You are on page 1of 8

Procee~ngs of the Eleventh (2001) International Offshore and Polar Engineering Conference

Stavanger, Norway, June 17-22, 2001

Copyright 2001 by The International Society of Offshore and Polar Engineers
ISBN 1-880653-51-6 (Set); ISBN 1-880653-55-9 (VoL IV); ISSN 1098-6189 (Set)

Recommended Hot Spot Analysis Procedure for Structural Details of FPSOs

and Ships Based on Round-Robin FE Analyses
Wolfgang Fricke
Technical University Hamburg
Harburg, G e r m a n y

curve, which implicitly considers the effects of the localized notch and,
furthermore, is valid for a certain class of weld shapes and materials.
Regarding the hot spot stress, experimental and analytical procedures
have been derived for its determination by extrapolating the structural
stress outside the localized notch-affected zone to the weld toe. The
approach was firstly applied in the 1970's to tubular joints of offshore
structures, where the increase in structural stress can be very high due to
local bending of the tubular shell close to the connection between brace
and chord (Almar-N~ess, 1985).
The hot spot stress approach was later applied also to welded plate
structures (Radaj, 1990)being typical for FPSO's (Floating Production,
Storage and Offloading Units) and ships. Here, three different types of
hot spots at weld toes can be identified which are exemplified in Fig. 1:
a) at the weld toe on the plate surface at an ending attachment
b) at the weld toe around the plate edge of an ending attachment
c) along the wdd of an attached plate (weld toes on both the plate and
attachment surface).


As part of the Joint Industry Project 'FPSO Fatigue and Fracture Capacity', a Special Task Group with nine participants performed investigations regarding finite element (FE) modelling and
analysis of typical structural details in FPSO's (Floating Production, Storage and Offioading Units) as well as in ships. The purpose
of this special effort was to develop recommendations on appropriate hot spot stress methods and S-N data for fatigue strength
design. In total, five details with different characteristics, from both
geometry and fatigue loading perspectives, were selected for which
stress measurements and fatigue tests are available. Various finite
element models were developed by participants, using different
types and sizes of elements, modelling and stress evaluation techniques as well as FE programs. Comparisons between the analysis
results and measured stresses near the weld toes allow conclusions
to be drawn. Three different stress extrapolation techniques for
predicting hot spot stresses at the weld toes were investigated. The
resulting hot spot stresses, together with the estimated fatigue lives,
are compared against the existing design S-N curves published by
the International Institute of Welding (IIW). It was concluded that
the hot spot stresses predicted using the three stress extrapolation
techniques, where the element sizes and stress evaluation points are
determined by the plate thickness, can be used with the current
design S-N curves. Most significantly, one of the recommended
methods requires no stress extrapolation, which is considered an
attractive and practical alternative to the existing practices developed by class societies. This method will provide significant saving
on analysis efforts during design.

Figure 1. Types of Hot Spots in Welded Structures

KEY WORDS: Stress analysis, hot spot stress, finite element method,
fatigue, welded joint

For weld toes on a plate or shell surface, i. e. types a) and c), the
structural stress can be defined as the sum of the axial and bending part
of the stress distribution in the thickness direction (Niemi, 1993). However, a unique definition of the structural stress at plate edges (weld toe
type b), which can be used for fatigue strength assessment, is not possible. Therefore, extrapolation of edge stresses is currently regarded as
the only practical way for the determination of hot spot stresses.
The hot spot stress approach has often been criticized. Especially the
problem of defining appropriate reference points for the stress
extrapolation created many debates. Several codes and guidelines in
various industrial sectors recommend different procedures for the

The hot spot stress approach for the fatigue strength assessment of
welded joints is based on the assumption that the local stress increase at
the weld toe can be subdivided into two parts, one governed by the
structural (i. e. macro-geometrical) stress increase and the other created
by the localized notch stress due to the weld toe itself which is restricted
to a region of approx. 2 - 3 mm around the toe. Furthermore, it is
assumed that the fatigue assessment can be based on the first part, i. e.
the structural stress at the 'hot spot', together with an appropriate S-N


