P.J. et al.: "Size Effects in Machine Components and Welded Joints", Paper
1017, Houston, Texas, 1988.
[14] Maddox, S.J. "The Effect of Plate Thickness on the Fatigue Strength of Fillet Welded
Joints", The Welding Institute, 1987.
Table 1 Evaluation of Improvement Methods
GROUP

METHOD

ADVANTAGE DISADVANTAGE COST

COMPARATOR

156

GEOMETRY
GRINDING
IMPROVEMENT METHODS
METHODS
General

Relatively
Applicably mainly
simple
and to planar joints that
easy
to can be expected to
perform.
fail from the toe.
All
grinding
Give
large techniques give a
improvement poor
working
environment
regarding
noise
and dust. Access to
weld may be a
limiting factor.
Relatively
Marginal increase
simple
to can be expected for
perform,
large size welds
inexpensive.
tubular joints due
to
stress
Simple
in concentration
section
effect of groove.
criterion (depth
min. 0,5 mm
below
plate
surface
or
undercut).

Full profile
grinding

burr Very
slow.
Expensive due
to high labour
costs and high
tool wear rate.

Disc grinding

Very
fast
compared with
burr grinding.
Can
cover
large areas.

Large
20
improvements to
be expected for all
types of welds.

Score marks give 2

lower
improvements than
burr grinding.
Improper use may
introduce serious
defects.

REMELTING
METHODS
General

Large
Operator
needs
improvements special training
are
possible.
Suitable
for

157

mechanisation.
TIG dressing

Small physical Careful cleaning of 1

effort required. weld and plate
Inexpensive
necessary.
High hardness may
result in C-Mn
steels due to low
heat input.

Plasma dressing

Easy
to Lower
hardness N/A
perform due to than TIG dressing
large weld pool
Somewhat
large
improvement
than
TIG
dressing

Heavy,
cumbersome
equipment.
Accessibility may
limit use

WELD
PROFILING
METHODS Genera
l

The
Defects at weld toe
improvement is not removed.
introduced in
the
welding
process itself.

AWS
profile

Very large scatter N/A

in test results due
to variation in
microgeometry at
weld
toe.
Suitable
for Consistent
large
welds improvements only
and
tubular possible if method
joints
is combined with
others, e.g. toe
grinding, hammer
peening or shot
peening.

improved Well defined

inspection
criterion (the
"dime test")

158

Special electrodes

Easy
to
perform.
Suitable
for
small
joints.
Inexpensive.

Improvement
N/A
smaller than, e.g.
grinding or TIG
dressing.

Table 1 (Continued)
GROUP

METHOD

RESIDUAL General
STRESS
METHODS

ADVANTAGE

DISADVANTAGE

COST
COMPARATOR

Large improvements Not suitable for low cycle

possible.
fatigue applications.
Beneficial effects may
disappear under variable
amplitude
loading
involving
peak
compressive loads.

Hammer
peening

Very
large Limited to toe treatment 2
improvements
only.
possible for poor
quality welds.
Simple inspection Excessive peening may
criterion (depth of cause cracking.
groove > 0,6mm).

Shot peening Well

developed Practicable application to N/A
procedures for small large scale structures not
parts. Covers large demonstrated.
areas.
Simple methods for Best suited
quality control.
notches.

for

mild

Improves resistance Very thin surface layer

to stress corrosion deformed; corrosion may
cracking.
quickly remove beneficial
effects.

159

Lecture 12.6: Fatigue Behaviour of Bolted Connections

OBJECTIVE
Introduction to the design of bolted connections under fatigue loading.
PREREQUISITES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 11.4: Analysis of Connections
RELATED LECTURES:
Lecture 12.2: Advanced Introduction to Fatigue
SUMMARY
The basic principles of fatigue resistance of bolts and bolted connections are established. The
load transmission is described in shear and tension connections. In each case, the bolts can be
non-preloaded or preloaded. The positive effect of the preload of the bolts on the fatigue
behaviour in both shear and tension is discussed. Some economical solutions are proposed.
NOTATION
A Nominal area of a bolt [mm2]
Aa Stress area of a bolt [mm2]
160

db Nominal diameter of a bolt [mm]

da Diameter of the stress area [mm]
dr Shank diameter [mm]
dc Core diameter [mm]
m Slope of a strength fatigue curve [-]
N Number of stress cycles [-]
Normal stress range [MPa]
Shear stress range [MPa]
Fb Normal force in a bolt [N]
Fp Preload in a bolt [N]
1. INTRODUCTION
All the concepts given in Lectures 12.1 and 12.2 relating to the design of structures against
fatigue loading and fatigue assessment procedures are applicable to bolted connections.
However, the presence of geometrical discontinuities (holes, changes of section) causes stress
concentrations which increase the stresses locally and influence resistance to fatigue. Stress
concentrations occur in bolts at the thread roots, thread run-out and at the radius under the head.
Fatigue failures in bolts in fluctuating tension commonly occur at this last location or in the first
thread under the nut.
The design of the joint is very important; the fatigue strength finally depends on the real path of
the loads through the connection, and the fluctuation in stresses of the fatigue sensitive regions.
Two types of load cases on a bolted connection can be discriminated. One where the load is in
the axial direction of the bolts and the other where the load transfer is perpendicular to the bolt
axis. In this Lecture these two types are referred to as:
I Bolted connections loaded in tension
II Bolted connections loaded in shear.
An example of the first type is a bolted flange connection as shown in Figure 1. An example of
the second type is a bolted coverplate connection in a flange of a beam section or a simple strip,
see Figure 2. In the latter case the load is transferred by shear either in the bolts (for non
preloaded bolts) or at the plate surfaces (for preloaded bolts).

161

In addition to these two load situations, combinations are possible.

162

2. FATIGUE BEHAVIOUR OF BOLTS LOADED IN TENSION

Before discussing bolted connections loaded in tension and their specific requirements to
prevent fatigue failure, the fatigue behaviour of the bolt (or thread) is discussed.
2.1 Location of Failure
The thread in a bolt acts as a notch and therefore a high stress concentration is caused at the root
of the thread. At two locations of the thread the stress concentration can be even higher, i.e. at
the runout of the thread and where the thread of the nut first engages the thread of the bolt.
In addition, the head-shank transition is also a stress concentration.
There are, therefore, basically three locations in a bolt with nut axially loaded, where a fatigue
crack can initiate in a bolt with nut axially loaded. These locations are:
a. head-shank transition
b. runout of thread
c. thread at nut.
In standard bolts the radius at the bolt-head shank transition is large enough to prevent fatigue
cracks at this point.
Normally, if fatigue cracks occur, they will be located at the first engagement of the threads of
the bolt and nut (c in Figure 3). This is due to the load transfer from nut to bolt.

The load transfer at the contacting thread faces of the bolt and nut give rise to extra bending
stresses in the threads, as shown in Figure 4.

163

Moreover the load is not equally distributed between the contacting faces of the thread of the
bolt and the nut. In most situations the load transfer is concentrated at the first engagement of
the thread faces and can be 2 to 4 times the mean value [1]. However this depends on the thread
form, pitch difference, difference in Young's modulus where different materials are used etc.
The load transfer distribution can become more uniform, by plastic deformation of the nut.
2.2 Influence of Mean Stress and Material
The fatigue behaviour of the thread of a bolt is more or less comparable to the fatigue behaviour
of a weld. In both cases there is a notch where a fatigue crack initiates. For the weld it is the
weld toe and for the bolt it is the thread root.
Due to the presence of the notch and the resulting high stress concentration factor, the fatigue
behaviour is in most cases hardly affected by:

The mean stress level

The material quality.

The negligible influence of the mean stress level is caused by the high stress concentration. At
the first occurrence of the maximum load level of a cycle, yielding at the notch occurs. The
following cycles then cause a stress variation at the notch which has a maximum equal to the
yield strength independent of the mean stress level of the load itself. An exception to this is the
situation where the bolt thread is made by rolling after the heat treatment of the bolts which
results in residual compressive stresses at the thread roots. In that case the fatigue performance
is better at low mean load level.
The phenomenon that the material has a negligible effect is explained by the fact that as material
strength improves the sensitivity to notches increases. This effect is illustrated in Figure 5 [5]
where the influence of the ultimate tensile strength on the fatigue strength for different notch
cases is given [5].

164

2.3 The Fatigue Design Curve for Bolts in Tension

Although the notch at a weld (with its undercuts and slag inclusions) is possibly more severe
than the machined or rolled notch at the thread root, the concentrated load transfer between bolt
thread and nut on top of the inherent stress concentration can cause a relatively poor fatigue
performance.
Therefore, in the Eurocode 3 classification [4], axially loaded threads and bolts fall in the
category equal to the lowest category for weld details, being class 36. The relevant design line
for this category is given in Figure 6. The stress range given on the vertical axis should be based
on tensile stress area of the bolt.

165

It is only mentioned here that according to Clause 9.7.3 of Eurocode 3 a modified design curve
may be used for threads and bolts.
As can be seen from the design curve in Figure 6, the constant amplitude fatigue limit for the
bolts is 26MPa. This means that, for a constant amplitude loading, there is no fatigue damage
where the stress range is less than 26MPa. For a variable amplitude the fatigue limit is 15MPa.
2.4 Comparison Between Ultimate and Fatigue Load Resistance of a Bolt
The following example illustrates that the fatigue load bearing resistance is very low compared
to the static strength of a bolt. For a bolt under static loading the tension resistance
Ft.Rd according to Clause 6.5.5 of Eurocode 3 is given by:
Ft.Rd = 0,9fub As /Mb
Substituting the appropriate values for a bolt M24 grade 10.9 gives the following result:

Ft.Rb =

= 254 kN

For a constant amplitude fatigue loading at zero mean level containing more than 10 7 cycles, the
allowable maximum force on the bolt will be:
Fmax = F/2 = DAs /2 = 0,5 x 26 x 353 = 4,6 kN
In other words, a bolt designed to transfer a tension force of 254kN may not be fatigue loaded
with a maximum force higher than 4,6kN (under the circumstance of zero mean level and more

166

than 107 cycles). This example illustrates the relatively weak fatigue performance of an axially
loaded bolt.
3. FATIGUE BEHAVIOUR OF BOLTED CONNECTIONS LOADED IN TENSION
Although the fatigue performance of the axially loaded bolt itself is poor, that is not necessarily
the case for axially loaded connections. For these connections, the fatigue performance depends
on the structural detailing and the applied preload in the bolt.
3.1 The Principle of the Effect of Preloading
The effect of preloading, where there is a tensile loading on a bolted connection is illustrated for
a flange connection, in Figures 7 and 8. For example, the connection can be a flange connection
in a tubular section (chimney or tower). The distribution of the forces is compared for the
situation with and without preloading of the bolts. The thickness of the flange is assumed to be
large enough to neglect bending flexibility and possible prying forces.

Without Preload
Where there is no preload (Figure 7) and thus no contact force F c on the facing surfaces of the
flanges, the external tensile force F t applied on the connection will be transferred by the force in
the bolts Fb. Therefore the variation of the tensile force F t will result in a variation of the force in
the bolts and at the same time a displacement of the flanges. The connection can be considered
as a two spring system as indicated.
167

With Preload
In case of preload with a force F v, this force will initially be in equilibrium with a contact force
Fc on the contact area of the flanges, Figure 8. The two flanges now act as one piece as long as
the external load Ft is less than the preload Fv.
As a result, when the external load is applied the forces in the bolts will change little. Only the
elastic deformation (mainly change in thickness) of the two flanges will cause a change in bolt
load. The flanges however are relatively stiff due to their much larger area compared with the
bolt area.
However, the load in the bolts will increase rapidly as soon as the contact surfaces separate due
to the external force surpassing the preload force F v. At that moment the situation is equivalent
to the non preload case.
As long as the external load F t is below the preload force F v the situation can be considered as a
three spring system. Two small springs being the bolts and one stiff spring being the two
flanges, Figure 8.
The diagram at the right-hand side of Figure 8 gives the relation between the different forces. At
no external load (Ft = 0) the elongation due to the preload of the bolts is at point A in this
diagram. When an external load Ft is applied, the connection will stretch, resulting in an
elongation of the bolts and flange thickness thus resulting in an increased F b and at the same
time a reduction of the compressive force F c in the flanges as indicated. At each stage the
following relation yields:
Fb = Fc + Ft
It follows from the diagram that the increase in the external force is compensated for the larger
part by a decrease of the contact force F c and a small increase in the forces in the bolts Fb.
The amount of variation of forces in the bolts due to the variation in the external forces is
dependent on the stiffness ratio of the flanges and the bolts. Therefore, the more flexible the
bolts the less force variation they will undergo. Increasing the length of the bolts by inserting
washers or using spring washers will be beneficial because it means that the two springs in the
diagram (being the bolts and possible washer etc.) are more flexible. Inserting gaskets between
the flanges will make the flange assembly more flexible and would have a detrimental effect.
The flanges must be thick to reduce the bending flexibility, otherwise the location of the contact
area becomes critical.
3.2 The Effect of the Location of the Contact Area
In the previous section it was shown that preloading the bolts in a tensile loaded connection
reduces the force variation in the bolts and therefore can avoid fatigue failure of the bolts. The
preload in the bolts must be greater than the external load.
However, preload alone is not always a guarantee for a reduced force variation in the bolts. The
contact force of the connection, which is developed by tightening of the bolts to its preload,
must also be located in a favourable position as well.
168

This is illustrated in Figure 9 by a flange connection, where the thickness of the flange is much
smaller than in the previous example and is therefore flexible in bending. In the flange
connection of two T-sections the location of the contact forces has been established by
introducing loose shims in two different ways. The location of the shims defines the location of
the contact forces. In both cases the bolts are tightened to the same preload.

The schematic models of the relevant spring system are also shown in Figure 9.
Contact Area at the Centre
Where the shims and thus the contact force are in the centre, Figure 9a, there is effectively a
very stiff spring in the middle compared to the two flexible springs representing the flexibility
of the bolt and the bending flexibility of the flanges (the latter in this case being the most
important part of the total flexibility). This case is similar to the situation in the previous section
with a much larger difference in the stiffness between the contact area and bolts + flanges due to
the bending of the flanges.
Contact Area at the End of Flanges
Where the contact area is at the end of the flanges, Figure 9b, the springs representing the
contact area and the bending flexibility of the total flanges, are very flexible. Therefore the
springs, representing the bolts plus a part of the flanges, have a relatively much higher stiffness.
As a result the variation of the external force F t will result in a variation in the bolt forces of
nearly equal magnitude.
Results of Measurements

169

For the examples above actual measurements of the bolt forces have been carried out [2]. The
measured bolt forces in these two situations are given in Figure 10. In each case the bolts were
tightened to a preload Fv of 100kN each. In Figure 10 the force in the bolt F b is plotted by the
thick line as a function of the external load F t. At an Ft of zero, Fb starts at the preload of 100kN.
The external force Ft is also given by the dashed line under an angle of 45 . From the
equilibrium of the forces it follows that the vertical distance between this line and the thick line
of the measured bolt force is equal to the contact force F c.

In the situation with the contact in the centre, Figure 10a, the forces in the bolts are almost
constant until the external force surpasses the preload F v. This means that the part of the
connection including the contact area (the middle spring in the spring model) is extremely stiff
compared to the flexibility of the bolts plus bending of the flanges (side springs in the spring
model). As a result the variation in the forces in the bolt is negligible as long as the preload is
greater than the external load. Fatigue failure in this case is not to be expected.
This is in contrast to the situation with the contact forces at the end of the flanges, Figure 10b.
In this situation the stiffness of the flanges is negligible compared to the stiffness of the bolts.
All external load is now transferred by the bolts. Where the external load is a cyclic loading the
load variation must be very small, otherwise fatigue failure of the bolts occurs very soon.
In general the most favourable situation with respect to fatigue resistance is obtained when the
contact area is as close as possible to the components in which the tensile force is acting.
In Figure 11 some examples of favourable and less favourable situations are given. More
examples are given in [2] and [3].

170

4. FATIGUE OF BOLTED CONNECTIONS LOADED IN SHEAR

A simplified example of a bolted connection loaded in shear is shown in Figure 2. The load is
transferred from one strip to the other by the coverplates. The connection can be assembled by
bolts which are not preloaded and by bolts which are preloaded. Both situations have their own
way of load transfer and failure mechanism.
4.1 The Principle of Load Transfer
Non Preloaded Bolts
In case of non preloaded bolts the forces are transferred by bearing of the plates against the
shank of the bolt and consequently shear in the bolt shank as indicated in Figure 2a. This type of
joint can not be used where the variable load changes sign since the clearance between the holes
and shank allows large displacements to occur repeatedly.
The load transfer in this type of joint is very concentrated at the location where the shank bears
against the holes as indicated in Figures 2a and 12a.

171

Preloaded Bolts
Where the bolts are preloaded, the forces are transferred by friction of the plate surfaces. The
bolts which transfer the load by friction are known as High Strength Friction Grip Bolts
(HSFG), Figure 12b. High strength bolts and controlled tightening are necessary to obtain
sufficient compressive stresses to enable the load to be transferred by friction.
The load transfer by friction takes place over the whole area where compressive stresses are
present due to the bolt preloading as indicated in Figures 2b and 12b. Therefore, the load
transfer is not as concentrated as with non preloaded bolts. Connections with HSFG bolts can
also be used where the variable load changes sign.
4.2 Stress Concentration Around the Holes
Non Preloaded Bolts
For non preloaded bolts there will be a stress concentration at the holes as indicated in Figure
12a.
172

The stress concentration results from the fact that there is a hole in a stressed plate. Moreover
the load is introduced by the bolt shank in a very concentrated way.
Preloaded Bolts
In case of preloaded bolts there is no stress concentration at the holes. The stresses may even be
lower than the nominal stress as indicated in Figure 12b.
This is due to the fact that at the hole a part of the load has already been transferred. Moreover
the bolt head and nut will reduce the deformation of the hole.
4.3 Location of Failure
Non Preloaded Bolts
Due to the stress concentration at the hole a fatigue crack can occur there (see Figure 12a).
Another possibility is the failure of the bolt as a result of the variable shear load in the shank at
the shear plane. The threaded part of the bolt should not be in the shear plane because the notch
effect of the thread would reduce the fatigue resistance drastically.
Preloaded Bolts
For preloaded bolts the stresses at the holes are low. Fatigue cracks do not, therefore, occur at
the holes. The fatigue crack normally occurs in the gross section of the plates, see Figure 12b.
The contact pressure applied by the preload of the bolt gradually decreases around the hole. The
crack initiates where the contact pressure is not high enough to prevent slip between the plate,
resulting in crack initiation by fretting.
5. FATIGUE DESIGN CURVES FOR CONNECTIONS LOADED IN SHEAR
5.1 Non Preloaded Bolts
In this case there are two possible failures - the shank of the bolts and the plates. Both should be
verified against the relevant design curves.
For the bolt shank loaded in shear, the design strength according to Eurocode 3 [4] is given in
Figure 13. No thread is allowed in the shear plane.

173

For the plates, the stresses should be calculated for the net section and the detail category 112,
according to the Eurocode 3 classification, should be used [4].
5.2 Preloaded Bolts
In the case of preloaded bolts, the bolts themselves will not fail provided that the preload in the
bolts prevents total slip.
The plates fall into the same category as in the non preloaded case. However, since failure does
not occur in the net section, the gross section of the plate can be used for calculating the fatigue
stresses.
6. REMARKS CONCERNING THE MAGNITUDE OF THE PRELOAD
The magnitude of the total preload must be large enough to prevent slip (shear connection) or
disappearance of the contact forces (connection loaded in tension) at the maximum possible
load on the connection.
Where the connection is loaded in shear, any slip of the connection due to an extreme load can
reduce the friction coefficient by an unknown factor. Thus the preload has to be designed on the
basis of the maximum extreme load case. The calculation procedure to prevent this is given in
ESDEPLecture 11.3.2. Another way of preventing the slip due to accidental extreme load cases
is to use injection bolts [6].
Where the connection is loaded in tension, an "overload" cancelling the contact forces will
result in a force variation in the bolt. This in itself will not cause fatigue failure since the
number of cycles is limited. However, after this loading, the preload in the bolt can be reduced
due to local yielding and resulting plastic deformation of the bolt or contact areas between bolts
and flanges.
7. FATIGUE STRENGTH OF ANCHOR BOLTS
The dimensioning of holding down bolts under static loading and their anchorage into the
foundation are described in Lecture 11.3.2.

174

Concerning fatigue strength, anchor bolts do not behave in the same way as normal bolts; some
parameters are different: the thread size, the diameter and the method of forming the thread.
Test results have shown [7] that the bolt diameter and the thread size do not influence fatigue
behaviour; the fatigue lives were almost identical for tested specimens as for normal bolts.
On the other hand, the method of forming the thread influences the fatigue strength. Tests were
carried out on anchor bolts with rolled threads or with cut threads. The specimens with rolled
threads provided the longer fatigue life. This better performance may be due to the compressive
residual stresses at the thread root generated by the thread-rolling operation.
When the threads are cut automatically, this operation leaves a transition at their termination. It
is a sharp notch adjacent to a region of smooth bar. It has been shown that there is an important
stress concentration at the notch which induces fatigue cracks.
Consequently rolled threads appear to improve the fatigue performance of the bolt and are
recommended for use when available. Note that the fatigue life of a bolt is a function not only of
the bolt thread, but also of the nut.
As already mentioned for common bolts, the use of a double nut increases fatigue resistance and
its influence seems to be larger for anchor bolts.
All considerations described in Section 5 concerning the influence of the prying effect are
applicable for connections made with holding down bolts. For instance, tests on site have shown
that the designs in Figures 14b and d give better fatigue behaviour than the solutions in Figures
14a and c.

8. CONCLUDING SUMMARY

A bolt loaded in tension has a low fatigue performance.

In a bolted connection loaded in tension, fatigue of the bolts can be prevented by

preloading the bolts and taking care that the contact area is favourably located.

175

In general a favourable position of the contact area with respect to the fatigue of the
bolts is obtained when it is located as close as possible to the components in which the
tensile force is acting.

For a bolted connection fatigue loaded in shear, preloading the bolts leads to the
following advantages over the non preloaded case:

The variable load may change sign.

The thread of the bolt is allowed in the shear plane.
The fatigue strength of the connection is enhanced, since the stresses are based on the gross
section instead of the net section.
9. REFERENCES
[1] Frost, N.E., March, K.J., Pook, L.P., Metal fatigue, Oxford
University Press 1974, ISBN 0198561148
[2] Bouwman, L.P., Bolted connections dynamically loaded in tension, ASCE, J. of the
Structural Division, Vol. 108, No. ST, September 1982.
[3] European recommendations for bolted connections in structural steelwork, No. 38, March
1985.
[4] Eurocode 3: "Design of steel structures": ENV 1993-1-1: Part 1.1, General rules and rules
for buildings, CEN, 1992.
[5] Gurney, T.R., Fatigue of welded structures, Cambridge 1968.
[6] Bouwman, L.P., Gresnigt, A.M., Dubois, G.A., European Recommendations for Bolted
Connections with Injection Bolts, ECCS-TC10 draft.
[7] Frank, K.H., Fatigue Strength of Anchor Bolts, ASCE, Journal of the Structural Division,
vol. 106, ST 6, June 1980.

176

Lecture 12.7: Reliability Analysis and

Safety Factors Applied to Fatigue
Design
OBJECTIVE
To introduce the main concepts and derivation regarding both the statistical evaluation of the
fatigue strength of structural details and the determination of partial safety coefficients on which
are founded the fatigue assessment rules in Eurocode 3 [1].
PREREQUISITES
None.
RELATED LECTURES:
Lecture 12.8: Basic Fatigue Design Concepts in Eurocode 3
SUMMARY
This lecture presents an overview of the statistical analysis procedure applied to fatigue test data
of a particular detail in order to derive the most appropriate S-N curve. Special attention is given
to the definition of the partial safety factors which are in Eurocode 3 for fatigue design
assessment [1]. The use of a proper fatigue damage-tolerant level is discussed which is based on
sufficient residual strength and stiffness in remaining members between inspection intervals
until the fatigue crack can be detected and repaired.
1. INTRODUCTION
Fatigue failure may arise in many engineering structures submitted to repeated loadings. Many
failures occurring in structures are due to the process of fatigue crack propagation. Fracture may
be in encountered in many civil engineering structures such as bridges, cranes, gantry girders,
offshore or marine structures, transmission towers, chimneys, ski lifts, etc. The safe-life
prediction of structures subjected to fatigue loading is recognised as being a very difficult
problem. The stresses in any civil engineering structure are often caused by random loading; the
properties of the materials, and the fabrication conditions for the structure may also vary in a
random manner. The integrated influence of all these variables yields a wide scatter in life
prediction.
As a result, in spite of the progress made in the understanding of the fatigue mechanism, and of
the conservatism introduced in design against fatigue failure, design rules are still based on a
"damage-tolerant approach" which is more or less substantiated by reliability analyses.
2. STATISTICAL ANALYSIS OF S-N CURVES
The S-N curves are evaluated from a series of fatigue tests performed on specimens which
comprise the structural detail for which assessment of the fatigue strength is required. When the
fatigue test results are plotted on a log-log scale, i.e. log (stress range, ) versus log

177

(number of cycles to failure) a considerable statistical scatter exists in the fatigue data as
illustrated in Figure 1.

