You are on page 1of 10

John Dalsgaard Srensen

Aalborg University,
Sohngaardsholmsvej 57,
DK-9000, Aalborg,
Denmark
e-mail: jds@civil.auc.dk

Jesper Tychsen
Maersk Oil and Gas AS,
Kanalen 1,
DK-6700 Esbjerg,
Denmark
e-mail: jpt@maerskoil.dk

Jens Ulfkjr Andersen


Det Norske Veritas,
Tuborg Parkvej 8,
DK-2900 Hellerup,
Denmark
e-mail: Jens.ulfkjaer.andersen@dnv.com

Ronnie D. Brandstrup
COWI A/S,
Stormgade 2,
DK-6700 Esbjerg,
Denmark
e-mail: rdb@cowi.dk

Fatigue Analysis of Load-Carrying


Fillet Welds
The fatigue strength of load-carrying fillet welds is, in most codes of practice, performed
neglecting the influence of bending in the weld throat section. However, some commonly
applied structural details give rise to significant bending in the weld throat section. An
example of such a detail is a doubler plate connection, which is often applied in connection with modifications of offshore structures. As a part of the present work, fatigue tests
have been performed with test specimens fabricated by the current industry standard for
welded offshore steel structures. The fatigue tests show that the degree of bending (DOB)
has an influence on the fatigue lifetime. The fatigue lifetime decreases significantly when
increasing the bending stress. In order to take into account the effect of the bending, a
new fatigue stress definition applicable for fillet welds failing through the weld is presented. Using the test results, it is shown that the new definition of fatigue stress can be
used for a wide range of DOB with a low standard deviation of the resulting SN
curve. DOI: 10.1115/1.2163876

Introduction
This paper focuses on determination of the fatigue strength of
the load-carrying fillet welds. Fatigue design of fillet welds is, in
most codes of practice, performed neglecting the influence of
bending in the weld throat section. The design stress is obtained
by dividing the relevant component of the force transmitted per
unit length of the weld by the throat weld area. Although most
codes of practice, generally, do not allow use of single-sided fillet
welds, some commonly applied structural details, e.g., doubler
plates Fig. 1, give rise to significant bending in the weld throat
section.
The current design guidance for fatigue assessment of fillet
welds is based on tests of cruciform specimen, see Fig. 2. In
cruciform joints, the degree of bending is small. Considering that
limited bending is present in all tests making the background for
the current code guidance and the fact that the bending stress is
not part of the fatigue stress definition, it can be discussed if the
current guidance is applicable for fillet welds subject to a larger
degree of bending.
The present work includes six fatigue test series 28 tests in
total with a varying degree of bending in the fillet throat section.
In the tests, the bending stress ranges from 30% of total stress in
cruciform joint specimen to around 60%-95% in planar doublerplate joints. The tests include ten cruciform joints to allow direct
comparison to available test results from the literature see 1.
The test specimens are fabricated to the current industry standard
for welded offshore steel structures.
Statistical analysis of the new test data is performed using classical statistical methods and the maximum likelihood method
MLM. The resulting statistical uncertainties are used to determine characteristic fatigue design curves. In order to take into
Contributed by the Ocean Offshore and Arctic Engineering Division of ASME for
publication in the JOURNAL OF OFFSHORE MECHANICS AND ARCTIC ENGINEERING. Manuscript received January 21, 2005; final manuscript received September 23, 2005.
Assoc. Editor: Pingsha Dong.

account the effect of the bending, a new fatigue stress is defined


and validated. The new fatigue stress is based on the largest principal stress and is applicable for fillet welds loaded primarily in a
plane perpendicular to the weld axis mode I loading at the fillet
root.
The effect of statistical uncertainty is evaluated, and the significance for fatigue reliability analysis of load-carrying fillet welds
subject to bending is discussed. Finally, a recommendation for a
new design SN curve for load-carrying fillet welds is proposed
together with specifications for calculation of the fatigue stress.
The stress field in the fillet welds is too complex to allow desk
calculation of the hot-spot stress. For this reason, the new specification is based on finite element FE analyses.