Detail 2: Gussets on Plate Edge

The second model, gussets on plate edges shown in Fig. 2a, has been
taken from tests performed by HHI in the IIP (Kim, 1999). Two
150 mm long and I0 mm thick gussets, representing face bars with
tapered ends, are fillet-welded to the edges of a plate strip 60 x 10 ram,
which is subjected to tensile stresses. The critical weld toe belongs to
type b) in Fig. 1.
Detail 3" Doubling Plate
The third detail is the doubling plate shown in Fig. 2b, where the
critical weld toe on the plate surface belongs to type c) in Fig. 1. The
model considered here was investigated in the Japanese research project
SR 202 (Yagi et al., 1991), so that strain measurements and fatigue tests
are available. When the parent plate is subjected to tension, the onesided doubling plate causes secondary bending. This has to be considered in the analysis, creating some modelling problems in connection
with shell elements.
In the course of the analysis, relatively large differences between the
Japanese measurements and the analytical results were found. These are
most likely caused by angular distortion during the welding of the
doubling plate on the plate strip. As pronounced stress magnification
due to misalignment shall be considered in the applied stress rather than
in the design S-N curve, a stress magnification factor Km= 1.2 and a
certain stress gradient were estimated for the local stress on the basis of
additional tests and calculations performed.
Detail 4: Hopper Corner Model
The fourth detail is the hopper comer shown in Fig. 2c which was
also tested in the JIP. The critical weld is located at the transition from
the 10 mm thick flange to the sloped hopper plate. The structure is
subjected to a vertical force, producing bending and shear in the beam.
The stress increase will be referred to the nominal bending stress at the
knuckle under the assumption of full effective breadth. The weld toe
belongs to type c) in Fig. 1.
Detail 5: Load Carrying Fillet Welds
The last detail is again a model tested in the JIP, i. e. the filletwelded connection between a vertical I-beam and a vertical plate, which
represents a stiffener connection being subjected to shear and bending,
see Fig. 2d. Reference stress is again the nominal bending stress at the
upper edge of the vertical plate. The weld toe at the hot spot belongs to
type b) in Fig. 1.

determination of hot spot stresses and S-N curves. Even in the

shipbuilding and offshore industry, the procedures established by the
classification societies and other authorities are diverging. Uncertainties
about the suitability of the hot spot stress approach have been raised by
round-robin analyses such as ISSC (1997) showing large scatter and
differences between analysis results and measured stresses.
The industry, however, considers the hot spot stress approach as a
very practical approach, offering a better alternative compared to the
traditional nominal stress approach for the assessment of individual
joint geometries in ships and FPSO's, which vary from detail to detail
due to different scantlings and geometrical configurations. As large
differences in fatigue assessment procedures as well as under-predicted
fatigue lives cannot be tolerated, much effort has been spent into further
investigations of the hot spot stress approach within the Joint Industry
Project (JIP) "FPSO Fatigue Capacity" with 18 partners, coordinated by
Det Norske Veritas. Within this project, a special task group performed
investigations to answer the following questions:
1. Which ways of modelling yield hot spot stresses with sufficient
accuracy for typical structural details?
2. How correlate certain types of stress extrapolation with S-N curves
It was intended to perform round-robin stress analyses on a number
of typical details where local strain measurements and fatigue test data
are available, allowing comparison with stress and life calculations. In
total, nine partners have participated in the work, offering a wide range
of modelling and evaluation techniques as well as finite element
analysis programs. As certain procedures and ideas about stress
determination and evaluation exist, it was decided to include all of them
in the analysis rather than systematically varying certain modelling
parameters. Insofar, the round-robin stress analysis reflects current
techniques from which recommendations for appropriate procedures
can be derived. Further questions arising from the analysis are intended
to be investigated in Phase II of the project.
It should be noted that the investigation is focussed on the fatigue
assessment of weld toes only and not on possible cracks originating
from the root of welds with incomplete penetration which are usually
assessed on the basis of the nominal stresses in the weld throat. Also the
assessment of stress peaks at rounded plate edges are outside the scope
of this paper.
The structural details were selected considering the following

all three types of hot spots shown in Fig. 1 should be included

plate bending caused by the attachment should be included, which

is quite typical for details in ships and FPSO's

strain measurements and fatigue tests should be available; models

tested within the JIP are preferred because detailed or additional
information is available
Five details have been selected for the round-robin hot spot stress
analysis, which are shortly described in the following:

d)Oetail 5

c) Detail 4

Detail l: ISSC Model

The first detail, the model for the study of ISSC (1997) already
mentioned and further described by lwahashi et al. (1998), is the connection of a buckling stiffener 100x12 to the flange of a T-shaped longitudinal (web 350x12, flange 150x20). This model being typical for
FPSO and tanker structures was tested experimentally in the Japanese
research project SR 219 so that local strain measurements are available,
however no fatigue test data, as turned out later. The finite element
model shown in Fig. 4 gives an impression of the structure. The longitudinal is subjected to bending and shear, creating an increased hot spot
stress at the welded connection with the buckling stiffener. It should be
noted that the fillet weld has an increased leg length of 13 mm to avoid
root failure. The hot spot belongs to type a) in Fig. 1.

