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

net/publication/266038561

Guideline for the Fatigue Assessment by Notch Stress Analysis for Welded
Structures

Conference Paper · July 2008

CITATIONS READS

170 6,211

1 author:

Wolfgang Fricke
Hamburg University of Technology
350 PUBLICATIONS 5,157 CITATIONS

SEE PROFILE

All content following this page was uploaded by Wolfgang Fricke on 11 January 2024.

The user has requested enhancement of the downloaded file.

 IIW-Doc. XIII-2240r2-08/XV-1289r2-08

IIW Recommendations for the Fatigue Assessment

by Notch Stress Analysis for Welded Structures

Wolfgang Fricke

Hamburg University of Technology

Ship Structural Design and Analysis

July, 2010

Abstract

The notch stress approach for fatigue assessment of welded joints is based on the highest elastic
stress at the weld toe or root. In order to avoid arbitrary or infinite stress results, a rounded shape
with a reference radius instead of the actual sharp toe or root is usually assumed. Different pro-
posals for reference radii exist, e.g. by Radaj who proposed a fictitious radius of 1 mm to consider
micro-structural support effects for steel. The present guideline reviews different proposals for ref-
erence radii together with associated S-N curves. Detailed recommendations are given for the nu-
merical analysis of the notch stress by the finite or boundary element method. Several aspects are
discussed, such as the structural weakening by keyhole-shaped notches and the consideration of
multiaxial stress states. Regarding the fatigue strength, appropriate S-N curves are presented for
different materials. Finally, four examples illustrate the application of the approach as well as the
variety of structures which can the analysed and the scatter of results obtained from different mod-
els.

Keywords: notch stress, fatigue strength, welded joint, notch radius, FEM
 2

Table of Contents
1 Introduction ................................................................................................................... 1
2 Background of the Approach ........................................................................................ 2
2.1 Overview ............................................................................................................. 2
2.2 Notch Rounding Approach .................................................................................. 2
2.3 Modified Notch Rounding Approach ................................................................... 3
2.4 Small Size Notch Approach for Plates with Thickness < 5 mm ........................... 4
2.5 Approaches for Multiaxial Stress States ............................................................. 4
3 Numerical Analysis of Notch Stresses .......................................................................... 6
3.1 Numerical Methods (FEM, BEM) ........................................................................ 6
3.2 Links to the Structural Stress Analysis ................................................................ 6
3.3 Modelling of the Weld for Notch Stress Analysis................................................. 7
3.3.1 Modelling of idealized weld profiles .......................................................... 7
3.3.2 Modelling of the actual weld profile .......................................................... 8
3.4 Stress Analysis using Mesh Refinement ............................................................. 9
3.5 Stress Analysis using Sub-models or Super-elements ..................................... 10
3.6 Special Aspects ................................................................................................ 12
3.6.1 2D vs. 3D Analysis ................................................................................. 12
3.6.2 Compensation for fictitious radius in thin-walled structures .................... 12
3.6.3 Consideration of misalignment and other imperfections ......................... 13
3.7 Parametric Formulae for Notch Stress Concentration Factors .......................... 14
4 Fatigue Strength ......................................................................................................... 15
4.1 Design S-N Curves for Reference Radius of 1 mm........................................... 15
4.2 Correction for Mild Weld Notches with Reference Radius of 1 mm................... 17
4.3 Design S-N Curves for Mild Weld Notches with Larger Radii............................ 18
4.4 Design S-N Curves for the Small Size Notch Approach with Reference Radius
of 0.05 mm for Plates < 5 mm Thick ................................................................. 18
5 Demonstration Examples............................................................................................ 20
5.1 Fillet Welded Cruciform Joint ............................................................................ 20
5.1.1 Description of the detail .......................................................................... 20
5.1.2 Description of the models and analysis results....................................... 21
5.1.3 Discussion of the results......................................................................... 23
5.1.4 Comparison with parametric formulae .................................................... 23
5.2 T-Joint between Two RHS Members with Weld Toe Failure ............................. 24
5.2.1 Description of the detail .......................................................................... 24
5.2.2 Description of the models and analyses results ..................................... 24
5.2.3 Discussion of the results......................................................................... 26
5.3 Fillet-Welded End Connection of a RHS Member with Non-Fused Weld Root
Faces ................................................................................................................ 27
5.3.1 Description of the detail .......................................................................... 27
5.3.2 Description of the models and analyses results ..................................... 28
5.3.3 Discussion of the results......................................................................... 30
5.4 Spot-Welds in an Automobile Door ................................................................... 31
5.4.1 Description of the structure..................................................................... 31
5.4.2 Description of the model ......................................................................... 32
5.4.3 Fatigue assessment ............................................................................... 33
6 Acknowledgement ...................................................................................................... 34
7 References ................................................................................................................. 35
 1

1 Introduction
Welded joints are frequently assessed with respect to fatigue by the S-N curve approach, using
S-N curves (also referred to as Wőhler curves) giving the design fatigue life for constant amplitude
loading and an appropriate damage accumulation rule to consider the effect of variable-amplitude
loading. The approach may be based on different types of stress at the critical welded joint, i. e.:

• nominal stress approach, based on the stress that excludes any stress increase due to the
structural detail or the weld

• structural hot-spot stress and other structural stress approaches, based on the stress con-
taining only the stress increase due to the structure, but not due to the local weld geometry
(usually refers to the weld toe)

• notch stress approach, based on the local stress at the weld toe or the weld root, assuming
ideal-elastic material behaviour and micro-structural support effects to a certain extent

• notch stress intensity approach, using the notch stress intensity factor (NSIF) at the weld
toe with zero radius as the fatigue parameter

The approaches use fracture of a specimen or structure as failure criterion. Alternatively, the elas-
tic-plastic notch strain approach may be applied, which is based on crack initiation in the material
and considers the plasticity effect on fatigue using relevant material properties. Additionally the
crack propagation approach is applied for computing fatigue lives up to a defined crack length. The
crack propagation approach based on fracture mechanics principles is in widespread use as an
alternative to the S-N curve approaches, assuming a fictitious or actual initial crack. The approach-
es have been described in various references, more recently to various degrees by Hobbacher
(2009), Niemi et al. (2006), and Radaj et al. (2006), showing their details, potentials and limits.

In comparison with the other approaches, the notch stress approach allows the effect of the local
weld geometry to be included directly in the stress so that different geometrical configurations can
be compared with each other and can even be optimised. Also, the so-called thickness effect is
implicitly considered, at least from the geometrical point of view.

As with all local approaches, experience is needed to apply the notch stress approach correctly, to
consider its limits and shortcomings and to draw reasonable conclusions from the results. This
guideline is intended to help in this respect. First the background of the approach, as outlined by
Radaj et al. (2006) and by further, more recent, publications, is summarized, with focus on the nu-
merical methods. Guidelines are then given for the numerical stress analysis and the fatigue clas-
ses or design S-N curves used in the fatigue strength assessment. Finally, some examples of prac-
tical applications that were analysed in a round-robin are described. One example includes the
application of selected approximation formulae for notch stress concentration factors which are still
sometimes applied, although the steadily improving numerical methods make direct analyses easi-
er.

The present guideline supplements the fatigue design recommendations of the International Insti-
tute of Welding (IIW), which have recently been updated (Hobbacher, 2009). Additional hints and
more recent findings are given regarding the notch stress approach, which is applied to an increas-
ing extent, but is still under development. The reader should keep this in mind when applying this
approach. Furthermore it is strongly recommended to make comparisons with the other approach-
es mentioned and to question any differences critically. In this way, the notch stress approach can
be expected to contribute to a rational fatigue assessment and to help in the design of sound, fa-
tigue-resistant welded structures.
 2

2 Background of the Approach

2.1 Overview
The notch stress approach considers the increase in local stress at the notch formed by the weld
toe or the weld root, based on theory of elasticity, i. e. without consideration of elastic-plastic mate-
rial behaviour. The micro-structural support effect of the material, which considers the effect on
fatigue behaviour of the inhomogeneous material structure under a stress gradient,, can be taken
into account by different hypotheses in the (elastic) notch stress approach:

• the stress gradient approach (Siebel and Stieler, 1955)

• the stress averaging approach, proposed by Neuber (Neuber, 1937, 1946 and 1968)
• the critical distance approach (Peterson, 1959)
• the highly stressed volume approach (Kuguel, 1961; Sonsino,1994 and 1995).

Only the last three hypotheses have found wide application to welded joints. The stress averaging
approach is mainly used in the form of fictitious notch rounding, Fig. 2.1, known also as effective
notch stress approach, while the critical distance approach employs the ratio of a material constant
and the notch radius to reduce the elastic stress concentration factor Kt to the fatigue notch factor
Kf.

In the following sections, approaches using fictitious notch rounding are described in more detail.