Fatigue Tests
Cruciform and doubler plate joints are sketched in Fig. 2. A
large number of fatigue results are available for cruciform joints,
whereas it is difficult to find test results for fillet welds subject to
bending in the literature. Figure 3 illustrates normal stresses in a
doubler-plate joint with transverse loading. Two points with large
stress concentrations are identified. Fatigue design of fillet welds
is in, e.g., Eurocode 3 2 performed without taking into account
the degree of bending. The fatigue stress S is obtained by dividing
the relevant component of the force F transmitted per unit length
of the weld, by the throat size a. The stress is also mentioned as
the nominal shear stress on the minimum weld throat area
see 2
S=

F/2
al

where l is the length of the weld. It is seen that the bending stress
is not a part of the fatigue stress definition. The degree of bending
DOB is defined by

Journal of Offshore Mechanics and Arctic Engineering


Copyright 2006 by ASME

FEBRUARY 2006, Vol. 128 / 65

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Fig. 4 Importance of degree of bending DOB

The cruciform test geometries have a DOB of 0.30, whereas


the DOB for the doubler plates are 0.60.95. Data for the test
specimens are as follows:

Fig. 1 Jacket structure, with doubler-plate connection

Width l of all test specimens is 100 mm.


Fabrication/assembly process is selected such that it is ensured that the two loading plates are perfectly aligned and
perpendicular to the doubler plates. To ensure an ideal twodimensional 2D weld geometry, 2 3 mm of the plate
width was ground away from the sides of the specimen as it
is difficult to keep the a measure at the plate edges during
welding
The steel material is FE510 with 3.1.B certificates. All materials of equal thickness are cut from the same plate. All
welds are shielded metal arc welding SMAW. 100% MPI
Welding Class C from the Danish code has been performed. Welding and nondestructive-testing NDT specification are representative for doubler-plate joints on offshore
structures.
Loading is with constant force amplitude, and the loading
frequency is 10 Hz. The complete load cycle is in tension.
See Tables 1 and 2 for maximum and minimum load-cycle
loads. Failure is taken as total rupture, see Fig. 5.
Extensive measurement of the actual as-built geometry was
performed prior to testing. This included measurement of
the a measure at five different locations in each fillet weld,
measurement of actual specimen width, plate thickness, and
alignment of loading plates.

DOB =

bending
membrane + bending

where the total stress is divided in membrane stress membran and


bending stress bending obtained by linearization of the stresses in
the throat section. It is noted that the DOB is not uniquely defined
from a FE analysis, as the singularity at the weld-root location
makes it impossible to obtain convergence on the linearized
stresses.
The new fatigue tests have been performed at Aalborg University; see Andersen and Brandstrup 3 and the next section. In Fig.
4, the results are plotted based on the stress definition from Eq.
1. It is seen that the degree of bending has a large influence on
the fatigue lifetime. The fatigue lifetime decreases significantly
with an increasing DOB. Figure 4 also shows a comparison of SN
curves from different standards. It is seen that for large DOB the
characteristic SN-curves from common codes of practice are unsafe and cannot be used without special attention. Therefore, the
bending stress has to be included in the design development of
new fatigue stress.

The tests are made at the Structural Research Laboratory, Aalborg University, Denmark. The test setup and procedure are illustrated in Fig. 5. In the tests, the following equipment is used: MTS
Material Test System test machine 810, MTS 418.91 Microprofiler, and MTS 458.20 Microconsole. The test machine was calibrated by the Force Institute, Denmark, prior to testing.