Figure 2. Details 2 - 5 selected for Round Robin Analysis

(Detail 1 is shown in Fig. 4)


In most cases, the element length and breadths close to the weld toe
were chosen as t x t (t = plate thickness). Fig. 4 exemplifies the modelling by Detail 1. Smaller element dimensions (typically t/2 or t/4) were
selected for meshes indicated by 2Solid2Ow(f) and 4Solid8w. Additional
remarks on specific modelling aspects are given in the following.


In order to find appropriate techniques for modelling welded structures as well as evaluating the hot spot stress increase due to structural
effects, a great variety of models and ways of evaluation was taken into
consideration. In principle, the different types of modelling can be
divided into two groups, illustrated in Fig. 3:
* using plate or shell dements without weld representation; the
elements are located in the mid-plane of the associated plate; in
some cases the weld is included in a simplified way
using solid elements with the possibility to model the fillet welds


Detail 1

Shell elements
t welds)


= 98


- - -

f /


\ I\1\|


w = attachmentwidth


Figure 3. Typical Finite Element Models for the Welded Joint

Shown in Fig. 1


Variations are seen in different element types offered by the finite

element programs as well as element sizes chosen by the analyst. The
following dement types were considered in the investigation, using
special short names as mentioned below:

Figure 4. Typical Finite Element Modelling

In Detail 1 (ISSC-Model), the reinforced fillet weld attracts additional stresses which is considered in one shell model containing the
weld in a simplified way (ShellSw). In almost all shell models of Detail
2 (Gusset on Plate Edge) the gussets were arranged directly at the edge
of the longitudinal plate edge except for one model where the shell
elements were arranged in the mid-plane of the gusset and connected by
a reinforced plate strip with sloped ends (Shell8w). Detail 3 (One-Sided
Doubling Plate) was generally modelled taking the offset into account.
In case of shell elements, either rigid links or vertical plate elements
were arranged, see Fig. 5a.

1Solid20w and 2Solid20w

20-noded isoparametric solid element, used with either one or two
element layers over full or half plate thickness modelled (prefix 1 and 2,
respectively). The welds are generally included (suffix w), in some
models very simply disregarding the root gap (see Fig. 3).

8-noded solid element, normally used with four layers over the plate
thickness. The welds are included in the models (suffix w).


a) D o u b l i n g F' "

Higher-order solid elements (geometric p-elements) with refined

mesh and optimised shape function in the critical area. The welds are
included in the models (suffix w).

Shell& Shell8r and Shell8p

rigid links or p l a t e

8-noded shell element without weld representation. In case of offsets, either rigid links (suffix r) or a plate connection between the midplanes of the plates (suffix p) are arranged.

b) H o p p e r C o r n e r

Shell4, Shell4r and Shell4p

Same as above, but with 4-noded shell elements, partly having constant stress state (css) and partly improved in-plane bending behaviour.

8-noded shell model with simplified weld representation by arranging a reinforced plate strip at the foot of the attached plate, having the
thickness increased by one leg length and extending to the actual weld
toe position. The endings are sloped, meeting the weld toe position on
the parent plate.