Instead of taking a lower bound limit approach, a statistical evaluation of the fatigue test results
is performed. Assuming a linear relationship between log and log N the following equation
may be written:
log N = log A - m . log (1.1)
Let
y = log N; x = log ; a = log A; b = - m (1.2)
where log is the logarithm to base 10.
The first step of the statistical analysis is to apply a technique of standard linear regression
analysis of the data to find estimates of a and b, denoted and
respectively. Each pair of
fatigue data, yi = log Ni and xi = log i, should satisfy the relation:
yi =

. xi + ei (1.3)

where ei is called the residual.

The estimates and are determined so that the sum of the squares of the residuals is a
minimum. This condition leads to the following estimates [2].

178

(1.4)

and
=

where

and

(1.5)
are respectively the means of yi and xi, i.e.

where n is the number of data points (sample size).

The second step is to find the variance of the distribution (yi) at a given xi. The variance of yi,
written Var(yi), is assumed to be constant for all values of x i=logi. The constant variance
assumption is sometimes questionable, but holds true for a majority of fatigue test data. It is also
usually assumed that, for any fixed value of x i, the corresponding value of yi forms a normal
distribution.
One of the main comparison indicators which is used in relation to the classification system
adopted in Eurocode 3 [1] is the characteristic strength at two million cycles, written x c =
log c. Let yc = log Nc be the random variable corresponding to a given value xc.
The sampling distribution of yc may be obtained from the following estimation [2]:

.xc (1.6)

and

(1.7)
where Var (yc/xc) is the variance of yc given that x is equal to xc.

(1.8)
where

(1.9)

179

A 95% confidence interval estimate for y c is given by:

(10)
where t95 is the 95% percentile of the Student's distribution with n-2 degrees of freedom [2].
Thus there is confidence that 95% of the yc Student's t population will have values above yck.
Very often in fatigue test analysis the sample size is small (n 30) and the value of the
estimation of the variance of yc as defined by Equation (1.7) fluctuates considerably from
sample to sample. To take this fact into account, the distribution of (yi) for sample size
n 30 has been assumed to follow a Student's t distribution.
Knowing yck from Equation (1.10), the characteristic strength at two million cycles may be
calculated from Equation (1.1). Finally the "one-sided" confidence estimate is performed for a
particular point referred to as the characteristic strength at two million cycles.
The characteristic S-N curve is given by the following equation:
log Nk = (log A - t95ycxc) - m.log k (1.11)
and the characteristic strength at two million cycles may be calculated from:
log kc = [(log A - t95ycxc) - log (2 106)]/m (1.12)
Therefore
kc =10logkc

(1.13)

Numerical Example
Assume a given detail has been tested under constant amplitude fatigue loading. The fatigue test
data is reported in the following Table. No failure was observed for the first five specimens for a
number of cycles over two million. The characteristic fatigue strength curve corresponding to
the given detail is determined as follows:
Stress Range
(N/mm)
74
74
108
108

180

108
108
108
139
139
202
202
202
265
265
265
Table Fatigue Test Data
First Step: Perform the linear regression analysis:
The evaluation of a and b according to Equations (1.4) and (1.5) may be obtained from the
following numerical Table which includes only the tests that have resulted in a fatigue failure.
No

(N/mm)

108

108

139

139

202

202

202

265

181

265

10

265

b=

a=
therefore, the linear regression equation is given by:
log N = 12,334 - 3,102 log (1.11)
for Nc = 2000 000 cycles, from (1.11):
log c = 1,945 and c = 88,05 N/mm2
Second Step: To find the variance of y = log N
from the calculation table above.

and the variance:

The standard deviation is then

yc/xc = 0,151
182

The Student's t95 value with n - 2 = 8 degrees of freedom can be obtained from a statistical table
of the t distribution [3]:
t95 = 1,860
The characteristic strength at two million cycles is then:

log kc =
Therefore:
kc = 101,854 = 71,5
In conclusion, referring to the basic S-N curves in Eurocode 3 (Fig. 9.6.1 in Eurocode 3), the
detail analysed under the preceding statistical evaluation may be classified under category 71,
Figure 2.

The standard deviations vary from test to test and from the type of detail studied. In general, the
higher the stress concentration factor, the lower the standard deviation of the fatigue test results.
The following table gives some indication of the standard deviations which were obtained when
performing the statistical analysis of various types of detail category. When mixing fatigue test
results from various sources, the standard deviation tends to increase and care should always be
exercised to minimise problems arising from inhomogeneity of data, Figure 3. These values of
the standard deviation of the number of cycles given in the table are somewhat different to the
values appearing in Annex C of reference [4] due to the fact that more complete sources of
fatigue data where analysed when reviewing the classification for Eurocode 3 [5].

183

Type of detail

Rolled beam
Welded beam
Vertical stiffener
Transverse attachment
Longitudinal attachment
Cover plate on flange
Bolted connection in shear
3. SAFETY CONCEPT AND PARTIAL SAFETY COEFFICIENTS
A structural element submitted to fatigue loading is subjected to several uncertainties. The
variability of the parameters governing the fatigue life of a structural element, i.e. fatigue
loading and fatigue resistance are largely unknown. A level II reliability model has been
implemented for the derivation of recommended partial safety factors in relation to the
following fatigue strength assessment equation:
f s = R/M

(3.1)

where
184

s is the equivalent constant applied stress range which, for the given number of cycles,
leads to the same cumulative damage as the design spectrum.
R is the fatigue strength as given by the S-N curve of the relevant detail category.
f and M are the partial safety factors applied respectively to the spectrum loading and to the
resistance.
3.1 Derivation of Partial Safety Factors
It will be assumed that log s and log R are both random variables following a normal
distribution law. Therefore the random variables s and R are said to follow a lognormal distribution.
The fatigue limit state function may be written as:
g = log R - log s (3.2)
Introducing the normalised basic variables u and v as:
u = {log R -

} / SlogR (3.3)

v = {log S -

} / SlogS

is the mean value and SlogR is the standard deviation of the variable log R.
The limit state function, Equation (3.2), re-written with the normalised basic variables becomes:
g(u,v) = SlogR.u - SlogS.v +

(3.4)

Having assumed that the random variables log S and log R are normally distributed,
then the safety index is related to the probability of failure as:
= (-) (3.5)
in which (-) is the standardised normal distribution function and is given by (it is
assumed for the sake of simplicity that the variables u and v are uncorrelated):
={

} / (S2logR + S2logS)

(3.6)

It is recalled that can be geometrically interpreted as the shortest distance from the origin to
the failure surface in the standard normal space, Figure 4.

185

The coordinates of the design point D represent the values of u and v with the highest
probability of failure. These coordinates are expressed by:
u = -SlogR / (S2logR + S2logS)

(3.7)

v = SlogS / (S2logR + S2logS)

Rearranging Equation (3.7) in terms of the basic variables log S and log R gives:
R =

- S2logR / ( S2logR + S2logS)

s =

+ S2logS / ( S2logR + S2logS)

(3.8)

The characteristic values of the random variables (log Rk and log sk) are then expressed
by:
log Rk = (log R)k =
log sk = (log s)k =

(1 - kR CR) (3.9)
(1 + ks Cs)

where Cs and CR are respectively the coefficients of variation of log s and log R (i.e. the
ratio of standard deviation over the mean value).
To determine the partial safety factors in a semi-probabilistic format (level I reliability format),
which corresponds to the same degree of safety (represented by the safety index ) as a level II
reliability format, one has to utilize the Equations (3.8) and (3.9). Assuming:
= Slog R/Slog R
then the partial factors can be obtained from the following relations:
186

log M = Slog R /[(1+ 2)] - kR

(3.10)

log f = Slog s /[(1+ 2)] - ks

For a defined safety index the partial safety coefficients may be calculated from Equations
(3.10), knowing the standard deviations of the resistance and the loading (S log R and Slog s)
and the coefficients kR and kS related to the definition of the characteristic values adopted for
(logR)k and (log s)k.
Remarks

As shown in Section 2, there is sufficient experimental data to determine adequate

values of the standard deviation of the resistance Slog R.

On the other hand there is little information concerning the variation of the fatigue
loading. The standard deviation of the fatigue loading, (S log s), must be evaluated or
estimated and depends very much upon the type of traffic load (railways, roadway or
highway). Since the distribution of s is lognormal, then the standard deviation of
log s may be expressed in terms of the coefficient of variation of the loading as:

Slog s = log e [ln(1 + Vs2)] (3.11)

with by definition:
Vs = Ss /
where e is the Euler's number and ln is the Naperian logarithm.

The formal difference in the ECCS Fatigue Recommendations is due to the fact that the
characteristic value of the loading has been introduced in a probabilistic format, which
is not the case in the ECCS Recommendations where the fatigue loading is defined by
its mean value.

The partial safety coefficients depend upon the required safety index. From the
viewpoint of the fatigue reliability assessment, there are many "critical" structural
components or structural details which must be considered in any civil engineering
structure. However, it must be recognised that failure of a particular structural
component in a structure does not necessarily imply a complete failure of the structure.
A distinction must be made between the notion of "fail-safe" and "non fail-safe"
structural components. This notional concept is exemplified in Figure 5: It has to be
understood that in a "fail-safe" assembly, the result of a normal failure is a loss of
rigidity, but the structure retains its integrity. Nevertheless, it is necessary to assess a
safe life capability of structural components whose fracture may potentially give rise to
a catastrophic failure. Safe-life design should be required especially for structural
components or details for which inspection is difficult, or cannot be properly carried
out.

The periodic inspection and maintenance of the construction in conjunction with the
variations in accessibility of the structural detail for inspection or repair must be taken
into consideration when evaluating the risk and assessing the proper safety indices.

187

In the same structure, there are components which may be classified as "fail-safe" and
others as "non fail-safe" from the viewpoint of failure consequences.

It must be understood that the safety indices which were proposed (see Table 1) in Eurocode 3
(Chapter 9) are mainly based on an engineering judgement of what may be called a potential
risk of acceptance of losses and damages. It is the responsibility of each concerned authority to
make the decision on the proper choice of these values on the basis of a realistic risk
assessment.

Periodic inspection and maintenance. Accessible joint detail.

Periodic inspection and maintenance. Poor accessibility.
Table 1 Recommended values of safety indices
Figure 6 gives the partial safety coefficient R in terms of and . These curves have been
drawn for kR=2 and kS=1,645 and for slogR=0,07 (which corresponds to slogN=0,210.

188

In Chapter 9 of Eurocode 3 [1] values of the product f.M have been proposed (Table 2)
based on the values of safety indices given in Table 1. There is little information concerning
fatigue loadings and if the partial safety factor f is known, then the partial safety
factor M related to the strength must be adjusted.

Periodic inspection and maintenance. Accessible joint detail.

Periodic inspection and maintenance. Poor accessibility.
Table 2 Recommended values of the product = f.M
Remarks
Discontinuities play a major role in the fatigue strength, particularly for welded details. Careful
consideration must be given to the weld quality since it significantly affects fatigue strength
variation. Moreover, the measures that can be taken to achieve the required degree of structural
reliability include not only the justification of relevant design rules and the choice of associated
partial safety factors, but also an appropriate level of execution quality and proper standards for
workmanship which are developed in pr EN 1090-1 [6].
4. CONCLUDING SUMMARY

Opinions regarding the fatigue resistance of a structure vary from the extreme that the
structure should be safe during the design life under any circumstances (or during
damage that may be inflicted) to the view that safe-life cannot be predicted, or a
reasonably economic life goal cannot be assured.
189

The purpose of a code is to set a set of partial safety factors which minimise the rate of
fatigue damage in service, and to take advantage of suitable inspection and proper
maintenance procedures.

In structures designed to Eurocode 3, since a completely fatigue-resistant structure is

unlikely to be economically feasible, some fatigue cracks during service should be
expected.

"Fail-safe design" can best serve a useful purpose when supplemented by suitable
inspection and maintenance. The design concept in Eurocode 3 may be described as
follows:

(i) The structure must have an adequate life either crack-free or during which the growth rate of
cracks is sufficiently low to allow detection during an inspection period.
(ii) Visual inspection of all critical areas is possible in service.
5. REFERENCES
[1] Eurocode 3: "Design of Steel Structures": ENV 1993-1-1: Part 1.1, General Rules and Rules
for Buildings, CEN, 1992.
[2] Walpole E.R. and Myers R.H., Probability and statistics for engineers and scientists,
MacMillan Publishing Co. Inc., New York, 2nd Ed., 1978
[3] Natrella M.G., Experimental Statistics, National Bureau of Standards Handbook 91, Issued
August 1 1963. Reprinted October 1966 with corrections.
[4] Recommendations pour la vrification la fatigue des structures en acier. CECM - Comit
Technique no. 6: "Fatigue". CECM no. 43, 1987, Premire dition.
[5] Background Documentation to Eurocode 3: Chapter 9 - Fatigue, Background information on
fatigue design rules: Statistical Evaluation, December 1989.
[6] prEN 1090-1:1993 Execution of steel structures, Part 1: General rules and rules for
buildings.

190

Lecture 12.8: Basic Fatigue Design

Concepts in Eurocode 3
OBJECTIVE/SCOPE:
This lecture contains the background information of the basis of the Eurocode 3 rules
concerning the fatigue design of structural elements.
PREREQUISITES
None.
RELATED LECTURES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.2: Advanced Introduction to Fatigue
SUMMARY
The lecture discusses the main fatigue design rules contained in Eurocode 3 [1]. These fatigue
design rules are based on fatigue test results obtained mainly under constant amplitude loading.
The classification of a given detail, either welded or bolted, results from a statistical evaluation
of the fatigue test data with a 95% probability of survival for a 75% confidence interval. The
evaluation is compared with a set of equally spaced S-N curves with a slope constant of m = 3.
Explanation is given on the choice of a normalised double-slopes S-N curve. Then several
factors, introduced in Eurocode 3 [1], affecting the fatigue strength are also discussed.
1. INTRODUCTION
The principal objective of this lecture is to review the main rules which are the basis for Chapter
9 of Eurocode 3 [1] concerning the fatigue strength assessment of steel structural details.
The main provisions of Eurocode 3 [1] rely upon a set of fatigue resistance curves, equally
spaced, upon which are classified a set of constructional details. The concept for fatigue strength
design follows the Recommendations of the European Convention for Constructional Steelwork
(ECCS). The Recommendations [2] define a set of equally spaced fatigue strength curves with a
constant slope of m = 3 (for normal stress), or m = 5 (for shear stress, hollow section joints, and
some particular details).
In addition to this approach another concept supported mainly by recent developments and
research in the field of fatigue for "offshore" structures is referred to in Eurocode 3 as the
geometrical stress concentration concept (also called the "hot spot stress" method).
To determine the fatigue strength provisions given in Eurocode 3, a compilation of fatigue data
of various sources was carried out. This work has provided an opportunity to re-evaluate
existing fatigue test data and allowed for a more consistent approach to the classification of
detail categories.
191

2. PRACTICAL IMPLICATION OF DESIGN CRITERIA

2.1 Main Factors Affecting the Fatigue Strength
Fatigue of steel structural components, especially welded steel details, is a particularly complex
problem, and many factors may exert an influence on the fatigue life. Table 1 lists a nonexhaustive inventory of these various factors and those which are taken into account either
explicitly or implicitly in Chapter 9 of Eurocode 3 are indicated.
Whilst some factors are dealt with in Chapter 9 of Eurocode 3, other factors, particularly those
related to fabrication are considered in an implicit manner through defined discontinuities or
weld defects acceptance criteria and quality control requirements. These general requirements
will be defined in a standard concerning the "Execution of steel structures".
Table 1 The main factors affecting fatigue strength

Designation of the factors affecting the fatigue strength

Taken
into
account
in
Eurocode 3

Stress
Stress or strain range

Stress sequence
Frequency (no significant effect when < 40 Hz in a non
corrosive environment)
Mean stress (no effect in heat affected zone due to residual
*
stresses)
Residual stresses

Geometry
Nominal or geometrical stress
Local stress concentration

Small discontinues

- scratches

*(implicit)

192

- grinding marks
- surface pittings
- weld defects or misalignments
Size effect (or scale effect)
*

Material Properties and Fabrication

Stress-strain behaviour of materials
Hardness
Chemical composition of steels
Metallurgical homogeneity
Electrical potential
Micro structural
boundaries)

discontinuities

(grain

size,

grain

Welding process
Weld heat treatment
Weld surface treatment

Environment
Corrosive atmosphere

*(implicit)

Temperature

*(implicit)

Humidity (hydrogen embrittlement)

Irradiation

2.2 Fatigue Failure Criteria

193

In the preparation of Eurocode 3, classification into detail categories was established from a
statistical analysis of fatigue test data obtained from various laboratory sources. To obtain more
homogeneous samples of the test results, particular attention was paid to failure criteria
considered in these tests.
Several failure criteria may be adopted to characterize the experimental failure condition at the
end of a fatigue test in the laboratory. Three criteria are generally considered:

First appearance of a crack either detected visually or detected by means of a physical

measure, e.g. by the record of a change in the local strain condition.

Through-thickness crack: the fatigue crack starts from the front surface and grows
through the thickness of the test piece and reaches the back surface.

Complete fracture of the tested specimen or large displacement of the tested structural
element such that the displacement becomes so important that the applied "jack load"
cannot be maintained. When performing a fatigue test on a beam, the failure may be
conventionally defined as the point when the mid-span deflection reaches a certain
limit.

Generally for small scale specimens, the difference between the fatigue life at complete fracture
and at a more realistic tolerable fatigue crack size is negligible. However, in a large scale
structural element tested in fatigue the difference may be highly significant.
In Eurocode 3, the fatigue strength refers to the complete failure of the structural element. This
condition corresponds, usually, to the criterion generally adopted by structural laboratories or
reported in literature.
2.3 Design Stresses for Fatigue Assessment
Different stresses may affect the fatigue strength classification of a structural detail. For a
particular detail, the various origins of stresses have to be identified in order to define more
precisely the design stresses for the fatigue assessment concepts involved in Chapter 9 of
Eurocode 3.
a. Nominal Stress
Consider a uniform structural member subjected to a simple axial force or to a bending moment.
The nominal stress is the stress resultant calculated according to the basic strength of material
(Figure 1).

194

The nominal stress of a member under uni-axial stress is:

N =

(2.1)

where N is the normal force and A the gross section area.

For a prismatic member section under a bending moment, the stress resultant is:

M =

(2.2)

where:
M is the applied bending moment
I is the moment of inertia of the section
v is the distance from the neutral axis to the outmost fibre.
b. Stress concentration effect due to geometrical discontinuities
There are three main sources which can create a state of stress concentration in a structural
detail:
The global geometry of the structural element which contains the structural detail, e.g.
attachments on a beam web or gusset plates on a beam flange.
The local stress concentration due to local disturbance of the weld geometry, bolt holes, local
variation in stiffness, etc... For example, if a hole is drilled in a plate, the stress distribution
across the section containing the hole will be different from the nominal stress distribution
existing in the plain plate cross-section. An important stress gradient will occur in the vicinity of
195

the hole. This geometrical stress concentration is due to both the decrease from the gross section
to the net section and to the stress "raiser" (concentrator) caused by the presence of the hole
(Figure 1).
The local stress concentration due to local discontinuities occurring during fabrication
(misalignment, surface scratch, pitting, weld defect, etc).
In many cases, and by simplification, the geometric stress concentration is usually calculated on
the basis of the nominal stress applied to the gross section area and the stress concentration
factor kG, as:
G = kG . nom (2.3)
This structural geometrical stress concentration, which is defined as the maximum principal
stress existing in the vicinity of the detail, may be evaluated from experimental tests or from
finite element methods.
The local stress concentration is present in addition to the structural geometric stress
concentration and may be due to local disturbances of the local geometry of the detail such as:
local cross-section change (geometry of welds for example).
local geometrical imperfections such as misalignment.
small local discontinuities inherent to the action of the environment or of the fabrication
process such corrosion pits, surface scratches, drag lines due to flame cutting, grinding marks,
welding process defects such as undercut, lack of penetration, lack of fusion, slag inclusions,
porosities, hydrogen-induced cracking, etc. These very small discontinuities are present in every
element of engineering structures. Their presence determines a potential location for initiation
of a fatigue crack.
Local stress concentrations are taken into account in an implicit manner in the derivation of the
S-N curve from fatigue test results. Great care must be taken when assessing fatigue strength
from tests on small scale specimens instead of large scale specimens. The scale effect due to
weld geometry may have a greater influence on the fatigue strength in small test specimens than
in large test specimens.
Usually, fatigue specimens have been tested with inherent discontinuities, and fatigue strength
curves, so derived, make allowance for tolerable defects. The acceptance criteria for weld
discontinuities which will be proposed in the "Execution of steel structures" standard would
guarantee the fitness for purpose of the fatigue strength design rules of Eurocode 3. In other
words, the quality assurance system which covers the fabrication process should ensure that the
fabricated constructional detail complies with the relevant quality requirement specified in the
standard for the "Execution of steel structures".
When assessing the fatigue strength by the so-called geometric stress range method, according
to Clause 9.5.3 of Eurocode 3, the geometric stress concentration as defined by Equation (2.3)
must be properly evaluated. The local geometry of the weld must not be taken into account in
196

the calculation procedure of the design stress range, since the local discontinuity effect is
already introduced in the derivation of the S-N curves. However, when determining the design
stress, secondary stresses arising from joint eccentricity or due to joint stiffness, stress
redistribution due to buckling or shear lag, and effects such as prying action, should be taken
into account.
3. DESIGN STRESS SPECTRUM
3.1 Stress History
A fluctuating stress to which a structural detail is subjected may have a stress history of constant
amplitude or of variable amplitude (Figures 2 and 3).

For cumulative damage analysis, the stress history is split up into individual cycles and related
stress ranges which are summed up to a distribution of stress ranges. This distribution of stress
ranges is called a stress spectrum, see Lecture 12.2.
For a variable amplitude stress history, there is a need to define such a stress cycle associated
with a particular stress range. There are several procedures for cycle counting methods.
Eurocode 3 refers to the "reservoir method" which gives a sound representation of the stress
197

variation characteristic by allowing a proper contribution of each stress range to the fatigue
damage process. This stress range counting method is the most commonly accepted. This
counting method is somewhat similar to the well known "rainflow counting method". The
"rainflow" and the "reservoir" counting methods do not lead to exactly the same result.
However, in terms of fatigue damage both counting procedures give very close results, and for
"long" stress histories they give nearly the same result.
3.2 Stress Histogram
The most common way of representing irregular stress histories for fatigue analysis is to sum up
the stress ranges of equal amplitude, and to obtain a distribution of stress range blocks which is
called a stress histogram (or a stress spectrum) consisting of a number of constant stress range
blocks. Each block is characterized by its number of cycles n i and stress range i (Figure 4).
The ordering of the different blocks does not make any difference since the damage calculation
rules specified in Eurocode 3 refers to the linear cumulative damage rule of Palmgren-Miner.
However for convenience the stress histogram is commonly presented with stress blocks ranked
in decreasing order (Figure 5) which often can be approximated by a two-parameters Weibull
distribution such as:

198

 = 0

(2.4)

4. FATIGUE DESIGN CURVES CLASSIFICATION CONCEPT

The classified fatigue design curves adopted in Eurocode 3, are the same as proposed in the
"European Convention for Construction Steelwork Fatigue Recommendations" [2]. The ECCS
Fatigue Recommendations were one of the first attempts to provide uniformity to the
determination of the fatigue strength design curves.
The ECCS Recommendations define a set of equally spaced S-N curves plotted on a log-log
scale. Reference to these curves allows a detail category to be classified (representative) of a
particular structural detail which corresponds to a notch effect or a characteristic geometrical
discontinuity). This classification has been determined by a series of fatigue test results, from
which a statistical and a probabilistic evaluation is performed, see Lecture 12.7.
Each individual fatigue strength curve is defined in a conventional way (Figure 6) by a slope
constant of m = 3 (slope = -1/3). The constant amplitude limit is set at 5 million cycles. The
slope constant m = 3 was a best fit for a large number of different structural details tested in
fatigue. The figure of 5 million cycles for the constant amplitude fatigue limit is a compromise
between 2 million cycles for "good" details and 10 million cycles for details which create a
severe notch effect. For any stress range of constant amplitude below this limit, no fatigue
damage is expected to occur.

199

When a detail is subjected to variable stress ranges, which is generally the case in reality,
several options may occur:

If no stress range of variable amplitude exceeds the fatigue limit, no fatigue damage
assessment has to be carried out.

If at least one stress range block exceeds this fatigue limit, a damage calculation has to
be performed on the basis of the linear cumulative damage rule, referred as the
Palmgren-Miner's rule.

In this last option, two cases have to be considered for the cumulative damage calculation when
some stress ranges are below the constant amplitude fatigue limit:

Either the damage calculation is made simply assuming that the S-N curve of slope
constant m = 3 is extended beyond the constant amplitude fatigue limit.