Fig. 2 Cruciform and doubler-plate joint

Table 1 Test results with cruciform joints with fillet welding F:


applied force, a: mean throat width of ruptured welds
ID

Fmin / Fmax / Frange


kN

a
mm

S
MPa

1A
1B
1C
1D
1E
1F
1G
1H
1I
1J

10.5/ 67.4/ 56.9


8.1/ 85.3/ 77.2
10.6/ 86.0/ 75.4
8.8/ 102.6/ 93.8
10.0/ 127.4/ 117.4
10.4/ 142.8/ 132.4
12.3/ 161.3/ 149.0
9.3/ 183.9/ 174.5
9.0/ 201.6/ 192.6
9.9/ 253.0/ 243.2

4.76
4.89
5.08
5.12
4.92
4.94
4.96
5.01
5.19
5.11

64.16
81.22
77.08
102.21
127.52
142.94
155.78
185.26
198.23
263.24

2,057,708
995,783
1,089,816
535,911
268,067
227,872
115,163
75,237
41,956
18,766

Fig. 3 Typical stress distribution in doubler-plate joint

66 / Vol. 128, FEBRUARY 2006

Transactions of the ASME

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Table 2 Test results for doubler plates

Table 3

ID

Emln / Fmax / Frange


kN

a
mm

S
MPa

2A
2B
2C
3A
3B
3C
4A
4B
4C
5A
5B
5C
6A
6B
6C
6D
6E
6F

9.8/ 103.6/ 93.9


11.3/ 169.5/ 158.4
9.1/ 120.7/ 111.6
10.0/ 93.8/ 83.8
10.3/ 143.7/ 133.5
10.4/ 109.8/ 99.4
9.3/ 84.7/ 75.4
12.0/ 139.5/ 127.6
10.5/ 99.9/ 89.5
9.7/ 89.1/ 79.5
12.0/ 129.5/ 117.5
10.8/ 65.3/ 54.5
12.3/ 124.0/ 111.7
12.2/ 93.7/ 81.5
11.8/ 80.2/ 68.4
10.2/ 64.0/ 53.8
14.4/ 204.2/ 189.8
13.4/ 109.1/ 95.7

4.98
5.35
5.15
4.86
4.84
5.02
5.23
5.24
5.04
5.31
5.46
5.33
5.30
5.46
5.02
5.15
5.16
5.16

93.52
142.37
103.84
81.14
141.97
103.18
70.06
116.72
86.43
70.88
103.86
53.05
101.67
72.57
67.41
52.01
184.64
88.34

1,584,794
139,816
475,971
664,267
110,400
623,224
2,369,666
93,533
328,190
330,459
42,966
3,100,000a
22,387
97,169
197,772
651,476
2363
31,476

Runner

The tests are planned and performed according to the general


specifications in Eurocode 0 4. The planning of the experiments
is based on preliminary computational models.
The test results are analyzed using the maximum-likelihood
method. The basic linear SN curve is written
N = ASm

or

logN = logA m logS

where N is the number of cycles to failure with stress range S and


A and m are parameters to be fitted from the test results using the
maximum-likelihood method. Equation 3 is rewritten

Definition of characteristic fatigue SN curve

Document

Basis for Design SN curve

Eurocode 3

75% confidence interval of 95%


probability of survival for logN test
data, accounting for standard deviation
and sample size
95% lower confidence limit on logN
test data
Mean minus two standard deviations on
logN test data

Background
document for EC3
NORSOK N-004
HSE 4th ed. GNs
BS 7608: 1993
API RP2A

Lower bound on test data

nR

PN logA m logS +
i

i=1

where Ni is the number of stress cycles to fail with stress range Si


in test number i. nF is the number of tests where failure occurs,
and nR is the number of tests where failure did not occur after Ni
stress cycles runners. n = nF + nR is the total number of tests.
A, m, and are obtained from the optimization problem
maxA,mLA , m , , which is solved using a standard nonlinear
optimizer e.g., the NLPQL algorithm, see 5.
Because the parameters A, m, and are determined using a
limited amount of data, they are subject to statistical uncertainty.
Since the parameters are estimated by the maximum-likelihood
technique, they become asymptotically number of data should be
25 30 normally distributed stochastic variables with expected
values equal to the maximum-likelihood estimators and covariance matrix equal to see, e.g., 6
CA,m, = HA,m,1 =