Figure 5. Modelling of Offsets in Shell Models


One specific problem of Detail 4 (Hopper Corner Model) is the offset between the intersecting horizontal, vertical and sloped plate at the
critical point. This is automatically accounted for in solid models, while
it is considered only in part of the shell models by a plate connection
between the two intersection points, see Fig. 5b. These models are
denoted by Shell8p and Shell4p. An additional study showed that this
offset reduces the stresses in the actual structure by up to 10 % in the
vicinity of the hot spot. For Detail 5 (Load Carrying Fillet Welds),
again one shell model contains the weld in simplified form (Shell8w).
The stresses were evaluated with two objectives, (a) to compare the
computed stresses with measured values in the area close to the hot spot
and (b) to extrapolate the stresses to the hot spot using common techniques, from which the following three were selected:
1. Linear extrapolation over reference points 0.5 and 1.5 x plate
thickness t away from the hot spot (technique preferred by most
classification societies)
2. Linear extrapolation over reference points 0.4 and 1.0 x plate
thickness t away from the hot spot (technique recommended by the
International Institute of Welding, see Hobbacher, 1996)
3. No extrapolation, but considering the stress value at 0.5 x plate
thickness t as the relevant hot spot stress
Typical stress evaluation paths are indicated in Fig. 3. First comparison of the analysis results showed that a better agreement between the
results from the different models and from the measurements is
achieved if the distance of the read-out-points (ROP's) is measured to
the hot spot as modelled, i. e. to the weld toe if the weld is modelled or,
else, to the structural intersection point, if the weld is not modelled. The
same applies to the point where the stress is extrapolated to. The justification for this procedure, which affects the results for sheU models
without weld representation, is that the stress at the fictitious weld toe
position is in many cases too low due to the reduced stiffness compared
to the real structure. This means on the other hand, that the proposed
extrapolation to the structural intersection point may yield conservative
results - a tribute to simple modelling.
A great variability exists in selecting the type of stress and in evaluating the stresses at the desired locations. The selected details do not
show significant differences between principal and directional stress
(perpendicular to the weld toe) so that the type of stress does not affect
the results here. Either nodal stresses or element stresses are evaluated,
the latter normally extrapolated linearly from the integration points to
the plate surface or edge and in some cases averaged to obtain a midside stress. Particularly for 4-noded shell elements, different results
were obtained depending on the shape function and the consideration of

in-plane stress components. By comparing all results, conclusions are

drawn with respect to appropriate ways of determining hot spot stresses.
The evaluated stresses are summarized in Figs. 6 - 10, where also
the measured stresses are shown. The nominal stress has been set to
unity for details 2 - 5, so that the results are actually hot spot stress
concentration factors (SCF's). It should be kept in mind that also the
measured stresses contain some uncertainties which was revealed in the
tests by scattering strains obtained from different comparable locations
and test models. The right part of the figures shows the stresses at readout-points (ROP's) selected by the participants. These ROP's are
located at certain distances from the hot spot modelled (i. e. weld toe or
structural intersection point, if the weld is not modelled, see previous
chapter) and can be identified as nodal or element stresses (at mid-side
or integration points) in those cases where the element length equals the
plate thickness t (exceptions are finer meshes denoted by 2Solid2Ow,
4Solid8w and Solidpw). The left part gives the stresses extrapolated to
the hot spot according to the three methods mentioned above. The
stresses at the reference points were interpolated by the participants by
curve fitting.
In order to facilitate the interpretation of the results, different types
of symbols were chosen for the results from solid and shell models. The
symbols have been connected by straight dotted lines with smaller dots
for shell models. These straight lines appear to increase the scatter,
however, relevant are only the symbols. The measurement results are
plotted using continuous lines. They have not been extrapolated. Fine
mesh solid models with higher-order elements, which play a special
role, can be recognized by:
open squares (p-elements Solidpw)

full squares (2Solid20w(39)

In the following, the results are discussed in detail tbr the different

Detail I (ISSC-Model), Fig. 6

The measurement shows a stress of approx. 140 MPa at a distance
10 mm away from the hot spot (ISSC, 1997). Only one coarse solid
model is included (1Solid2Ow), yielding slightly higher stresses. Here,
the results are somewhat affected by the singularity at the hot spot,
although the stress was averaged over the attachment width (buckling
stiffener plus welds) by arranging relatively wide solid elements. On the
contrary, the very fine solid mesh (2Solid2Ow(f)) shows smaller stresses
with a steep stress increase to the hot spot.


,,-I-- ,Shell4



"~- .Shell8
- .0- .Shell8w


- ~ - .Shell8


- -I-- .Shell4





- "~- .Shell8
- ,o- .Shell4(css)

- -('-- .SheU4

~ 10


Re~ P~. (t=2Omm)





distance ROP's to hot spot [mm]

Figure 6. Stress Results for Detail 1 (ISSC-Model)



The results of the shell models are fairly scattered. Looking again at
the location 10 mm away from the hot spot, three results are relatively
close to the measured ones, while four others are remarkably lower. The
reason for this could not be clarified as the stress evaluation points and
methods are quite different (at nodes, at element centres, and extrapolated from integration points to the element edge).
The general trend, that solid elements overestimate and shell elements underestimate the stresses, has been observed also in other studies (e. g. ISSC, 2000). Poutiainen and Niemi (2000) showed that this
happens in those cases where a web is arranged below the plate with a
longitudinal attachment. In order to limit the under-estimation of
stresses, the in-plane bending behaviour of plates should be improved
(the 4-noded elements showing the lowest results have only constant
stress state (css)) and the element width might have to be restricted.
Niemi (1995) proposed the 'attachment width' (thickness of attachment
plus twice weld leg length) for the breadth over the two elements in
front of the weld. Furthermore, the consideration of the relatively thick
weld slightly increases the stress (by approx. 2 %).