Or the damage calculation is made assuming that beyond the constant amplitude fatigue
limit, the S-N curve of slope constant m = 3 is extended by a straight line of slope
constant m = 5. The intercept of this straight line with the vertical line at 10 million
cycles provides a cut-off limit. The reason for using an S-N curve with two slopes to
cumulative damage calculations is that it is an approximate way to take into account the
progressive reduction of the constant amplitude fatigue limit as a result of the damage
caused by the stress ranges above that limit. In this way, eventually all stress ranges in
the spectrum become damaging. Fracture mechanics confirm this decrease of the slope
of the S-N curve in the long fatigue life range.

In both cases, all cycles below cut-off limit can be ignored when evaluating the fatigue damage.
It should be noted that Eurocode 3 leaves the design engineer free to use either the single-slope
S-N curve or the double-slope S-N curve.
Experimental results have indicated that within the range of high numbers of cycles, a change in
the slope of the fatigue strength occur due to a decrease of the crack growth rate. The
introduction of a double-slope concept and a constant amplitude fatigue limit at 5 million cycles
is still a matter of controversy. Despite a number of criticisms, particularly concerning the
increase in complexity of the analysis, Eurocode 3 has kept the double-slope curve because this
rule may, for some detail categories, improve the accuracy of the fatigue check. However, this
200

improvement can not expected for all types of structural detail, and all stress spectra. In some
cases, especially for those details with a very severe notch effect, the double-slope curve may
not lead to a conservative result.
Some details, for example, cover-plated beams, have shown a constant amplitude fatigue limit
of almost 10 million cycles. To avoid non-conservative conditions, some details (which
generally have severe notch effect) have been classified in categories slightly lower than their
fatigue strength at 2 million cycles would have required. The concept of the specified ECCS
fatigue design curves, which consists of 14 equally spaced curves, a new design fatigue strength
curve is not required for each new structural detail.
The "grid system" of S-N curves has been established as follows. The vertical distance of the
ordinate log-scale between each fatigue strength curves has been obtained by dividing the
difference between one order of magnitude into 20 equal spaces (Figure 7). For example, taking
two reference values as c=100MPa and c = 1000 MPa at 2 million cycles, the
calculation of the spacing is determined from the following:

The general S-N curve equation may be written as:

log N = log a - 3 log (4.1)
so with c = 100 MPa (log 2 000 000 = 6,30103)
log a = 6,30103 + 3 log 100 = 12,301 (4.2)
and for c = 1000 MPa
log a = 6,30103 + 3 log 1000 = 15,301 (4.3)
The spacing between two contiguous curves represents
log a = (15,301 - 12,301)/20 = 0,15 (4.4)
So starting from the reference values of c = 100 MPa, with log a = 12,301, the subsequent
values of c may be obtained from Equation (4.1) as given in Table 2.
201

Table 2 Characteristic fatigue strength at 2 million cycles

log a

c (rounded value)

...

...

12,601

125

12,451

112

12,301

100

12,151

90

12,001

80

...

...

Table 2 shows that the number defining the characteristic fatigue strength at 2 million cycles,
used as a detail category identification, is a rounded value.
5. FATIGUE TEST RESULTS
Generally fatigue strength curves are evaluated from series of fatigue tests performed on
specimens which typically reproduce the detail to be studied. The fatigue strength curves (S-N
curves) can be most accurately determined when a group of fatigue specimens are tested at
different stress range levels. However, there is no recognized standard method for fatigue testing
and design experiments. As a result, the fatigue test data found in the literature are somewhat
non-homogeneous.
It is clear that, under such circumstances, a review of existing fatigue data and their statistical
evaluation, even when limited to the same detail category, may lead to large discrepancies in the
results. Such differences may be attributed, not only to the fatigue testing practice in each
laboratory, but also to the detailed fabrication procedure and quality achieved in the preparation
of the specimens. Discontinuities play a major role in fatigue strength, particularly for welded
details and careful consideration must be given to the weld quality which may considerably
affect the variation in fatigue strength.
Fatigue specimens are fabricated with certain inherent discontinuities which are not fully known
or may not be properly evaluated in laboratory reports. In such cases, it is generally rather
difficult to appreciate if the fabrication quality of specimens is representative of current
workshop practice. Moreover, when performing a statistical analysis on fatigue test data from
different origins, a rather large variation of fatigue strength may result. Careful attention must
be paid to the homogeneity of the fatigue resistance.
These considerations were borne in mind during the preparation of Eurocode 3. The fatigue test
results which were statistically analyzed and then classified according to the procedure
described fulfil certain requirements:
202

Priority was given to test results from full size specimens compared to small scale
specimens simulating the same structural detail. For a comparable quality of weldments,
smaller welded test specimens exhibit a higher fatigue strength (and a higher constant
slope) than full size test specimens. This difference in fatigue behaviour is mainly due
to the fact that full size specimens lock in more residual welding stresses than small size
specimens do. This difference is residual stress magnitude is the result of variations in
mechanical constraints during welding.

In welded specimens the stress range () and the number of cycles to failure (N)
were considered as the main parameter controlling the fatigue strength curve.

A minimum of 12 fatigue test results were required to reach a certain significance level
and to lead statistically to a confident interpretation of the test results.

6. CUMULATIVE DAMAGE RULE, EQUIVALENT STRESS RANGE CONCEPT

6.1 Palmgren-Miner Summation
In real life, structural elements are subjected to varying fatigue loads, and not to constant
amplitude fatigue loadings. Eurocode 3 refers to the Palmgren-Miner summation to evaluate the
cumulative damage (Figure 8). This rule is based on the assumption that the total damage
accumulated by a structural element under varying stress ranges, is obtained by the linear
summation of the damage of each individual stress range, i.e:

D=

(6.1)

where:
ni is the number of cycles of constant amplitude stress ranges i
Ni is the total number of cycles to failure under constant amplitude stress range i.
The structural element is designed safely against fatigue if:
D 1 (6.2)
203

No account of the damage is taken for any varying stress ranges falling below the cut-off limit.
6.2 Equivalent Stress Range
The concept of equivalent stress range has been introduced in the ECCS Recommendations [2]
and is also referred to in Eurocode 3. The definition of the equivalent stress range is
conventional. It can be said that the equivalent stress range concept is simpler than a direct
Palmgren-Miner summation when the S-N curve is of unique slope (-1/m). The expression is, in
this case, quite simple and the recalculation of the damage for each S-N curve is therefore
avoided:

equ =

(6.3)

with m = 3 or m = 5 as appropriate.
The equivalent stress range equ depends only on the fatigue load spectrum and the slope
constant m. In such a case, knowing equ evaluated according to Equation (6.3), it is easy to
choose directly a detail category which will have an adequate fatigue resistance.
6.3 Equivalent Stress Range for an S-N Curve with a double Slopes Constant
When the basic S-N curve is of double slope, the expression of the equivalent stress range
becomes more unwieldy. The practicability of its application is questionable, except if using the
limit state function as defined by the following equation:
f . equ Rd / f (6.3)
The derivation of equ when the S-N curve has a double slope is given below:
a. Damage calculation for a double slope S-N curve when the stress range is below and
above D
Suppose there are some stress range blocks where the range is below the value of D and
some above D (Figure 9); it is assumed that the proper partial safety coefficients have
introduced in i and j.

204

 block i when i > D

block j when j > D
From the definition the damage is given by:

D=

(6.5)

taking into account the S-N curve slope for each set of stress range blocks:

D=

(6.6)

Equation (6.6) may be written as:

D=

(6.7)

From Figure 9:
ND = a D-3 = b D-5
ND corresponds to the fatigue limit of the S-N curve at 5 million cycles.
a/b = 1/D2 (6.8)

205

Hence:

D=

(6.9)

where:
Q = ni i3 + nj j3 (j /D)2
The damage may be calculated using either Equation (6.5) or Equation (6.9) directly.
b. Calculation of the equivalent stress range equ for a double slope S-N curve
In this particular case, a decision must be made as to which slope the definition of equ refers.
The choice of a slope constant of 3 or 5 makes absolutely no difference to the final result of the
calculation of equ when the load spectrum straddles both parts of the double slope S-N
curve. The calculation of the equivalent stress range equ is derived below from a slope
constant of m = 3 of the double slope S-N curve (noted as equ.3). The same demonstration
holds for a slope constant of m = 5. By definition:

D=

(6.10)

where:
Nequ is the equivalent number of cycles at failure under the equivalent stress range equ
N is equal to ni + nj
Evaluating Nequ on the basis of the S-N curve of slope constant m=3:

D=

(6.11)

by equating Equations (6.6) and (6.11), the damage is:

D=

(6.12)

then Equations (6.11) and (6.12) give:

equ3 =

(6.13)

therefore:
206

equ.3 =

(6.14)

Rd.3 is defined as the fatigue resistance corresponding to equ.3 on the S-N curve of
constant slope m = 3.
Rd.3 = D (ND / N)1/3 (6.15)
From Equations (6.14) and (6.15):

(6.16)

This expression is equal to the damage as given by Equation (6.9):

(6.17)

Remarks:
1. Both fatigue assessment formats, the Palmgren-Miner summation, and the equivalent
stress range concept, are rigorously equivalent in terms of damage.
2. Reference in the above demonstration is made to D and ND corresponding to the
"knee" point of the double slope S-N curve. Since the S-N curve is written as:
N (Rd)m = a = constant
another reference value may be taken, for example:
D3 ND = C3 NC = constant
C, being the stress range at NC = 2 million cycles.
3. Special care must be taken when calculating equ.3 and Rd.3: both expressions
must be evaluated with the same slope constant.
4. The values of equ.3 and Rd.3 are clearly different and may not be used
indiscriminately when plotting fatigue test results on a log versus log N diagram.
Generally when fatigue tests have been performed under variable stress range
amplitude, the equivalent stress range as given by Equation (6.3) has been used to plot
the experimental results.
7. RESIDUAL STRESS EFFECT

207

Welded joints in structural details contain tensile residual stresses in the vicinity of the weld
bead. Figure 10 shows that their magnitude may be as high as the yield stress of the weldment
metal. Figure 10 also shows high tensile residual stresses near the edges which were flame-cut.

It is well established that the presence of residual stresses of such magnitude makes the fatigue
strength of a welded joint independent of the applied load ratio, and dependent only on the
applied stress range. The full significance of the tensile residual stresses due to welding was not
appreciated originally, since many fatigue test results were obtained from welded specimens
which were too small to retain the major part of the welding residual stresses such as would
occur in large structural components.
It is evident that tensile stresses play a significant role in the propagation of a crack, since they
tend to act as a opening mode due to tensile stresses applied at the crack lips. The crack
propagation rate is likely to be reduced, when the crack grows into a zone of compression
residual stress.
It is in recognition of this physical crack propagation behaviour that the R ratio (R = min/max)
has been considered in Eurocode 3 Chapter 9 for non-welded or stress relieved details. Figure
11 shows the comparison between fatigue test results and two "bonus factor" rules which were
studied when drafting Chapter 9. The rule which was finally selected takes into account of the
208

effect of compressive stress ranges by multiplying the part of the stress range in compression by
a factor of 0,6. The validity of this rule has been compared with fatigue test results performed
on non-load carrying weld cruciform joints for various R ratios ranging from -3,0 to 0,8. These
fatigue tests were carried out on small specimens.

8. CONCLUDING SUMMARY

Fatigue behaviour of structural details is governed by many factors which are by nature
random.

The present state of knowledge provides sufficient information for reasonably

comprehensive and safe fatigue design rules. It is recognized that the extreme life
region of the fatigue strength curve is not well established. The current availability test
data in this region are very sparse.

The quality of fatigue design is closely related to the attention given to structural
details, i.e. not only to the geometrical shape and the dimensions, but also to the quality
of fabrication and the acceptable defects, etc.

In carrying out the fatigue assessment of structures, the designer must, first, carry out a
proper fatigue load analysis in order to evaluate correctly the stress resultants acting on
details.

The designer must then select the proper fatigue strength curve related to each of these
details.

Both the analysis and curve selection require skill in recognition and interpretation of
the main design factors affecting fatigue resistance.

9. REFERENCES
209

[1] Eurocode 3: "Design of Steel Structures": ENV1993-1-1: Part 1.1, General rules and rules
for buildings, CEN, 1993.
[2] European Convention for Constructional Steelwork: Recommendations for the Fatigue
Design of Steel Structures. ECCS Publication 43, 1985.
[3] Eurocode 1: "Basis of Design and Actions on Structures", CEN (in preparation).

210

Lecture 12.9: Eurocode 3 Classification of Constructional Details

OBJECTIVE/SCOPE:
To assist with an understanding of the proper detail category and of the importance of detail
execution. To present an insight into the significance of details from the study of individual
cases.
PREREQUISITES
None.
RELATED LECTURES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.8 Basic Fatigue Design Concepts in Eurocode 3
SUMMARY
Based on Chapter 9 of Eurocode 3, Part 1, this Lecture presents an examination of typical
structural details in an imaginary bridge structure. This enables a designer to compare various
details when designing a structure, and to be aware of specific fabrication requirements. It also
enables the designer to decide the proper category of details for calculations of fatigue strength.
1. INTRODUCTION
In Lecture 12.1 the basic concept of classifying structural details into categories depending on
their fatigue resistance is introduced. Each category is defined by the design constant stress
range which can be endured with adequate reliability for 2 106 cycles (e.g. a category 112
detail would endure 2 106 cycles of 112 N/mm with an adequate degree of reliability). For
each such category, fatigue strength curves (commonly referred to as "S-N" curves) are
presented in Figs. 9.6.1, 9.6.2, 9.6.3 and 9.7.1 of Eurocode 3 Part 1. These curves provide a
complete relationship between the stress range and endurance for each category.
A large (but necessarily limited) selection of structural details is illustrated in Chapter 9 (Section
9.8, Tables 9.8.1 to 9.8.7) of Eurocode 3 Part 1, together with the categories (determined from a
large number of test results ) into which the details fit. The first task facing a designer checking
a structure for fatigue is therefore the determination of the appropriate categories for the details
of the structure. This may appear a comparatively trivial task, but in practice a civil engineering
structure will contain a large number of structural details which may be prone to fatigue
cracking; problems in allocating appropriate categories can arise from several sources, the most
important of which are:
a. The actual details may not correspond exactly to any of those described in Eurocode 3 Part 1.
b. The detailed geometry, stress distribution and direction, workmanship, etc. may alter the basic
category.

211

In order to assist a designer in selecting the correct detail category, this lecture presents a case
study of a particular civil engineering structure, and shows how the details can be classified.
2. GENERAL PRESENTATION OF THE CASE STUDY
The case study comprises an imaginary steel bridge, as shown in an exploded isometric view in
Figure 1. In order to illustrate as many different details as possible, two forms of construction
are shown with a box girder on the left hand side and a braced plate girder on the right hand
side. Furthermore, there are differences between the two sides in bearing arrangement,
connection of cross girders, etc. It is not suggested that these arrangements, nor even some of
the details, are necessarily representative of good design practice; they are presented for the
purpose of illustrating a point of discussion.

This imaginary structure is then subdivided as shown in Figure 1 into several close-up details in
Figures 2 - 7, which in some cases are further subdivided where necessary for clarity. The
detailed figures indicate the direction of principal stress, and the potential crack location and
direction; the category into which the detail should be classified is shown as a number in a circle
beside the detail. Occasionally the slope of the S-N curve departs from the "standard" value of
3; this is indicated where "m=.." is noted beside the detail category.
In a number of instances, where indicated by a small letter inside a square, additional notes are
required to enlarge on some important factors which may affect the classification. This may
happen where some guidance is needed in respect of the calculation of the design stress or of the
fabrication procedure or requirements. Sometimes more than one form of cracking may occur at
a single site, or more than one factor may contribute to cracking. Such matters are also covered,
where relevant, by notes.
Where a detail in the case study does not correspond closely to a specified detail in Chapter 9 of
Eurocode 3 Part 1, the category quoted is based on that specified in other codes or derived from
212

tests known to the authors of this lecture; when this knowledge has been used it is described in a
note. It is also expected that the Tables of details will be extended or modified for steel bridges
when Eurocode 3 Part 2 is published in due course, and in a few cases reference is made to what
may appear there. Furthermore, even when a detail is ostensibly covered by Chapter 9, it is
sometimes necessary to take into account particular design aspects or fabrication procedures
when determining the category. Such considerations (many of which are outside the scope of
this lecture, but are covered in other lectures) are:
a. Design Stress Evaluation
The importance of knowing the direction of the principal stress fluctuation cannot be overemphasised; in most cases this is straightforward, but sometimes it is made more difficult by the
uncertainty of prediction of exactly where the crack will occur and in what direction it may
propagate. Since the stresses to be used must be based on elastic calculations, account must be
taken of secondary stresses which in complex details may be difficult to quantify.
b. The Quality of the Preparation of a Joint
The fit-up of a joint, and the method of terminating a butt weld (extensions or run-off plates)
can all affect the category, especially for site welds.
c. The Accessibility of a Location for Welding
Welds with difficult access are frequently of lesser quality and are more prone to defects and
hence fatigue than those which are easily accessible.
d. Proper specification of parent metal and welding consumables.
e. Operating Procedure
For example, butt welds without backing flats are of a better category than those with, provided
that full penetration can be obtained. There are differences between permanent and temporary
backing flats.
f. The Profile and Surface Finish of the Weld
Welds with excessive "reinforcement" or rough surface generally perform less well in fatigue
than if they are ground level and smooth.
g. The Acceptance Criteria for Weld Quality
In order to achieve the category in Chapter 9 of Eurocode 3 Part 1 certain minimum standards of
quality in regard to defect shape, location, type and size must be attained.
h. Improvement Techniques for Welded Joints
These include grinding, peening, gas tungsten arc remelting, etc. They are intended to improve
the category above that shown in Chapter 9, and are described in detail in Lecture 12.5.
213

j. Weld Defect Assessment and Repair

If the defects in a weld are greater than are permitted for a particular category, the weld may,
rather than being rejected, either be downgraded (if the fluctuations in stresses are low) or
repaired. Repair requires special care in specification and execution; not infrequently a repaired
weld is worse than the original weld. Whatever repair procedure is used, the final category
should be determined with caution.
3. NOTES ON DETAILED FIGURES 2 - 7 OF THE CASE STUDY
FIGURE 2

a. This detail is not given explicitly in Chapter 9; it should generally be avoided (it is usually
better, and probably easier, to detail the longitudinal stiffeners passing through "mouseholes" in
the transverse stiffeners). The nearest detail in Chapter 9 is shown under cruciform joints, where

214

root cracking of the weld should be checked as category 36*, and of the plate from the weld toe
as shown as category 71. The category shown of 50* results from other work.
b. The category of 112 shown is the "standard" one for automatic fillet welding carried out from
both sides, but containing stop-start positions (Table 9.8.2 (3) or (4)). If it contained no stopstart positions it could be upgraded to category 125, or even 140 if a specialist inspection shows
that the welds are free from significant flaws; conversely, if the fillet were placed manually, it
would be downgraded to category 100. It is likely that Eurocode 3 Part 2 will exclude the higher
categories of 125 and 140 for this detail since the necessary quality of workmanship is
impractical for bridge structures.
c. Stresses should be calculated using the gross section for slip resistant connections, or the net
section for all other connections (Table 9.8.1 (6) or (7)). The effects of eccentricity in the
connection should be taken into account when calculating the stresses in a single-sided
connection.
d. This detail (at the termination of a longitudinal stiffener) may be treated for cracking in the
main plate as a long (>100m) longitudinal attachment within the width of a plate with a nonload carrying weld (Table 9.8.4 (1)). Eurocode 3 Part 2 will probably add the requirement that
the weld should be carried round the end of the stiffener. Note that the weld may also require
checking in shear, with the stress range calculated from the weld throat area.
e. The gusset plate attached as shown to the leg of the angle may be treated as a cover plate
wider than the flange (with the leg of the angle representing the flange) (Table 9.8.5 (5)).
Provided all plate thicknesses are 20mm or less, this is category 50* for cracking in the angle;
this reduces to 36* if thicknesses exceed 20mm. The weld should be continued down the leg of
the angle, and ground to remove undercut if necessary.
FIGURE 2a

215

a & b These details show how cracks may grow in different directions in an area of complex
geometry and stress distribution. Considerations are similar to Figure 2, note e, but the bearing
plate > 20mm thick and so the category for plate cracking is reduced to 36* (Table 9.8.5 (5)).
c. See Figure 2, note b; as the weld to the bearing plate will almost certainly be placed manually,
the lower category of 100 is used (Table 9.8.2 (5) or (6).
d. The category of the plate edge depends on the method of production; if it is a rolled flat the
category could be increased to 160, or if machine flame cut with subsequent machining to 140.
The indicated category of 125 is for a machine flame cut edge without subsequent machining,
although Eurocode 3 Part 2 will probably specify the quality of the cut edges (Table 9.8.1 (5)).
It should contain no repairs by weld infill.
e. As for Figure 2, note b.
f. As for Figure 2, note c.
g. The category of this weld has been reduced from the 71 or 80 shown for web stiffeners (Table
9.8.4 (4)) since the stiffener is shown flush with the edge of the plate.

216

FIGURE 3

a. This is a rather poor detail, since because of the taper in the flange a good fit cannot be
guaranteed above the backing flat; hence the low category of 50 (Table 9.8.3 (11)).
b. At the top of the butt weld, provided the "reinforcement" does not exceed 0.1 times the width
of the weld bead, the category is 90 (Table 9.8.3 (4)); up to 0.2 it would be 80 (Table 9.8.3 (7)).
Run off pieces should also be used. (If the weld is ground flush the category could be 125 or
higher). Normally there would be little point in making the category of the top surface much
higher than that of the bottom, unless the eccentricity arising from the change of plate thickness
results in a higher stress range at the top.
c. The comparatively high category of this weld is only true for a gusset plate with a generous
radius as drawn ( > 150mm, and also > (width of main plate)/3) (Table 9.8.4 (2)). The radius has
to be formed by initial machining or gas cutting, with subsequent grinding of the weld area
parallel to the direction of stress. If the radius < (width of main plate)/6 the category falls to
45*, and between the two limits above to 71.

217

d. This is a straightforward application of the detail of Table 9.8.5 (4). It should be noted that the
weld should be held back 10mm from the end of the gusset. As it is a single sided connection,
the effects of eccentricity should be considered.
e. This is again a straightforward application, this time of Table 9.8.5 (3). The calculation of the
stress in the main plate requires care, and in a single sided application as shown may have to
allow for eccentricity.
f. This is a standard "bad" detail for increasing the area of a plate and the category is given in
Table 9.8.5 (5). The plates in the example are not thicker than 20mm so the category is 50*;
above this thickness the category is reduced to 36*. Contrary to what may be thought, tapering
the cover plate as shown, or rounding its end, does not, in itself, improve the detail; however, a
special detail with tapering welds and chamfered cover plates, being developed by German
Railways, may raise the category to 80. This may appear in Eurocode 3 Part 2.
g. This is a two sided butt weld, with the surface ground flush (Table 9.8.3 (1)). Significant
quality control and inspection is required to permit the use of this high category.
h. As for Figure 2, note d.
FIGURE 4

218

a. This is a butt weld, without backing flat and ground flush, between plates of different
thickness (Table 9.8.3 (3)). Provided the difference in thickness is taken up by tapering the
thicker member with a slope of not greater than 1:4, this still qualifies as a category 112 weld.
b. As Figure 3, note g.
c. As Figure 3, note g, but it is a single sided weld without backing flat and because very high
quality of execution and inspection is specified, the category can be raised to 125 (Table 9.8.3
(1)).
d. As for Figure 2, note c, but as the connection is double sided no eccentricity occurs.
FIGURE 4a

219

a. This is the standard detail for fillet welds in shear (Table 9.8.5 (6)). The stress range should be
calculated from the weld throat area.
b. As for Figure 2, note c, but as the connection is double sided no eccentricity occurs. Note that
crack begins from edge of washer.
c. As for Figure 2, note c, but as the connection is double sided no eccentricity occurs. Note
carefully that direction of crack is related to direction of stress.
FIGURE 4b

220

a & b Are both as for Figure 2, note c, but note that the direction of stress, and hence the
direction of cracking, may differ from hole to hole.
c. As for Figure 4a, note a.
d. This is the standard detail of Table 9.8.4 (4) (left hand diagram), provided the thickness of the
stiffener does not exceed 12mm. (If it does exceed 12mm, the category is reduced to 71). Note
that the stiffener should terminate at least 10mm above the flange, and the weld should be
returned round the bottom of the stiffener. Some recent evidence suggests that out-of-plane
flexure of the web plate at the termination of the stiffener could degrade this detail, but research
continues on it.
FIGURE 5

221

a. As for Figure 4a, note a. Note crack propagating across direction of principal tensile stress.
b. This is the standard detail for the welding of diaphragms in box girders to the webs and
flanges, where the diaphragm thickness is not greater than 12 mm (Table 9.8.4 (5)). If the
thickness were greater the category would be reduced to 71.
c. This is the standard detail for corner welds of box girders (Table 9.8.2(6)). Note that a good
fit between flange and web is essential, so that a one sided weld can be placed without blow
through. In certain forms of construction and loading this weld is also prone to bending about its
longitudinal axis due either to local traffic loading or distortional effects in the box girder. It is
virtually impossible to give a category for such effects, and considerable experience is
necessary.
d. As for Figure 2a, note d.
e. See Figure 2, note b. As the weld will be placed manually, it is category 100 (Table 9.8.2 (5)).
FIGURE 5a

222

a. This weld is being stressed by flexure of the web plate and is not readily classifiable from the
details in Eurocode 3 Part 1. It is similar, however, to the long attachment, Table 9.8.4 (1), and it
is probably safe to use that category (50*).
b. As for Figure 2, note c.
c. See Figure 3, note c. Because the main plate (the flange of the box girder) is wide, the radius
of the gusset plate is more severe than it appears and hence the weld falls into the lowest
category, for this detail, of 45*.
d, e & f These welds are very difficult to categorise and are not covered explicitly in Eurocode 3
Part 1. Furthermore, although the direction of stress is shown by arrows on the detail, the welds
may also be subjected to flexural effects in the web and flange. Considerable caution should
therefore be used in attempting to classify them.
Detail d can be thought of as an incomplete penetration butt weld placed from one side only
(Table 9.8.3(8)), and hence classified as category 36*.
Details e and f are analogous to the cruciform detail, Table 9.8.5 (2), and so are category 36* as
far as cracking from the root is concerned. Cracking in the parent plate from the toe of the weld

223

may be checked at the higher category of 71 in the gusset plate (detail e, Table 9.8.5 (1)) or 90
in the flange (detail f, Table 9.8.3 (4), provided the special requirements in the table are met).
FIGURE 5b

a. As this will be a machined plate, the high category of 140 may be used (Table 9.8.1 (4)).
However, as there is a re-entrant corner, stress concentrations will occur and the magnified
stresses should be used in making the check.
b. This is similar to the category at the end of lengths of intermittent fillet weld where the gap is
less than 2.5 times the weld length (Table 9.8.2(8)). Hence the category may be taken as 80.
c. As for Figure 5, note b.