A2

A,mAm

A,A

A,mAm

m2

m,m

A,A m,m

where is assumed to be normal distributed with mean value= 0


and standard deviation= . It is noted that the test data do not
allow a bilinear SN curve to be fitted.
The log-likelihood function for the parameters A, m, and
becomes

where HA,m, is the Hessian matrix with second-order derivatives


of the log-likelihood function. A, m, and denote the standard deviations of A, m, and , respectively, e.g., A,m is the
correlation coefficient between A and m.
The characteristic SN curve can be defined in different ways,
see, e.g., 7 and Table 3, where the number of data is taken into
account using the Student-t distribution. In the following, the SN
curve is defined as 95% probability of survival for logN test
data:

logN = logA m logS +

nF

LA,m, =

PN = logA m logS +
i

i=1

PN logA m logS + = 0.05

Fig. 5 Test setup and procedure

Journal of Offshore Mechanics and Arctic Engineering

Given the stress range S, the number of stress ranges N can be


determined from 7. The fatigue strength c corresponding to
Nc = 2 106 cycles is also determined from 7. If statistical uncertainty is included, then A, m, and are modeled as stochastic
variables. The probability PN logA mlogS + can be estimated by the first-order reliability method FORM, see 8,9,
and/or by Monte Carlo simulation.
From Table 3, it is noted that the fatigue strength curve from
Eurocode 3 is defined slightly differently. In Eurocode 3 2, the
characteristic fatigue strength curve is defined as the lower 75%
confidence interval of 95% probability of survival for logN test
data, accounting for standard deviation of the test results and
sample size. The least-squares method LSM is used. In the background document for Eurocode 3 10, the characteristic fatigue
strength curve is defined as the lower 95% confidence interval on
the mean SN curve. In other codes of practice, the characteristic
SN curve is defined as the mean minus two standard deviations on
logN data.
FEBRUARY 2006, Vol. 128 / 67

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Fig. 6 Tests specimens measurements are in millimeters.


The nominal value of a is 5 mm.

Cruciform Joints With Fillet Welding. The cruciform geometry is shown in Fig. 6 and the test results in Table 1.
The results are analyzed as described above. In Table 4, results
using the maximum-likelihood method MLM are shown and in
Table 5 the results using the least-squares method are shown. Results are obtained both with m free and with m fixed to 3.0. Results with and without statistical uncertainty included.
In Fig. 7, the test results for the cruciform joints are shown
together with the associated SN curves from the maximumlikelihood method. In Table 5, results are shown for the leastsquares method and EC3 definition of characteristic SN curves. It
is seen that the standard deviation of the test results is very low,
also for the case with m = 3, and that almost the same statistical
results are obtained using the MLM and LSM. m fixed to m = 3
results in lower fatigue strengths c, and furthermore, it is seen
that it is not important to include statistical uncertainty in this
case.
It is noted that the test procedure implies that fatigue failure can
take place in the symmetrically placed fillet welds. This situation
can easily be modeled in the likelihood function by adding a multiplicative term expressing no failure in the other welds than the
one failing see discussion in relation to Table 10.
In order to verify the tests, a comparison has been made to test
results for cruciform joints available in the literature from Gurney
et al. 1 see Fig. 8. It is seen that the test results from Table 1
are representative since they are almost placed on the mean SN
curve for all tests. Therefore, the same test procedure is used also

Fig. 7 Test results series 1 for transverse fillet weld connection and characteristic SN curve maximum-likelihood
estimate

for the tests with doubler plates.


