International Journal of Fatigue 25 (2003) 359–369

www.elsevier.com/locate/ijfatigue

Comparison of different calculation methods for structural stresses

at welded joints
O. Doerk, W. Fricke ∗, C. Weissenborn
Technical University Hamburg-Harburg, Laemmersieth 90, Hamburg 22305, Germany

Received 24 May 2002; received in revised form 29 October 2002; accepted 18 November 2002

Abstract

Different methods and procedures exist for the computation of the structural hot-spot stress at welded joints. These are either
based on the extrapolation of stresses at certain reference points on the plate surface (or edge) close to the weld toe—as known
from experimental investigations—or on the linearization of stresses in the through-thickness direction. Procedures for the application
of both methods to finite element analysis have recently been proposed in the literature. In the present paper, the different methods
are reviewed and applied to four different details in order to compare the methods with each other and to illustrate the differences.
Conclusions are drawn with respect to their accuracy and sensitivity to finite element meshing.
 2003 Elsevier Science Ltd. All rights reserved.

Keywords: Welded joint; Structural stress; Hot-spot stress; Finite element method; Stress analysis

1. Introduction ation and extrapolation at certain distances away from

the weld, which depend on the plate or shell thickness.
The crack initiation and early propagation at weld toes This development, which was reviewed a. o. by van
is governed by the local stress distribution around the Wingerde et al. [19], was particularly successful for the
weld. Its analysis and assessment with respect to fatigue fatigue strength assessment of tubular joints due to their
has already a rather long history. According to [18], first complex joint geometry and high local bending of the
investigations were performed in the 1960’s by several tubular walls.
researchers, including Peterson, Manson and Haibach, to First attempts to apply the approach to welded joints
relate the fatigue strength to a local stress or strain meas- at plates were already seen in the early 1980’s. Remark-
ured at a certain point close to the weld toe, for example able investigations were performed in Japan to analyse
at a distance of 2 mm [7]. Although the characteristic the stress concentration due to the local structural
fatigue strength related to this local stress shows fairly geometry of ship hull details, which were summarized
small scatter it has been shown e.g. in [1] that it is still a. o. by Matoba et al. [11]. The design stress was
affected by the local notch at the weld toe and, therefore, obtained from finite element analyses by linearization of
not independent from local notch geometry. Investi- the stress through the plate thickness. Radaj [17] summa-
gations of relatively thick tubular joints have shown that rized these and other investigations and defined the
the local notch effect of the weld toe affects the stress structural stress at the hot spot (weld toe) as the surface
in the region up to 0.3⫺0.4·t (t ⫽ plate thickness) away stress which can be calculated at the hot spot in accord-
from the weld toe. This resulted in the 1970’s in the ance with structural theories used in engineering. He
development of the well-known hot-spot stress approach demonstrated that the structural stress can be analysed
with the definition of reference points for stress evalu- either by surface extrapolation or by linearization, e.g.
through the wall thickness, in order to exclude the local
non-linear stress peak caused by the weld toe.
∗
Corresponding author. Tel.: +49-40-428-32-3148; fax: +49-40- In the early 1990’s, Petershagen et al. [16] derived a
428-32-3337. generalized hot-spot stress approach for plate structures
E-mail address: w.fricke@tu-harburg.de (W. Fricke). using Radaj’s effective notch stress approach [17] and

0142-1123/03/$ - see front matter  2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0142-1123(02)00167-6
360 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369

Nomenclature
b width of doubling plate w attachment width
B width of parent plate x, y, z coordinates
F force d distance
l element length s normal stress
M bending moment sm membrane stress
SCF stress concentration factor sb bending stress
t plate thickness t shear stress