224

d. As for Figure 4a, note a. Note crack propagating across direction of principal tensile stress.
FIGURE 6

a. This detail is not classified in Eurocode 3 Part 1. It is clearly of a very low category and
should not be used if the stress range is significant. It would appear appropriate to classify it as
the lowest category available, 36*.
b. The effect of the shear connectors on the base plate is to cause a category 80 detail (Table
9,8.4 (6)).
c. The weld connecting the shear studs is classified in Table 9.8.5 (8) with the shear stress
calculated on the nominal cross section of the stud. Further information on fatigue of studs is
available in Eurocode 4.
d. As for Figure 4a, note a.
FIGURE 7

225

a. This detail represents a butt weld on a permanent backing flat, where the backing flat fillet
weld terminates closer than 10mm from the plate edge (Table 9.8.3 (11)).
b. As for Figure 2, note b.
c. As for Figure 2, note c.
d. This connection is effectively a welded transverse attachment with a non-load carrying weld
(Table 9.8.4(3)). However, the weld terminates at the plate edge, and so the detail is a worse
category than in the table. Category 50 appears appropriate. It should be pointed out from this
how an apparently minor, non-structural, detail can seriously degrade the fatigue capacity of the
structure. If it has to be used, it should be positioned in an area of low stress fluctuation.
FIGURE 7a

226

a. These welds are similar to those shown on Table 9.8.5 (1) for cracking in the parent plate
from the toe of the weld in cruciform joints.
b. This detail is effectively a gusset with zero plan radius (Table 9.8.4 (2)) and so falls into
category 45*.
FIGURE 7b

227

a. This is a detail which is not explicitly classified in Eurocode 3 Part 1. It is close to the
cruciform detail (Table 9.8.5 (2)) but probably rather less severe. An appropriate category is
50*.
b. These welds are all effectively the worst possible cruciform details (Table 9.8.3 (2)). Note
that if the welds are made of sufficiently large section to avoid root cracking there are other
mechanisms which may govern.
FIGURE 7c

228

a. As for Figure 4a, note a. Note crack propagating across direction of principal tensile stress.
b. These welds are similar to those shown on Table 9.8.5 (1) for cracking in the parent plate
from the toe of the weld in cruciform joints.
c. This weld is likely to be placed manually - see comments at note b for Figure 2.
d. As for Figure 2a, note d.
e. This detail is intended to represent what happens with a bolt in tension (Table 9.8.1 (8))
through an endplate. The category for the bolt itself is the low one of 36*, and the stress in
tension in it should be calculated using its stress area. Account should also be taken of any
prying action resulting from flexing of the endplate; it should be noted, however, that the stress
range in the bolt may be reduced substantially by appropriate preloading. The crack position in
the endplate shown on Figure 7c should also be checked under the flexural stresses resulting
from prying action.
FIGURE 7d

229

a. This is a straightforward instance of the detail for the ends of a continuous weld at a cope hole
(Table 9.8.2 (9)).
b. This detail is a straightforward instance of the end of an intermittent fillet weld. Note that
where it occurs close to (but not actually at) a cope hole, it permits use of the higher category of
80, compared with detail a above where terminating the weld actually at the cope hole requires
use of category 71.
c. This detail is not explicitly covered in Eurocode 3 Part 1. The weld is non-load carrying, and
hence there are some similarities with the detail shown in Table 9.8.4 (3). However, the
"transverse attachment" is a load carrying plate, hence the detail is not fully appropriate. Tests
have indicated a somewhat lower category (50) is reasonable.
d. This detail is equivalent to the standard one for cracking in the main plate at the end of a fillet
welded lap joint (Table 9.8.5 (3)). Note the specified rule for the calculation of the stress in the
main plate.
e. This detail is equivalent to the standard one for cracking in the lap plates in a fillet welded lap
joint (Table 9.8.5 (4)). Note that the weld termination should be held back at least 10mm from
the plate edge, and that shear cracking in the weld should also be checked according to Table
9.8.5 (7).
f. As for Figure 3, detail g.

230

g. Whilst this detail belongs in the relatively high category of 140 for a machine gas cut edge
with all edge discontinuities removed (Table 9.8.1(4)), the stresses should be calculated using
the appropriate stress concentration factor for the radius which is used.
h. This is the standard category for web stiffeners where the thickness of the stiffener does not
exceed 12mm and the welds do not come within 10mm of a plate edge (Table 9.8.4 (4) and (5)).
i. Whilst this detail is not explicitly covered in Eurocode 3 Part 1, it shows a number of
similarities to the "wide cover plate" detail of Table 9.8.5(5)). It is clear that a low category is
appropriate, and 45* is proposed.
4. CONCLUDING SUMMARY

The classification for fatigue of all the details of a practical structure can cause
considerable problems, even to an experienced designer. It is hoped that the typical case
study presented in this lecture will help designers in this task, particularly in cases
where it is not immediately obvious into which category a particular detail falls, or
where specific requirements are necessary to ensure compliance with a category.

It is not suggested that all critical details need a full fatigue calculation; frequently low
category details are very lowly stressed and hence of no importance. However, the
classification gives a designer some insight into which details should be avoided if
stress ranges are high or, if unavoidable, gives a means of calculating the endurance.

Finally, it is worth pointing out that work on classifying details is far from complete.
Some of those given in Eurocode 3 Part 1 may well be reclassified (although probably
not by more than one category up or down) and some may be deleted, or additional
details added, when Eurocode 3 Part 2 is published. A designer must, therefore, keep
abreast of latest developments.

5. ADDITIONAL READING
1. Eurocode 3: "Design of Steel Structures" ENV 1993-1-1: Part 1.1, General Rules and
Rules for Buildings, CEN, 1992.

231

Lecture 12.10: Basics of Fracture

Mechanics
OBJECTIVE/SCOPE:
To introduce the basic concepts of linear elastic fracture mechanics.
PREREQUISITES
None.
RELATED LECTURES
Lecture 12.11: Stress Analysis of Cracked Bodies
Lecture 12.12: Determination of Stress Intensity Factors
Lecture 12.13: Fracture Mechanics Applied to Fatigue
Lecture 12.15: Fracture Mechanics Applied to Fitness for Purpose
SUMMARY
The lecture describes the origins of fracture mechanics treatments based on strain energy
concepts and the link to modern treatments based on crack tip stress analysis and the stress
intensity factor. The effects of finite crack and component geometry are described together with
the effects of small scale yielding and plasticity. The lecture concludes with a brief statement of
the significance of the stress intensity factor.
1. INTRODUCTION
Standard design methods for engineering structures and components under static loading are
usually based on avoiding failure by yielding/plastic collapse or buckling. The derivation of
loading resistance is based on conventional solid mechanics theories of stress analysis.
Conventional design procedures against fatigue failure are based on experimental results for
particular geometric details and materials. None of these procedures are capable of allowing for
the effects of severe stress concentrations or crack-like flaws. The presence of such flaws is
more or less inevitable to some extent in practical fabrications.
The modes of failure which are most affected by the presence of crack-like flaws are fracture
and fatigue. The study of the effects of cracks on local stress and strain fields in the
neighbourhood of the crack tip and the consequent effect on failure is the subject of fracture
mechanics. The application of fracture mechanics methods allows analyses to be carried out to
predict the effects of flaws on failure in a wide range of geometries to give complementary
information to that obtained from experimental testing. For fatigue of welded structures the
performance is significantly affected by the tiny flaws inherent to welding. Fracture mechanics
analyses can be very helpful in predicting the effects of geometrical variations on basic fatigue
behaviour.

232

Fracture mechanics methods are particularly useful in making fitness for purpose assessments of
the effects of flaws, and for helping to decide on inspection procedures for fabricated structures
and acceptance levels for any flaws which may be found during such inspections.
2. BACKGROUND TO MODERN FRACTURE MECHANICS
The origins of modern fracture mechanics go back to the work of A A Griffith [1] in 1920
investigating the strength of glass. Griffith used the linear elastic stress analysis solution for the
stresses around an elliptical hole in a plate subject to uniform tension. He allowed the ellipse to
degenerate to a crack and derived an expression for the energy released when an element of
material at the end of the crack fractured to give incremental extension of the crack. He then
suggested that, if the energy released was greater than the surface tension or cohesive force
energy which had been holding the element together, then the situation was unstable, and
continued unstable crack extension (i.e. fracture) would occur.
Griffith's expression for the change in strain energy from a plate with no crack to a plate with a
crack length of 2a was:
U = -22a2/E

(1)

His expression for the strain energy release rate (now known as crack extension force), for a
crack of length 2a in an infinite plate of unit thickness under uniform tension was given by:

(2)
Griffith suggested that an existing crack would propagate unstably if the strain energy release
rate, G, exceeded the energy required to create new fracture surfaces, 2 da, for crack growth
da at each end of the crack, where is the surface tension for glass. Hence he suggested that
fracture would occur when:
2a/E 2(3)
i.e. at a critical value of the strain energy release rate G c. Using this approach Griffith was able
to explain that the reason that the observed strength of glass was much lower than theoretical
estimates linked to the modulus was due to the presence of inherent tiny crack-like flaws in the
material.
Irwin [2], and Orowan [3,4], extended original concepts of energy to create new surfaces, to
include work of plastic deformation prior to fracture, provided the disturbance to the overall
elastic stress field was small.

233

Irwin, used classical stress analysis methods (see Lecture 12.12) to investigate the detailed
stress distributions near to the crack tip. Based on the complex stress function approach of
Westergaard, Irwin showed that the elastic stress field in the neighbourhood of the crack tip (see
Figure 1) was given by:

x =

(4)

y =

(5)

xy =

(6)

It should be noted that these stress distributions are inversely proportional to the square root of
distance from the crack tip. At the crack tip itself, (r = 0), the stress distributions predict infinite
stresses, but this is an idealised situation, known as a stress singularity, resulting from the
assumption of elastic behaviour without any limiting failure criterion. On the plane of the crack,
( = 0, y = 0) the shear stress is zero and the direct stress components are given by:

x = y =

(7)

The term
is dependent only on the applied stress and crack size, and defines the
gradient of stress with inverse square root of distance away from the singularity at the crack tip.

234

The term
was defined by Irwin as the stress intensity factor and given the symbol K. It
should be noted that K is not a stress concentration factor, and that K has dimensions and units
of stress x
. Although the definition of stress intensity factor as K=
is the
one generally used for the case of a central crack in an infinite plate subject to remote tension,
there are some papers in the literature where an alternative definition has been adopted without
the , namely K=

, and care must be taken to check which definition is being used in

any particular case. In these notes the Irwin definition of K =

is used throughout.

It is very important to recognise that the stress singularity and stress intensity factor which
dominate the stress field at the crack tip are features of tension loading which arise because
tension forces cannot be transferred across the free surfaces of the crack and are redistributed
around the ends of the crack in a non-uniform manner. When compression loading is applied to
the cracked plate, if the crack surfaces are in contact, forces can be transmitted directly through
the crack so that there is no requirement for redistribution and hence no stress singularity and
the stress intensity factor is zero. This has important consequences when fatigue loading is
applied to a cracked component.
Irwin showed that the stress intensity factor K was directly related to Griffith's crack extension
force (or strain energy release rate) by the following expressions:
K2 = EG (plane stress) (8)
K2 = EG/(1 - 2) (plane strain) (9)
3. EFFECTS OF MODE OF LOADING
The description of the stress intensity factor given above is based on the simple case of an
infinite plate with a central crack of length 2a subject to remote tension stress. This mode of
loading is known as Mode 1 and the stress intensity factor resulting from this loading is strictly
K1. There are two other forms of loading which produce a similar effect of a stress singularity
because forces cannot be transmitted across the free surfaces of a crack. These forms are shear
loadings parallel to the crack surfaces either in the plane of the plate, also known as edge
sliding, (Mode II stress intensity factor K II), or perpendicular to the plane of the plate, also
known as skew sliding or antiplane strain, (Mode III stress intensity factor K III). These three
different forms of loading are shown in Figure 2. In practice in structural components there may
be combinations of the different modes to consider.

235

Another important case is that of a crack subject to internal pressure loading within the crack.
For the notional case of a through-thickness crack in an infinite plate subject to internal pressure
p, the stress intensity factor is given by:
K=p

(10)

4. EFFECTS OF CRACK GEOMETRY

In the above considerations the crack has been a complete separation throughout the thickness
of the plate and its geometry has been defined by the crack length 2a. In practice cracks in
structural components may occur in a number of different forms. These forms are conveniently
summarised for fracture mechanics analysis purposes in three categories, namely, throughthickness, part-thickness surface breaking, and part-thickness embedded cracks. A case of
special importance is that of an elliptical crack embedded in an infinite body and subject to
remote tension stress as shown in Figure 3.

Irwin obtained an analytical solution for the stress distribution in the neighbourhood of the
crack and found that a stress singularity occurred all round the perimeter of the crack front
characterised by the stress intensity factor, but the magnitude of the stress intensity factor varied
around the crack front. Irwin's solution for the variation of K around the crack was as follows:
236

(11)
where a, c, , are as shown in Figure 3, and E() is the elliptical integral:

E() =

(12)

The maximum value of the stress intensity factor occurs at the ends of the minor axis for this
solution under uniform tension loading. The ratio of the height of the crack (2a) to the length of
the crack (2c) is known as the aspect ratio. As this ratio decreases the solution for the elliptical
embedded crack tends to a value K=
, i.e. the same expression as for the central
through-thickness crack of length 2a but with the elliptical crack having a height of 2a. Thus, for
an embedded crack of this shape under tension loading, the dimension of the crack which has
the greatest effect on the stress intensity factor in the height. Once the length is greater than
about ten times the height, further increases in length make little or no difference to the K value
unless they affect the cross-section area. For an embedded circular crack, the Irwin solution for
the elliptical crack reduces to the same as the "penny shaped crack" case of Sneddon, i.e.

K=

(13)

Irwin applied similar arguments to derive the stress intensity factor for semi-elliptical surface
cracks in a semi-infinite body with remote tension stress as shown in Figure 4.

He suggested that this must be effectively one half of the elliptical embedded crack case divided
on a plane of symmetry, but with a free surface correction factor. The result for the semielliptical surface crack is therefore:

(14)
237

Again, as for the embedded elliptical crack case, it is the crack height which has the major effect
on the maximum stress intensity factor. This maximum occurs at the end of the minor axis, i.e.
the deepest point, for tension loading.
It should be noted that under applied bending stresses the variation of the stress intensity factor
around the perimeter of the crack is different from the tension loading case. The maximum
value can then occur at the ends of the crack depending on the aspect ratio.
5. EFFECTS OF FINITE COMPONENT GEOMETRY
The results described above for effects of crack shape are for the case of a body of infinite size.
In practice there are also effects of finite size brought about by the proximity of boundaries or
free surfaces. The effect has already been seen for the case of a semi-elliptical surface crack
compared to the embedded elliptical crack where a free surface correction factor was included
in Equation (14).
An important effect arises when the crack affects the net cross-section area either in the case of
a through-thickness crack in a plate of finite width, or in the case of remaining ligaments
between the crack front and a free surface for part-thickness cracks. The free surface and finite
width corrections apply to all bodies of finite geometry. In general these correction factors can
only be determined by numerical methods or experimental techniques as discussed in later
lectures. There is also an effect of yielding in real materials leading to a further correction factor
for small amounts of plasticity as discussed in the next section of this lecture.
The general form of the stress intensity factor for remote tension stresses applied to partthickness elliptical or semi-elliptical cracks can be written as:

K=

(15)

where:
MD is the finite width and thickness correction factors
MS is the free surface correction factor
MP is a correction for local plasticity at the crack tip (discussed in Lecture 12.12)
MG is a correction for local stress concentration
E () is the elliptic integral dependant on crack shape aspect ratio.
The overall correction factors for finite geometry are dependent on the type of stressing.
Solutions have been obtained by numerical methods for a range of part thickness elliptical crack
shapes, subject to tension stresses m and to bending stresses b. It is usually possible to
approximate actual stress fields by a combination of direct and bending stress components. The
results can be presented as parametric equations or families of curves for coefficients M m and

238

Mb against a/t for different aspect ratios a/c or a/2c, where the expression for the stress intensity
factor K value is given by:

K = (Mm m - Mb b)

(16)

where Q is a crack shape parameter based on the elliptical integral E().

It should be noted that the M m and Mb values vary around the perimeter of the crack, and the
parametric equations include a term for position around the crack front. The most accurate
results published openly at present appear to be those due to Newman and Raju in a series of
papers, although there a number of stress intensity factor handbooks now available.
A further important effect is that of stress concentration regions at which a crack may lie. For
example, fatigue cracks often develop from initial stress concentration regions and grow
through a changing stress field. Two important examples of this kind are cracks at the edge of
holes and cracks at the toe of welded joints. The case of cracks at the edge of a hole was solved
by Bowie and is shown in Figure 5. For cracks which are small compared to the radius of the
hole, the cracks behave as if they are surface cracks in a uniform stress field equal to three times
the remote tension stress since the stress concentration factor at the edge of the hole in the
absence of the crack is three. For cracks which are large compared to the radius of the hole, the
two cracks behave as a single crack of total length equal to the sum of the two individual cracks
plus the diameter of the hole.

In the case of cracks at the toe of welded joints, the stress intensity factor can be expressed by
the use of a magnification factor M x times the result which would be obtained for a crack of the
same geometry subject to the same loading but without the stress concentration effect of the
weld present. This case is discussed further in Lectures 12.13.
A convenient way to express the effects of all the coefficients/correction factors which can
affect the stress intensity factor is to combine them all together into a single term generally
know as the Y factor. This leads to the following overall expression for the stress intensity
factor:
239

K=Y

(17)

6. LOCAL YIELDING EFFECTS AT A CRACK TIP

In real materials used for structural purposes, such as structural steels, the infinite stresses
predicted by elastic theory at a crack tip are relieved by the occurrence of yielding. A first
approximation to the size of the plastic zone at a crack tip is given by finding the distance
ry from the crack tip at which the elastic stress level is equal to the yield strength. This distance
is given by:

ry =

(18)

Limiting the stress at the crack tip to the yield strength means that the load bearing capacity on
the crack plane is changed. This change leads to a redistribution of stresses locally. The effect of
the redistribution is that for limited plasticity, (r y<<a), the real crack with plastic zone is
equivalent to a crack in an elastic material of length 2(a + r y), as shown in Figure 6. The concept
of 'plastic zone corrections' to elastic stress fields at cracks is useful as a limited extension to
linear elastic fracture mechanics, but the effects of any more extensive plasticity have to be
taken into account by different types of analysis such as crack tip opening displacement
(CTOD) or J contour integral methods.

240

For the small scale yielding situation, substituting the modified equivalent crack length 2(a + r y)
for 2a in the stress intensity factor formula for a central crack in an infinite plate gives the
following plastic zone correction formula:

K=

(19)

7. SIGNIFICANCE OF THE STRESS INTENSITY FACTOR

It has been seen that, for components stressed in tension or bending, the stress field near to a
crack tip under elastic conditions is predicted to be a singularity following an inverse square
root relationship with distance from the crack tip, with the strength of this singularity described
by the stress intensity factor. Furthermore in real materials the singularity is relieved by local
yielding and the size of the plastic zone is directly related to the stress intensity factor. Since the
whole of the stress and displacement field near to the crack tip is controlled by the stress
intensity factor it would be reasonable to expect that any modes of failure which depend on
reaching some critical stress, strain or displacement condition at the crack tip will be described
by a critical level of the stress intensity factor. The two modes of failure most affected by the
presence of cracks are fracture and fatigue. It will be seen in later lectures that the stress
intensity factor is a most useful parameter for determining the effects of cracks on these modes
of failure.
8. CONCLUDING SUMMARY

Crack-like flaws are present to some extent in practical fabrications. They have the
greatest effects on fracture and fatigue performance.

Fracture mechanics treatments based on strain energy originated from the work of A A
Griffith on the strength of glass. They are particularly useful for assessing the effects of
flaws.

Modern treatments use crack tip stress analysis and the stress intensity factor. The
effects of mode of loading, crack geometry and component geometry are taken into
account. Local yielding at the crack tip also affects performance and its effects can be
included in the assessment.

The stress intensity factor is a most useful parameter for determining the effects of
cracks on fracture and fatigue.

9. REFERENCES
[1] Griffith, A.A. 'The phenomena of rupture and flow in solids', Phil. Trans. Roy. Soc. 1921
A221, 163.
[2] Irwin, G.R. 'Fracture dynamics', in Fracturing of metals ASM Cleveland, 1948.
[3] Orowan, E. 'Fracture and strength of solids', Rep. Prog. Phys. 1949 12, 185.

241

[4] Orowan, E. 'Energy criteria of fracture', Weld. J. Res. Suppl. 1955 20, 157s.
10. ADDITIONAL READING
1. Simpson, R. 'Brittle Fracture, Chap. 8 - Steel Designers' Manual', Oxford, Blackwell
Scientific Publications, 1992.
2. Meguid, S.A. 'Engineering Fracture Mechanics', Elsevier Applied Science, 1989.
3. Barsom, J.M. 'Fracture Design, Chap. 5.5 Constructional Steel Design: An International
Guide', London, Elsevier Applied Science, 1992.
4. Knott, J.F. 'Fundamentals of Fracture Mechanics', Butterworths, 1973.

242

Lecture 12.11: Stress Analysis of Cracked

Bodies
OBJECTIVE/SCOPE:
This lecture gives an introduction to the methods of analysis of cracked bodies and reviews the
main concepts of fracture mechanics. These concepts are necessary to understand the design
considerations for fracture assessment under static loading and fatigue under repeated loading as
specified in Annex C and Chapter 9 of Eurocode 3, [1].
PREREQUISITES:
An understanding of plasticity [2,3].
RELATED LECTURES:
Lecture 12.10: Basics of Fracture Mechanics
Lecture 12.12: Determination of Stress Intensity Factors
Lecture 12.14: Fracture Mechanics Structural Engineering Applications
Lecture 12.15: Fracture Mechanics Applied to Fitness for Purpose
SUMMARY:
The lecture begins by outlining the primary basis, within the framework of linear-elastic fracture
mechanics, for determining the stress field in the local region ahead of a crack tip embedded in
a solid body. Then, it briefly discusses models used to assess the local plastic zone existing in
front of the crack tip.
1. INTRODUCTION
All structural elements have discontinuities, either created unintentionally during fabrication or
developed during service conditions under repeated load to which the structure is subjected.
Discontinuities will affect the strength of a structural element and may lead to fracture under a
particular state of stress induced by static or fatigue loading.
Various fracture mechanic quantities have been introduced to give a measure of the severity of
an existing crack in a body, e.g. the stress intensity factor K, the Rice's J integral, the energy
release rate G (which is the amount of energy released during an incremental increase of the
crack area) the crack opening displacement COD. All these quantities are related to one another
under specific assumptions. The most used of these quantities, which characterize the local
stress and strain field in the vicinity of a crack is the stress intensity factor K. This parameter,
fundamentally derived from linear elastic fracture mechanics, is a function of the applied stress
and the geometrical configuration and dimensions of both the crack and the body which
contains the crack.