When looking at the hot spot stresses extrapolated over 0.5t/1.5t, the
scatter is moderate within 10 % (152 MPa 15 MPa). It increases if
the extrapolation is performed over 0.4t/1.0t. This is obviously a result
of the relatively coarse meshes, matching better the first extrapolation

Detail 2 (Gussets on Plate Edge), Fig. 7

The measurement shows a mean structural stress increase by 1.7 at a
location 5 mm away from the hot spot, however with a large scatter.
If the two smallest stresses resulting from the fine meshes with
higher-order solid elements (2Solid2Ow(j') and Solidpw) and one from
the 4-noded shell elements with constant stress state (css) are excluded,
the results are fairly close together. This can also be seen in the hot spot
stresses extrapolated over 0.St/1.5t, which are between [.85 and 2.08,
i. e. _+6 %. The different idealisations of the root gap and stress evaluations in solid models (averaged over thickness or taken at comer) have
little effect on the results. Again, the scatter is much higher if the stress
is extrapolated over 0.4t/1.0t, which is due to the steep stress increase in
front of the hot spot.






' !~Experiment
- - ] - - S olidpw

: 4Solid8w
0 4Solid8w
~a~.- 2Solid20w

(n 2'6

~, 2.6



~ 2.2




--Jk.-- 2Solid20w
= 2Solid2Ow(f)
- .0- .Shell8w
- .-t..- .Shell4


~ 1.4








- -I--,Shell4





distance ROP's to hot spot [mm]

Ref. Pts. (t=10mm)

Figure 7. Stress Results for Detail 2 (Gussets on Plate E~dge)




--E~-- Solidpw









\ . '.
: ~'~
~ ~,
- ,,=.,w,,~


- -. ,~ ,~j

i-- ,so, 2ow j


-)<- .Shell4r
.)K- 'Shell8r

.o. 'She 4r(css)

-X- . Shell4r



Ref. Ots. (t= 1gram)

5 ....

10 . . . .

15 . . . .



distance ROP's to hot s p o t [ m m ]

Figure 8. Stress Results for Detail 3 (Doubling Plate)





models with higher-order solid elements (2SoIid2Ow09 and Solidpw) are

excluded, the scatter is fairly small, resulting in a hot spot stress
increase between 1.96 and 2.16 if extrapolated over 0.5t/1.5t i. e. +5%
from the mean.

Detail 3 (One-Sided Doubling Plate), Fig. 8

The results are fairly close together at the hot spot and match the
measured values containing the stress magnification factor Km
explained earlier. The scatter in the results is mainly due to the problem
of correctly modelling the offset in the shell element models. Most of
the results originating from shell models with connections by plates or
rigid links are slightly above the other ones.
The remaining results for the coarse solid models are very close
together, also when looking at the extrapolated stresses.

Detail 5 (Load Carrying Fillet Welds), Fig. 10

Also for this model, the results reveal a high scatter. As for detail 2,
the stress singularity plays a part, increasing additionally the stress
-values close to the hot spot. It is unclear why the measurements show
rather small values.
The scatter is much reduced, if the results for the fine meshes with
higher-order solid elements (2Solid2Ow(D and Solidpw) as well as the
lowest results for the 4-noded shell elements with constant stress state
(css) are again excluded. Then the hot spot stress increase varies
between 1.64 and 2.18 (_+10%) if extrapolated over 0.5t/1.5t, where the
lowest result refers to a model with simplified weld representation

Detail 4 (Hopper Corner Model), Fig. 9

At a first glance, the results appear rather scattered. Only the mean
value from the measurement, which also showed a large scatter, has
been plotted.
Five models, where the misalignment illustrated in Fig. 5b has not
been considered (Shell4 and Shell8) show the highest results at 5 mm
distance from the hot spot. If these results as well as those from the

- -I-- ,Shell4
- -I-- ,Shell4
- "~.- ,ShellSp
- SK- ,Shell8
- -X- .Shell4p
- -&- .ShellSp
. .o. ,Shell4(css)
- .-P- ,Shell4
.--El-. Solidpw
~ k - - 2 Solid20w









~ 1.4

~-. _ : " " " ~ . .