applied it to complex welded structures [4]. Detailed rec- 2. Evaluation of structural stresses from finite
ommendations concerning stress determination for element models
fatigue analysis of welded components were given by
Niemi [12]. 2.1. Finite element modelling of welded structures
However, several applications showed that the stress
results are still affected by the finite element meshing As mentioned in the introduction, different types of
and element properties. Additional recommendations for weld toes can be identified, see Fig. 1, which require
finite element modelling and hot-spot stress evaluation different stress evaluation techniques:
were given by Huther et al. [9] and by Fricke [6], the
latter based on extensive round-robin stress analyses of a) weld toe on the plate surface at the end of an attach-
several details. Special considerations have been shown ment
to be necessary for in-plane notches such as welded edge b) weld toe at the plate edge at the end of an attachment
gussets, where plate thickness is no more a relevant para- c) weld toe along the weld of an attachment (the more
meter for the definition of the reference points for stress highly stressed of both weld toes)
evaluation. Niemi and Tanskanen [13] as well as Fricke
and Bogdan [5] proposed alternative procedures for the Types a) and c) are in principle similar, however, the
hot-spot stress analysis in such cases, using absolute dis- influence of modelling is particularly large at the ends
tances for the reference points. A comprehensive IIW- of welded attachments, i.e. at type a) and b), where the
local stress singularity is more pronounced due to the
guidance for the structural hot-spot stress approach is
additional stress concentration at the V-shaped corner.
currently under preparation [14].
In order to limit the computational effort, relatively
Dong [2] utilized the structural stress definition by
simple models and coarse meshes are preferred in prac-
Radaj [17] and evaluated the structural stress directly at
tice. Basically, two types of finite element modelling are
the weld toe position from finite element results by using
usual, which are illustrated in Fig. 2 by the example
principles of elementary structural mechanics. Mesh shown above:
insensitivity is claimed and demonstrated by several
examples, however, mainly on 2D basic joints [2], [3].
In this paper, the different methods for structural
stress evaluation are explained in more detail and com-
pared with each other. Afterwards, their application is
illustrated by several 2D and 3D examples, showing the
similarities of the methods and answering the question,
how far mesh-insensitivity can be reached.
It should be emphasized that the structural stress
approach is restricted to the fatigue strength assessment
of weld toes, where cracks start from the surface of the
structure. Cracks starting from the root of not fully pen-
etrated welds are not covered and require a different
assessment procedure.

Fig. 1. Types of weld toes.

 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369 361

For type a) and c) weld toes, the IIW recommen-

dations [8,14] propose a linear extrapolation over two
reference points, which are located 0.4·t and 1.0·t away
from the hot spot, where t is the thickness of the adjacent
plate (Fig. 3.1). The stresses are typically evaluated at
nodal points, so that the length of the first element is
0.4·t and the second 0.6·t. In case of a coarser mesh with
higher order elements, having lengths equal to t, the
stresses in the surface centres of solid elements or at
mid-side nodes of shell elements may be evaluated and
extrapolated over 0.5·t and 1.5·t (see Figs. 2 and 3.2),
as proposed by some ship classification societies.
At type a) weld toes, however, the width of the solid
element or the two shell elements in front of the hot spot
should not exceed either two times the plate thickness t
or the attachment width w (=attachment thickness plus
two weld leg lengths).
The situation is different for type b) weld toes, i.e. at
plate edges. As plate thickness is not relevant for the
element size nor the location of the reference points,
fixed reference points are proposed. Following the pro-
posal by Niemi and Tanskanen [13] to apply quadratic
extrapolation over three points, 4 mm, 8 mm and 12 mm
away from the hot spot, element lengths of 4 mm or
even better 2 mm are required to obtain stresses at nodal
Fig. 2. Typical finite element models and stress evaluation paths. points not affected by the stress singularity (Fig. 3.3).
The alternative proposal by Fricke and Bogdan [5]
1. using plate or shell elements which are arranged in
implies a linear extrapolation of stresses obtained from
the middle plane of the plates. The weld is frequently
the mid-side points of higher-order elements (e.g. isopar-
omitted, except in cases with plate offsets (e.g.
ametric 8-node shell elements) with 10 mm length and
doubler plates) or welds close to each other, where
depth, which means that the stresses are extrapolated
interaction effects occur. In such cases the weld can
over points 5 mm and 15 mm away from the hot spot
be modelled by vertical or inclined plate elements or
(Fig. 3.4).
by rigid links (constrained equations). The plate or
shell elements should generally contain improved in-
plane behaviour to model steep stress gradients.
2. using solid elements allowing the weld to be easily
modelled with prismatic elements. If isoparametric
20-node elements are applied, one element is suf-
ficient in thickness direction due to the quadratic dis-
placement function and linear stress distribution. In
connection with reduced integration, the linear part of
the stresses can directly be evaluated.