243

This lecture is an introductory lecture to the two main applications of fracture mechanics, which
will be developed in subsequent lectures. The first application is the design against fracture of
cracked elements subjected to particular temperature and loading conditions. The safe design is
achieved when the stress intensity factor K is less than a critical value K c known as the fracture
toughness. The second and more common application is related to fatigue crack growth
conditions under repeated loading cycles. It is proved experimentally that, for steel structural
material, the rate of crack propagation is a function of the range of stress intensity factor. Slow
stable crack propagation which mainly governs the fatigue life of welded structural elements
takes place for values of K smaller than a critical value K d somewhat different from the
previously defined fracture toughness Kc.
Having in mind these two applications this lecture is mainly concerned with linear fracture
mechanic concepts. The analytical basis in assessing local stress field near the flaw is described,
and then some considerations concerning plasticity at the crack tip are introduced. Finally,
approximate methods for the evaluation of stress intensity factors in more general cases are
given.
2. BASIC SOLUTION FOR A STRESS FIELD IN A PROBLEM OF PLANE
ELASTICITY
The stresses and strains at any point near a crack tip (Figure 1) can be derived from the theory
of elasticity. Stresses and strains within the interior of a solid body subjected to external load
and/or displacement conditions are known to satisfy a set of fundamental differential equations
resulting from equilibrium, compatibility conditions and physical properties of the material
which constitutes the solid body.

The process of solution of a problem of plane elasticity consists of finding a particular

mathematical function of the distribution of stresses (or strains) which fulfils not only these
differential equations, but also satisfies the boundary conditions expressed in term of forces or
displacements specified over the surface of the body. In plane stress problems
(i.e. zz = zx = zy = 0; which is the case of thin plates) or plane strain problems
(i.e. xz = yz = zz = 0; which corresponds to the condition of strains existing at mid244

thickness of a thick plate) the usual method of solving the set of differential equations is to
introduce the so-called Airy stress function.
The solution of a plane elasticity problem is reduced to finding a biharmonic function F
satisfying the conditions expressed previously. It can be demonstrated [4] that, for zero body
forces, the stresses may be derived in the plane elasticity problem from the following relations:
xx = 2F/y2
yy = 2F/x2 (2.1)
xy = 2F/xy
For discontinuous boundary conditions, the classical polynomial method is not suitable if
boundary conditions are to be precisely satisfied. Westergaard and Mushkelishvili have
developed independently general methods for stress function solutions which are well suited to
solve plane problem of elasticity for cracked body. The Westergaard method is more particularly
applicable to crack problems in infinite plate.
2.1 Method of Westergaard
The method proposed by Westergaard [5] systematically expresses the Airy stress function F in
terms of general harmonic functions to a desirable form:
F = f1 + x.f2 + y.f3 (2.2)
where
F is biharmonic when f1, f2, and f3 are harmonic functions.
Westergaard chooses the functions f1, f2, or f3 as the real and imaginary parts of an analytical
function and its derivatives (named the Z functions). Taking into account the particularity of the
problem, such as the symmetry of the stress field, F can be simplified by taking f 2 or f3 = 0.
Assuming Z as the analytical function, its derivatives are:
Z = dZ/dz; Z = dZ /dz; Z'" = dZ /dz (2.3)
If there is a displacement continuity in the y axis direction (Figure 2) and a crack parallel to the
x direction, due to symmetry of the stress field, the Airy stress function as expressed by
Equation (2.2) may be written:
F = Re Z + y . Im Z (2.4)
Re and Im are respectively the real and imaginary parts of the Z functions. The polar coordinate
r and (Figure 2) are used to locate any point in the local region ahead of the crack tip,
therefore
z = iei = x + iy
245

From Equation (2.4) following expressions are derived:

2F/x2 = Re Z" + y . Im Z'"
2F/y2 = Re Z" - y . Im Z'" (2.5)
2F/xy = y . Re Z"
Therefore, taking into account Equations (2.5), the stresses as given by Equation (2.1) are
expressed by:
xx = Re Z" - y . Im Z'"
yy = Re Z" + y . Im Z'" (2.6)
xy = - y . Re Z"
It can be seen, that this type of Airy stress function satisfies certain problems for which
symmetry conditions hold, e.g. for y = 0; xx = yy and xy = 0.
2.2 Definition of z (or its derivatives) for the Case of a Through-Thickness
Crack of Length 2a in an Infinite Plate (Figure 2)
The Z functions, must fulfil the boundary conditions of the problem. The boundary conditions to
be met at the crack tip are:
for y = 0 and - a x a ; yy = xy = 0
246

Assuming that Z" takes the following form:

(2.7)
In the previous equation, let ao, a real constant, be assumed as:
(2.8)
then, from Equation (2.7)
lim Zz0z.ao = (2)

(2.9)

Therefore, from Equation (2.3)

Z
(2z)
Z
2z

(2.10)

Z'" -(2z).1/z
And from Equation (2.6)

(2.11)
Hence, if the Z function is known (or its derivative Z"), the stress field near the crack tip is
given by Equations (2.11)
where (2)lim Zz0z
2.3 Determination of z in the Case of the Griffith Problem
The Griffith problem (1920) is defined as the case of a sharp centred through-thickness crack in
a plate subjected to plane stress perpendicular to the crack at the infinity (Figure 2). In addition
to the boundary conditions previously stated, the Z" function should also verify the following
particular boundary condition at the infinity.
yy = and xy = 0 when z
Translating the Cartesian coordinate system at the centre of the straight front crack, the function
as given by Equation (2.7) takes the following form:

247

(2.12)
for z Z" a1 + a...
taking a1 = and a2 = a3 = ... = 0
the unknown coefficients a1, a2, a3, ... are determined by matching the specified boundary
conditions.
Then yy = xx = ReZ" = a1 = since Im Z'" = 0 (2.13)
and yy = - y Re Z'" = 0
Therefore, from Equation (2.12) and (2.13)
(2.14)
If Z" is developed at the vicinity of z = a, then from Equation (2.9)
/(2) = limza /(z2 - a2) (a) /(2)
Hence = (a)

(2.15)

Equation (2.15) is known as the Griffith reference solution to the straight front crack in-plane
problem. This solution was first established by Irwin (1954).
3 CRACK TIP ELASTIC STRESS FIELD FOR GENERAL CRACKING CONDITIONS
In general, three modes of cracking (mode I, mode II and mode III) exist due to different basic
stress conditions exerted at the crack tip (Figure 3). Within the framework of the linear elastic
fracture mechanic assumption, the general cracking conditions are governed by the
superposition of these three different modes.

248

However, the predominant and most influential mode of cracking refers to mode I named the
opening mode. This mode of cracking governs most of the fatigue crack behaviour encountered
in practice for steel structures.
Applying the same methods developed earlier, similar expressions for the stress field derived for
mode I, may be obtained for mode II and mode III. These expressions are given in a schematic
format below; KII and KIII are respectively the stress intensity factors for mode II and mode III.
These stress intensity factors depend only upon loading and geometric configuration. Relations
for the displacement and the strain fields are not given, but they are easily obtained from the
equations of the theory of elasticity in terms of the stresses.
Mode II

249

(3.1)
for plane strain zz = (xx + yy) ; xz = yz = 0
for plane stress zz = xz = yz = 0
Mode III

(3.2)
xx = yy = zz = xy = 0
4. PLASTICITY
The solutions which were derived, for determining the stress field at the crack tip, are based on
linear-elastic theory. Equation (2.11) shows that, when r 0, the stress at the crack tip becomes
infinite; this has no physical meaning. Therefore, at the front of the crack tip there will always
be a localized plastic deformation which will affect the crack behaviour, i.e. the crack growth
rate. The size of this plastic deformation at the front of the crack is discussed below.
4.1 Irwin's Model
At the crack front there is always a localized plastic deformation which will affect the growth of
the crack. The first attempt which was made to predict the size of the plastic zone in the front of
a crack was reported by Irwin. He proposed a very simple model. Referring to the same Griffith
reference case, the elastic stress yy along the x axis at the front of the crack (Figure 4a) using
plane stress theory is given by (see Equation (2.11)):
(4.1)
where is the stress in the y direction at infinity.

250

Let fy be the elastic limit of the steel plate material; the elastic stresses are, therefore, limited by:
(4.2)
from which the size of the plastic zone can be predicted
ry = 1/2 (KI/fy)2 (4.3)
For plane strain theory, ry becomes:
ry = 1/2 [KI (1 - 2 )/fy]2
However, this reasoning is very simple, and is not representative of reality. In setting up the
inequality (Equation (4.2)) it was considered that no stress redistribution due to plastic
deformation takes place within the plate. Consequently the plastic deformation size along the x
axis cannot be the one represented on Figure 4a. The size of the plastic deformation zone,
allowing for stress redistribution, must be greater. Suppose that the elastic stress curve has been
shifted to the right by a quantity 00 (the stress curve being translated by the vector (00 )
(Figure 4b). The stress distributionyy along the x axis is expressed by:
yy = fy for x rp
and (4.4)

251

for x > rp
In order to estimate the plastic deformation zone allowing for stresses redistribution, Irwin made
the assumption that the total elastic area GBE, in other words the elastic energy, is equal to the
elasto-plastic energy represented by the area ACF.
From consideration of the geometry the following equality may be written:
(4.5)
Therefore

Hence
(4.6)
Taking into account Equation (4.3), Equation (4.6) becomes:

Figure 4c gives a schematic representation of the plastic zone, and the associated effective crack
length (a + ry), the size of the plastic zone being equal to 2 ry.
4.2 Plastic Zone Contour from Von Mises and Tresca Criteria (Figure 5)

252

If the simple model of Irwin empirically takes into account the stress redistribution, it ignores
completely the presence of the other stresses xx and yy.
4.2.1 Von Mises Criterion
Designating 1, 2 and 3 the principal stresses, the von Mises criterion is
(1-2)2 + (2-3)2 + (3-1)2 = 2 fy2 (4.7)
in which

1 = 1/2 (xx + yy) + 1/2

2 = 1/2 (xx + yy) + 1/2

(4.8)

and 3 = 0 (plane stress)

3 = (1 + 2) (plane stress)
Substituting Equation (2.11) for mode I in Equation (4.8) and in Equation (4.7) gives:

for plane stress

rp = (1/2) (KI/fy)2 cos2 /2 (1+3 sin2 /2) (4.9)

253

for plane strain

rp = (1/2) (KI/fy)2 [(1-2)2 + 3 sin2 /2] (4.10)

4.2.2 Tresca Criterion
The Tresca criterion is based on the maximum shear max and is written as:
max = 1/2 fy (4.11)
where

for plane stress max = 1/2 1

for plane strain max = 1/2 |(1-3)/2 ; (1-2)/2|

(4.12)

A similar derivation to that above leads to the following equations:

for plane stress

rp = (1/2) (KI/fy)2 cos2 /2 (1+ sin /2)2

(4.13)

for plane strain

rp = (1/2) (KI/fy)2 cos2 /2 max [1, (1-2 +sin /2)2]

(4.14)

5. FRONT CRACK YIELDING MODEL OF D S DUGDALE (1960) AND BAREN

BLATT (1962)
5.1 Plastic Zone Dimension
This yielding model ahead of a front crack due to D S Dugdale (1960) deserves to be presented
since it introduces the crack opening displacement concept (COD). Consider again the Griffith
problem, and suppose that the yielding spreads at both crack lips on a distance w (Figure 6a).
This short yielded distance w may be easily determined, considering the following superposition
of stress fields.

254

Support that the crack of length 2(a + w) is subjected to a tensile stress applied at the infinity.
This is referred in Figure 6 as stress condition 1. A second stress condition may be considered
where two distributed compressive field stresses are opening both crack lips over the distance w,
referred in Figure 6c as stress condition 2. Superposing the stress condition 2 with the stress
condition shown in Figure 6a, gives stress condition 1. The stress intensity factor of a throughcrack length 2(a + w) in an infinite body is:
(5.1)
In the case of stress condition 2, the stress intensity factor may be easily obtained from the
Green function of the solution relative to a pair of concentrated opening forces acting
symmetrically on the two lips of a crack (cf. Lecture 12.12). Therefore, the solution is:
255

(5.2)
The sum of the stress intensity factors K I1 and KI2 must be equal to zero. Then the plastic zone
dimension w is:
w = -a + a / cos |/2fy|

(5.3)

By developing cos |/2fy| in series when << fy, and taking the first term of the series
gives:
cos |/2fy| {1 - |/fy|2./8}

(5.4)

5.2 Crack Tip Opening Displacement

The dimension of the plastic zone may be obtained from Equation (5.2) taking into account
Equations (5.3) and (5.4), then
w = /8 |K1/fy|2 w = /8 {K1/fy}2

(5.5)

This value is very near to the solution obtained by Irwin (see Equation (4.3)). When the ratio
( /fy) becomes greater the difference between the two solutions increases.
Burdekin and Stone [6], according to the approach outlined in paragraph 2.1, calculated the
displacement at the end of the real crack (i.e. for x = a) and found:

(5.6)
In the concept of the crack opening displacement approach, the crack propagates
when reaches its critical opening displacement c which is a characteristic of the material.
From Equation (5.6) the critical stress corresponding to the Griffith problem is obtained
knowing the critical opening displacement. Developing cos {/2fy} in series, and keeping
the first term Equation, (5.6) reduces to

(5.7)
The COD definition, being the crack opening displacement at the interface of the elastic and
plastic zones, is convenient for calculation purpose. However, it is not very easy to measure
experimentally the value of the crack opening displacement.
6. CONCLUDING SUMMARY
For a stress field in a problem of plane elasticity:-

256

Stresses near the crack tip may be calculated in first approximation from:

(2)

lim Sz Z''z
A proper Z function must be selected, which satisfies the boundary conditions.

The stress field of the Griffith reference problem is expressed by:-

2 = (a)

The elastic stresses distribution in the vicinity of the crack tip depends only on r and ,
while their magnitude at any given point defined by (r, ) depends only on KI.

Therefore, the distribution of stresses in a cracked body is an invariant with respect to the
loading and geometries of the crack and the body. However, the magnitude of these stresses
depends upon these two parameters which are taken globally in the K I value called the stress
intensity factor in mode I. KI depends on the external loading, the overall geometry of the crack
body, and on the crack geometry (dimensions and shape).

The stress intensity factor KI is a basic concept of fracture mechanics.

This factor should not be confused with the geometric stress concentration factor K t which is,
for a particular notch, the ratio of the maximum stress to the nominal stress. K t is, therefore, a
purely conventional coefficient, giving an indication of the stress concentration at a notch for
particular conditions of geometry and loading.
In reality there will always be localised plastic deformation at the front of the crack tip which
affects crack behaviour:

In general plastic zone size at crack tips can be assessed and its effect can be taken into
account through a correction term of the crack length (r y) as suggested by Irwin (Figure
4c).

The crack tip plastic zones are smaller in-plane strain than in-plane stress hypothesis,
and the Tresca criterion gives larger plastic zone than Von Mises criterion. These
remarks justify the recommendations concerning the size of the specimen for the
determination of KIC.

Figure 7 represents in thick plates the plastic zone configuration which can be expected
from the previous finding since the surface is in a plane stress condition and the midthickness is in a plane strain condition.

However the plastic zone, for normal constructional steel, is rather small compared to
the dimensions of the structural part and of the crack size. In practical high cycle fatigue
problems, this plastic zone is expected to be small due to triaxiality of the crack tip
stress field and cracking occurring under quasi-elastic stress range conditions.

257

The dimensions of the plastic zone ahead of a crack may be found using the yielding model of
D S Dugdale:

When the tensile stress is small compared to the yield stress of the material ( << fy),
the plastic zone dimension found by the Dugdale model is very close to the Irwn
solution, the difference being a factor 4/.

The Dugdale model of the plastic zone leads to the concept of crack opening
displacement which may be theoretically calculated if the crack size and the tensile
stress away from the crack are known. However, experimental measures of the crack
opening displacement are rather difficult to obtain accurately since they depend on the
precise location where the measurements are made.

The relationships between COD, strains and crack size are not readily known for most
geometries; the COD method has the advantage that it takes into account the assessment
of plasticity at the crack tip. However the COD method has not yet reached the same
level of recognition in engineering practice as the K based method.

7. REFERENCES
[1] Eurocode 3: "Design of Steel Structures": ENV 1993-1-1: Part 1.1, General principles and
rules for buildings, CEN, 1992.
[2] Timoshenko S, Goodier N.J., "Theory of Plasticity", McGraw Hill Book
Company, 2nd Edition, New York 1951
[3] Germain P, "Mcanique des Milieux Continus", Masson et Cie, Paris, 1962.
[4] Westergaard H.M., "Theory of Elasticity and Plasticity", Dover Publications, inc. New York,
1964
[5] Francois D., Joly L. (Ed.), La rupture des mtaux, Masson et Cie, 1972

258

[6] Burdekin E.M., Stone D.E.W., The crack opening displacement approach to fracture
mechanics in yielding materials Journal of Strain Analysis, vol. 1, n o 2, 1966.

259

Lecture 12.12: Determination of Stress

Intensity Factors
OBJECTIVE/SCOPE:
To describe the basic principles of the different methods of determining stress intensity factors
for different geometries of component.
PRE-REQUISITES
Lecture 12.10 : Basics of Fracture Mechanics
RELATED LECTURES
Lecture 12.11 : Stress Analysis of Cracked Bodies
Lecture 12.14 : Fracture Mechanics Structural Engineering Applications
Lecture 12.15 : Fracture Mechanics Applied to Fitness for Purpose
SUMMARY
The lecture starts with a reminder of the basic definition of the stress intensity factor as the
parameter controlling crack tip stress, displacement and energy conditions for linear elastic
materials. Some classical cases for which analytical solutions for the stress intensity factor are
available are described together with the basis for combining solutions by superposition. The
two main numerical analysis methods for determining stress intensity factors are then described,
namely weight function methods and finite element methods including some details of the way
in which the stress intensity factor is derived from the basic analysis results. Treatments for
stress concentration regions are also described with particular reference to welded joints.
1. INTRODUCTION
In Lecture 12.10: Basics of Fracture Mechanics, the concept of the stress intensity factor (K)
was introduced as the parameter which defines the stress and strain fields at the tip of a crack,
i.e. on the plane of the crack ahead of the crack tip the stress normal to the crack is given by:
y = K / (2r)

(1)

where r is a small distance ahead of the crack tip on the crack plane.
The stress intensity factor K also controls the opening displacements of the crack faces within
the crack close to the tip for elastic conditions as follows:
d = (4/E) K(2r)

(2)

where r is a small distance back from the crack tip within the crack.

260

In addition, the stress intensity factor K is linked to the Griffith strain energy release rate G
(crack extension force) by the relationship:
K2 = E'G

(3)

where E' = E for plane stress

and E' = E/(1-2) for plane strain
For non-linear materials a further important crack tip parameter is the J contour integral, J. As
originally defined by Eshelby and by Rice, J was shown to be a path independent integral of
strain energy density terms taken around the crack tip, with the path starting on the lower face of
the crack and finishing on the upper face as shown in Figure 1.

The definition of J in these terms is as follows:

(4a)
where
W() = Jij dij

261

where W is the strain energy density,

Ti are components of surface tractions moving through displacements du i,. The integration is
carried out around any contour path anti-clockwise from one surface to the other enclosing the
crack tip.
In addition to demonstrating path independence for this integral for two dimensional geometries
with externally applied loading, Rice also showed that J was related to the change in potential
energy with incremental increase in crack length, i.e. J can be thought of as a non-linear version
of G and reduced to G for linear elastic materials. For a linear elastic material:

(4b)
The importance of all of the above observations is that, if methods are available to determine
stress fields ahead of the crack tip, displacement fields within the crack, or strain energy fields
related to G or J, then the stress intensity factor K can be determined from these results.
2. ANALYTICAL SOLUTIONS
A number of classical cases where the stress intensity factor can be determined analytically
were described in Lecture 12.10 on Basics of Fracture Mechanics. These cases included
solutions for a central through-thickness crack, an embedded elliptical crack and a semielliptical surface crack all located in infinite plates and subject to remote tension loading.
Reference was also made to the general form of expressions for the stress intensity factor for
cases subject to remote stresses as either:

(5)
where
MD is the finite width and thickness correction factors (Dimensions correction factor)
MS is the free surface correction factor
MP is the local plasticity at crack tip correction factor
MG is the stress gradient correction factor for local effect
E() is the crack shape correction factor

262

An example of using this expression for a stress concentration region at the end of a welded
cover plate is given in the Appendix to this lecture.

or

(6)

where
Mm and Mb are as defined in Section 5.
Two other cases where the results of classical analytical solutions are relevant to practical
applications are the cases of a multiple array of co-linear cracks subject to remote loading and
of a single crack subject to a pair of splitting forces applied to the crack faces. The multiple
array case is shown in Figure 2, where the cracks are each of length 2a, and the spacing between
their mid-length positions is W. The solution for the stress field for this case, where the cracks
are subject to remote tension stress in an infinite plate, can be obtained from the complex
stress function approach of Westergaard used by Irwin to give the stress intensity factor as
follows:

(7)
This case of multiple cracks can be used to investigate the effects of interactions between
adjacent cracks and of finite width of plates. It can be seen that when the ratio of a/W is small,
this expression reduces to the same result as that for a single crack in an infinite plate, i.e. there
is no interaction between the cracks. As a/W increases so the stress intensity factors increase
above the corresponding value for single cracks of the same size. It will be left to the student to
check that when the spacing between the tips of the adjacent cracks is equal to the crack length,
i.e. W-2a = 2a, the stress intensity factor is increased by about 13% compared to the case of the
single crack of the same length. This result is the basis for the recommendation in a number of
defect acceptance codes that cracks may be treated as separate provided the ligament between
263

them is at least equal to the adjacent crack size. The spacing W may also be considered as
equivalent to the width of a series of finite plates each containing a single crack. Hence as the
ratio a/W increases, the increase in K represents the effect of net section stresses and of the
approaching edge of the plate. This interpretation is not strictly valid, however, as the stress
distribution on the line of symmetry between the cracks does not reproduce fully the free edge
condition. It does give reasonably good answers for ratios of a/W up to about 0,6.
The case of a single crack subject to a pair of splitting forces on the crack line is shown in
Figure 3.

Again the solution can be obtained by use of the Westergaard complex stress functions and the
resulting stress intensity factor is as follows:

(8)
This result can be used to represent the effects of internal pressure inside a crack by integrating
the splitting forces p applied at positions of b from 0 to a. As indicated in Lecture 12.10 the
stress intensity factor for the internal pressure case is K =
of remotely applied tension.

, i.e. the same as for the case

This is one example of a more general result which forms the basis of the weight function
approach for determination of stress intensity factors.
3. BUECKNER'S PRINCIPLE AND WEIGHT FUNCTIONS
If a cracked body subject to external loading or prescribed displacements at the boundary has
forces applied to the crack surfaces to close the crack together, these forces must be equivalent
to the stress distribution in an uncracked body of the same geometry subject to the same
external loading. In the condition when the crack is closed together there is effectively no crack
present and the stress intensity factor is zero. This can be regarded as equivalent to superposing
the case of the cracked body with external loading minus the case of the same geometry of
cracked body subject to the reverse of the crack surface loading necessary to close the crack
264

together again. Since the total resultant stress intensity factor is zero, the stress intensity factor
for the crack body subject to external loading must be the same as that for the same body
subject to crack surface loading of the same distribution as would be present in the uncracked
body over the region of the crack position. For many cases the cracked body with remote
loading can be replaced by the same geometry with crack surface loading only to determine the
stress intensity factor. This principle was first expounded by H F Bueckner and is known as
Bueckner's principle.
When a pair of point forces P are applied to the opposite surfaces of a crack, they produce a
stress intensity factor K at the crack tip as described for the case of two balanced pairs of
splitting forces on a crack in an infinite plate in Equation (8). A further example is the case of an
edge crack in a semi-infinite plate subject to concentrated forces P as shown in Figure 4.

The stress intensity factor for this case is given by:

(9)
where F(x/a) is a tabulated function given by Hartranft and Sih [1].
A weight function is a function which gives the ratio of (the stress intensity factor at a crack tip
due to the application of a stress to an element of area dA on the crack surface) to (the stress
itself).
Equation (9) may be used as a fundamental Green's function for generating solutions for stress
intensity factors in crack problems involving an arbitrary distribution of tractions on the crack
surfaces as shown in Figure 4b. Integrating Equation (9) from x=0 to x=a, the stress intensity
factor for the edge crack subject to an arbitrary stress distribution p on the crack surfaces gives:

(10)
This can be turned into a general summation for a two dimensional weight function expression
in the form developed by Albrecht [2] as follows:
265

A more general weight function expression is given by the 'O' integral weight function
developed by Core and Burns [3], as follows:

(12)
where
is the stress applied to an element of crack surface area dA
I is the distance from this point to the position where the K value is required
{} represents the sum around the crack perimeter s of elemental lengths divided by distance to
the point of stress application squared as shown in Figure 5.