~ 1.2







distance ROP's to hot spot [mm]

Ref. Pts. (t=lOmm)

Figure 9. Stress Results for Detail 4 (Hopper Comer Model)



- -+- ,SheU4
-I-- ,Shell4
~ - .Shell8



.o. ,Shell4(css)

- N.-,, 'Shell4

\\~ \ \,1

<>" .Shell8w


--@=- 4 SolidSw
- ' = " - 2 S olid2Ow(t)



" ,



u2. ~. m.

Ref. Pts. (t=lOmm)


'--0--" S olidpw




distance ROP's to hot spot [mm]

Figure 10. Stress Results for Detail 5 (Load-Carrying Fillet Welds)





alignment. The latter have to be considered in certain joints where their

pronounced misalignment effects are possible, such as plate butt welds,
cruciform joints and transverse fillet welds on free plates, .by appropriate Kin-factors defined e. g. by Hobbacher (1996). The extent of the.
local finite element model has to be chosen such that effects on the
structural detail considered are sufficiently small and reasonable boundary conditions can be formulated.
In plate structures, three types of weld toes can be identified, which
are exemplified in Fig. I. The relevant hot spot stress is the principal
stress on the surface or at the edge of the plate acting approximately
perpendicular to the weld toe (45 to 90).
Models with thin plate or shell elements or alternatively with solid
elements may be used. It should be noted that on the one hand the
arrangement and type of elements have to allow for steep stress gradients as well as for the formation of plate bending, and on the other
hand, only the linear stress distribution in the plate thickness direction
needs to be evaluated with respect to the definition of structural stress.
The following methods of modelling are recommended, see also Fig. 3:

Available fatigue test data, i. e. fatigue lives obtained for given load
levels, can be used to verify if the computed hot spot stresses correlate
with standard S-N curves. Fatigue test data are available for details 2, 4
and 5 (Kim, 1999) and also for detail 3 (Yagi et al., 1991). In addition,
models similar to details 1 and 3 have been analysed and tested by HHI,
which are included in the present analysis as Model 1 (Longitudinal
Gussets) and Model 3 (Two-Sided Doubling Plates). It should be noted
that only tests with high R-ratio are taken from the data by Kim (1999).
The verification is performed on the basis of lower-bound hot spot
SCF's from the analysis. In this way the comparison with standard S-N
curves is conservative. As discussed before, the results from the fine
meshes with higher-order solid elements as well as those from the 4noded shells with constant in-plane stress state have been excluded
which means that S-N curves derived are associated to analyses with
certain modelling and extrapolation techniques.
The lowest SCF-values are mostly determined by solid models using
8-noded elements with dimensions t/4 x t/4 x t/4. The additional models
1 and 3 were also analysed by this technique and also by shell elements.
It is interesting to note that the results from the very fine solid models (2Solid2Ow(f)), extrapolated over 0.4t and 1.0t, fit quite well with
the other values extrapolated over 0.5t/1.5t. However, the results
obtained from the p-elements are below these values.
The S-N plot in Fig. 11 shows the fatigue lives in relation to the hot
spot stress range for the first extrapolation technique, i.e. nominal
stress range multiplied by the lowest calculated hot spot SCF's extrapolated over 0.5t/1.5t. The details 4 and 5 show generally higher lives
which are assumed to be due to beneficial residual stresses found in
detail 5 and to a failure criterion defined by a relatively long crack.
Furthermore, the results for model 1 and details 2 - 3 are closer to the
lower bound due to the fact that only the worst of four competing hot
spots in the models are represented.
In addition, the S-N curve FAT 100, characterized by a slope exponent m = 3 and a reference value of 100 MPa at 2 million cycles, is
plotted in the figure, representing the lower bound of all results. The
same evaluation for the other extrapolation techniques, which are omitted here, shows also FAT 100 as lower bound for the results extrapolated over 0.4t/1.0t, while FAT 90 is the lower bound for the results at
0.5t. It should be noted that all test data used here is related to joints
with plate thickness ranging from 10 to 15 mm.

Plate and Shell Models.