2.2. Structural stress evaluation by surface stress

extrapolation

The ‘classical’ way of evaluating the structural stress

at the hot spot is the linear or quadratic extrapolation
over two or three reference points in a similar way as
done experimentally with strain gauges. Fig. 2 shows
typical stress evaluation paths. In case of shell models
without weld representation it is recommended to
extrapolate the stress to the structural intersection point
as modelled in order to avoid stress under-estimation due
to the decreased stiffness of the model [6]. Fig. 3. Extrapolation of surface stresses to the hot spot acc. to [14].
362 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369

2.3. Structural stress evaluation according to Dong [2] gated only through a part of the thickness. In this case,
the stresses acting at the lower boundary of the area, i.e.
The structural stress evaluation method proposed by in the depth t1, have to be included in the a.m. equations,
Dong [2] generally focusses on the linearization through because the lower boundary is no more a free surface.
the wall thickness directly at the hot spot, however, In thick section joints and some other joint configur-
depends on the type of modelling. ation, such as fillet welds that are symmetric with respect
For solid models, where the element stresses might be to geometry and loading, there is a non-monotonic
disturbed by the singularity at the weld toe, the element trough-thickness stress distribution. In these cases the
stresses are evaluated at a certain distance d away from linearization is also performed to a finite depth t1, which
the weld toe, e.g. equal to the element length, see Fig. is equal to t/2 in case of symmetry.
4. Assuming equilibrium between the axial and shear
stresses acting here (Section B-B) and in the section For a shell model, the structural stress can be evalu-
directly at the weld toe (Section A-A), the linear part of ated directly at the hot spot because the linear stress dis-
the latter can directly be derived (stresses acting on the tribution is already assumed in the elements, see Fig. 5.
other sides of the element are neglected). Using trap- In order to avoid inaccuracies due to stress distribution
ezoidal integration for n ⫹ 1 equally spaced nodes over assumed in the element formulation, the structural stress
the plate thickness yields two equations for sm and sb: is calculated directly from the nodal forces and moments
at the element edge in question.

冕
t
1 1 A multi-linear stress distribution is assumed for sev-
sm ⫽ sxx(z)·dz ⫽ [s ⫹ 2·sxx,1 ⫹ … eral elements along the weld which is derived from an
t 2·n xx,0
0 equation system for the stress values at the element cor-
ners.
⫹ 2·sxx,n⫺1 ⫹ sxx,n]
By using these stresses, mesh insensitivity is claimed

冕 冕
t t by Dong [2] even for hot spots with high stress singular-
t2 t2 t2 ity, i.e. types a) and b) in Fig. 1.
sm ⫹ sb ⫽ sxx(z)·dz⫺d· txz(z)·dz ⫽ [s
2 6 6·n2 xx,0
0 0

⫹ 6·sxx,1·z2 ⫹ 12·sxx,2 ⫹ … ⫹ (n⫺1)·6·sxx,n⫺1 3. Examples

t
⫹ (3n⫺1)·sxx,n]⫺d [txz,0 ⫹ 2·txz,1 ⫹ … ⫹ 2·txz,n⫺1 In the following, four examples with different types
2·n of weld toes are described, where the methods men-
⫹ txz,n] tioned above are applied to derive the structural hot-spot
stress, i.e.
Fig. 4 shows the stress linearization through the whole
plate thickness t, resulting in the structural stress as 앫 surface stress extrapolation acc. to IIW [8,14], i.e. lin-
defined by Radaj [17]. Alternatively, the linear stress can early over 0.4 t /1.0 t for type a) and c) joints and
be derived for part of the thickness t1, which allows the quadratically over 4 mm /8 mm /12 mm for type b)
structural stress to be derived for a crack having propa- joints in connection with element lengths of at least
0.4 t or 4 mm, respectively (Figs. 3.1 and 3.3)
앫 surface stress extrapolation over 0.5 t /1.5 t (5 mm
and 15 mm for type b) joints) in connection with rela-
tively coarse meshes, having elements with quadratic