Evaluation of the 'O' integral is carried out by summing up the effects of different stress levels
according to the uncracked stress distribution applied over the whole crack surface divided up
into a network of elements.
It will be seen that the expression given by Equation (12) involves two singularities, namely
when 1=0 and when p=0. The evaluation of the numerical integration is carried out by dividing
the crack area into three zones as shown in Figure 6. Zone A is a single element immediately at
the position on the crack front where K is to be determined. Zone B is a series of elements
around the remainder of the crack front perimeter. Zone C covers the remainder of the interior
of the crack surface area. Careful attention is required to element sizes and aspect ratios to
achieve accurate results. The 'O' integral formulation applies strictly to an embedded crack of
arbitrary shape in an infinite body, but it is possible to apply correction factors for the effects of
free surface and finite geometry effects.

266

Thus, one general method for determining stress intensity factors is to determine firstly the
stress distribution in an uncracked body over the region where a crack is to be considered, e.g.
by finite element analysis, and to use weight function approaches to determine the K value for
any required crack geometry. Although the approach requires repeated use of the numerical
integration of the weight function for a series of cracks as would be required to be considered
during fatigue crack growth, it only requires one finite element analysis for the uncracked body
and is generally an efficient overall process.
4. FINITE ELEMENT ANALYSIS OF CRACKED BODIES
When finite element analysis is carried out, the first steps are to decide on the type of element to
be used and on the mesh to be used to discretise the volume of the body concerned. Different
types of element use different interpolating functions for displacements throughout each
element. Each type has an inherent ability to deal with stress gradients within the element. For
example, the simplest types of element are the constant stress or constant strain triangular
elements. If an attempt is made to use these elements to analyse a body with steep stress
gradients, a very fine mesh consisting of many elements must be used. This will be very
inefficient on computing time. This problem can be reduced by using higher order elements
which have the ability to represent linear or parabolic stress distributions within them. For
example, 8 noded or 12 noded solid elements are often used when accurate stress analysis
results are required from two dimensional complex geometries.
For example, 8 noded or 20 noded solid elements are often used when accurate stress analysis
results are required from three dimensional complex geometries.
As indicated previously, the stress distribution at a crack tip subject to tension loading in elastic
material, shows a singularity with infinite stresses. The stress intensity factor describes the way
in which the stress declines from the singularity and clearly involves very steep stress gradients.
For normal finite element analysis, it would appear at first sight that it would be very difficult to
obtain anything like a realistic representation of elastic crack tip stress fields unless an
inordinately fine mesh is used involving a very large number of elements and associated costs.
This problem can be overcome by use of special crack tip elements [4]. For two dimensional
analyses (plane stress or plane strain) there are elements of degenerated triangular 20 moded
type or 52 moded type based on a 12 isoparametric quadrilateral finite element. These elements
267

are arranged into a fan focused at the crack tip by degenerating them into triangular elements by
bringing two corners together at the same co-ordinates. In addition, the analysis moves the midpoints of the sides of the elements adjacent to the crack tip to the quarter points, as shown in
Figure 7. The effect is to create an inverse square root singularity of stress ahead of the crack tip
without having to go to a very fine mesh. It then remains to determine the value of the stress
intensity factor from the results of this arrangement.

It has been pointed out that the stress intensity factor controls the stress ahead of the crack tip,
the displacement with the crack just behind the tip and the strain energy fields around the crack
tip. Because the stress gradients are so steep away from the singularity, the determination of K
from stresses ahead of the crack is not very accurate. The displacement field within the crack
has much more gradual variations and can be used to determine K quite accurately. The usual
method is to determine the displacements at the first and second nodes back from the crack tip
and use the formula of Equation (2) to derive K. A number of standard computer programs have
been written which include special crack tip elements and the facility to carry out a J contour
integral determination using any chosen contour around the crack tip. The normal procedure is
to determine J on three paths around the crack, to check path independence, and to determine K
from J using Equation (3). A variation on this procedure is the Virtual Crack Extension method
developed by Parks. In this case the energy change associated with a small (virtual) increase in
length of the crack is determined and this is equivalent to J (or G for linear elastic materials).
The advantage of the strain energy or contour integral approaches is that they are evaluated at
some distance from the crack tip itself and hence in regions where the stress and strain gradients
are less severe.
The disadvantage of using finite element analysis of the crack body is that a new analysis of the
whole body is required for each crack geometry to be considered. Hence, when one is
considering fatigue crack growth, a series of results for progressively increasing crack sizes is
required. Nevertheless, with modern computing power and facilities, there are many results for
stress intensity factors obtained by finite element analysis of the complete cracked body.
5. REFERENCE SOLUTIONS FROM PARAMETRIC EQUATIONS

268

A number of handbooks have been published which give solutions for stress intensity factors for
a range of practical problems in the form of parametric equations or analytical expressions.
Some of these stress intensity factors are given in Table 1 and others are listed in [1-10].
An extremely useful set of parametric equations has been developed by Raju and Newman [9]
for the case of semi-elliptical surface cracks in flat plates of finite thickness and width, subject
to either tension or bending stresses. These results were derived primarily by an extensive set of
finite element analyses of different crack geometries followed by curve fitting to obtain the
parametric equations. These equations enable the stress intensity factor to be obtained at any
position around the crack front. This is particularly important when it is necessary to know the
K value at the deepest point and at the ends of the crack at the surface to determine crack
growth rates in fatigue analysis and allow for crack shape changes. The results are also very
useful when the overall stress distribution can be divided into membrane and bending
components since the K values due to the separate components can be determined and then
added together to give the total K. The following general expression for the stress intensity
factor is obtained:

(13)
where Mm, Mb are coefficients determined from the Raju and Newman equations dependent on
a/T and a/2c, m, b are membrane and bending components of stress, a is the crack height or
depth, and Q is a crack shape parameter, given to sufficient accuracy by the expressions:

(14)

(15)
When the crack is located in a region of stress concentration there will be an increase in the
stress intensity factor compared to regions away from the stress concentration. The overall
effect depends on the relative size of the crack and the zone of stress concentration. Where the
crack is small compared to the stress concentration zone, it will behave almost as if it is located
in a region of uniform stress of magnitude equal to the stress concentration value. Where the
crack is larger than the zone of stress concentration, it passes through a region of stress gradient
and the final stress intensity factor is less than the result for treating the stress as uniform and
equal to the SCF value. One important example of this which is particularly important in fatigue
of welded structures is the stress intensity factor for cracks at the toes of fillet and T-butt welds
as shown in Figure 8. For this case, a stress intensity magnification factor M t is introduced to
represent the multiplication necessary compared to the result for a crack of the same geometry
subject to the same remote stresses in a flat plate without the weld being present.

269

A significant amount of work has been carried out at the Welding Institute (TWI) and at UMIST
to derive parametric equations for M t. The TWI results which are included also in PD
6493:1991 [10], have the following form:

(16)
where p and q are constants for a particular geometry but depend on the ratio of weld
attachment length on the surface to the main plate thickness T.
The UMIST results allow also for the effect of weld angle and weld toe radius and take the
general form:

(17)
where the coefficients Ci are given in parametric equations for tension or bending, involving the
attachment length to thickness ratio, the weld angle and the toe radius.
Another important case for analysis of welded joints is the case of the stress intensity factor for
the root of the unpenetrated land of a fillet weld subject to tension loading. The only work
which seems to have been done on this is by Frank and Fisher whose results are included in PD
6493 1991, and by Saket at UMIST. Saket's results are as follows:
K = (2,8817 a/W - 0,074)Ktm for 0,25<a/W<0,45 (18)
where W = tp + 2tw, with tp being the attachment plate thickness, and t w the fillet throat thickness
on one side of the plate, (two welds being assumed), and K tm is given by:

(19)
where p is the remote tension stress in the attachment plate.

270

A summary of some cases of stress concentration and gradient factors for cracks at welded
details is given in Table 2.
6. PLASTICITY EFFECTS
As indicated in Lecture 12.10, the effect of plasticity is to cause a real crack to have a stress
intensity factor equivalent to a slightly longer notional crack in elastic material. The effect
depends on the size of the plastic zone at the tip of the real crack. Under plane stress conditions
the plastic zone is larger than under plane strain conditions where constraint effects inhibit
yielding until higher stresses are reached.
Under compression loading, if the crack faces are forced into contact, the singularity effect is
lost and the stress intensity factor is zero, since forces can be transmitted in bearing and there is
no need for them to be diverted around the end of the crack. One of the significant effects of
plasticity in fatigue is on the occurrence of crack closure behaviour when the crack faces come
into contact and compression stress is then transmitted in bearing. Plasticity allows the crack tip
to be stretched open so that on unloading the original crack surfaces would remain apart. As the
crack propagates through the plastic zone under fatigue loading, however, the stretched material
of the plastic zone forms a wake down the faces of the crack. These regions fill the gap on
unloading despite the stretching at the tip. The net result is that the crack closes on itself at some
stage during the unloading part of a fatigue cycle and does not re-open until the same load is
reached on re-loading. Since there is no stress intensity factor when the crack is closed, fatigue
damage only occurs when the crack faces separate again. This leads to the concept of an
effective stress intensity factor given by:
Keff = U.K (20)
where U is a factor dependant on the stress ratio R and the yield strength through crack closure
effects.
7. CONCLUDING SUMMARY

In many cases, stress intensity factors can be estimated from known solutions, or
parametric equations in published handbooks with adjustment or correction factors as
required.

Where it is necessary to determine stress intensity factors without prior information this
can be done either using weight function methods or using finite element analysis of the
cracked body.

For the weight function method it is necessary to have available the results of stress
analysis of the uncracked body and to apply the stress distribution for this case as
tractions on the crack surfaces, summing up their effects using the appropriate weight
function to give the stress intensity factor at any required position.

For the finite element analysis of the cracked body, it is preferable to use special crack
tip elements to deal with the stress singularity at the tip and to use methods for
estimating the stress intensity factor based on remote strain energy density estimates, or
on displacements within the crack rather than the steep stress gradients ahead of the
crack.
271

The Mt method for dealing with stress intensity factors at weld toes has been described
and attention drawn to the need to have available solutions for at least the deepest point
and the ends of the crack when fatigue crack propagation studies are undertaken to
enable account to be taken of changes in the shape of the crack as it grows.

8. REFERENCES
[1] Sih G.C. Handbook of stress intensity factors, Lehigh University, Bethlehem, Pennsylvania,
USA, 1973.
[2] Albrecht P. and Yamada K. Rapid calculation of stress intensity factors, Journal of the Struct.
Div., ASCE, Proc. 12742, Vol 103, No. St.2, 1977.
[3] Core M. and Burns D.J. Estimation of stress intensity factors for irregular cracks subject to
arbitrary normal stress fields, Proc. 4th International Conference on Pressure Vessel Technology,
I.Mech.E., London, Vol. 1, pp 139-147, 1980.
[4] Gifford L.N., Hilton P.D. Stress Intensity Factors by Enriched Finite Elements Engineering
Fracture Mechanics, Vol. 10, pp 485-496, 1978.
[5] Tada H, Paris P and Irwin G.R. The stress analysis of cracks handbook, Del Research
Corporation, Hellertown, Pa., USA, 1973.
[6] Rooke D.P. and Cartwright D.J. Compendium of stress intensity factors, HMSO, London,
1976.
[7] Rooke D.P., Baratte F.I. and Cartwright D.J. Simple methods of determining stress intensity
factors, Engineering Fracture Mechanics, Vol. 14, 1981.
[8] Labbens R., Pellisier-Tanon A. and Heliot J. Practical methods for calculating stress
intensity factors through weight functions. Mechanics of crack growth, ASTM STP 590, 1976.
[9] Raju I.S. and Newman J.C. Stress intensity factors for a wide range of semi-elliptical surface
cracks in finite thickness plates. Eng. Fract. Mech. Vol. 11, pp 817-829, 1979.
[10] PD 6493, Guidance on methods for assessing the acceptability of flaws in fusion welded
structures. BSI London, 2nd edition, 1991.
APPENDIX
NUMERICAL EXAMPLE
Estimate the stress intensity factor related to a semi-elliptical crack shape at the toe of a cover
plate end weld (Figure 9)

272

Numerical Data

crack depth, a = 3mm

crack width, c = 8mm
weld flank angle = 45
weld leg = 15mm
flange thickness = 40mm
flange width = 600mm
cover plate thickness = 25mm
nominal tension stress max = 300 N/mm2

Calculation
(a) Stress Gradient

kt =
(See Table 2)
(b) Stress Gradient Correction Factor
FG =

(See Table 2)

(c) Free Surface Correction Factor

273

Fs =
(d) Finite Dimension Correction Factor
FD = 1
(e) Free Surface Correction Factor

Fs =
(f) Local Plasticity at Crack Tip
Fp = ^ (no correction for local plasticity considered)
Ki =
=
=

Table 1 - Some Solutions of Stress Intensity Factors

TWO DIMENSIONAL EDGE CRACK PROBLEMS

See Fig 2-1

See Fig 2-2

274

See Fig 2-3

Fig T1-1

Fig T1-2

275

Fig T1-3

Fig T1-4

Fig T1-5

Fig T1-6

276

Table 2 - Stress concentration and stress gradient correction factors

CASE

Transverse butt
weld

STRESS CONCENTRATION FACTOR kt

FG STRESS GRADIENT FACTO

Conditions: =180 - 135

circular arc overfill

See Fig T2-1

where q = log (11,584 - 0,0588 )

Transverse
stiffener weld
(non-loadcarrying fillet
welds)

where a = a/t d = 0,3602 q = 0,2487

See Fig T2-2

Gusset plate

See Fig T2-3

where a = a/t d = 1,158 q = 0,6051

Cover plate

See Fig T2-4

where a = a/t d = 0,1473 q = 0,4398

277

Transverse
load-carrying
fillet welds

Conditions:
crack at the weld root

See Fig T2-5

Note: for crack at weld toe, see case

Fig T2-1

278

Fig T2-2

Fig T2-3

Fig T2-4

Fig T2-5

279

Lecture 12.13: Fracture Mechanics

Applied to Fatigue
OBJECTIVE/SCOPE:
To summarize the main factors affecting the fatigue strength, as opposed to the static strength,
of welded joints and to illustrate the method of carrying out a fatigue check.
PREREQUISITES
Lecture 12.10: Basics of Fracture Mechanics
Lecture 12.11: Stress Analysis of Cracked Bodies
Lecture 12.12: Determination of Stress Intensity Factors
RELATED LECTURES
Lecture 12.14: Fracture Mechanics: Structural Engineering Applications
Lecture 12.15: Fracture Mechanics Applied to Fitness for Purpose
SUMMARY
After separating out initiation and propagation aspects of fatigue behaviour the lecture
introduces the Paris law which describes the rate of fatigue crack propagation in terms of stress
intensity factor range. Integration of the Paris law to give life for crack growth from initial to
final sizes is described for constant amplitude loading and then extended to variable amplitude
loading to show consistency with Miners law. The effects of threshold behaviour, residual
stresses and effective stress intensity factor in crack closure behaviour are described.
1. INTRODUCTION
When a component or structural member is subject to fluctuations in stress, it may lead to the
development of fatigue cracks. Fatigue cracks extend slowly, generally with a very small
increment of crack growth occurring with each cycle, and with little or no evidence of plastic
deformation. The cracks can continue to grow until they cause complete failure of the
component, member or structure by fast fracture, plastic collapse or other mode which prevents
service duties being performed.
In laboratory tests on smooth polished specimens, each stress fluctuations causes a movement of
dislocations until some slip planes lock and microscopic slip protrusions occur. Some authorities
describe the formation of persistent slip bands as Stage 1, and the development of protrusions
and minute fissures as Stage 2. The further development of these fissures into a macroscopic
crack and its subsequent propagation is Stage 3.
The stages prior to the formation of a macroscopic crack represent the initiation life of the
component whilst the remaining life after formation of such a crack is the propagation life. Thus
the overall fatigue life of a component in general is as follows:
280

Total life = Initiation life + Propagation life

For cases where initial cracks are present the initiation life disappears and the whole life is
occupied by the propagation of the crack from the initial size to a final size which is determined
either by instantaneous fracture or by plastic collapse of the remaining cross-section. In some
applications the presence of initial crack-like flaws is inherent to the manufacturing process. A
particularly important case where this applies is in welded structures where tiny slag intrusions
of the order of 0,1 to 0,4 mm depth occur in the partially melted region at the weld toe. The
fatigue behaviour of welded joints is therefore dominated by the propagation life.
As indicated in Lecture 12.11 the topic of fracture mechanics is concerned with analysis of
crack tip stress fields and their effects on failure mechanisms. The application of fracture
mechanics to fatigue behaviour is concerned with the propagation life since it is only during this
stage that a macroscopic crack is present. Thus for situations where no initial flaw is present,
fracture mechanics analyses can be used to estimate the propagation part of the total life from
assumed initial to final flaw size. It is then necessary also to estimate the initiation life and add
this to the propagation life to get an estimate of the total life. For welded structures however the
propagation life gives a good estimate of the total life provided realistic assumptions are made
about initial and final crack sizes.
2. FRACTURE MECHANICS CRACK PROPAGATION BEHAVIOUR
The proposal that the rate of crack propagation per cycle should be controlled by the range of
the stress intensity factor for the cycle was first made by P C Paris as part of his research work.
The general relationship now known universally as the Paris Law is as follows:

= C(K)m (1)
where:
da/dN is the rate of crack growth per cycle
K is the range of stress intensity factor at the crack tip
C and m are material constants (although not necessarily truly constant)
Experimental work to investigate the relationship between fatigue crack growth rate and range
of stress intensity factor can be presented on a graph of da/dN against K. In general, such a
graph on log-log scales shows three regions. At the bottom end there is a threshold region
of K below which cracks do not propagate. This threshold value, Km, is dependent on both
mean stress and environmental conditions. At the top end of the graph the rate of fatigue crack
propagation may be increased if the upper end of the applied stress intensity factor range
approaches the material fracture toughness. In between these regions the graph is generally
linear on logarithmic scales. By taking logarithms of both sides of Equation (1) it can be seen
that it predicts that log da/dN should be proportional to log K, so that the slope of the straight
line is the constant m, and the position of the line is determined by the constant C. Thus the

281

predictions of the Paris law are confirmed by experiment for the central region of behaviour as
shown in Figure 1.

There have been various attempts to extend the validity of the Paris law to cover the whole
range of behaviour. Some of these attempts will be discussed later. Many valuable results can be
derived using the simple Paris law without the complications of the other proposals. These
aspects will be examined first.
The value of the crack propagation law exponent m is found to lie between about 2,6 and 3,6 for
different materials and conditions. A typical value often taken for structural steels is m = 3,0.
The corresponding value of C for crack propagation of steels in air is about 2,0 10-13 in
N/mm-3/2units giving crack propagation rate in mm/cycle. (The equivalent value of C in
MPa

units is 6,32 10-15 for K in the same units, giving da/dN in m/cycle).

3. DETERMINATION OF LIFE UNDER CONSTANT AMPLITUDE

For the simple case of a central crack of length 2a in an infinite plate subject to remote
fluctuating tension stress range , the stress intensity factor range is given by:
K =

(2)

Putting this value into the expression of the Paris law, the following result is obtained:

= C (

)m (3)

This can be re-arranged as a simple differential equation as follows:

282

= (S

)m dN (4)

This expression can be integrated directly, and for the case of m = 3 for example, this gives:

= (S

)3 N (5)

For given initial and final crack sizes and material properties this is equivalent to:
S3 N = Constant (6)
Use of this approach is shown in Example 1 below.
Example 1
Problem A thick plate has an extended length surface crack of height 2 mm perpendicular to the
surface and to fluctuating applied stresses with a range of 100 N/mm 2. Assuming that the Paris
law is valid with C = 2 10-13 and m=3, determine the life for the crack to grow to a height of
10 mm.
Solution Integration of the Paris law as in Equation (5) gives:

= 2 10-13 (1,12 100

)3 N (7)

therefore:
N = 0,782 / [2 10-13 198,53] = 500,000 cycles (8)
It should be noted that the above analysis is only strictly valid for the case of the surface crack
in an infinite plate. In real finite geometries the expression for the stress intensity factor includes
the Y coefficient which may itself be a function of the crack size dimension a. This effects the
integration of the left hand side of Equation (4) which then becomes:

= C (

)m dN (9)

S-N design curves for welded details are effectively of the form of Equation (6), the constant
being different for different geometries (equivalent to different Y values).
For weld toe geometries the magnification of the stress intensity factor which occurs over crack
depths up to 20% of the thickness can be represented by the term M k, where:
K = Mk Yu

(10)
283

and Yu is the appropriate Y value for a crack of the same shape in a plate without stress
concentration effects.
Research at the Welding Institute and at UMIST in the United Kingdom has produced
parametric equations for Mk in terms of weld attachment length-to-thickness ratio, weld angle
and weld toe radius for geometries of the form shown in Figure 2. An approximate expression
for Mk is as follows:

Mk = p

(11)

where p and q are constants dependent on the detailed weld geometry. More complex
expressions for Mk have been derived in the UMIST work which allow for all three of the weld
toe geometric parameters. This work has also shown the need for a further coefficient/correction
factor to allow for the crack shape aspect ratio and for the 'undershoot' effect of the stress
distribution across the thickness which is necessary to compensate for the stress concentration at
the surface as shown in Figure 2.
Provided the overall effect of geometry can be expressed in the form of parametric equations,
the crack propagation integral of Equation (9) can be evaluated numerically in incremental
steps. Furthermore since the stress intensity factor varies around the perimeter of the crack front
of a semi-elliptical surface crack, and the crack growth rate depends on (K)m, it can be seen
that the crack will grow at different rates at different positions on its perimeter. Thus the crack
will change its shape as it grows. Therefore the stress intensity factor has to be re-evaluated
incrementally due to both crack growth and change of shape. A number of researchers have
developed computer programs to evaluate the crack propagation integral progressively for
different geometric applications and have found good agreement with experimental data. The
power of the approach is its ability to evaluate a wide range of geometric effects and to predict
the significance of imperfections and defects on fatigue performance.
284

4. VARIABLE AMPLITUDE LOADING

Sequential block loading
Consider the propagation of a fatigue crack under a sequence of blocks of different levels of
constant amplitude loading as shown in Figure 3.

The crack grows from size a0 to a1 under n1 cycles of stress range 1, from a1 to a2 under
n2 cycles of stress range 2, from a2 to a3 under n3 cycles of stress range 3, etc., up to the
final block of cycles mf at stress range f taking the crack up to its final size af. Crack growth
for each stage will be described by the following equation:

(12)

(13)

(14)

(15)
etc.
285

Examination of these equations shows that in the terms on the left hand side the top limit for
one integral is the same as the bottom limit for the next integral as they represent the crack size
at the change from one block of loading to the next. If all the equations are added together then
the intermediate limits on the left hand side all cancel out, and the following equation results:

(16)
If each of the different stress range blocks had been applied as constant amplitude loading to
grow the crack completely from its initial size a 0 to its final size af, with lives Ni corresponding
to stress ranges i the crack growth equations would be:

= C(
) f3 Nf (16)

)3(13N1 =

C(

)323N2 =

C(

)333N3 =...