Fig. 2.1: Fictitious notch rounding (graph according to Hobbacher, 1996)

2.2 Notch Rounding Approach

The basic idea behind this approach is that the stress reduction in a notch due to averaging the
stress over a certain depth can alternatively be achieved by a fictitious enlargement of the notch
radius. Neuber (1968) proposed the following formula for the fictitious radius ρf:

ρf = ρ + s·ρ* (2.1)

where ρ = actual notch radius

s = factor for stress multiaxiality and strength criterion
ρ* = substitute micro-structural length
 3

In the approach proposed by Radaj (1990) for welded joints, the factor s is assumed to be 2.5 for
plane strain conditions at the roots of sharp notches, combined with the von Mises multiaxial
strength criterion for ductile materials. Values for the substitute micro-structural length ρ* for some
materials are given in Fig. 2.2. Considering typical welds in (low strength) steel, the choice of ρ* =
0.4 mm (for cast steel in the welded zone) is appropriate. Both factors result in an increase of the
actual radius by 1 mm to obtain the fictitious radius ρf according to eq. (2.1). The rounding is ap-
plied to both the weld toe and the weld root.

In a 'worst case' or conservative way, Radaj's approach is applied assuming an actual radius of
zero so that the fictitious radius, now considered as the reference radius, is rref = 1 mm, see Fig.
2.1. As the stress analysis results in the fatigue-effective stress, the approach is also called effec-
tive notch stress approach. S-N curves related to the reference radius r1.0 are discussed in Section
4.

Attempts have been made to derive a corresponding fictitious radius for aluminium alloys, which
would be expected to be much smaller than that for steels assuming the substitute micro-structural
length according to Fig. 2.2 (Sonsino et al., 1999). However, more recent investigations have
shown that within a rough approximation the same reference radius as for welded joints in steel
can also be used for welded joints in aluminium alloys (Morgenstern et al., 2004).

Fig. 2.2: Substitute micro-structural length dependent on yield limit for various materials
(Neuber, 1968)

2.3 Modified Notch Rounding Approach

Seeger and co-workers modified the notch stress approach in an attempt to obtain better defini-
tions of mean fatigue strengths and scatter ranges (Köttgen et al., 1991; Olivier et al., 1989 and
1994). Taking structural mild steel (St52-3 or S355) as an example, they fixed the notch rounding
to r = 1 mm independently of the actual radius which varied around this value. The notch stress
was determined for this value both at weld toes and roots without further corrections regarding the
micro-structural support effect.

The resulting fatigue strength, expressed in terms of elastic notch stresses, was derived from fa-
tigue tests and notch stress analyses of various T- and Y-joints, led to a simple design S-N curve,
which could be applied to the structural component under consideration. Further details are given
in Section 4.
 4

2.4 Small Size Notch Approach for Plates with Thickness < 5 mm
The approaches described above assume a notch rounding with a reference radius of 1 mm, which
may cause problems in thin structures, less than 5mm thick. In particular, the groove created by
this radius at the weld root causes a substantial reduction in cross-section and hence modifies the
stress distribution, which affects the results of the fatigue assessment.

Therefore, a small-size notch approach which uses a reference radius rref = 0.05 mm has been
proposed for assessing welded joints in thin-sheet material (steel or aluminium alloys). In a similar
way to the rref = 1 mm approach, alternative design S-N curves based on notch stresses calculated
assuming rref = 0.05 mm have been derived from fatigue test results from welded specimens, which
are applicable to relevant well-defined groups of welded joints and materials, as shown in Section
4.

2.5 Approaches for Multiaxial Stress States

If the welded joint is subjected to more than one stress component, e. g. to axial and shear stress-
es acting in the plate adjacent to the weld, it is necessary to distinguish between:

1. loading that produces constant principal stress directions, which is usually associated with
proportional stress components;

2. loading that produces principal stresses that change direction , i. e. non-proportional stress
components, as produced, and indeed generally simulated in tests, by out-of-phase loading
components

In the first case, the superimposed notch stress range can be computed directly and assessed in a
similar way to the case of uni-axial loading. For simplification, in such cases the largest principal
stress range which is expected to act within a sector of ±45 deg. approximately perpendicular to
the weld line may be considered as the relevant fatigue parameter. In this case, the equivalent
stress range, computed with the ranges of the notch stress components, is usually smaller. How-
ever, if the welded component is subjected to pre-dominantly in-plane shear load, the equivalent
von-Mises stress can be larger than the maximum principal stress. For such cases, which are
characterized by the second principal stress having a different sign, use of the equivalent von-
Mises stress is recommended, see also Table 2.1. Alternatively, the individual stress components
may be assessed by interaction formulae, e. g. the Gough-Pollard ellipsis for combined normal and
shear stresses (Sonsino and Wiebesiek, 2007) or even longitudinal stress. Reference is made to
Hobbacher (2009) and to more recent literature in this field as it is still under development.

If the notch stresses are computed with stress concentration factors applied to the different stress
components, account should be taken of the fact that the locations of the corresponding peak
stresses might differ.

Table 2.1: Stress Types to be used in the Case of Multiaxiality

First (max.) principal Equivalent von-Mises Interaction formula with nor-

stress range acting ap- stress using ranges mal and shear stress ranges
prox. ⊥ to weld line of stress components
Proportional loading Yes, if second principal Yes (with reduced Yes (see Hobbacher, 2009;
stress has same sign S-N curve, see 4.1) Sonsino & Wiebesiek, 2007)
Non-proportional loading No No Yes (see Hobbacher, 2009;
with changing principal Sonsino & Wiebesiek, 2007)
stress directions
 5

In the second case mentioned above, the fatigue damage may be different due to the changing
directions of the principal stresses with time. Therefore, at present it is generally recommended
that the notch stress ranges due to the individual stress components are assessed by an interac-
tion formula, as indicated in Table 2.1.
 6

3 Numerical Analysis of Notch Stresses

Usually, the notch stress is computed numerically using the finite element or the boundary element
method. As an alternative, published parametric formulae are available for standard cases which
have been derived mainly from numerical analyses. Both cases are considered in this Section.

3.1 Numerical Methods (FEM, BEM)

The objective of the numerical analysis is the computation of the stress concentration in the fa-
tigue-critical notch under specified loads assuming linear-elastic material behaviour. This is a rela-
tively simple task in view of the powerful methods available today, requiring mainly a sufficiently
fine discretization of the structure in the notch area. A linear-elastic analysis is in most cases suffi-
cient. However, effects of large displacements on the structural stress might be important, particu-
larly in thin-walled structures, in which case geometrically non-linear structural stress analysis
would be required. Contact problems may also require a non-linear analysis. However, contact
between non-welded root faces is not usually assumed, leading mostly to conservative results.

The finite element method (FEM) is the most likely choice for stress analysis today, where the
structure is subdivided into a large number of small elements which are connected at nodes. The
elements are characterized by definite shape functions for the displacements between the nodes
and by the elastic material properties. An equation system is established and solved with respect
to the nodal displacements (inclusive of rotations where applicable). These are later used to com-
pute the stresses in the elements at the integration points depending on the shape function (e. g.
linear or quadratic) and extrapolated to the nodal points. Details of the method and the different
element types can be found in various textbooks, such as Hughes (1987), Bathe (1995), Zienkie-
wicz and Taylor (2000) or Cook et al. (2002).

The boundary element method (BEM) utilizes the possibility of mapping the elastic behaviour of a
body on its surface so that only this has to be described and discretized by elements. It allows a
much simpler input of geometrical data. The unknowns are either the tangential and normal dis-
placements at the boundaries, which can be linear or quadratic within the elements, or, if these are
prescribed, the tangential and normal external stresses. After solving the equation system, the dis-
placements and internal stresses are determined at the boundaries as well as within the body. De-
tails of the method are described among others by Brebbia et al. (1984), Cruse (1988), Banerjee
(1994) and Gaul et al. (2003).

In view of its widespread application, most of the following descriptions will refer to the finite ele-
ment method. The recommendations can accordingly also be transferred to the boundary element
method by using models with corresponding element sizes and properties at the boundaries.

The stresses may be solved by a 3D or a 2D analysis, the latter being restricted to special cases,
of the kind mentioned in Section 3.5, where variations of the geometry and loading in the 3rd direc-
tion can be neglected. In this case, plane strain conditions are usually assumed as biaxial stresses
occur in the notches due to restraint in the 3rd direction.

3.2 Links to the Structural Stress Analysis

The stress concentration effect of a weld toe is limited to the vicinity of the local notch, so that it
can be considered by a weld factor Kw superimposed on the undisturbed structural stress σhs at the
hot spot. This allows the use of a structural hot-spot stress analysis with its corresponding model
as the first step in the notch stress analysis. The latter can be performed with a refined local model
using the sub-model technique where the stresses or displacements taken from the structural
stress analysis are applied as boundary conditions. Alternatively, the super-element technique may
be applied where the refined local model is inserted in the overall model. Also structural and/or
notch stress concentration factors can be applied if they are available for the welded detail under
consideration.
 7

3.3 Modelling of the Weld for Notch Stress Analysis

When modelling the weld, two cases of application can be distinguished:

1. an idealized weld profile, characterized by a constant flank angle and the radius of the weld
toe or root either by a reference radius (e. g. rref = 1 mm) or the actual notch radius

2. the actual weld profile, originating from measurement, which is idealized by a shape con-
sisting of circular and straight parts

These two cases are dealt with in more detail in the following.