Fig. 4. Structural stress evaluation for solid models (acc. to [2]). Fig. 5. Structural stress evaluation for shell models (acc. to [2]).
 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369 363

shape function and lengths of 1.0t or 10 mm, respect- The application of the structural stress approach
ively (Figs. 3.2 and 3.4) according to Dong [2] yields almost the same structural
앫 structural stress evaluation acc. to Dong [2], using SCF for several mesh densities, as shown in Fig. 6d. As
meshes with different element sizes. All calculations no shear stress is acting in the plate, the stress evaluation
were performed by the authors on the basis of the can simply be reduced to a linearization through the
references given. thickness at any section in the right part, yielding a struc-
tural stress SCF of approximately 1.19.
The element type and weld representation have not
been varied within each comparison. The same value is achieved by extrapolating the sur-
face stresses, see Fig. 7. As expected, the mesh density
plays almost no role also in the case of surface stress
3.1. Plate lap fillet weld extrapolation. The constant structural stress distribution
would even allow any location of the reference points,
The first example concerns a 2D example, the plate as long as they are beyond 0.4 t.
lap fillet joint described in [2]. Fig. 6a illustrates the one-
sided lap joint, which is subjected to an axial force F.
The weld toe belongs to type c) according to Fig. 1.
Due to the eccentricity of the lap joint and the bound-
ary conditions at the ends, a constant bending moment
without any shear force is acting in the plate in front of
the weld. Therefore, a constant structural stress is acting
which is determined by the stiffness of the actual struc-
ture.

Fig. 7. Plate fillet lap joint and results obtained for surface stress
extrapolation.
Fig. 6. Plate fillet lap joint and results obtained by Dong [2].
364 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369

3.2. One-sided doubling plate

The second example is the one-sided doubling plate

shown in Fig. 8, where the critical weld toe on the plate
surface belongs to type c) in Fig. 1. The model was
investigated in a Japanese research project [20]. The
example is similar to the first one, however, the round
doubling plate causes a non-uniform stress distribution
in the transverse direction.
The example was also investigated in the round-robin
analysis described by Fricke [6], where different tech-
niques of modelling the one-sided doubling plate by
shell elements were applied. In the present analysis, the
doubling plate was modelled by solid elements, allowing
the weld to be realistically considered. Fig. 9 shows three
different finite element models. In all cases, 20-node
solid elements with reduced integration order were used.
One element was arranged over the plate thickness,
while the element lengths in front of the weld toe ranged
from approx. 0.4–2 t.
The computed stress distribution in front of the weld
toe is plotted in Fig. 10. In contrast to the previous study
[6], no stress magnification due to weld distortion was
considered. For this reason, the measurement results
from Yagi et al. [20] have not been included in Fig. 10,
because these were obviously affected by this.
Although the resulting stresses are fairly close
together, a slight influence of the element size can be
observed. The extrapolation of the surface stresses to the
hot spot, performed for the associated models and indi-
cated by arrows in Fig. 10, yields hot-spot stress ratios
of 1.25 (over 0.4 t /1.0 t) and 1.26 (over 0.5 t /1.5 t).
The round-robin study [6] showed a higher scatter (±6%)
due to the application of different element types and
particularly due to simplified weld modelling in case of
shell models, where plate connections and rigid links
were used. Fig. 9. Different finite element meshes for modelling the one-sided
doubling plate (1/2-model).
A scatter of approximately 10% is contained in the
results based on the approach by Dong [2], which are
plotted on the left side of Fig. 10. The structural stress
in this example is obviously not insensitive to the mesh
density. The aforementioned scatter due to different
element types and simplified modelling may addition-
ally occur.

Fig. 10. Surface stress and structural stress ratio for one-sided doub-
Fig. 8. One-sided doubling plate investigated by Yagi et al. [17]. ling plate.
 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369 365

In order to clarify the reasons for this mesh-sensi-

tivity, the geometry of the one-sided doubling plate has
been varied. For the sake of simplification, a rectangular
doubling plate with constant length (60 mm), but varying
width b has been chosen. The thickness of the doubling
plate is 10 mm. The dimensions of the parent plate are
B ⫽ 240 mm and t ⫽ 15 mm.
Fig. 11 shows three models with different ratios b/B,
ranging from 1/12 (shallow longitudinal stiffener) to 1/1
(2D case). The element length in front of the doubling
plate was again varied from 0.4 t to 2.0 t.
Fig. 12 shows the structural stress evaluated at the
centre line according to Dong [2]. It can clearly be seen Fig. 12. Structural stress according to Dong [2] evaluated from dif-
that the difference between the results becomes larger if ferent meshes of rectangular doubling plates.

the concentration becomes more localized. The reason

is seen in the neglect of vertical shear stresses acting on
the transverse element sides in the equilibrium equation
described in section 2.3.