C(

If Equation (14) is divided by Equation (15), using the appropriate term on the right hand side
of Equation (15) containing the same i as the successive terms in Equation (14), the result
is the well known linear damage relationship, Miner's Law:

=1
It should be noted that this fracture mechanics version of Miner's Law has been derived on the
assumption that there is no interaction between successive blocks of loading at different stress
ranges and without taking any account of effects of mean stress or stress ratio on the crack
propagation constants for each block of loading. However, in principle, the length of each block
could be reduced to a single cycle, so that the analysis predicts that Miner's Law should hold for
random variable amplitude loading. The fact that in practice significant retardation effects are
observed for occasional overloads and acceleration effects for underloads, and that Miner's Law
damage summation figures differing significantly from 1 are often obtained for different loading
spectra, suggests that some of the underlying assumptions of the fracture mechanics analysis are
not valid. In fact these effects can be taken into account in more sophisticated fracture
mechanics analyses by considering crack closure and plasticity effects.
Example 2
Problem The plate of example 1 with initial extended length crack of height
2 mm is subject to the following combination of stress ranges and cycles under variable
amplitude loading for a period of five years. Assuming that the Paris Law is valid, with
constants C = 2 10-13 and m = 3, determine the final crack height when all cycles have been
applied.
Stress range
286

N/mm2
120
100
80
60
40
20
Solution Integration of Equation (16) gives the following results:

= 1,114 10-12 (1203 102 + 1003 103 + 803 104 - 603 105 + 403 106 +
203 107) 5 (19)
therefore:

= 2 10-13 5,568 1,7189 10" = 0,957 (20)

therefore af = 19,16 mm.
5. THRESHOLD EFFECTS
For constant amplitude loading, threshold effects are straightforward; if the applied stress
intensity factor range is less than the threshold value, the rate of fatigue crack propagation is
zero. When the inherent size of initial flaw or imperfection associated with a particular material
and fabricated geometry is considered, it can be seen that the threshold stress intensity factor
will correspond to a fatigue limit stress level below which cracks will not develop. Thus fracture
mechanics and conventional S-N approaches are again matched.
For variable amplitude loading, the effects of the threshold are more complex. At short crack
lengths the lower stress ranges may be insufficient to make the stress intensity factor exceed the
threshold but the higher stress ranges may be sufficient to cause crack propagation to occur. As
the crack grows longer under the action of the higher stress ranges, so the stress intensity factor
due to the lower stress ranges increases. These lower stress ranges will 'clip in' progressively to
start driving the crack. It is then necessary to calculate the crack length at which each of the
stress ranges 'clips in' and to carry out the integration of crack growth between limits for each of
the these stages with the appropriate set of stress ranges active as shown in Example 3.
Example 3
287

Problem For the plate and loading of Example 2 the threshold stress intensity factor is 100
N/mm-3/2 and the loading is applied for eight years. Determine the crack sizes at which different
stress levels 'clip in', and the resulting final crack size allowing for threshold effects.
Solution To determine the crack sizes at which the different stress levels 'clip in', it is necessary
to use the equation:
Kth =

(21)

This equation gives the following results for the stress ranges concerned, with a threshold value
of 100 N/mm-3/2.
Stress range
N/mm2
120
100
80
60
40
20
From the above table it can be seen that for an initial crack size of 2 mm, all the stress ranges of
40 N/mm2 and above exceed the threshold from the start, but the stress range of 20 N/mm 2 does
not become effective until the crack has grown to a size of 6,34 mm. The crack propagation
integration now has to be divided into two stages, from 2 mm to 6,34 mm, and from 6,34 mm to
the final size as follows:

= 1,11410-12 (1203102 + 1003103 + 803104 + 603105

+ 403106)xy

(22)

where y is the number of years for the crack to grow to 6,34 mm under the stress ranges from 40
N/mm2 upwards. Solution of this equation leads to y = 6,05 years. For the total life of 8 years,
the remaining life for the second stage with all stress ranges driving is 1,95 years. The integrated
crack growth behaviour then becomes:

288

= 1,114 10-12 (1203102 + 1003103 + 803104 + 603105

+ 403106 + 203107) 1,95

(23)

This leads to a final crack size of 22,5 mm. The effect of the threshold has been
to remove the damage caused by the lowest stress range until the crack has grown to 6,34 mm,
and hence to extend the life for the crack to grow to the order of 20 mm from 5 to 8 years.
6. EFFECTS OF RESIDUAL STRESSES
Residual stresses are an inevitable consequence of a number of fabrication processes and
procedures, and are particularly important in welded steel structures. They have significant
effects on fracture and fatigue. The effect on fatigue is to change the mean stress and stress ratio
locally at the weld compared to the applied loading conditions. Thus at a welded joint where
residual stresses are often at yield level in tension, constant amplitude fatigue loading leads to
actual stresses at the weld which cycle from the yield strength downwards by the magnitude of
the applied stress range. If the applied loading included stresses in the compressive region or
which would cause crack closure so that part of the stress range is non-damaging, the effect of
the residual stresses is to hold the crack open so that the whole of the stress range becomes
damaging. As the crack grows away from the weld however it may run out of the tensile
residual stress zone and into a compressive region. In these circumstances part of the applied
stress range will become non-effective. To deal with these types of situation it is useful to use
the fracture mechanics crack propagation equations with an 'effective' stress intensity factor
range allowing for crack closure effects, and to divide the crack growth into stages defined by
crack sizes at which the effective stress intensity factor range changes due to the crack tip
moving into different residual stress regions.
7. CONCLUDING SUMMARY

The total fatigue life of a component is in general made up of an initiation life for the
formation of a crack and a propagation life to the final size determined by instantaneous
fracture, or by plastic collapse of the remaining cross-section.

The Paris Law describes the rate of fatigue crack propagation in terms of stress intensity
factor range.

Using the Paris Law, the life for crack growth from initial to final sizes can be
calculated for constant amplitude loading and also for variable amplitude loading. The
Paris Law is consistent with Miner's Law.

The fatigue behaviour is influenced by threshold effects, residual stresses and the
effective stress intensity factor.

289

Lecture 12.14: Fracture Mechanics:

Structural Engineering Applications
OBJECTIVE/SCOPE
Introduction to design concepts in the presence of structural damage, and to life prediction
methodologies.
PRE-REQUISITES
None
RELATED LECTURES
Lecture 12.7: Reliability Analysis and Safety Factors Applied to Fatigue
Lecture 12.10: Basics of Fracture Mechanics
Lecture 12.12: Determination of Stress Intensity Factors
SUMMARY
The basic concepts of safety, durability and reliability are briefly introduced and discussed with
reference to structural engineering applications. The basic principles for fail-safe and safe-life
design are presented. The main parameters to be accounted for in a fatigue life prediction are
briefly presented and discussed.
1. SAFETY, DURABILITY AND RELIABILITY
1.1 Introduction to Safety and Durability Concepts
If a structural member or connection is subjected to a cyclically varying load, it may fail after a
certain number of load applications even if the maximum nominal stress in a single cycle is
much less than the yield stress of either the material or the weld metal or the fastener.
A crack may in fact be initiated at some mechanical or metallurgical discontinuity (stress
concentration) and be propagated through the material with successive load repetitions until the
affected part loses its ability to carry the load. This failure mechanism is known as "fatigue".
Many failures of engineering structures are attributed to the consequences of pre-existing cracks
or crack-like discontinuities. Unfortunately, standard design models for beams, columns and
trusses which use conventional stress analyses and strength of materials methods, cannot
evaluate the effects of cracks in structural elements. Fracture mechanics principles enable a
more rational assessment of structures containing cracks to be carried out.
Originally, most structural engineering applications of fracture mechanics were limited to inservice assessment of structures, mainly aircraft. Recently, the size and shape of discontinuities
have been correlated to fabrication quality levels [1]. As a result, code-drafting committees are
beginning to employ fracture mechanics calculations in order to evaluate specifications for
290

quality assurance. Furthermore, when no specifications are applicable, fracture mechanics can
be used directly by the engineer for evaluating non-destructive inspection results.
The concept of damage-tolerant design is perhaps the major justification for the use of fracture
mechanics. This concept accepts the possibility that an element remains useful when it has been
subject to damage upon fabrication, transportation or even after several years in service. Such
damage, is often manifested by cracking.
Damage tolerance has now become a standard design concept in some engineering fields;
acceptance of certain aspects is apparent in structural engineering, where new concepts are
becoming more and more common.
Of particular importance, when adopting a damage-tolerant-design approach are the concepts of
safety and durability.
Both safety and durability requirements can be satisfied by considering initial damage in the
structure (or in the structural element) and assuring that such damage does not grow and reach
specific limits in prescribed time intervals.
Safety is associated with the extreme (i.e. the worst case) damage that may be missed during
manufacturing inspection. This damage must not grow and degrade the strength of the structure
(or of the member) below specified limits throughout the service life. Durability limits are
associated with the economics of in-service repair, i.e. initial damage typical of common
manufacturing procedures is not allowed to grow to a size sufficient to require extensive rework
or repair in less than one design service lifetime.
Current practice involves the use of deterministic methods and fracture mechanics concepts to
predict the accumulation of crack growth damage. Engineering experience with the use of these
methods has allowed the proper emphasis to be placed on design decisions affecting such
factors as allowable stress. Most of the input data for the deterministic methods are available in
statistical format (e.g. initial flaw sizes, material properties...) and thus it is possible to make
estimates of the probability of damage growth exceeding a specified level. Both the
deterministic and probabilistic approaches require a model of the damage accumulation process.
Lack of essential data for important variables and inexperience of the use of fracture mechanics
and reliability methods for making conventional design decisions have been the main reasons
for these techniques not being used in current practice.
When dealing with structural damage, two major concepts must be taken into consideration:
damage growth and damage containment. These concepts give the two criteria for:
a) Safety - the assurance that safety (that is the strength) of the structure will achieve and
maintain a specified residual level (in the presence of undetected damage) through the
anticipated service life.
b) Durability - the assurance that the structure can operate effectively with a minimum of
structural maintenance, inspection, costly retrofit, repair and replacement of major parts due to
the degrading influence of general cracking, corrosion, wear, etc.

291

Safety and durability limits are specified, designs chosen to comply, tests conducted to verify
and management procedures established to maintain and/or adjust the limits in service. Each
structure must meet both the safety and durability requirements.
1.2 Damage Growth Concepts
Past experience with testing of structural members under simulated service loading conditions
indicated that time to initiation of cracks from most structural details, such as sharp corners or
holes, is relatively short and that the majority of the life (i.e. as much as 95%) is spent growing
the resultant cracks to failure [2]. Likewise, analyses of in-service fracture, cracking instances,
etc. have indicated that a major source of cracks is the occurrence of initial manufacturing
defects such as sharp corners, tool marks, weld inclusions and the like [3]. Thus it is now
common practice to consider the damage accumulation process as entirely crack growth with
zero time to initiate the crack. Although this assumption may seem unduly severe, the
consideration of initial damage in the form of cracks or equivalent damage is absolutely
necessary to ensure structural safety.
Figure 1 (from [4]) includes a schematic of the damage model associated with structural safety.

Safety is ensured by designing to specific requirements of damage tolerance in which initial

damage is never allowed to grow and reduce the residual static strength of the structure below a
prescribed level, Pxx, throughout the life of the structure (or of the structural element). As
indicated in Figure 1, P xx is the greater of the design limit load or the maximum load that might
be encountered during the specified minimum period of unrepaired service usage. If in-service
inspections are required to ensure safety (e.g. for fail-safe designs) then the residual strength
level, Pxx, is the maximum load likely to occur during the inspection lifetime. The assumed
initial damage, ai, is that associated with the capability of manufacturing inspection.

292

When the concept of structural durability is considered, it includes "fatigue" procedures. In

addition it includes also the requirements for protection against such degrading factors as
corrosion, stress corrosion, wear, fretting, etc.
To prevent widespread cracking, a flow growth model can be considered such as that shown in
Figure 2 (from [4]).

In Figure 2 the limit, ae, is not failure as in the safety requirement but is associated with a size of
crack that can be repaired. The economic life, N e, is defined as that time when the population of
such cracks in each structure (or structural element) just reaches sufficient size that rework or
repair is no longer cost-effective. The durability specification requires that the "economic life"
be in excess of the design service life. In the example shown in Figure 2, the initial population
of flaws represents "apparent initial quality", or the conditions of structural details following
typical manufacturing operations such as hole drilling, shop or field welding, etc.
Since the limit for economic rework is strictly related to the manufacturing procedures (for
example, for fastener holes, it is of the order of the next minimal hole size), it is clear that the
limit of growth for durability requirements is significantly smaller than for safety.
Durability is associated with total flaw population, or total number of flaws in excess of the
economic rework limit for each member (or joint) in the structure. Safety is concerned with the
"worst case" of flaw size likely to be missed during an inspection.
In general it is not possible to anticipate the extent of usage of a structure (or structural detail).
Design is accomplished with the best estimate of typical usage and variance in expected usage.
Thus safety and durability limits of some individual structural detail may be shorter than
estimated in design. This condition places a tremendous emphasis on life management and inservice inspection of individual structures. In practice, design estimates of safety and durability
limits, as well as of remaining fatigue life, are adjusted to reflect usage severity based on
analyses of the accumulation of crack growth damage performed on individual structures.
1.3 Design Concepts Accounting for Damage

293

In the design of a structure which is to be subjected only to static loading, the design process
consists essentially of choosing the structural dimensions so as to satisfy at all points the
condition:

(1)
where the function of the cross-section size may be represented by an area, a section modulus,
etc., corresponding to functions of the applied loading such as axial load, bending moment, etc.,
and the design stress is exclusively a function of the mechanical property of the material, such
as yield strength or ultimate tensile strength. Equation (1) is quite independent of the detailed
shape of the structure at any particular point, and its solution requires only the application of
simple mathematics.
The design of a structure which may suffer fatigue failures is, on the contrary, a somewhat more
complex process. In fact the fatigue strength depends on many more variables than the static
strength of the material, and particularly on the detailed design of the joints in the structure.
Thus, although the equation which has to be satisfied is still the same, there is the additional
problem of selecting a suitable value for the design stress. This selection becomes the main
issue when designing a structure against fatigue.
The first difficulty is to decide what one is trying to achieve in terms of life, i.e. how long
should the structure last. In the past, even though detailed design to prevent fatigue damage was
not common, infinite life was often achieved (it is sufficient to think as an example of Roman
bridges, some of which are still standing at present because the static design of structures
involved the application of large safety factors and the use of approximate design methods,
which resulted both in the use of pessimistic estimates of design resistance and in the
overestimation of the service loads.
However, design stresses have steadily increased and the use of more sophisticated design
methods means that, nowadays, design stresses tend actually to be present in structures in
service. This level of stressing is the reason why fatigue has become a problem. The only way
that a designer can ensure infinite life is to use working stresses below the fatigue limit of the
joints in the structure. This approach implies the use of very low working stresses. However, if
any stress cycle has an amplitude greater than the fatigue limit, then it must be accepted that
failure will eventually occur and the life be finite.
The designer is therefore in a dilemma. Is he to keep the stresses very low, inevitably at the
expense of a heavy structure, and try to obtain infinite life, or is he to accept a finite life? And if
he does accept a finite life, on what basis is he then to choose the design stresses?
Much of the thinking on this subject has originated in the aircraft industry (where weight saving
is of particular importance). It is of particular interest to consider the two design philosophies
which were developed by aircraft engineers, which are known as the "safe-life" and the "failsafe" concepts of structural design.
In the "safe-life" method the designer starts by making an estimate of the load spectrum to
which the critical structural components are likely to be subjected in service. These components
294

are then analysed or tested under that load spectrum, so as to obtain its expected life. Finally a
factor of safety is applied so as to give a safe life during which the possibility of fatigue failure
is considered to be sufficiently remote. At the end of that period, the structural components
concerned are automatically replaced. It is important to realize that, even though the estimate of
the laboratory life can be made by practical tests, the safe-life method is ultimately theoretical in
nature. The critical step is to derive from the expected life the safe life. The factor of safety
introduced at this stage has to allow for a large number of unknowns, such as scatter obtained in
fatigue test results, errors in the estimate of the load spectrum and the effect of extraneous
variables such as environmental conditions, corrosion, etc. It is clear that by making the factor
of safety sufficiently large, the designer can govern the probability of failure associated with his
design. On the other hand, if a fatigue crack does occur, it may well be catastrophic, and safety
depends on achieving a specified life without a fatigue crack developing. With this design
method, the emphasis is on prevention of crack initiation.
With the "fail-safe" concept the basis of design is that, even if failure of part of the main
structure does occur, there will always be sufficient strength and stiffness in the remaining part
to enable the structure to be used safely until the crack is discovered. This concept implies that
periodic in-service inspection is a necessity, and that the methods used must be such as to ensure
that cracked members will be discovered, so that repairs or replacements can be made.
It is clear that with this method of design the probability of partial failure is much greater than
with the "safe-life" design. In developing a fail-safe structure, the safe-life should also be
evaluated, in order to make sure that it is of the right order of magnitude. However, the
emphasis, instead of being on the prevention of crack initiation, is on producing a structure in
which a crack will propagate slowly, and which is capable of supporting the full design load
after partial failure. The basic principle of fail-safe design is therefore to produce a multiple
load-path structure, and preferably a structure containing crack arresters. In addition, the
structural elements must be arranged so as to make inspection as easy as possible. In areas
where that is not possible, the elements must be oversized so that either fatigue cracking does
not occur in them, or fatigue crack growth is so slow that there is no risk of failure.
It must be emphasized that the objective of both philosophies is the same, namely to reduce to a
negligible level the risk of catastrophic failure. It is now generally accepted that neither method
is satisfactory when applied in isolation [5]. Both are necessary and they are not mutually
exclusive. In fact, a safe-life design would produce a structure with a high level of freedom
from fatigue cracking; however, the same structure should possess fail-safe characteristics.
As a consequence, even in as critical and expensive a structure as an aircraft, designers are
prepared to accept a philosophy which results not only in a finite life from the point of view of
fatigue cracking, but also, with sufficient safeguards, in a structure which may contain fatigue
cracks during part of its working life.
Once the idea is accepted that it is reasonable to design a structure on a finite life basis, it is
clear that the exact design philosophy to be adopted depends on the consequences of failure.
Sometimes, a satisfactory design method is provided by a modified safe-life philosophy, the so
called "working-life" method [5].

295

The objective is not to reduce the risk of catastrophic failure to a negligible level, but to produce
a structure which will usually (but not necessarily always) be free from fatigue cracks for a
specified life. As compared with the true "safe-life" philosophy, less importance is placed on
making an accurate estimate of the safe life. In formal terms, this means that two important
changes are made in that philosophy: (a) that the structure is not automatically replaced at the
end of its "safe working life", and (b) that it is accepted that some cracking may develop before
the "safe life" is reached. It therefore becomes reasonable, instead of estimating the safe life
accurately by carrying out programmed fatigue tests on specimens representing the particular
structure, to estimate it analytically by using fatigue test results obtained on laboratory
specimens representing standard details typical of those used in the structure.
It is important to notice that each of the previously described methodologies essentially imply a
different probability of failure. In fact, for a safe-life design, it might be decided to use an S-N
curve relating to the mean minus 2 standard deviations of Log N (corresponding to the 97,7%
probability of survival), while for fail-safe design the S-N curve corresponding to the mean
minus 1 standard deviation (representing the 84,1% probability of survival) might be considered
to be sufficiently conservative.
1.4 Reliability and Risk Analyses
In current practice, designs are chosen to satisfy specific safety and durability life and strength
requirements by deterministic analyses of damage growth and residual strength. Decisions on
materials, structural configurations, allowable stresses, etc., are based upon the results of these
analyses.
Durability of metallic structures is influenced by the local details of joints, splices, welds,
fasteners, eccentricities, etc. For a given design, the lower the gross stress, the more durable the
design.
Analyses of crack growth damage, while generally deterministic, rely on input data such as
initial flaw sizes, material and usage variability, etc., which are available in a statistical format.
Thus, in addition to predicting mean values of damage accumulation, the methods can be used
to predict the distribution of damage with time. For example, under specific usage and assumed
distributions of the other variables, it is possible to estimate the distribution of lives, growth
rates, final flaw sizes, etc. and thus the probability of equalling or exceeding a certain
requirement. If only one variable is considered, the estimation can be straightforward. When
joint probabilities are involved, the procedures become more complex. The results of the
probabilistic analyses are extremely sensitive to the initial distributions of the variables and the
functions used to approximate them. Among the major variables to be considered are:

material strength

crack growth rate

critical crack size

usage and maximum operational loads

crack detection capability

296

inspection techniques

frequency of inspection

Although published reliability methods can be found in the literature, current practice does not
in general reflect their use. In fact, the sparsity of data associated with the major variables listed
above, and the extreme sensitivity of the results to the distribution functions has limited the
effectiveness of, and the confidence in, the results. Furthermore, the design to a specified
reliability level requires that the acceptable failure rate be established in advance, and that
relationships be established between reliability and normal design decision factors (i.e.
allowable stress, structural configuration, etc.). Since overall structural reliability can be
significantly altered by factors other than those normally considered in design (e.g. inspection
frequency), it is rather difficult to assign the proper weighting function to any particular
variable.
Actually, reliability techniques have been more successful in assessing older structures (i.e. in
assessing remaining fatigue life) where extensive test and service failure data are available to
establish the mean time to failure with some degree of confidence [6].
For the concept of durability described in the previous paragraphs, probabilistic techniques may
have the most direct impact, since the ultimate decision on the economic life is connected to the
global population of cracks just below the rework limit. Safety requirements, on the other hand,
account for growth of the extremes of the initial flaw population and other variables (i.e. the
worst case). It is anticipated that the requirements will continue to be satisfied with conventional
practices based on deterministic methods.
In summary, methods for calculating the accumulation of crack growth damage are required for
both deterministic and probabilistic analyses. Lack of essential data for important variables and
inexperience with the use of reliability methods for making conventional design decisions have
been the main reasons for the nearly complete absence of these techniques in current design
practice.
2. LIFE PREDICTION METHODOLOGY
2.1 Introduction
Currently, life prediction for a structure subject to variable loading is generally based on crack
growth damage integration models which use a data base and analysis to interrelate the
following elements:
(a) initial flaw size and distribution
(b) loading conditions
(c) basic crack growth material properties
(d) crack and structural properties
(e) damage model
297

(f) fracture or life limiting criteria.

These models are always calibrated on the basis of test results, and the confidence normally
associated with life predictions is usually derived from the ability of the model to predict the
laboratory generated crack growth behaviour.
The only quantifiable measure of fatigue damage is a crack. Cracks impair the load-carrying
characteristics of a structure. A crack can be characterized for length and configuration using the
stress intensity factor K, a structural parameter interrelating the local stresses in the region of
the crack tip with (a) crack geometry, (b) structural geometry, and (c) level of load on the
structure.
The crack grows in response to the cyclic loading applied to the structure. Any crack (a) will
grow a given increment (a) when subjected to a given number of cycles (N), the rate being
measured by
. When the crack length reaches a critical value (a cr), the growth becomes
unstable, thereby inducing failure. The life (N F) is the measure of accumulated cycles required
to drive the crack from its initial length a i to the critical length a cr. The interrelationship between
crack length, loading and residual strength of a structure is illustrated in Figure 3 (from [4]).

As shown in Figure 3, the monotonic increase in crack length is induced by a continuous

sequence of cyclic loads. The residual strength (res), that is the load carrying capacity of the
298

cracked structure, is shown to monotonically decrease with increasing crack length following an
expression of the type:
res = Kc/f(a) (2)
where:
Kc is a material property, termed fracture toughness, which is a constant for a specific geometry
f(a) = Y(a)

is a structural property (the stress intensity factor).

When the residual strength decays to the level of the maximum stress in the service load history,
fracture occurs. The crack length associated with fracture (i.e. a cr) is determined by solving
Equation (2) for crack length, assuming that the residual strength equals the maximum stress
level in the spectrum, or that it equals the design limit load stress level (whichever is greater). It
is interesting to notice that the crack growth rate is directly related to the rate of loss of residual
strength through Equation (2), thus justifying the selection of crack length as the main
parameter to quantify structural fatigue damage.
2.2 Crack Growth Behaviour Effects
A crack of length ai will grow to acr in some service life, N F. Experiments have shown that
several parameters affect NF; the most important of these are:

initial crack size, ai

loading history and conditions

material properties

structural properties

critical crack size, acr.

2.2.1 Initial Crack Size

The role of initial crack size on fatigue crack growth is of paramount importance; in fact, as
illustrated in Figure 4, given a configuration and loading, the smaller the initial crack size, the
longer the fatigue life. Furthermore, the shape of the a-N curve for a given configuration and
loading remains essentially constant for any given crack growth increment. Thus, given the
crack growth curve from a i1 in Figure 5 (from [4]), it is possible to determine the crack growth
curve from ai2>ai1 by simply shifting the previous curve on the left, as shown in the figure,
where Ni represents the number of cycles required to grow the crack from a i1 to ai2. It follows
that NF2 = NF1 - Ni.

299

2.2.2 Loading
The stress history experienced at each location of the structure will differ due to changes in
bending and twisting moments, shear, etc. It follows that similar structural details placed in
different locations within the structure will experience different loading histories. Given a
particular joint and initial crack configuration, it is clear that, when subjected to a more severe
loading spectrum, a shorter life (N F1) will result than under a less severe spectrum (resulting in a
longer fatigue life NF2).
2.2.3 Material Properties
Experimentally, it has been shown that for the same loading condition (i.e. the same number and
amplitude of stress cycles) cracks will grow faster in certain materials (or alloys) than in others.
The crack growth rate (a/N) can be derived experimentally for each material. Given the
300

same load and geometric conditions, the material (or alloy) having the slower growth rate
characteristics will have a longer life (NF), as shown in Figure 6 (from [4]).

2.2.4 Structural Properties

The most complex of the parameters affecting crack growth behaviour are the structural
properties. The structural properties involve such things as crack configuration, load transfer
mechanisms, fastener hole size, dimensions and extent of weldments, part thickness, stress
concentrations due to geometric effects, etc. A substantial amount of both experimental and
numerical work has been performed to characterize the geometrical effects on fatigue life see,
for example [2, 3, 5-15].
2.2.5 Critical Crack Length
The critical crack length (acr) is a function of material, structural geometry and loading. As
shown in Figure 7 (from [4]), for acr/ai sufficiently large (e.g. >5), the relative effect of a cr on
NF is typically small. The primary advantage of designing for a large critical crack length is the
increased inspectability that it provides. A large critical crack length increases the probability of
locating the crack before it becomes critical, thereby enhancing structural safety.