3.3.1 Modelling of idealized weld profiles

Fig. 3.1 shows the rounding of weld toes of butt- and fillet welds, characterized by the weld toe
radius r and the weld flank angle θ. The rounding of the roots of non-penetrating fillet welds is
shown in Fig. 3.2. The length of the non-welded root faces is retained by locating the vertex point
of the circle at the weld root, see Figs. 3.2 and 3.3. Two shapes are possible, the keyhole notch
and the U-shaped notch, see Fig. 3.2b and c. The latter reduces the high stress concentration in
the keyhole notch for loading parallel to the non-welded root faces but it should be noted that it can
also lead to an underestimation of the required notch stress for assessing potential fatigue failure
in the weld throat (Fricke et al., 2009). An idealisation for a permanent backing bar of a butt joint
with keyhole notches is shown in Fig. 3.4.

In thin-walled structures it might be necessary to compensate for the increase in net section stress
due to the notch depth, see sub-section 3.5.2.

Fig. 3.1: Rounding of weld toes of a butt-weld, a fillet weld and a butt weld with undercuts
of depth d

Fig. 3.2: Rounding of the weld root of a non-penetrating fillet weld

by a keyhole and an U-shaped notch
 8

Fig. 3.3: Notch rounding of the weld root of a Y-joint

Fig. 3.4: Rounding of the weld root of a butt joint with

permanent backing bar with keyhole notches

3.3.2 Modelling of the actual weld profile

A point-wise measured, actual weld profile is usually smoothened in a first step by arranging
curved and straight parts of a surface with tangential transitions to avoid unrealistic stress peaks in
concave corners. A radius followed by a straight line may be enforced at the weld toe to derive the
primary geometric parameters, i. e. the weld toe radius r and angle θ, Fig. 3.5. Procedures for the
derivation of these parameters have been developed (Lieurade et al., 2003).

Fig. 3.5: Approximation of a measured weld profile by an idealized shape

Depending on the type of the approach chosen, the radius might have to be fictitiously enlarged to
consider micro-support effects of the material.
 9

3.4 Stress Analysis using Mesh Refinement

Usually, the discretization of the structure is performed such that a relatively coarse overall mesh is
established, which is locally refined in the neighbourhood of the notches under consideration. The
overall meshing must take into account the force flow and deformation behaviour of the overall
structure in order to obtain appropriate loading of the notched areas. In particular, the bending be-
haviour of structures has to be considered by appropriate elements and subdivision.

The mesh should be gradually refined towards the notched area, avoiding large steps in element
size and excessive element distortion. The objective is to produce a mesh at the notch which is
fine enough to model the steep stress increase normal and tangential to the notch surface and so
yield the notch stress with sufficient accuracy. Figs. 3.6 and 3.7 show examples of meshes that
achieve this objective.

For weld toes with an angle θ = 45° (Fig. 3.6), it is recommended that at least three elements with
quadratic displacement function are arranged along the curve of the radius, which means a maxi-
mum element length of 0.25 mm in the case of a 1mm radius. The number of elements should
substantially be increased if they are of the linear displacement function type. Since the stress
peak might be close to the transition between the radius and adjacent straight parts, it is advisable
to continue with the same element length on the straight side before gradually increasing it.

Similarly, because there is also a steep stress gradient normal to the surface, the element length
should be the same or better smaller in this direction. By arranging approximately quadrilateral
elements, the mesh will become gradually coarser as shown in Fig. 3.6.

Fig. 3.6: Typical mesh for the notch stress analysis with elements
having quadratic displacement function at a weld toe

In the case of a 3D model, the element subdivision in the third direction, i. e. along the weld length,
should be chosen in accordance with the stress gradient expected. If the gradient is small, relative-
ly long elements should be suitable.

The recommended element sizes for the two radii mentioned in the previous Section are summa-
rized in Table 3.1. The numerical error is of the order of just a few percent if these limit values are
used (investigations showed up to 2 % in the case of quadrilateral elements with quadratic dis-
placement function, see Baumgartner and Bruder, 2011).

Corresponding recommendations are embodied in the mesh shown in Fig. 3.7, where the weld root
region is modelled by a keyhole shape.
 10

Fig. 3.7: Typical mesh for a notch stress analysis with elements
having a quadratic shape function at a weld root modelled with a keyhole shape

Notch stresses are evaluated on the rounded surface of the notch, where a plane stress state ex-
ists, using the tangential and normal stresses in the section and the shear stresses on the notch
surface. From these, principal or equivalent stresses can be derived.

Usually, nodal stresses are obtained from the computer program output. Special attention is drawn
to the use of simple linear elements with constant stress distribution, where appropriate stress ex-
trapolation to the free notch surface might be necessary. A check should be made that the stress
distribution approaching the surface is smooth, otherwise the mesh might be too coarse.

Table 3.1: Recommendations for element sizes (along and normal to notch surface)

Element type Relative size for size for No. of ele- No. of ele- Estimated
(displacement func- size r = 1 mm r= ments over ments over error*
tion) 0.05 mm 45 deg arc 360 deg
arc
quadratic (e.g. with ≤ r/4 ≤ 0.25 mm ≤ 0.012 mm ≥3 ≥ 24 ≈ 2%
mid-side nodes)
linear ≤ r/6 ≤ 0.15 mm ≤ 0.008 mm ≥5 ≥ 40 ≈ 10%
*) acc. to Baumgartner and Bruder (2011); may be improved by refinement in depth direction

3.5 Stress Analysis using Sub-models or Super-elements

In view of the extensive mesh generation effort required for mesh refinement, the analysis is fre-
quently performed using the sub-model or alternatively the super-element technique.

In the sub-model technique, the analysis is performed in two steps:

1. Computation of deformations and stresses in an overall, coarse model suitable for the
computation of structural stresses (example in Fig. 3.8a, being further described in 5.3)

2. Computation of notch stresses in a fine-meshed sub-model, which is loaded by prescribed

displacements or stresses at the boundaries taken from the overall model; in the example
in Fig. 3.8, the uppermost flat part is considered because it is the most highly stressed un-
der bending of the hollow section about its horizontal axis.
 11

Fig. 3.8: a) Overall finite element model of a fillet-welded end joint of a rectangular
hollow section and b) extent of a volumetric sub-model and c) 2D-sub-model

The sub-model may be a volumetric model (Fig. 3.8b) or, if the stresses are acting mainly in the
plane normal to the weld line, a 2D-model assuming plane strain conditions (Fig. 3.8c). The re-
quirements on element sizes etc. are the same as those given in the preceding sub-section. The
extension of the sub-model should be chosen such that boundary effects on the notch stresses are
negligible.

The loading of the sub-model consists in most cases of the application of the nodal displacements
originating from the overall model. Local loads such as pressure also need to be considered. As
the overall model consists of a coarser mesh, the displacements must be interpolated between the
nodal points in an appropriate way. The sub-model technique that includes the interpolation to
achieve compatible displacements at the boundary of the sub-model is supported by some finite
element programs.

The alternative to the prescription of boundary displacements is the application of sectional forces
and moments or stresses at the boundaries, again taken from the overall model. This method,
which has frequently been applied to 2D sub-models, has the advantage that the loading of the
connected structural components is clearly evident. The loading has to be in equilibrium as far as
possible. Non-equilibrated forces or moments can occur due to shear forces acting in the cross
section. These have to be taken by supports which should be arranged in an area where their ef-
fect on the notches is minimal.

A very important requirement for obtaining the same results as with mesh refinement is that the
overall model should have the same stiffness in the relevant part as the sub-model. Otherwise, the
notch stresses will be wrong. For example, if the overall model has a higher stiffness in the area
under investigation than the sub-model, which is frequently the case, the notch stresses

• will be too small if prescribed displacements are arranged at the boundaries, because the
latter are under-estimated by the stiffer overall model

• might be too large if forces and moments (or internal stresses) are applied at the bounda-
ries, because these can be overestimated by the overall model (particularly in the case of
statically indeterminate structures).
 12

The relevant part of the overall model should be created in such a way that local deformations can
fully develop, in particular local bending which might be partly prevented by the element behaviour.
In the case of non-penetrating welds, the non-fused region should be included in the overall model
and the weld should be able to deform in the same way that it does in the sub-model. Fig. 3.8 gives
an example of such a global model, showing the refined mesh of the fillet weld.

Appropriate modelling can be checked by comparing the stresses at the boundaries of the sub-
model with those in the overall model, which have to be the same if the displacements at the
boundaries are prescribed. If the forces and moments (or internal stresses) are prescribed, the
boundary displacements must be the same.

The super-element technique (and the historically earlier substructure technique) differs from the
sub-model technique insofar as the local model is inserted into the overall model as a ‘super-
element’. This avoids the problems mentioned with respect to different stiffness of the sub-model
and the respective part of the overall model. Here it is also important that the extent of the super-
element is chosen in such a way that boundary effects on the notch stresses are negligible. Fur-
thermore, compatible displacement functions have to be ensured at the connection between the
super-element and the surrounding structure, which is possible with special coupling elements in
some programs. It is recommended to check the accuracy of the results.