A simple modelling is offered by thin plate and shell elements which
have to be arranged in the mid-plane of the structural components. 8noded elements are recommended particularly in case of steep stress
gradients. Care should be given to possible stress underestimation
especially at type b) weld toes in connection with 4-noded elements,
which should contain at least improved in-plane bending modes.
The welds are usually not modelled except for special cases where
the results are affected by high local bending, e. g. due to an offset
between plates or due to a small free plate length between adjacent
welds. Here, the weld may be included by vertical or inclined plate
elements having appropriate stiffness or by introducing constrained
equations to couple the node displacements. Particularly for weld toes
of types b) and c), a simple alternative for weld modelling is offered by
the arrangement of a reinforced plate strip at the foot of the attachment,
having a thickness increased by one leg length and sloped e n d s ,
extending to the actual weld toe positions.

Solid Element Models:

An alternative particularly for complex cases is offered by solid elements which need to have a displacement function allowing steep stress
gradients as well as plate bending with linear stress distribution in the
plate thickness direction. This is offered, e. g., by isoparametric 20-node
elements (with mid-side nodes at the edges) which means that only one
element in plate thickness direction needs to be arranged. An easy
evaluation of the membrane and bending stress components is then
possible if a reduced integration order with only two integration points
in the thickness direction is chosen. Modelling of the welds is generally
recommended and easily possible as shown in Fig. 3.

Based on the results of the round-robin finite element analysis
described above, the following recommendations can be given.
Hot spot stresses are calculated assuming linear material behaviour
using an idealised structural model with no fabrication-related mis-






N (cycles)

Figure 11. Fatigue test results vs. hot spot stress range based on lowest calculated SCF's extrapolated over 0.5t/1.5t


Element Sizes:
For both types of modelling, the dimensions of the first two or three
elements in front of the weld toe should be chosen as follows:
Weld toes of types a) and c): The dement length should correspond
to the plate thickness. In the transverse direction, the plate thickness
may be chosen again for the breadth of the plate elements. However, the
breadth over the first two elements should not exceed the 'attachment
width', i. e. the thickness of the attached plate plus 2 x the weld leg
length (in case of type c: the thickness of the web plate behind plus 2 x
weld leg length). This attachment width may also be taken for the width
of solid elements in front of the weld toe, see Fig. 3. Still open is the
question how the element size has to be defined e. g. for bulb profiles.
Furthermore it should be noted that a much finer mesh is needed in case
of 8-node solid elements.
Weld toes of type b): For weld toes at the plate edge, the plate thickness is principally not a suitable parameter to determine the element
size and stress extrapolation points. A further study (Fricke and Bogdan, 2001) shows that fixed values can better be used to describe the
stress increase at in-plane notches. From this study and from details 2
and 5 of the current study it can be concluded that higher-order elements with element lengths of 10 x 10 mm together with a linear extrapolation over the mid-side points yield conservative results. Therefore,
it is recommended to use element dimensions fixed to these dimensions.

The present study shows that further investigations are required to

clarify especially the following items:
Modelling and stress evaluation for very thick components, such as
bulbs of profiles
Extent of plate bending to be considered at type b) weld toes
Type of stress and evaluation in case of complex stress fields
Finally it should be noted that root cracking at fillet welds is not
covered by the hot spot stress analysis procedure. Appropriate and
practical procedures for the analysis of the relevant stresses in complex
structures and their fatigue assessment are still to be developed.
The JIP has been supported by 19 participants under coordination of
Det Norske Veritas. An overview over the project is given by Lotsberg
(2001). The present study has been managed by Germanischer Lloyd.
Hyundai Heavy Industries, as a participant of the JIP, performed the
fatigue tests with small scale specimens of typical fillet welded joints in
ship structures, which was one of the main tasks in the JIP forming the
basis of the investigations described in the paper.
Nine participants contributed with great enthusiasm to the roundrobin analyses: Weimin Chen (Umoe Technology), Ben Feron (Bluewater), W.S. Kim (Hyundai Heavy Industries), Helena Polezhaeva
(Lloyd's Register of Shipping), Philippe Rucho (Bureau Veritas),
Armin S~ibel (Germanischer Lloyd), Tore Ulldand (Aker Maritime),
Richard Yee (American Bureau of Shipping) and also Y.-S. Choo
(National University of Singapore). The author wishes to thank all of
them for their work and all participants of the JIP for their valuable
comments and permission to publish the results.