3.3. Bracket toe

The third example concerns a bracket toe, which was

investigated within the European Research Project
FatHTS [15]. Fig. 13 shows the test model with a diag-
onally acting hydraulic cylinder, which produces a com-
bination of axial force, shear force and bending moment
in two horizontal and vertical I-beams.
The critical position is the bracket toe, which exists
four times in each test model. The plate thickness of the
flange is 20 mm, while the bracket is 12 mm thick. Full
penetration welding was applied with a leg length of the
fillet weld reinforcement of 8.5 mm.
During the investigation, strain measurements and
finite element calculations were performed. Fig. 14
shows two different finite element models of the critical
area based on above described recommendations, where
the element length in front of the bracket corresponds
to the flange thickness. Fig. 15 compares the computed

Fig. 11. Finite element models of rectangular doubling plates having

different width. Fig. 13. Bracket investigated by Paetzold et al. [15].
366 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369

longitudinal stress in front of the hot spot with measured

values for a cylinder force of 100 kN. Apart from the
measured stress close to the hot spot, which is affected
by the local notch, the agreement is very good.
Only the shell model is considered for the present
comparison of the different methods. 8-noded quadratic
shell elements have been chosen to represent the I-beam,
the bracket and the flange. The weld was not modelled
as frequently done in practice. In total six meshes have
been created with element sizes in front of the bracket
toe ranging from 0.4 t × t / 2 to 2 t × 2 t. Half the attach-
ment width (w / 2 ⫽ 14.5 mm) was partly taken for the
element width as recommended by Niemi [14] and
Fricke [6].
The resulting stress distribution is shown in Fig. 16.
The stress singularity influences the results close to the
hot spot. However, the structural hot-spot stress derived
from surface extrapolation is almost the same for both
alternative methods mentioned above. A slight stress
under-estimation can be observed when comparing the
results with Fig. 15—an effect which has frequently
been found in connection with shell models. The restric-
tion of the element width to w/2 has only a small effect
on the results in this example.
The results obtained by application of Dong’s method
are generally higher and show a very large scatter. This
is obviously due to the stress singularity, as the local
stress becomes infinite if the element size approaches
zero. The method [2] as applied to this model is highly
mesh-dependent and not able to yield a reasonable struc-
tural stress for simplified models. The surface stress
extrapolation method has, of course, also problems in
such cases, however they seem to be less severe.
The mesh density effect is normally related only to
the elements in front of the hot spot. However, Fig. 17
shows that also the modelling of other areas—in this
case the bracket—may strongly affect the results. A
Fig. 14. Shell and solid finite element models of the bracket with coarse modelling of the bracket toe would increase the
longitudinal stress distribution.
local stress by approximately 10% and, thus, closing the
gap between shell and solid models. This means that we

Fig. 15. Stress distribution in front of the bracket from measurements Fig. 16. Stress distribution in front of the bracket toe and structural
and f.e. models (Fig. 14). hot-spot stresses for various shell models.
 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369 367