3.6 Special Aspects

3.6.1 2D vs. 3D Analysis
In several cases, a 2D analysis is sufficient, which greatly simplifies the modelling. The following
pre-requisites should be fulfilled for a 2D analysis:

• The loading of the weld should act mainly in the plane perpendicular to the weld line, which
means that normal and shear stresses along the weld can be neglected

• The geometry of the weld should not change in the area considered so that a representa-
tive geometry can be chosen

As mentioned before, plane strain conditions should be imposed on the 2D model as the contrac-
tion in high stress concentration notches is restrained normal to the plane modelled. As a conse-
quence, a biaxial stress state occurs at the notch surface.

Methods for determining the notch stress concentration at welds in cases of multiaxial loading,
inclined welds and weld ends based on 2D models (mainly cross-sectional) are described by Radaj
et al. (2006), pp. 135-142 and 150-152. These methods may be appropriate where a 2D notch
stress analysis by FEM or BEM is too expensive or time-consuming.

3.6.2 Compensation for fictitious radius in thin-walled structures

As noted earlier, the introduction of the fictitious 1mm radius can increase the net section stress in
thin material and thus lead to an over-estimate of the notch stress. The problem is particularly
acute at the weld root. To compensate, Radaj et al. (2006) give the following reduction factors to
be applied to the fatigue notch factor Kf (and hence, to the notch stress) for cases based on inter-
nal forces and moments, where it is necessary to distinguish between a one-sided notch (eq. 3.1
and Fig. 3.9a,b) and a two-sided notch (eq. 3.2 and Fig. 3.9c,d):

(1 − ρ f / t )2
Kf * = Kf (3.1)
1 + (1 + σ l / σ u ) ρ f / t
 13

2 (1 − ρ f / t )2
Kf * = Kf (3.2)
1 + (1 + σ l / σ u ) ρ f / t

where Kf* is the reduced fatigue notch factor and the other terms are described in Fig. 3.9. The
amount of the upper stress σu has to be larger than or equal to the lower stress σl. It should be
noted that the correction might be too large e. g. in cases with additional restraint of the surround-
ing structure. In such cases an additional computation with a shifted notch, keeping the net section,
is recommended to derive the correction factor.

Additional modifications have been proposed to consider the effect of gap closure in overlap joints
and for the interaction of fictitious notch radii that are too close together.

Fig. 3.9: Modification of the membrane and bending stress by fictitious notch rounding with
undercut at one side (a,b) or at both sides (c,d), after Radaj et al. (2006)

3.6.3 Consideration of misalignment and other imperfections

In numerical stress analyses, perfectly-aligned structures are usually assumed, However, axial and
angular misalignments may substantially increase the local stresses and have, therefore, to be
considered in the notch stress analysis (as also in the structural hot-spot stress analysis), see
Hobbacher (2009).

Generally, two alternatives exist for the consideration of misalignment effects:

1. Inclusion of the misalignments in the model, based for example on specified fabrication tol-
erances, measured values or worst-case assumptions

2. The application of stress magnification factors, which are usually related to the applied axial
stress components. Typical stress magnification factors for those joints which are particu-
larly affected by misalignment are given by Hobbacher (2009)

If the numerical analysis is performed in two steps, i. e. the computation of the structural stresses
or overall deformations in the first step, from which the notch stresses are derived in the second
step, the magnification factors are only applied once.

Concerning other imperfections, it should be noted that the fatigue classes presented in Section 4
were derived on the basis of idealized numerical models of welds with relatively good quality toe
profiles, with no significant undercut or other features that introduce sharp section changes. More
 14

severe imperfections such as deeper undercut or cold laps need to be modelled accordingly, e. g.
as shown in Fig. 3.1c (Gosch and Petershagen, 1997). Both types of imperfections can also be
analysed using the crack propagation approach.

3.7 Parametric Formulae for Notch Stress Concentration Factors

Parametric formulae have been published for several standard cases, such as butt joints, T-joints
and both load-carrying and non-load-carrying cruciform joints. Examples include Anthes et al.
(1993), Brennan et al. (2000), Iida and Uemura (1995), Lawrence et al. (1981), Lehrke (1999),
Radaj and Zhang (1991), Rainer (1983) and Tsuji (1990). An application is shown in Section 5.1.4.
 15

4 Fatigue Strength
4.1 Design S-N Curves for Reference Radius of 1 mm
The notch stress approach was first linked only to the endurance limit, defined as the fatigue
strength at 2·106 cycles. In this respect, for consideration of welded joints, Radaj (1990) proposed
application of the fatigue strength of low strength steels in the non-machined condition, which cor-
responds to a stress range of 240 MPa at 2·106 cycles for a stress ratio R = 0 and a survival prob-
ability Ps = 90%.

The extensive programme of fatigue tests on welded joints performed in connection with Seeger's
approach (Olivier et al., 1989 and 1994), see Section 2.2, provided further information about the
scatter in fatigue lives and the effect of the stress ratio R. Fig. 4.1 shows good correspondence to a
Gaussian distribution.

Fig. 4.1: Notch stress fatigue strength at 2⋅106 cycles for stress-relieved T- and Y-joints
from test results by Olivier et al. (1989 and 1994)

The original endurance limit approach was subsequently converted to one based on S-N curves on
the basis that such curves should be of the form ∆σ m·N = C, where ∆σ is the notch stress range
and the constant C = (FAT)m·2·106, FAT being the fatigue strength at 2·106 cycles. As generally
assumed for welded joints, the slope exponent of m = 3 was selected. This allows the IIW fatigue
classes defined for the nominal, structural hot-spot and effective notch stress approaches to be
compatible with each other, as their S-N curves all have the same slope.

The FAT value to be used for the effective notch stress approach S-N curve was derived from a
statistical analysis of relevant fatigue test results obtained from welded joints. In order to allow for
the presence of high tensile residual stresses in actual welded components and structures, particu-
lar attention was paid to fatigue data obtained under high stress ratios, such as those at R = 0.4 in
Fig. 4.1. Further evaluation of these and other results (Hobbacher, 2008, see Fig. 4.2), including
also large-scale models (Fricke and Paetzold, 2010), shows a characteristic fatigue strength
(stress range) of FAT 225 for a survival probability Ps = 97.7%. However, some evaluations have
shown a number of fatigue test results, particularly for butt welds, below the FAT 225 design S-N
curve (Pedersen et al., 2010; Kranz and Sonsino, 2010). This might be due to a lower stress con-
centration and/or correspondingly shallower S-N curve for the joints concerned (see also 4.2).
 16

For aluminium alloys, a corresponding value of FAT 71 (Morgenstern et al., 2004) and for magne-
sium a value of FAT 28 may be assumed (Karakas et al., 2007). The values are summarized in
Table 4.1. It should be noted that these fatigue classes and the assumption that the S-N curve has
a slope of m = 3 are only valid for relatively sharp notches with a fictitious rounding of 1 mm, as
proposed in Seeger's approach and as 'worst case' in Radaj's approach.

Fig. 4.2: Re-analysis by Hobbacher (2008) of test Fig. 4.3: Stress distribution
data by Olivier et al. (1989) and Köttgen et al. at a weld toe
(1991) for T- and Y-joints subject to different
loading modes with R ≥ 0.4, root and toe failure

Table 4.1: Characteristic fatigue strength for welds of different materials

based on effective notch stress with rref = 1 mm (maximum principal stress)

Material Characteristic fatigue strength Reference

6
(Ps = 97.7%, N = 2·10 )
Steel FAT 225 Olivier et al. (1989 and 1994)
and Hobbacher (2008)
Aluminium alloys FAT 71 Morgenstern et al. (2004)
Magnesium FAT 28 Karakas et al. (2007)

The scatter in the fatigue data that led to the recommendations in Table 4.1 amounted to a scatter
ratio of 1.5 between the fatigue strengths for survival probabilities of 90% and 10%, which corre-
sponds to a standard deviation of log∆σ of 0.0687 and with the slope exponent of the design S-N
curve of m = 3 to a standard deviation of logN of 0.206. This allows alternative S-N curves based
on other survival probabilities to be derived.

It is emphasized that the FAT classes are based on the maximum principal stress in the notch. If
the von Mises equivalent stress is used, a correspondingly smaller FAT class applies, because the
equivalent stress is usually smaller in sharp notches. This subject is currently under investigation
but meanwhile a reduction by at least one fatigue class is recommended.

It should be noted that the so-called thickness correction, which is applied in the nominal and struc-
tural hot-spot stress approaches to welded plates more than 25mm thick, is already included in the
effective notch stress. This is because the notch stress is governed by the ratio r/t, such that it in-
creases with increased thickness t. Regarding consideration of misalignment and weld imperfec-
tions, note the remarks in 3.6.3.
 17

4.2 Correction for Mild Weld Notches with Reference Radius of 1 mm

A problem arises for relatively mild notches which, despite the radius of 1 mm, have a low notch
stress concentration factor. Such cases have not been verified by the aforementioned fatigue tests.
The problem can easily be understood by considering transverse butt welds with almost no weld
overfill. The corresponding weld toe notch stress concentration factor will be close to unity but the
fatigue class is certainly far below FAT 225. Mild notches may generally occur at weld toes that
have been ground, have small flank angles and/or are in thin plates.