Stress Evaluation:
The structural stress components on the plate surface should be
evaluated along the paths shown in Fig. 3 and extrapolated to the hot
spot. Recommended are stress evaluation points located at distances
0,5 t and 1.5 t away from the hot spot modelled, where t is the plate
thickness at the weld toe. If the weld is not modelled, the hot spot is the
structural intersection point modelled. In case of type b) weld toes,
fixed distances of 5 mm and 15 mm are recommended.
If a relatively coarse mesh with the element sizes mentioned above is
chosen, the stresses may be evaluated as follows:
* In case of plate or shell elements the surface stress may be evaluated at the corresponding mid-side points (i. e. taking the mean
value of the element stresses at the adjacent corners). At weld toes
of type c) the stress may be averaged over the width of the attachment behind by taking the surface stress in the element centres as
also shown in Fig. 3.
* In case of solid elements the stress may be extrapolated linearly to
the surface centre (usually after averaging the stress components at
each of the the two layers of integration points)
At weld toes of type b), only part of plate bending obviously needs
to be considered. Further investigations are considered to be necessary as such cases are outside the scope of the present analysis.

Almar-N~ss, A. (1985). "Fatigue Handbook - Offshore Structures."
Tapir Publishers, Trondheim.
Fricke, W. and Bogdan, R. (2001). "Determination of Hot Spot Stress in
Structural Members with In-Plane Notches Using a Coarse Mesh."
IIW-Doc. XIII- 1870-01, International Institute of Welding.
Hobbacher, A. (1996). "Fatigue Design of Welded Joints and Components." Abington Publishing, Cambridge (UK).
ISSC (1997). "Report of Committee ILl - Quasi-Static Response."
Proc. of 13th Int. Ship and Offshore Structures Congress (Ed. T.
Moan and S. Berge), Elsevier Science.
ISSC (2000). "Report of Committee II.1 - Quasi-Static Response."
Proc. of 14th Int. Ship and Offshore Structures Congress (Ed. H.
Ohtsubo and Y. Sumi), Elsevier Science.
Iwahashi, Y. et al. (1998). "Finite Element Comparative Study of Ship
Structural Detail." Marine Structures 11, pp. 127 - 139.
Kim, W. S (1999): Fatigue Tests of Typical Welded Joints. Hyundai
Heavy Industries Co., Ltd. Ulsan (unpublished).
Lotsberg, I. (2001). "Overview of the FPSO - Fatigue Capacity JIP."
Proc. of OMAE'01, ASME, Rio de Janeiro.
Maddox, S. (2001). "Recommended Design S-N Curves for Fatigue
Assessment of FPSOs. ISOPE' 2001, Stavanger.
Niemi, E. (1993). "Recommendations Concerning Stress Determination
for Fatigue Analysis of Welded Components". IIS-IIW-1221-93,
Abington Publishing, Cambridge (UK).
Poutiainen, I. and Niemi, E. (2000). The Determination of Hot Spot
Stress in Gusset Structures Using a Coarse Element Mesh. IIW-Doc.
XIII- 1820-2000, International Institute of Welding.
Radaj, D. (1990). Design and Analysis of Fatigue Resistant Structures.
Abington Publishing, Cambridge (UK).
Yagi, J.; Machida, S.; Tomita, Y.; Matoba, M.; Kawasaki, T. (1991).
"Definition of Hot Spot Stresses in Welded Plate Type Structure for
Fatigue Assessment." IIW-Document IIW-XIII-1414-91, International Institute of Welding.

Fatigue Assessment:
If the hot spot stress is evaluated by linear extrapolation in the way
described, the fatigue strength may be assessed with a usual design S-N
curve based on hot-spot stresses (e. g. Hobbacher, 1996).
Alternatively a simplified approach without any stress extrapolation
seems to be reasonable where the stress is taken at the location 0.5 t
(resp. 5 mm for weld toes of type b) away from the hot spot modelled
and assessed with a design S-N curve reduced by one fatigue class.
If the hot spot stress is evaluated from strain measurements or from
refined models with improved finite elements, a stress extrapolation
over reference points at distances 0.4 and 1.0 x plate thickness or a
quadratic extrapolation is recommended in line with Hobbacher (1996).
The present study has been used in the JIP together with other data
to derive a design hot spot S-N curve (Maddox, 2001). Here, a lower
S-N curve was found, which might be due to additional test data and in
particular to the inclusion of larger plate thickness (up to 25 ram).