according to Fig. 1. The model was investigated exper-

imentally by Kim et al. [10]. It was also included in the
round-robin study [6].
Again, shell modelling with 8-noded elements was
chosen for the present finite element analysis. The weld
was modelled in a simplified way as illustrated in Fig.
18. In this way, the correct weld toe position was kept.
The area in front of the weld toe was modelled in three
different ways by choosing element lengths ᐉ=2 mm,
ᐉ=5 mm and ᐉ=10 mm, respectively.
Fig. 19 shows the computed stress distribution at the
plate edge of the flat bar close to the weld. The force F
was chosen such that a unit nominal stress is acting at
Fig. 17. Stress distribution in front of the bracket toe for two different the welded toe. As expected for in-plane notches, the
meshes of the bracket.
stress distribution is affected by the stress singularity,
have to accept an additional scatter in the finite element showing increased stresses in the elements adjacent to
results due to the meshing around the critical area. the notch. The stress extrapolation yields a stress value
of 1.77 MPa for the fine mesh (quadratic extrapolation)
3.4. Fillet weld around plate edge and 1.68 MPa for the coarse mesh (linear extrapolation),
which means a slight difference between the two
The last example is a flat bar welded to an I-beam, methods. The difference is higher than expected from
where the critical hot spot is located at the plate edge, the former investigation [5], where only 2D structures
see upper part of Fig. 18, i.e. it belongs to type b) with 135° and 90° corners have been analysed.
Dong’s method was applied for an assumed crack
depth of 10 mm, defining the end of the fatigue life for
this specimen. The structural stress computed for the
three meshes in accordance with 2.3 shows only little
scatter, however, the stress is higher that that obtained
from surface extrapolation.
It should be mentioned here that the calculated struc-
tural stress is higher than the measured one and that the
corresponding fatigue life prediction has shown to be
very conservative for this example [6].

4. Conclusions

From the application of different structural stress

evaluation methods to four examples of welded plate

Fig. 18. Flat bar welded to an I-beam and modelling of the critical Fig. 19. Stress distribution in front of the fillet weld and structural
area around the weld toe. hot-spot stresses for various models.
368 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369

structures, the following conclusions are drawn and rec- ated with the structural stress by Dong [2] seem not
ommendations are given: to be in contradiction to this.

Although the examples chosen cover a variety of dif-

1. The two alternative methods for surface stress extra- ferent types of weld toes and practical situations, they
polation (as shown on the right and left side of Fig. are still relatively simple. Several questions remain open,
3) yield almost the same results. The first procedure e. g. the applicability of the methods to complex, bi-
with reference points 0.4 t / 1.0 t away from the weld axial stress states or to very thick structural members,
toe (or 4/8/12 mm at plate edges) requires a finer e. g. bulbs of profiles, where it is difficult to select an
mesh with element lengths of at least 0.4 t (or 4 mm, appropriate thickness for the definition of stress extrapol-
respectively), if higher-order elements are used. How- ation points or for the depth for stress linearization. All
ever, finer mesh densities are also allowed. The the aforementioned aspects should be considered when
second procedure with reference points 0.5t /1.5 t assessing the reliability of the different methods. In
away from the weld toe (or 5/15 mm at plate edges), addition the practicability is very important for the
which is preferred by several ship classification industrial application.
societies, requires fixed sizes of higher-order elements Furthermore it should be noted that the fatigue predic-
to achieve consistent results. tion may strongly be affected by other influence factors
2. The procedure proposed by Dong [2] for the evalu- such as positive (compressive) residual stresses or large
ation of the structural stress directly at the weld toe variations in the local weld profile, which should be
shows mesh-insensitivity for 2D problems. However, taken into account when assessing the methods. In this
in the case of 3D stress concentration, some scatter is sense, the structural hot-spot stress approach remains to
observed in the results evaluated from different mesh be a relatively coarse, however, very practical approach.
densities. This seems to be due to the neglect of
stresses in the equilibrium equations acting at the
transverse element sides. The scatter increases if the References
notch is very localized, as exemplified by the toe of
a bracket oriented in stress direction, which is in prac- [1] Atztori G, Meneghetti G. Fatigue strength of filled welded struc-
tice often very simply modelled by using shell tural steels: finite elements, strain gauges and reality. Inter-
national Journal of Fatigue 2001;23:713–21.
elements. In case of in-plane notches, i.e. at plate [2] Dong P. A structural stress definition and numerical implemen-
edges, the structural stress depends on the assumption tation for fatigue analyses. Int J Fatigue 2001;23(10):865–76.
of a crack depth, which defines the range of stress lin- [3] Dong P, Hong JK, Cao Z. A mesh-insensitive structural stress
earization. procedure for fatigue evaluation of welded structures. IIW-Doc.
3. Additional scatter of the stress results is expected due XIII-1902-01/XV-1089-01, International Institute of Welding,
2001.
to the usage of different element types offered by [4] Fricke W, Petershagen H. Detail design of welded ship structures
finite element programs and due to different tech- based on hot spot stresses. In: Caldwell JB, Ward G, editors.
niques of modelling the weld, particularly if shell Practical design of ships and mobile units. Elsevier Science;
elements are applied. This scatter is typically between 1992.
±5% and ±10% [6]. In addition, the investigations [5] Fricke W, Bogdan R. Determination of hot spot stress in struc-
tural members with in-plane notches using a coarse element
have shown that also the meshing outside the stress mesh. IIW-Doc. XIII-1870-01, International Institute of Weld-
evaluation area in front of the weld toe can further ing, 2001.
affect the results, so that mesh-insensitivity remains [6] Fricke W. Recommended hot spot analysis procedure for struc-
generally questionable. The analyst should always be tural details of ships and FPSOs based on round-robin FE analy-
aware of the limitations set by the finite element ses. Int J of Offshore and Polar Engng 2002;12(1):40–7.
[7] Haibach E. Fatigue Strength of Welded Joints from Viewpoint
model as well as by the evaluation method of the of Local Strain Measurement (in German). Report FB-77, Fraun-
structural hot-spot stress. hofer-Institut für Betriebsfestigkeit (LBF), Darmstadt 1968.
4. The definition of the structural stress is principally [8] Hobbacher A, editor. Recommendations for fatigue strength of
the same in all methods, except for cases with non- welded components. Cambridge: Abington Publishing; 1996.
monotonic trough-thickness stress distributions (see [9] Huther I, Gorski S, Lieurade HP, Laborde S, Recho N. Longitudi-
nal non loaded welded joints geometrical stress approach. Weld-
section 2.3). Therefore, the fatigue assessment by S- ing in the World 1999;43(3):20–6.
N curves should also be comparable. Niemi [14] rec- [10] Kim WS, Kim DH, Lee SG, Lee YK. Fatigue strength of load-
ommends for welds at steel FAT 100 for normal cases carrying box fillet weldment in ship structure. In: Wu Y-S, Cui
and FAT 90 for cases with full-load carrying fillet W-C, Zhou G-J, editors. Practical design of ships and other float-
welds (example 4) and side attachments longer than ing structures. Elsevier Science; 2001.
[11] Matoba M, Kawasaki T, Fujii T, Yamauchi T. Evaluation of
100 mm. The FAT number corresponds to the charac- fatigue strength of welded structures—hull’s members, hollow
teristic fatigue strength reference value of the design section joints, piping and vessel joints. IIW-Doc. XIII-1082-83,
S-N curve at 2 million cycles. The fatigue lives evalu- International Institute of Welding, 1983.
 O. Doerk et al. / International Journal of Fatigue 25 (2003) 359–369 369