The notch factor Kw of a weld is defined as the ratio of the effective notch stress σk to the structural
stress σs, see also Fig. 4.3:

Kw = σk/σs (4.1)

where the structural stress can be determined either with the methods proposed elsewhere (e. g.
Niemi et al., 2006) or from the notch stress distribution, taking the stress value at a distance of
2 mm from the transition between straight and curved part (without further thickness correction).

It is proposed as a preliminary approach to assume a notch factor Kw of at least 1.6 in connection

with the aforementioned fatigue classes. This means that both the structural hot-spot stress and
the effective notch stress at a weld toe need to be checked and the latter assumed to be 1.6 times
the former if it is found to be less.

As with the nominal and structural hot-spot stress approaches, the fatigue strength of a welded
joint is limited by the parent material S-N curve in the lower endurance regime (e.g. FAT 160 S-N
curve with a slope exponent m = 5 for steel). To allow for this in the effective notch stress ap-
proach, the fatigue strength of the parent material must additionally be checked using the structural
stress σs at the weld toe and the corresponding material S-N curve.

Fig. 4.4: S-N curves of fatigue classes in terms of nominal,

structural (FAT 160) and notch stress (FAT 225) compared with
FAT class 160 x weld shape factor Kw (Radaj et al. 2008)
 18

Fig. 4.4 compares the various IIW S-N curves for steel to illustrate how the S-N curve governing
the fatigue strength varies with endurance. The parent material FAT160 curve multiplied by the
notch factor Kw intersects the effective notch stress FAT 225 curve in this example at approximate-
ly 200,000 cycles, which means that the parent material governs design for smaller numbers of
cycles. Obviously, this situation is particularly relevant in cases where Kw is small.

Finally it should be noted that the slope of the S-N curve is generally shallower for mild notches.
This effect is at least partly considered by the aforementioned approach. Nevertheless, an ap-
proach with slope exponents changing with notch severity might be developed in future to better
consideration of the local influencing factors.

4.3 Design S-N Curves for Mild Weld Notches with Larger Radii
Larger notch radii may be relevant for some welding processes or for welds subject to post-weld
improvement treatment.

The S-N curves and fatigue classes mentioned in Section 4.1 are only valid for the reference notch
radius of 1 mm. In view of the larger reference radius that results from the addition of 1mm to the
actual radius (eq. 2.1), the notch stress needs to be assessed against a slightly reduced fatigue
class. The characteristic fatigue strength proposed by Radaj for non-machined steel, i. e. ∆σk =
240 MPa for Ps = 90%, yields a value of 215 MPa for Ps = 97.7% if the usual standard deviation is
assumed. A suitable fatigue class in the present IIW scheme would be FAT 200. A similar value
was found in an evaluation of fatigue tests with brackets using the actual notch radius enlarged by
1 mm (Fricke and Kahl, 2007). It should be noted that such a fatigue class is still only valid for rela-
tively sharp weld toe radii, e. g. 1 - 3 mm. This means that the procedure outlined in Section 4.2
and the assumption of a minimum notch factor Kw = 1.6 is still recommended for use in conjunction
with the lower FAT200 S-N curve.

The approach described in Sections 4.1 and 4.2 has not been verified for larger weld toe radii as
might arise in, for example, joints improved by grinding, peening or TIG-dressing. The logical ap-
proach is to assess the notch stress computed for such cases using the S-N curve for the weld or
parent metal, whatever is relevant. However, since this may be affected by the post-weld treat-
ment, a general recommendation is difficult and relevant published literature should be consulted.
In addition, as described in Section 4.2, the parent material in front of the weld toe must be
checked using the structural hot-spot stress and the parent material S-N curve.

Alternatively, the nominal or structural hot-spot stress approach with corresponding FAT classes
may be applied to the welded joint (Hobbacher, 2009).

4.4 Design S-N Curves for the Small Size Notch Approach with Refer-
ence Radius of 0.05 mm for Plates < 5 mm Thick
The small-size notch approach was originally developed for spot welds in thin sheets and later ex-
tended to thin-sheet lap joints. Corresponding design S-N curves have been derived from fatigue
test results obtained from welded specimens by Eibl et al. (2003) and Karakas et al. (2007), yield-
ing the FAT classes given in Table 4.2. Again it should be born in mind that the notch stresses are
theoretical values without consideration of local yielding of the material.

As before, the fatigue classes are based on the maximum principal stress in the notch of a 3D
structure. Equivalent stresses should be assessed using a lower fatigue class.
 19

Table 4.2: Characteristic fatigue strength for welds of different materials

based on theoretical notch stress with rref = 0.05 mm (maximum principal stress)

Material Characteristic fatigue strength Reference

6
(Ps = 97.7%, N = 2·10 )
Steel FAT 630 Eibl et al. (2003), Sonsino
(2009)
Aluminium FAT 180 Eibl et al. (2003), Karakas et al.
alloys (2007) , Sonsino (2009)
Magnesium FAT 71 Karakas et al. (2007) , Sonsino
(2009)
 20

5 Demonstration Examples
The following presents the results of a round-robin study (Fricke, 2006) performed by members of
IIW Working Group XIII-3 ‘Stress Analysis’. The objective was to identify sources of errors and
scatter in results by performing the stress analysis for well-defined cases using different modelling
techniques, analysis methods and computer programs. The following three structural details were
considered:

1. A cruciform joint with fillet welds similar to the left part of Fig. 2.1 (2D case)
2. A T-joint between two rectangular hollow section members (3D case)
3. A fillet welded end connection of a rectangular hollow section member with non-fused weld
root faces as shown in Fig. 3.7 (3D case)

A fourth example from the automotive industry illustrates the application of the small size notch
approach. Further demonstration examples for notch stress analyses on welded joints are given by
Radaj et al. (2006). Here, several design-related notch stress evaluations show the differences
between the notch stresses in different weld types applied to the same type of welded joint.

5.1 Fillet Welded Cruciform Joint

5.1.1 Description of the detail
Fig. 5.1 shows the geometry and loading of the cruciform joint made from 12 mm thick plates. The
fillet welds have a flank angle of 45 degrees and a throat thickness of 5 mm. The non-fused root
faces, indicated by horizontal lines in Fig. 5.1, are assumed to have a length equal to the plate
thickness (12 mm). Two alternative load cases (LC) are considered:

LC 1: Unit nominal stress σ1 in the continuous plate, i.e. the fillet welds are non load-carrying
and the most likely sites for fatigue cracking are the weld toes on that plate

LC 2: Unit nominal stress σ2 in the attached plates, i.e. the fillet welds are load-carrying, in
which case the potential sites for fatigue cracking are the weld toes in these plates or
the weld roots.

Notch stresses are therefore required at the weld toes and at the weld root for a reference notch
radius of rref = 1 mm in each case.
 21

σ2

12

5
σ1 σ1

12

45°

σ2

Fig. 5.1: Fillet welded cruciform joint

5.1.2 Description of the models and analysis results

In total seven analyses were performed, using the 2D finite element (FEM) and boundary element
(BEM) methods, see Table 5.1. In all cases, the double-symmetry was utilised by analysing only a
quarter of the structure. In five analyses, the weld root was rounded with a keyhole shape similar to
Fig. 2.1 (also shown in the left part of Fig. 5.3), retaining the actual length of the gap. In the two
remaining analyses, a U-shape with a gap width of 2 mm was used (see also the right part of Fig.
5.3).

Fig. 5.2 shows some typical meshes. The notches rounded with a radius of 1 mm were modelled
using a large number of elements for increased accuracy. Figs. 5.3 and 5.4 give examples of typi-
cal stress results. The values given were obtained for a unit stress, i.e. they can be interpreted as
stress concentration factors (SCFs) or fatigue notch factors Kf. It is interesting to note that the dif-
ferences at the weld root between the keyhole and U-shape notches are rather large for LC 1,
where the stress acts parallel to the gap and the weld is nominally non-load-carrying. The keyhole
notch may overestimate the real notch effect as the upper and lower parts of the notch do not exist
in the real structure. Compared e.g. with the crack propagation approach, where the stress intensi-
ty factor is rather small in this case, the U-shape seems to be the more realistic. On the other
hand, this possible weld root stress overestimation is not relevant in the detail investigated, as the
stress at the weld toe is more critical.
 22

Fig. 5.2: Examples of FE meshes with keyhole shape (quarter models)

Fig. 5.3: Maximum principal stress distribution and notch stresses for LC 1 (Participant A)

Fig. 5.4: First principal stress distribution and notch stresses for LC 2 (Participant A)
 23

Table 5.1: Analysis results for the cruciform joint with fillet welds

Element SCFs for LC 1 SCFs for LC 2

Partici- Weld Method / Element
length at
pant root Program type Toe Root Toe Root
toe / root
BEM / 0.05mm /
A Keyhole quadratic 2.56 1.95 4.58 5.56
BEASY ~ 0.16mm
FEM / 0.04mm /
B Keyhole linear 2.58 1.89 4.62 5.65
MSC Nastran 0.08mm
FEM / ~ 0.08mm /
C Keyhole quadratic 2.53 1.95 4.52 5.56
IDEAS 0.12mm
FEM / 0.017mm /
D Keyhole linear 2.42 1.99 4.05 5.40
MSC MARC ~ 0.16mm
FEM / 0.05mm /
E Keyhole quadratic 2.56 1.96 4.57 5.71
ANSYS 0.05mm
Coefficient of variation for meshes with keyhole root 2.5% 1.9% 5.3% 2.1%
0.05mm /
A U BEM / BEASY quadratic 2.57 1.35 4.80 5.64
~ 0.16mm
FEM / 0.05mm /
E U quadratic 2.56 1.47 4.80 5.72
ANSYS 0.05mm
Coefficient of variation for meshes with U-shaped root 0.3% 6.0% 0% 1.0%

For LC 2, the differences between the keyhole and U-notch are very small. It is interesting to note
that the notch stress at the weld toe is slightly affected by the modelling of the weld root. The U-
notch is connected to a weaker structure, causing increased stresses at the weld toe.