It is seen that the standard deviation of the test results from 1
is larger than the scatter from the cruciform tests performed as
part of the present work see Tables 4 and 5. The high scatter in
the tests reported in 1 is believed caused by:

Uncertainty in as-built dimensions. E.g., use of nominal


weld sizes instead of actual values may result in significant
scatter especially for small welds.
Different welding and NDT procedures. It is not unlikely
that some of the tests in 1 have been fabricated by inexperienced laboratory staff and without NDT.
Gurney et al. 1 reports that some of the specimens include
welding against painted plates highly scattered fabrication
quality.
Furthermore, when comparing with the EC3 36* and ISO
13819-2 W SN curves it is seen that these are acceptable com-

Table 4 Maximum-likelihood estimates


m free
Mean
SN curve
Standard deviation
Characteristic SN curve

logN=
12.38 3.3 log
0.071
logN=
12.24 3.33 log
c : + static uncertainty MPa 61
c : static uncertainty MPa 62

m=3
logN=
11.69 3 log
0.094
logN=
11.52 3 log
55
56

Table 5 Least-squares estimates

Fig. 8 Comparison to test results for test results for cruciform


joints from literature 1 maximum-likelihood estimates
Table 6 Maximum-likelihood estimates for all cruciform joints

m free

m=3

logN=
12.39 3.33 log
0.079
logN=
12.21 3.33 log
59
60

logN=
11.69 3 log
0.099
logN=
11.47 3 log
53
54

m free
Mean
SN curve
Standard deviation
Characteristic
SN curve
c: EC3 MPa
c: mean2 standard deviation MPa

68 / Vol. 128, FEBRUARY 2006

Mean
SN curve
Standard deviation
Characteristic SN curve

logN=
11.73 3.01 log
0.328
logN=
11.19 3.01 log
c : + static uncertainty MPa 42
c : static uncertainty MPa 42

m=3
logN=
11.71 3 log
0.328
logN=
11.19 3 log
42
42

Transactions of the ASME

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Fig. 9 Nominal test specimen geometry measured in


millimeters

pared to the test results and characteristic SN curves from tests.


The likely reason for this is that the W curve has been developed
based on the data in 1.
From the above, it is clear that a detailed knowledge of what
you have tested is vital in the effort to obtain reliable and low
scattered results e.g., any crude as-built measurements will turn
into scatter in the SN curve at a later stage. For this reason, as
described earlier, a large effort has been put into accurate as-built
measurements and representative fabrication.
Fillet Welds Subject to Bending. In Fig. 9 and Table 7 the
nominal geometry of the planar doubler-plate joint test specimens
is shown.
In Table 2, the test results are shown. In Fig. 10, they are shown
together with the characteristic SN curve in EC3 2 and ISO
13819-2 11.
In Table 8, results using MLM are shown. It is seen that the
standard deviation of the test results is very large, and the c
values are low compared to the results in Tables 46. The bending
stress decreases the fatigue strength and cannot be neglected. The
influence of the bending stress in relation to the statistical uncertainty is significant. The EC3 36* and ISO 13819-2 W SN curves
are nonconservative if applied for assessment of fillet welds with
a significant bending stress in the throat section and using the
standard fatigue stress definition from Eq. 1. For accurate assessment to be possible, a new assessment approach needs to be
developed.

Development of New Stress Measures


From the previous sections, it is clear that an improved assessment approach is required. The stress distribution within the fillet
weld is complex and will for many applications, e.g., doublerplate joints, be dependent on the local attachment stiffness. In
many cases, the fillet weld attachment is statically indeterminate,
making the balance between membrane and bending stress dependent on stiffness. Recognizing this complexity, it is clear that a
simple desk-calculation-based approach will never be accurate
enough for general applications. The obvious alternative is use of
the FE method to calculate the fatigue stress.
Investigations to determine the best possible stress parameter
have been performed in both 4 using ABAQUS and 12 using
ANSYS. The remaining of the present section is based on the results from 12.