[12] Niemi E, editor. Recommendations concerning stress determi- in ships. IIW Doc. XIII-1409-91, International Institute of Weld-
nation for fatigue analysis of welded components. Cambridge: ing, 1991.
Abington Publ; 1995. [17] Radaj D. Design and analysis of fatigue-resistant welded struc-
[13] Niemi E, Tanskanen P. Hot spot stress determination for welded tures. German Edition: DVS-Verlag, Düsseldorf 1985; English
edge gussets. Welding in the World 2000;44(5):31–7. Edition: Abington Publ., Cambridge, 1990
[14] Niemi E. Structural Stress Approach to Fatigue Analysis of [18] Radaj D, Sonsino CM. Fatigue assessment of welded joints by
Welded Components—Designer’s Guide. IIW-Doc. XIII-1819- local approaches. Cambridge: Abington Publ, 1998.
00/XV-1090-01 (Final Draft), International Institute of Weld- [19] van Wingerde AM, Packer JA, Wardenier J. Criteria for the
ing, 2001. fatigue assessment of hollow structural section connections. J
[15] Paetzold H, Doerk O, Kierkegaard H. Fatigue behaviour of differ- Construct Steel Res 1995;35:71–115.
ent bracket connectons. In: Wu Y-S, Cui W-C, Zhou G-J, editors. [20] Yagi J, Machida S, Tomita Y, Matoba M, Kawasaki T. Definition
Practical design of ships and other floating structures (Ed. Elsev- of hot spot stresses in welded plate type structure for fatigue
ier; 2001. assessment. IIW-Document XIII-1414-91, International Institute
[16] Petershagen H, Fricke W, Massel T. Application of the local of Welding, 1991.
approach to the fatigue strength assessment of welded structures