5.1.3 Discussion of the results

The results in Table 5.1 are fairly close together, yielding very small coefficients of variation. This is
due to the simple model and rather fine meshes. As mentioned before, fine meshes are required
not only along the notch surface, but also normal to it, in order to model the steep stress gradient in
the thickness direction, which was realised by all participants (see Fig. 5.2).

The stress deviates by more than 10% from the others in only one case, Participant D’s toe stress
for LC 2 (SCF = 4.05). The reason for this is not clear.

5.1.4 Comparison with parametric formulae

Parametric formulae given in the literature for the SCFs in cruciform joints may also be applied.
These can be applied directly to LC 2. However, it is possible to consider LC 1 as a cruciform joint
with full penetration welds and only calculate the stress at the weld toe. This is justified by the fact
that the non-fused root faces have only a small effect on the notch stress at the weld toe.

The results for three approximation formulae are compared in Table 5.2 with each other and with
the mean value of all computed SCFs from Table 5.1. The deviations are particularly large for the
results based on Lawrence’s formulae.

Table 5.2: Comparison between results from approximation formulae and computation

SCFs for LC 1 SCFs for LC 2

Method
Toe Toe Root
Formulae of Lawrence (1981) 2.21* 3.24 4.69
Formulae by Radaj and Zhang (1983) n.a. 4.15 5.45
 24

Formulae by Anthes et al. (1993), 2.77* 4.91 5.83

Mean value of computation (Table 5.1) 2.54 4.56 5.61
*) Computation for cruciform joint with full penetration welds

5.2 T-Joint between Two RHS Members with Weld Toe Failure
5.2.1 Description of the detail
The second detail comes from a competition organised by the American Society of Automotive
Engineers (weld challenge 1, ref. SAE, 2003) concerning the prediction of the fatigue life of welded
joints. The example, shown in Fig. 5.5, was a T-joint between rectangular hollow section (RHS)
members in 345 MPa yield steel. The main continuous member had a quadratic section with outer
dimensions of 4 x 4 inch (101.6 x 101.6 mm) while the connected member had a rectangular sec-
tion of 2 x 6 inch (50.8 x 152.4 mm). Both tubes were 0.312 inch (7.9 mm) thick and had corner
radii at mid-wall thickness of 0.486 inch (12.3 mm, or 16.3 mm at the outer surface. The weld leg
length was 5/16 inch (7.9 mm). No information was provided about the weld penetration.

A number of joints were fatigue tested under alternating loading (R = -1). The right end of the con-
tinuous RHS member and the end of the connected member were fixed as indicated in Fig. 5.6. A
horizontal force of 4,000 pounds (17.8 kN) was applied at the left end with a lever of 12.5 inch
(317.5 mm) related to the centre of the RHS, creating horizontal shear, bending and torsion in the
continuous member. The fatigue crack initiation point was at the upper weld toe in the connected
tube, as indicated by the arrow in Fig. 5.7. The mean fatigue life observed in the experiments was
about 75,000 cycles.

Numerical analysis and fatigue life assessment of the detail as part of the challenge were de-
scribed by Kyuba and Dong (2003).

5.2.2 Description of the models and analyses results

Due to the need for a 3D model of the joint, only two participants analysed this detail. As notch
rounding with 1 mm radius requires a very fine mesh, the finite element analysis was performed in
two steps, as described in Section 3.5:

1. Coarse mesh analysis with shell or solid elements resulting in structural hot-spot stresses
and deformations

2. Fine mesh analysis of the critical part using mesh refinement or the sub-model technique,
where the sub-model is loaded by the deformations at the boundaries taken from the
coarse model

Both participants performed the coarse mesh analysis using solid elements. Since precise details
of the weld were not known, it was modelled as both a full penetration and a fillet weld. Figs. 5.8 -
5.9 and Table 5.3 show details of the models.
 25

Fig. 5.5: Sketch of the RHS joint (SAE, 2003; units: inch)

Fig. 5.6: Boundary conditions and loading of the joint Fig 5.7: Crack location (SAE, 2003)

Table 5.3: Analysis results for the complex RHS joint (principal stress ranges)

Par- Program Element Element Full penetration weld Fillet weld (zero penetration)
tici- type length (at Struct. stress Notch stress Struct. stress Notch stress
pant notch) range [MPa] range [MPa] range [MPa] range [MPa]
A ANSYS Solid linear 0.1mm 350 702 410 877
B I-DEAS Solid quad. 0.07mm 340 729 450 971
Coefficient of variation 2.0% 2.7% 6.6% 7.2%
 26

Fig. 5.8: Solid model of participant A Fig. 5.9: Solid model of participant A
(fillet weld without penetration) (full penetration weld)
The resulting deformations and distribution of principal stresses are shown in Fig. 5.10. At the criti-
cal location, both normal and shear stresses act at the weld toe, resulting in changes in the direc-
tions of the principal stresses during load cycling, as illustrated in Fig. 5.11. Therefore, the individ-
ual stress components had to be extrapolated to the weld toe in this case, based on the recom-
mendations by Niemi et al. (2006). In the case of three elements over the plate thickness, the
structural stress was obtained by stress linearization in the plate thickness direction at the weld
toe.

The resulting structural hot-spot stresses in Table 5.3 are very similar. However, those for the fillet
weld case are significantly higher. This can be explained by changes in local bending due to the
increased eccentricity and flexibility of the fillet welded joint.

The effective notch stress was computed by participant A using the sub-model technique and by
participant B using a locally refined mesh. The sub-model incorporating the reference radius rref =
1 mm is shown in Fig. 5.12. The displacements at the boundaries were taken from the overall
mesh with three elements in the plate thickness direction (Fig. 5.9). The maximum principal stress
in the rounded weld toe was taken as the effective notch stress (Fig. 5.13). Again, the difference
between the stresses in the full penetration and fillet welds is relatively large (more than 20%).

5.2.3 Discussion of the results

The variation in the results for both the structural and notch stress analysis is considered to be
acceptable in view of the different models and analysis techniques. The type of welding in the ac-
tual test specimens (fillet or full penetration) was not clear so both types were analysed; quite dif-
ferent results were obtained. It seems most likely that the joints were made with fillet welds but that
some penetration occurred.

The corresponding mean fatigue lives for the two types of weld, based on the fatigue class FAT
225 and a standard deviation of logN = 0.2, were 85,000 and 63,000 cycles. These lives lay either
side of the actual mean life from the tests of 75,000 cycles, which is consistent with the actual
welds being partial penetration, as suggested above.
 27

Fig. 5.10: Deformations and principal Fig. 5.11: Distribution of principal stresses on
stresses in the RHS members the surface of the attached RHS member

Fig. 5.12: Sub-model of critical area Fig. 5.13: Stress distribution in sub-model
(Participant A) (Participant A, full penetration weld)

5.3 Fillet-Welded End Connection of a RHS Member with Non-Fused

Weld Root Faces
5.3.1 Description of the detail
The third detail is a fillet-welded connection between two rectangular hollow section (RHS) mem-
bers, see Fig. 5.14. The dimensions of the RHS were 120 mm x 80 mm x 6 mm (height x width x
thickness). The two hollow sections were connected to a 15mm thick intermediate plate using a
single-sided fillet weld. The nominal throat thickness a of the weld was 4.5 mm. The gap between
the intermediate plate and the hollow section was assumed to be zero, see “DET A” in Fig. 5.14.

The two load cases investigated are shown in Fig. 5.15, together with the boundary conditions. In
the stress analysis, the left end was fixed according to the symmetry conditions. In the tensile load
case, a uniaxial constant stress of 100 MPa was applied to the right end. In the bending load case
a distributed shear force of 17.25 N was applied to the right end creating the nominal bending
stress of 100 MPa.
 28

Fig. 5.14: Sketch of the fillet-welded end connection of an RHS member

Specimens of this type were fatigue tested either in tension or bending. Fatigue cracks generally
started at the weld root in both cases (Fricke and Kahl, 2006). Under tensile load, fatigue cracks
first appeared on the long side of the RHS, while under bending load the first crack appeared on
the upper short side, where the bending stress was highest..