Fig. 10 Test results for doubler plate together with stress definition from Eq. 1 maximum-likelihood estimates

A number of alternative stress parameters for the fatigue analysis have been investigated. In the following, comparisons are
made for the alternatives that turn out to be the best. Using the
terminology from 12, the alternatives for the new fatigue
stresses are, in the following named, SCF4 and SCF5. Both stress
measures are based on the numerically largest principal stress in
the fillet throat section. SCF4 is based on linear extrapolation of
the principal stress from two points in the fillet throat section to
the weld root see Fig. 11.
SCF5 is based on linear extrapolation of the six basic stress
components in the fillet throat section to the weld root. SCF5 is
the numerically largest principal stress calculated from these linear extrapolated stresses. Further details in relation to the stress
definition will follow. The main difference between SCF4 and
SCF5 is that SCF4 is based on extrapolation of the numerically
largest principal stresses evaluated in the extrapolation points. If
the direction of the principal stress is not the same in both extrapolation points and in case several load cases have to be added
and the principal stress directions are not the same between each
loadcase, SCF4 will lead to an error in the extrapolated stress. In
SCF5, this problem is overcome by first extrapolating the basic
six stress components for each load case, performing superposition of the basic stresses from different load cases, and finally,
calculating the main stress from the outcome of the superposition
of the basic stresses.
A FE model geometry has been prepared for each of the 28 test
specimens. Each model is based on the detailed as-built measurements performed. To determine requirements for meshing, element type, weld modeling, and stress sampling, 35 different
analysis series each series consist of 28 FE analyses have been
performed. A brief summary of this sensitivity investigation is
included later on.
In the following, the results for the base case analysis series
are presented. The base-case sets the crudest acceptable standard
for modeling, etc. The base-case specification is as follows:

20 node second-order isoparametric brick elements ANSYS


Solid 95
two elements along the fillet throat section
Table 8 Maximum-likelihood estimates
m free

Table 7 Length of doubler plates and degree of bending


Specimen ID
2A,
3A,
4A,
5A,
6A,

2B,
3B,
4B,
5B,
6B,

2C
3C
4C
5C
6C, 6D, 6E, 6F

Length of doubler plate


mm

DOB

60
80
100
120
140

0.67
0.80
0.94
0.97
0.93

Journal of Offshore Mechanics and Arctic Engineering

Mean
SN curve
Standard deviation
Characteristic SN curve

logN=
10.92 2.78 log
0.53
logN=
9.69 2.78 log
c : + static uncertainty MPa 17
c : static uncertainty MPa 22

m=3
logN=
11.36 3 log
0.53
logN=
10.44 3 log
24
25

FEBRUARY 2006, Vol. 128 / 69

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Fig. 11 Stress extrapolation points for SCF4 and SCF5 SEXP: extrapolated stress

two elements over the thickness of the surrounding plates


straight fillet weld surface profile
stress sampling from all elements in the model i.e., not only
from the fillet weld elements

The values of SCF4 and SCF5 from the 28 analyses performed


in the base-case analysis series are included in Table 9. Discussions and statistical calculations are included in the following sections for both SCF4 and SCF5.
Alternative Fatigue Stress Parameter SCF4. The stress parameter SCF4 is the numerically largest principal stress at the
weld root linearly extrapolated from the quarter points 4th in the
fillet throat section Fig. 11. The extrapolation is illustrated in
Fig. 12.
The quarter points are chosen on the basis of a convergence/
sensitivity analysis as there is limited influence of the fillet-root
notch stress 0.25 T away from the fillet root. Because the influence of the notch stress is limited, the stress parameter becomes
relatively independent of the element intensity. The stress parameter can typically quite easily be implemented in a finite element
analysis since almost all commercial finite element programs calculate principal stresses and directions. However, differences in
calculation procedures and stress interpolation, especially to nonode locations between different software programs, may exist.
For this reason, care should be taken with the use of the SCF4