2.3 Fatigue Crack Growth Prediction Models

301

Life predictions are usually performed by means of crack growth damage integration models,
which interrelate the following elements:
a) the initial flaw distribution, which accounts for size variations and location of cracks in a
given structure;
b) the usage of the structure, describing the load spectra data base;
c) material properties concerning constant amplitude crack growth rate material properties,
accounting for stress ratio and environmental effects;
d) crack tip stress intensity factor analyses which account for crack size, shape and structural
interactions;
e) damage integrator model which assigns a level of crack growth for each applied stress
application and accounts for load history interactions;
f) the fracture or life limiting criterion, which establishes the end point of the life calculation.
As expressed in a numerical form, the damage integrating equation is:
acr = ai + j = 1,Nf aj (3)
where aj is the growth increment associated with the j-th applied load cycle.
The purpose of Equation (3) is to determine the fatigue life (N F).
The results obtainable from Equation (3) depend on the interaction among the various elements
(a)-(f); such interaction takes place as follows:
1) acr is determined by interrelating (b), (d), and (f)
2) aj is a function of the interaction between (b), (c), (d) and (e).
2.3.1 Initial Flaw Distribution
A measure of initial quality in a structural component is given by the distribution of initial crack
sizes, Figure 8 (from [4]). For predictions of safety limits, the initial cracks larger than the non
destructive inspection (NDI) detectability limit are of main concern.

302

Presently, many specifications, in particular in the aeronautics field [4,7], detail NDI limits and
require verification and certification of the capability of the contractor to detect cracks smaller
than the specified NDI limits.
2.3.2 Usage
The sum of the load levels that a structure is expected to experience is determined by a
projection of the amount of usage expected over its service life, in the various possible service
conditions. The specific sequence of loads that is applied to the structure must be known for a
precise crack growth damage analysis. Each loading event in the design, analyses or test load
spectrum consists of a series of cycles that combine the deterministic and the probabilistic
events describing the type of loading. The deterministic events include usual loading conditions
(e.g. for a bridge structure, usual traffic loads, or for an aircraft takeoff, landing and some basic
manoeuvre loads during each flight). Probabilistic events such as oversized loads (legal or
illegal) for bridges, or rough field taxiing for aircraft, may occur periodically. Although it is
possible to estimate, at least roughly, the number of times these events occur, their position in
the load sequences may be determined only by means of probabilistic considerations.
In developing the load spectrum for crack growth damage analysis, it is necessary to determine
the stress history for each critical area in the structure. This is accomplished by determining, by
means of stress analysis, the relationship between the load history and the local stress response,
as illustrated in Figure 9 (from [4]).

2.3.3 Material Properties

303

The material properties enter the damage integration procedure in the form of constant
amplitude crack growth rate data. Crack growth data are generated in tests in the laboratory
under constant amplitude loading cycles, on simple specimens with accepted characterizing
stress intensity factors. Crack growth rate data are developed and correlated on the basis of
growth rate (da/dN) as a function of the stress intensity factor range K, (K = Kmax - Kmin).
For a given K, the crack growth rate increases with increasing stress ratio (R=min/max).
Hence, the constant amplitude crack growth rate properties for a given material or alloy consist
of a set of curves, as shown in Figure 10 (from [4]). It is important to notice that, for the fracture
mechanics approach, given a K:R combination, there is a unique value of the crack growth
rate (da/dN), which is independent of geometry. When necessary, thermal or environmental
effects can also be included in the crack growth rate data generated by laboratory testing.

2.3.4 Crack Tip Stress Intensity Factor Analysis

The crack tip stress intensity factor (K) interrelates the crack geometry, the structural geometry
and the load on the structure with the local stresses in the region of the crack tip. The stress
intensity factor takes the form:
(4)
where:

Y(a) is a correction factor which is a function of the crack length and accounts for the
structural configuration and loading conditions

is the applied stress

304

a is the crack length.

It can be seen that any number of combinations of the parameters Y(a), and a can give rise to
the same K. The crack growth analysis rests on the experimentally verified proposition that a
given K gives rise to a certain crack growth rate, regardless of the way in which the parameters
were combined to generate that K.
Considerable references are available which define experimental and mathematical solutions for
stress intensity factors, for various structural configurations, both in closed or approximate
form, e.g. [8].
2.3.5 Damage Integration Models
Rewriting Equation (3) such that the integration is conducted between the initial crack length
(ai) and any intermediate crack length (a n) between ai and acr, the following expression is
obtained:
an = ai + j = 1,N aj (5)
where N is the number of cycles corresponding to the intermediate crack length a n. The next
cycle of the spectrum stress induces a crack length increment aN+1. The damage integration
model provides the analysis capability to determine this crack length increment. The growth
incrementaN+1 is set equal to the constant amplitude crack growth rate, which in turn is
expressed as a function of stress intensity factor range (K) and stress ratio (R). These
functions are determined by using the maximum and minimum stresses in the N+1 cycle of the
given spectrum, and evaluating the correction factor Y(a), in Equation (4), associated with the
given structural geometry at crack length a n. After calculating the two crack tip parameters K
and R, they can be modified to account for the effect of prior loading history, using retardation
models (e.g. [9]). Retardation models account for high-to-low load interaction effects, i.e. the
phenomena whereby the growth of a crack is slowed by application of a high load in the
spectrum. Failure to account for high-to-low load interaction via retardation models leads to
conservative crack life predictions.
2.3.6 Final Crack Length
The final crack length, while estimating safety limits, may ordinarily be chosen as a cr. However,
durability considerations often dictate that the final crack size, a F, be chosen smaller than acr to
represent re-work or repair limits. A choice of a F along these lines is shown in Figure 11 (from
[4]).

305

3. CONCLUDING SUMMARY
o

Many failures of engineering structures are attributed to the consequences of

pre-existing cracks or crack-like discontinuities.

Standard design models for beams, columns and trusses using conventional
stress analyses and strength of materials methods, cannot evaluate the effect of
cracks in structural elements. Fracture mechanics principles enable a more
rational assessment of structures containing cracks.

Damage tolerance has now become a standard design concept in some

engineering fields. The concepts of safety and durability are particularly
important in this approach.

Safety is associated with the extremes damage that may be missed during
manufacturing inspection. Durability limits are associated with the economics
of in-service repair.

In current practice, designs are chosen to satisfy specific safety and durability
life and strength requirements by deterministic analyses of damage growth and
residual strength.

Currently, life prediction for a structure subject to variable loading is generally

based on integration models of crack growth damage. The models are calibrated
on the basis of test results. They interrelate initial flaw size and distribution,
loading conditions, crack growth material properties, crack and structural
properties, the damage model and the fracture or life limiting criteria.

4. REFERENCES
[1] Guidance on Some Methods for the Derivation of Acceptance Levels for Defects in Fusion
Welded Joints, British Standards Institution PD 6493, 1980.
[2] Rolfe and Barsom "Fatigue and Fracture Control in Structures", Prentice Hall Inc.,
Englewood Cliffs, N.J., 1977.
[3] Fisher J.W. "Fatigue and Fracture in Steel Bridges: Case Studies", J Wiley & Sons, New
York 1984.

306

[4] Wood H.A., Engle R.M., Gallagher J. and Potter J.M. "Current Practice on Estimating Crack
Growth Damage Accumulation with Specific Application to Structural Safety, Durability and
Reliability", AFFDL-TR-75-32, March 1975.
[5] Gurney T.R., "Fatigue of Welded Structures", Cambridge University Press, Cambridge, UK,
1979.
[6] Madsen H.O., Tallin A.G., "Fatigue Reliability Updating Based on Inspection and
Monitoring Results", Proceedings, IABSE Workshop on "Remaining Fatigue Life", Lausanne,
1990.
[7] ESACRACK User's Manual, ESA PSS-03-209, European Space Agency, Mechanical
Systems Department, Structures and Mechanisms Division, Structural Design Section,
Noordwijk, The Netherlands, 1989.
[8] Tada H., Irwin G.R., Paris P.C., "The Stress Analysis of Crack Handbook", Del. Research
Corp., Hellertown, PA, 1973.
[9] Willemborg J., Engle R.M., and Wood H.A., "A Crack Growth Retardation Model using an
Effective Stress Concept", AFFDL-TM-71-1-FBR, 1971.
5. ADDITIONAL READING
1. Fisher J.W. "Bridge Fatigue Guide: Design and Details", American Institute of Steel
Construction, New York, 1977.
2. Fisher J.W., Hirt M.A. and McNamee B.M., "Effect of Weldments on the Fatigue
Strength of Steel Beams", NCHRP Rept. 102, 1970.
3. Fisher J.W., Yen B.T., Klingerman D.J., McNamee B.M., "Fatigue Strength of Steel
Beams with Welded Stiffeners and Attachments", NCHRP Rept. 147, 1974.
4. Schilling C.G., Klippstein K.H., Barson J.M., Blake G.T., "Fatigue of Welded Steel
Bridge Members under Variable Amplitude Loading", NCHRP Rept. 188, 1978.
5. Fisher J.W., Barthelemy B.M., Mertz D.R., Edinger J.A., "Fatigue Behaviour of FullScale welded Bridge Attachments", NCHRP Rept. 227, 1980.
6. Fisher J.W., Hausamann H., Sullivan M.D., Pense A.W., "Detection and Repair of
Fatigue Damage in Welded Highway Bridges", NCHRP Rept. 206, 1979.
7. Fisher J.W., Yen B.T., Wang D., Mann J.E., "Fatigue and Fracture evaluation for Rating
Riveted Bridges", NCHRP Rept. 303.
8. Castiglioni C.A., Bremen U., "Influence of some geometrical parameters on stress
concentration in longitudinal attachments", Costruzioni Metalliche, n.4, 1989, Milano,
Italy.
9. Smith I.F.C., Gurney T.R., "Changes in the fatigue life of plates with attachments due to
geometrical effects", Welding Research Supplement, AWS, vol 65 no.9.
307

Lecture 12.15: Fracture Mechanics

Applied to Fitness for Purpose
OBJECTIVE/SCOPE:
To show methods for the practical application of fracture mechanics for predictions of fatigue
crack growth through to final failure including estimates of the significance of flaws.
PREREQUISITES
Lecture 12.10: Basics of Fracture Mechanics
Lecture 12.11: Stress Analysis of Cracked Bodies
Lecture 12.12: Determination of Stress Intensity Factors
Lecture 12.13: Fracture Mechanics Applied to Fatigue
RELATED LECTURES
Lecture 12.1: Basic Introduction to Fatigue
Lecture 12.2: Advanced Introduction to Fatigue
Lecture 12.8: Basic Design Concepts in Eurocode 3
SUMMARY
The lecture describes the application of fracture mechanics procedures for estimating crack
shape development and fatigue lives by considering incremental crack growth at the deepest
point and surface points of a crack. The effect of multiple initiation sites and the subsequent
coalescence is described. The failure assessment diagram method for determining final failure
by fracture or plastic collapse is explained. The extension of the fatigue treatments to variable
amplitude loading and to include residual stress, threshold, and crack closure effects is
described.
1. INTRODUCTION
The general background to the use of fracture mechanics for fatigue applications is described
in Lecture 12.13. This background showed that the Paris Law for instantaneous growth rate of
fatigue crack propagation could be integrated to determine the fatigue life between given initial
and final crack sizes. It was also shown that this approach is entirely consistent with the S-N
approach for welded details, and leads to relationships of the form S mN = constant, where m is
the crack growth exponent in the Paris Law, which typically has a value of about 3. For variable
amplitude loading it was also shown that the fracture mechanics approach gave the same
answers as Miner's Law for variable amplitude loading on the assumption that there is no
interaction between cycles of different stress levels.
308

2. CRACK SHAPE DEVELOPMENT

Cracks can be categorised in geometrical terms as:
(i) through-thickness cracks
(ii) embedded cracks
(iii) surface cracks
For the purpose of modelling the fracture mechanics, the part-thickness embedded or surface
cracks are idealised as elliptical or semi-elliptical in shape. Most fatigue cracks in welded
structures start either from surface stress concentrations such as weld toes, or from weld
discontinuities or internal flaws. Thus fatigue cracks grow from an initial part-thickness
condition to through-thickness and then grow on as through-thickness cracks if they have not
reached final failure.
During the stages at which a fatigue crack is part-thickness, the stress intensity factor varies
around the perimeter of the crack front, as described in Lectures 12.10 and 12.11. Since the
crack growth rate depends on the stress intensity factor raised to the power m, it follows that the
crack grows at different rates at different positions around the crack front and may therefore
change its shape progressively. The Y factor in the general expression for the stress intensity
factor, K = Y
, is itself a function of both a/T, the crack depth to thickness ratio, and
a/2c, the crack depth to crack length or aspect ratio for part thickness cracks. It is therefore no
longer possible to integrate the Paris Law equation directly as an analytical expression because
of the changes in Y as the crack grows and changes the a/T and a/2c ratios.
The procedure adopted to deal with this situation is to use incremental numerical methods.
Starting with the required initial crack depth and aspect ratio, the stress intensity factor is
calculated at the deepest point of the crack and at the ends of the crack for the stressing
conditions applied. This can be done using the parametric equations of Raju and Newman,
included for example also in BSI Document PD 6493 [1], for either pure tension or pure
bending stress conditions in a flat plate. It can be shown that conservative (high) estimates of
the stress intensity factor are always obtained if the applied stress field is represented by a linear
gradient of tension and bending stress over the region of the crack.
The Raju and Newman equations enable stress intensity factors to be calculated for unit stresses
in tension or bending at any position around the crack front, and the total stress intensity factor
can be calculated by adding together the separate tension and bending components
corresponding to the stress field considered. To determine the progressive change in crack
shape, the procedure is firstly to calculate the stress intensity factor for the deepest and surface
(end) positions on the crack front. The crack growth rate is then calculated at the deepest and
incremental distance in the deepest direction. This distance should be chosen as a small
proportion of the remaining ligament such that the crack growth rate can be considered sensibly
constant, and the number of cycles is the distance divided by the growth rate. This same number
of cycles is then applied to the growth rate at the ends of the crack to calculate the increment of
crack growth in this direction. Adding the increments of crack growth in each direction to the
original crack dimensions gives the new dimensions which determine the new shape.
309

The stress intensity factors at the deepest and end positions are then recalculated for the new
crack dimensions and the procedure repeated for a new increment of crack growth in the depth
direction. This calculation is repeated again and again until the crack reaches a pre-determined
final size or reaches a size to cause failure. The choice of size of increments of crack growth in
the depth direction is important in determining both accuracy and time/cost of carrying out the
analysis. The analysis is usually put into effect by writing a computer program to implement the
step by step calculations. The smaller the increments the more accurate the results but the more
steps are required to reach a given final crack size. The most efficient procedure is generally
found to be to use small increments from the initial crack size and increase these in steps on a
logarithmic basis over the remaining ligament of the thickness. The reason for this procedure is
that, as the crack grows larger, the growth rate increases substantially because the stress
intensity factor is raised to a power in the crack growth law. Thus most of the fatigue life is
spent whilst the crack is small and the crack growth rate accelerates to failure rapidly.
Because tension and bending stresses produce different ratios of the stress intensity factor at the
deepest and end positions of the crack, different crack shapes are produced for these conditions.
The fatigue crack tends to grow to a shape in which the stress intensity factor is uniform around
the crack front. For tension loading the stable crack shape tends towards semi-circular, whilst
for pure bending applied loads the stable shape is semi-elliptical. This is shown in Figure 1
where results of fracture mechanics analyses are compared with experiment in some work at
UMIST by Chu [2]. There can be some effects of variable amplitude loading and threshold on
crack shape development. These effects are discussed later.

310

The initial crack dimensions to be used in a fatigue fracture mechanics analysis depend on the
level of detailed accuracy of model used. Where assessment is being carried out of a known
detected flaw, clearly the dimensions of that flaw should be used. If an attempt is being made to
model the behaviour of notionally sound welds then consideration has to be given to the
imperfections and stress fields inherent to welded joints. The assumptions made in this respect
can make significant differences to the predicted results obtained. It is now well established that
virtually the whole of the fatigue life of as-welded joints is taken up with propagation of initial
tiny flaws to a final crack size for failure. It is necessary for the fracture mechanics models to
represent observed experimental behaviour both in respect of fatigue life to failure and crack
shape development. The models which do this best involve multiple crack initiation and
coalescence as discussed below.
3. MULTIPLE CRACK INITIATION
In welded structures it is commonly found that fatigue cracks initiate from tiny slag intrusions in
the partially melted region at the weld toe alongside the fusion line. These intrusions are
311

distributed randomly along the weld toe region. This distribution leads to multiple independent
initiation of a series of small fatigue cracks at the weld toe. The precise positions at which these
cracks are initiated depends both on the occurrence of these intrusions and on the detail of weld
toe geometry such as weld toe radius, weld size and weld angle. The cracks grow independently
until they interact with adjacent similar cracks. At this stage they coalesce to form a single crack
of the same depth as the individual single cracks but with a length covering the combined length
of all the cracks which have come together as shown in Figure 2. The effect is that the crack
shape development curve shows a discontinuity when coalescence occurs. This discontinuity is
shown in Figure 3 for the case of a fracture mechanics analysis of crack shape development in
the stress fields typical of offshore tubular joints. A comparison is also made in Figure 3 with
the case of a single crack of the same initial shape. A study of these effects by Thurlbeck [3] at
UMIST showed that the coalescence stage depended on the size, aspect ratio and pitch of the
initial weld toe intrusions, and that the fracture mechanics analysis could give a good prediction
of observed experimental behaviour provided the weld toe geometry, and variable amplitude
loading/threshold effects are taken into account.

312

The sizes of weld toe intrusions to be taken as initial cracks in the fracture mechanics analyses
of fatigue in welded joints are of the order of 0,1 to 0,4mm depth. For the multiple crack growth
case, the initial aspect ratio is taken as a typical a/2c ratio of 1, i.e. semi-circular. The initial
spacing of these intrusions depends on welding procedure, plate thickness and stress level, but is
typically of the order of 10 to 20mm.
In variable amplitude loading it is possible for the stress intensity factor at, say, the surface
(end) position of the crack to be above the threshold, whilst at the deepest position the stress
intensity factor may be below the threshold, particularly under bending stress conditions applied
to a weld toe region. If this situation occurs, crack growth can take place at the ends of the crack
but not at the deepest point for this load level. The fracture mechanics crack growth procedure
must take account of this behaviour. Thus the crack will change its shape in a different way
from that which would occur without threshold effects intervening. This effect tends sometimes
to drive cracks along the surface to give them a longer aspect ratio under variable amplitude
than for constant amplitude, but the effects can be predicted by a carefully set up fracture
mechanics model.
4. FINAL FAILURE CRITERIA
The final crack size to be used as the upper limit for the fracture mechanics analysis is usually
taken as either equal to the plate thickness or to the crack size which would cause an
unacceptable risk of failure. The failure modes usually considered are fracture or plastic
collapse. Fracture mechanics analyses are also used to determine these conditions and can be
built into an overall computer program calculating both fatigue and final failure behaviour. In
components containing liquids or gases, the development of through-thickness cracking clearly
represents failure by leakage, and this crack size represents a natural upper limit for such cases.
In the introductory Lecture 12.10 on fracture mechanics, the basis of the stress intensity factor
as the parameter controlling stress, strain, and energy fields in the neighbourhood of a crack tip
was explained, and methods of determining it were described in Lecture 12.12. It was also
explained that in materials which fail by fracture under linear elastic conditions the failure
occurs at a critical value of the stress intensity factor. This critical value is known as the fracture
toughness and given the symbol Kc. It is found that this fracture toughness depends on the
material, temperature, strain rate, and triaxiality of stress. This latter effect is usually manifested
by effects of geometry and thickness on the toughness values measured experimentally. The
toughness is found to be lower for thicker material. In standard bend or compact tension
specimens a minimum thickness requirement is given to ensure that plane strain conditions are
maintained throughout to obtain a minimum toughness level known as the plane strain fracture
toughness and given the symbol Klc. It was also explained in Lecture 12.10 that local yielding at
the crack tip had the same effect as a slight increase in crack length for elastic cases and the
concept of a plastic zone correction for the stress intensity factor was introduced. This effect of
plasticity in increasing the severity of crack tip conditions becomes more important as applied
stress conditions approach general yielding. The stress intensity factor loses its validity as a
linear elastic parameter, at least in small scale laboratory specimens, although in full scale
structures it is still possible to have a large absolute size of plastic region contained by
surrounding elastic material. To analyse such situations properly it is necessary to carry out
elastic-plastic stress analyses of cracked bodies and to use alternative parameters to measure the
severity of crack tip conditions. Two such parameters are the crack (tip) opening
313

displacement d, and the J contour integral. A full discussion of these parameters is beyond the
scope of the present lecture, but it is sufficient to know that both of these parameters are
commonly turned into equivalent notional stress intensity factor or fracture toughness values in
terms of K through the following relationships:
K2 = Ef J = Ef msyd

(1)

where Ef = E for plane stress, and Ef = E/(1- n2) for plane strain, and m is a constraint factor
lying between 1 and 2.
As well as failure by fracture including the effects of yielding on severity of crack tip
conditions, the effect of cracks on failure by plastic collapse on the remaining net cross-section
has to be considered. One of the best methods of combining together consideration of fracture
and plastic collapse is the use of failure assessment diagrams of the type developed originally in
the UK by the CEGB as their R6 approach, and now also included in PD 6493 [1] which
describes assessment of the significance of weld defects. An example of a level 2 assessment
diagram is shown in Figure 4. The vertical axis can be thought of as the fracture axis, and is the
ratio of the applied elastic stress intensity factor to the plane strain fracture toughness, K r.
Clearly, if this ratio reaches 1, then failure occurs by fracture. The horizontal axis can be
thought of as the plasticity axis, and is the ratio of the applied load to the plastic collapse load,
Sr (based on collapse at the flow strength, the average of yield and ultimate strengths). Again, if
this ratio reaches 1, failure is predicted, this time by plastic collapse. As the S r ratio increases so
does the amount of plasticity at the crack tip, which in turn increases the severity of crack tip
conditions from a fracture point of view. If the equivalent stress intensity factor of Equation (1)
is thought of as consisting of elastic and plastic parts, as the amount of plasticity increases so
the plastic part of the equivalent stress intensity factor must increase and the elastic proportion
must reduce. Since the Kr axis uses only the elastic stress intensity factor, the effect of plasticity
has to be taken into account by a reducing value of K r as Sr increases. Thus, in Figure 4, the
height of the assessment diagram curve at any value of S rrepresents the permitted value of
Kr based on elastic stress intensity factor, whilst the shaded area above the curve represents the
additional contribution to the equivalent stress intensity factor due to plasticity.

314

The procedure for assessing the acceptability of any given crack size is to calculate the
parameters Kr and Sr (or Lr for the level 3 assessment diagram based on yield and strain
hardening behaviour). In the 1991 edition of the BSI document PD 6493 [1] there are three
levels of assessment diagram as shown in Figure 5 and detailed procedures are given for
calculating the different parameters. The point representing the coordinates K r Sr (or Lr) is
plotted on the appropriate assessment diagram. If this point lies inside the assessment diagram
as shown at A on Figure 4 the crack is considered safe, whereas if it lies outside the curve as
shown at B the crack is considered unsafe.

315

It should be noted that, for fracture assessment purposes, the stress intensity factor must be
calculated based on the maximum value of the applied stress and must also include the effect of
any residual, thermal, or other secondary stresses. For fatigue crack growth calculations, it is the
range of stress intensity factor which must be used in Paris Law relationships and welding
residual stresses do not come into this directly although they may have indirect effects as
discussed below.
When computer programs are used to calculate fatigue crack growth behaviour, the same
program can be used to calculate the parameters K r and Sr (or Lr) at the end of each increment of
crack growth and check whether the values are acceptable for the assessment diagram
concerned.
5. ADVANCED TREATMENTS
Mention has been made in other lectures of crack closure, threshold and effective stress
intensity factors. There is considerable evidence to show that two of the effects of plasticity are
to make the crack tip propagate through stretched material and to create a wake of unloaded
plastic material behind the advancing crack tip. The effect of this behaviour is that the crack
faces may close together, and only that part of the stress intensity factor range that is damaging
for fatigue crack propagation is that due to stresses above the level at which the crack faces
separate again. A number of research workers have developed computer programs for fatigue
crack propagation based on use of effective stress intensity factor ranges, taking into account the
effects of stress ratio and residual stresses on crack closure behaviour, including stress cycle
interaction effects in variable amplitude loading. These treatments show considerable promise in
predicting the effects of these different variables on fatigue crack behaviour but detailed
consideration is beyond the scope of these lectures.
6. CONCLUDING SUMMARY

Fracture mechanics procedures may be used to estimate the development of crack shape
and fatigue life by considering incremental crack growth at the deepest point and
surface points of a crack.

The effects of multiple initiation sites and the subsequent coalescence can be taken into
account.

The failure assessment diagram methods may be used to determine final failure by
fracture or plastic collapse.

The fatigue treatments may be extended to variable amplitude loading and to include
residual stress, threshold and crack closure effects.

7. REFERENCES
[1] PD6493: Guidance on Methods for Assessing the Acceptability of Flaws in Fusion Welded
Structures, British Standards Institution, 1991.
[2] Chu, W. H., The Influence of Plate Thickness Geometry Variations on Fatigue Strength of
Fillet Welded Joints, MSc Thesis, UMIST, 1984.

316

[3] Thurlbeck, S. D., A Methodology for the Assessment of Weld Toe Cracks in Tubular
Offshore Joints, PhD Thesis, UMIST, 1991.

317

318