Fig. 5.15: Boundary conditions for tensile (left) and bending loads (right)

5.3.2 Description of the models and analyses results

In total seven models were analysed. In addition to members of the IIW work group mentioned, two
participants of the 'Network of Excellence on Marine Structures', which is funded by the European
Commission, joined the round robin.

All participants applied the FE method and modelled the structure with solid elements, using the
keyhole shape notch for the weld root and taking the theoretical root location as the outer point of
the circle with reference radius rref.= 1 mm. Participant A used the sub-model technique, while all
others used a refined mesh in the overall model. Symmetry conditions were partly utilised. Table
5.4 gives an overview of the models and the resulting notch stresses (principal stresses).

Figs. 5.16 to 5.19 show some examples of the models used. In most cases, the elements around
the keyhole notch were rather small. The longest elements in the circumferential direction, as cho-
sen by participant E (see Fig. 5.17), showed an anomalous distribution of nodal stresses (Fig.
5.18).
 29

Table 5.4: Analysis results for the fillet-welded RHS joint (for nominal stress of 100 MPa)

Partic- Element Type (Displace- Element length along Max. notch stress [MPa]
Program
ipant ment function and shape) circumference Tensile LC Bending LC
A ANSYS quadratic (hexahedral) ~ 0.1 mm 888 637
B ANSYS quadratic (hexahedral) ~ 0.25 mm 867 601
C ANSYS quadratic (tetrahedral) ~ 0.4 mm 913 650
D ANSYS quadratic Not reported 907 642
E I-DEAS quadratic (hexahedral) ~ 0.45 mm 914 644
F ANSYS quadratic (tetrahedral) ~ 0.2 mm 864 620
G ANSYS quadratic (tetrahedral) ~ 0.3 mm 700 600
Coefficient of variation (all results) 8.7% 3.3%
Coefficient of variation (all results without G) 2.5% 2.9%

Fig. 5.16: Finite element model of Participant B

Fig. 5.17: Local mesh Fig 5.18: Distribution of nodal stresses along circumference
of Participant E (clockwise) for bending load case (Participant E)
 30

The model in Fig. 5.19 is of particular interest. The number of elements around the notch radius
was quite high and the element length normal to the surface seems to be appropriate for modelling
the steep stress gradient. However, near the most highly stressed plane of the weld, highlighted in
the figure, very large elements were used close to the notch surface (indicated by an arrow). Also
the element length along the weld line seems to be too large. Such large elements were not able to
model the high stress increase in the highlighted plane, resulting in smaller notch stresses com-
pared to all other results.

Fig. 5.19: Local mesh of Fig. 5.20: Distribution of the notch stress along the
Participant G weld root line (Participant A)

In agreement with the failures observed in the tests carried out in tension, the highest notch stress
occurred halfway up the long side of the RHS, where more local bending is observed in the weld
throat than along the short sides which are more restrained by the corners, see Fig. 5.21. In the
bending load case, the highest notch stress occurred in the middle of the upper weld where the
largest bending stresses act.

Another interesting aspect was observed in connection with the sub-model technique used by par-
ticipant A (Fig. 5.21). Initially, a rather coarse overall finite element model of the RHS member was
chosen, with only one element in the thickness direction of the wall and the fillet weld. However,
the notch stresses obtained from the sub-models were obviously too small. The reason was that
the local bending of the weld shown in Fig. 5.22 was different from that in the coarse overall model,
which was too stiff, yielding smaller boundary displacements for the sub-model. Therefore, the
overall model had to be refined by modelling the weld throat with at least two elements along the
weld throat section. This resulted in almost 10% higher stresses, as given in Table 5.4.

5.3.3 Discussion of the results

Although different modelling techniques were used, variations in the results were fairly small if
those of participant G are excluded due to the influence of elements that were too large. Another
finding, already mentioned, is that the choice of sub-modelling technique is a source of inaccuracy.
To achieve reliable results, the sub-model and the respective part of the overall model need to
have the same stiffness, as already mentioned in Section 3.

Fatigue lives obtained from fatigue tests of the joint (Fricke and Kahl, 2006) were above the FAT
225 S-N curve when plotted in terms of the notch stress.
 31

Fig. 5.22: Deformation of the

Fig. 5.21: Overall model and sub-models of Participant A sub-model for bending load

5.4 Spot-Welds in an Automobile Door

5.4.1 Description of the structure
The fourth example is taken from the automotive industry where thin sheets with thicknesses be-
tween 0.6 and 3 mm are used. Spot-welding is frequently applied, laser welding not yet to the
same extent.

The durability of lightweight structures is of high interest. A typical example is the assessment of
the durability of joint techniques in automotive doors and lids, as presented by de Bruyne and
Hoppe (2006). Fig. 5.23 shows a spot welded door which is dynamically loaded during closing at a
speed of 1.5 m/s.

Fig. 5.23: Illustration of the automobile door without the inner frame
(de Bruyne and Hoppe, 2006)
 32

The example is selected to show the main procedural steps without judging the quality of modelling
which would require a deeper insight into several details. Generally it can be stated that weld spot
modelling for strength and stiffness assessment as well as structural optimisation is a highly devel-
oped procedure in automotive engineering, see also Radaj et a. (2006, ibis pp. 373-376 and 427-
429).

5.4.2 Description of the model

Because of the relatively thin material, the small-size notch approach has been applied.

The spot-welds were analysed using super-elements, as shown in Fig. 5.24, being gradually re-
fined at the root notch. It should be noted that the mesh is somewhat coarser than recommended
in Table 3.1. The connection to the overall finite element model made from shell elements was
provided by 8 nodes at the four corners of each sheet, having six degrees of freedom. Special
coupling elements along the boundaries are used to improve the compatibility of the displacements
with those of the overall model. The technique allows the analysis of large spot-welded structures.
The topology of the super-element is independent of the sheet-thickness. The parametric super-
element is automatically adapted to the actual geometry of the spot-weld and the plate thickness-
es.

Fig. 5.24: Super-element of a spot-weld with reference radius rref = 0.05 mm

(de Bruyne and Hoppe, 2006)

Fig. 5.25: Variation in the maximum principal notch stress and estimated fatigue life
(de Bruyne and Hoppe, 2006)
 33

5.4.3 Fatigue assessment

The fatigue assessment was based on the mean S-N curve derived from fatigue tests of simple
specimens.

The variation in the maximum principal stress in the notch of a selected spot-weld for one door
closing (left part of Fig. 5.25) shows one large pulsating tension cycle, and this is taken to estimate
the fatigue life (right part of Fig. 5.25), resulting in this case in 4,000 cycles.

Fig. 5.26 displays details of the calculation model, the location of the critical spot-weld and an ex-
ample of a fatigue test specimen that showed cracks after 10,000 cycles. Both the critical location
and the fatigue life obtained from the calculation agreed quite well with the experimental findings.

Fig. 5.26: Details of the calculation model and a fatigue test specimen
(de Bruyne and Hoppe, 2006)
 34

6 Acknowledgement
The document is a product of several years of study by Working Group 3, 'Stress analysis' of IIW
Commission XIII. The Support by the Chairmen of Commission XIII, Prof. S.J. Maddox (UK) and
later Prof. G. Marquis (Finland), and of the Joint Working Group of IIW Commissions XIII/XV, Prof.
A. Hobbacher (Germany) is appreciated. Special acknowledgement is made to Prof. D. Radaj,
Prof. T. Seeger and Prof. C.M. Sonsino (Germany) as well as to Prof. J. Samuelsson and his co-
workers (Sweden) for their contributions, support and critical reviews. Also the participants of the
IIW round-robin including the contributors of the MARSTRUCT Network of Excellence, coming
from several countries (see Fricke, 2006), are thanked for their numerical analyses and comments.
F. de Bruyne and A. Hoppe are thanked for providing the automotive example.
 35

7 References
Anthes R J, Köttgen V B and Seeger T (1993): Kerbformzahlen von Stumpfstößen und Doppel-T-
Stößen. Schweißen und Schneiden 45(12), 685-688.
Banerjee P K (1994): The Boundary Element Methods in Engineering. McGraw-Hill, New York.
Bathe, K J (1996): Finite Element Procedures. Prentice Hall, Englewood Cliffs (2nd Ed.).
Baumgartner J and Bruder T (2010): An efficient meshing approach for the calculation of notch
stresses. IIW-Doc. XIII-2369-11, International Institute of Welding.
Brebbia C A, Telles J C and Wrobel L C (1984): Boundary Element Techniques. Berlin: Springer-
Verlag.
Brennan F P, Peleties P and Hellier A K (2000): Predicting weld toe stress concentration factors for
T and skewed T-joint plate connections. Int. J. Fatigue, 22(7), 573-584.
de Bruyne F and Hoppe A (2006): Virtual durability assessment for joint techniques in automotive
doors and lids (in German). Proc. 13th Int. Kongress Berechnung und Simulation im Fahrzeug-
bau., VDI-Wissensforum Düsseldorf.
Cook R D, Malkus D S, Plesha M E and Witt R J (2002): Concepts and Applications of Finite Ele-
ment Analysis. John Wiley, New York (4th Ed.)
Cruse T A (1988): Boundary Element Analysis in Computational Fracture Mechanics. Dordrecht:
Kluwer Academic.
Eibl M; Sonsino CM; Kaufmann H and Zhang G (2003): Fatigue Assessment of Laser Welded Thin
Sheet Aluminium. Int J Fatigue, 25, 719–731
Fricke W (2006): Round-Robin Study on Stress Analysis for the Effective Notch Stress Approach.
Welding in the World 51, No. 3/4, 68-79.
Fricke W and Kahl A (2006): Fatigue assessment of weld root failure of hollow section joints by
structural and notch stress approaches. Proc. of Int. Symp. on Tubular Structures (ISTS'11),
Quebec.
Fricke W and Kahl A (2007): Local stress analysis of welded ship structural details under consider-
ation of the real weld profile. Proc. Int. Symp. on Practical Design of Ships and Floating Struc-
tures, Houston.
Fricke W, Paetzold H and Zipfel B (2009): Fatigue tests and numerical analyses of a connection of
steel sandwich plates. Welding in the World 53 (2009), No. 7/8, S. R151-R157
Fricke W and Paetzold H (2010): Full-scale fatigue tests of ship structures to validate the S-N ap-
proaches for fatigue strength assessment. Marine Structures 23(1), pp. 115-130 and IIW-Doc.
XIII-2279-09/XV-1321-09, International Institute of Welding.
Gaul L, Kögl M and Wagner M (2003): Boundary Element Methods for Engineers and Scientists.
Springer-Verlag, Berlin.
Gosch T and Petershagen H (1997): Effect of undercuts of the fatigue strength of butt joints (in
German). Schweißen & Schneiden 49(3), 158-163.
Hobbacher A (1996): Fatigue Design of Welded Joints and Components. Abington Publ., Cam-
bridge.
Hobbacher A (2008): Database for the Effective Notch Stress Method at Steel. IIW Joint Working
Group Doc. JWG-XIII-XV-197-08, International Institute of Welding
Hobbacher A (2009): Recommendations for Fatigue Design of Welded Joints and Components.
IIW doc.1823-07, Welding Research Council Bulletin 520, New York.
Hughes T R (1987): The Finite Element Method - Linear Static and Dynamic Finite Element Analy-
sis. Prentice Hall, Englewood Cliffs.
 36

Iida K and Uemura T (1995): Stress concentration factor formulas widely used in Japan. Fatigue
and Fracture in Engng. Materials and Structures, 19 (6), 779-786, and IIW Doc XIII-1530-94.
Karakas Ö; Morgenstern C; Sonsino C M; Hanselka H; Vogt M; Dilger K (2007): Grundlagen für die
praktische Anwendung des Kerbspannungskonzeptes zur Schwingfestigkeitsbewertung von ge-
schweißten Bauteilen aus Magnesiumknetlegierungen. LBF-Report FB-232, Fraunhofer-Inst. für
Betriebsfestigkeit & Systemzuverlässigkeit, Darmstadt and ifs-Report 17, Inst. für Schweißtech.
Fertigungsverfahren der TU Braunschweig, Braunschweig.
Köttgen R, Olivier R and Seeger T (1991): Schwingfestigkeitsanalyse für Schweißverbindungen
auf die Grundlage örtlicher Beanspruchungen (Fatigue analysis of welded connections based
on local stresses). Expert ’91 – Berechnung, Gestaltung und Fertigung von Schweißkonstruk-
tionen im Zeitalter der Expertensysteme, DVS-Bericht 133, 75-85, Düsseldorf, DVS, 1991, and
IIW Doc XIII-1408-91.
Kranz B and Sonsino C M (2010): Verification of the FAT values for the application of the notch
stress concept with reference radii rref = 1.00 and rref = 0.05. Welding in the World 54, No. 7/8,
pp. R218-R224.
Kuguel R (1961): A relation between theoretical stress concentration factor and fatigue notch factor
deduced from the concept of highly stressed volume. ASTM Proc 61, 732-744.
Kyuba, H. and Dong, P. (2003): Equilibrium-equivalent structural stress approach to fatigue analy-
sis of a tubular joint. IIW-Doc. XIII-1992-03/XV-1149-03, Int. Inst. of Welding.
Lawrence F V, Ho N J and Mazumdar P K (1981): Predicting the fatigue resistance of welds. Ann.
Rev. Material Science, Vol. 11, 401-425.
Lehrke H-P (1999): Berechnung von Formzahlen für Schweißverbindungen. Konstruktion 51:1/2,
47-52
Lieurade H-P, Huther I und Lebaillif D (2003): Weld quality assessment as regard to Fatigue. Proc.
of the IIW Fatigue Seminar 2003, Rep. 14, Lappeenranta University of Technology, Finland.
Morgenstern C, Sonsino C M, Hobbacher A and Sorbo F (2004): Fatigue design of aluminium
welded joints by the local stress concept with the fictitious notch radius of rf = 1 mm. IIW Doc
XIII-2009-04 and Int. J. Fatigue 28 (2006), pp. 881-890
Neuber H (1946): Theory of Notch Stresses, Ann Arbor Mi, Edwards.
Neuber H (1968): Über die Berücksichtigung der Spannungskonzentration bei Festigkeitsberech-
nungen. Konstruktion, 20 (7), 245-251.
Niemi E, Fricke W and Maddox S (2006): Fatigue analysis of welded joints - Designer's guide to
the structural hot-spot stress approach. Cambridge, Woodhead Publ.
Olivier R, Köttgen V B and Seeger T (1989): Schweißverbindung I (Welded Joints I), FKM-
Forschungshefte 143, Frankfurt/M, FKM.
Olivier R, Köttgen V B and Seeger T (1994): Schweißverbindung II, Schwingfestigkeitsnachweise
(Welded Joints II, Fatigue Assessments), FKM-Forschungsheft 180, Frankfurt/M, FKM.
Pedersen M M, Mouritsen O. Øm Mansen M R, Andersen J G and Wenderby J (2010): Re-analysis
of fatigue data for welded joints using the notch stress approach. Int. J. Fatigue 32(10), pp.
1620-1626
Peterson R E (1959): Relation between stress analysis and fatigue of metals’, Proc SESA, 11 (2),
199-206
Radaj D (1990): Design and Analysis of Fatigue Resistant Welded Structures, Cambridge, Abing-
ton Publ, 1990.
Radaj D and Zhang S (1991): Multiparameter design optimisation in respect of stress concentra-
tion. Engineering Optimisation in Design Processes, Berlin: Springer-Verlag, 181-189.
 37

Radaj D, Sonsino C M and Fricke W (2006): Fatigue Assessment of Welded Joints by Local Ap-
proaches. Abington: Woodhead Publishing (2nd Edition)
Radaj D, Sonsino C M and Fricke W (2009): Recent developments in local concepts of fatigue as-
sessment of welded joints. Int. J. Fatigue 31:1, 2-11.
Rainer G (1983): Parameterstudien mit finiten Elementen, Berechnung der Bauteilfestigkeit von
Schweißverbindungen unter äußeren Beanspruchungen. Konstruktion 37 (2), 45-52
SAE (2003): www.fatigue.org/weld/challenge-1
Siebel E and Stieler M (1955): Ungleichförmige Spannungsverteilung bei schwingender Beanspru-
chung. VDI-Zeitschrift, 97 (5), 121-126
Sonsino C M (1993): Zur Bewertung des Schwingfestigkeitsverhaltens von Bauteilen mit Hilfe örtli-
cher Beanspruchungen. Konstruktion 45/1, 25-33.
Sonsino C M (1995): Multiaxial fatigue of welded joints under in-phase and out-of-phase local
strains and stresses. Int. J. Fatigue 17, 55-70.
Sonsino C M (2009): A consideration of allowable equivalent stresses for fatigue design of welded
joints according to the notch stress concept with reference radii rref = 1.00 and 0.05 mm. Weld-
ing in the World 53, No. 3/4, R64-R75.
Sonsino, C M and Fricke, W (2006): Some remarks for improving the assessment of multiaxial
stress states and multiaxial spectrum loading in the IIW fatigue design recommendations. IIW
Doc. XIII-2128-06/XV-1222-06, International Institute of Welding
Sonsino C M, Radaj D, Brandt U and Lehrke H P (1999): Fatigue assessment of welded joints in
AlMg4.5Mn (5083) aluminium alloy by local approaches. Int J Fatigue, 21 (9), 985-999, and IIW
Doc XIII-1717-98
Sonsino C M and Wiebesiek J (2007): Assessment of multiaxial spectrum loading of welded steel
and aluminium joints by modified equivalent stress. IIW-Doc. XIII-2158r1-07/XV-1250r1-07, In-
ternational Institute of Welding
Tsuji I (1990): Estimation of stress concentration factors at weld toe of non-load carrying fillet
welded joints (in Japanese). West Japan Soc. of Naval Arch., 80, 241-251.
Zienkiewicz O C und Taylor R L (2000): The Finite Element Method, Vol. I: The Basis. Oxford, But-
terworth-Heinemann.

View publication stats