fatigue stress.
In Tables 10 and 11, results for SCF4 using MLM and LSM are
shown. The runner is included in the MLM estimates but not in
the LSM estimates. In Fig. 13, the test results are shown together
with the characteristic SN curve. It is seen that the standard deviation of the test results is much smaller, and the c values are
larger than those calculated using Eq. 1. For fixed m = 3, the
standard deviation increases. In Fig. 14, the characteristic SN
curve is shown together with the ISO 13819-2 W and the MOGGA and MOG-FA SN curves used for in -air environmental
conditions by Mrsk Oil and Gas see 13. The m = 3.0 part of
the MOG-GA curve is equal to the F2 curve in, e.g., ISO 19902
11.
As mentioned above, failure will first take place in the weld
with the shortest fatigue life. If the MLM likelihood function is
modified to take account of these runners, then for the case with m
free, the mean SN curve and the standard deviation changes to
logN = 14.25 3.94 log and = 0.19, i.e., the effect is
small. Furthermore, the results obtained using only the shortest
fatigue life in the statistical analysis are on the safe side.
Alternative Fatigue Stress Parameter SCF5. The stress pa-

Table 9 Test and analysis results for doubler plates


Specimen
ID
1A
1B
1C
1D
1E
1F
1G
1H
1I
1J
2A
2B
2C
3A
3B
3C
4A
4B
4C
5A
5B
5C
6A
6B
6C
6D
6E
6F
a

F
kN

a
mm

S SCF4
MPa

S SCF5
MPa

56.89
77.19
75.42
93.82
117.41
132.36
148.97
174.52
192.59
243.22
93.85
158.42
111.60
83.84
133.49
99.42
75.41
127.55
89.48
79.46
117.51
54.49
111.73
81.52
68.36
53.82
189.80
95.67

4.76
4.89
5.08
5.12
4.92
4.94
4.96
5.01
5.19
5.11
4.98
5.35
5.15
4.86
4.84
5.02
5.23
5.24
5.04
5.31
5.46
5.33
5.30
5.46
5.02
5.15
5.16
5.16

2,057,708
995,783
1,089,816
535,911
268,067
227,872
115,163
75,237
41,956
18,766
1,584,794
139,816
475,971
664,267
110,400
623,224
2,369,666
93,533
328,190
330,459
42,966
3,100,000a
22,387
97,169
197,772
651,476
2,363
31,476

92.92
108.33
103.44
135.67
182.30
204.82
223.98
283.54
265.53
349.78
118.15
190.05
136.44
121.49
196.43
144.22
135.21
229.38
158.15
169.78
254.33
115.24
296.92
215.69
179.20
141.23
496.93
254.76

93.98
109.21
104.29
136.78
184.38
207.13
226.51
287.62
267.68
352.64
119.43
193.79
138.81
118.87
191.77
141.11
134.18
227.37
156.64
169.77
254.27
115.24
294.00
213.82
176.88
139.19
489.63
252.15

Fig. 12

Alternative fatigue stress parameter SCF4

Table 10 Maximum-likelihood estimates-SCF


m free
Mean
SN curve
Standard deviation
Characteristic SN curve

logN=
14.18 3.93 log
0.18
logN=
13.85 3.93 log
c : + static uncertainty MPa 83
c : static uncertainty MPa 85

M =3
logN=
12.08 3 log
0.24
logN=
11.67 3 log
62
62

Runner

70 / Vol. 128, FEBRUARY 2006

Transactions of the ASME

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Table 11 Least-squares estimates-SCF4

Mean
SN curve
Standard deviation
Characteristic
SN curve
c EC3 MPa
c mean2 standard deviation MPa

m free

m=3

logN=
14.16 3.92 log
0.19
logN=
13.81 3.92 log
82
80

logN=
12.08 3 log
0.25
logN=
11.62 3 log
59
57

rameter SCF5 is obtained by linear extrapolation of the normal


and shear stress components 11, 22, 33, 12, 13, and 23 in
throat of the fillet weld to the root, as illustrated in Fig. 15. SCF5
is then calculated as the numerically largest principal stress using
the extrapolated stresses. Since the stress parameter is based on
fixed stress directions, it is possible to use superposition of the
extrapolated stresses from several load cases before evaluating the
principal stresses. In Table 9, the test results are shown and the
statistical analysis is presented in Table 12 and Fig. 16 and 17. It
is seen that the fits are almost as good as those obtained using
SCF4.
Figure 15 gives the full three-dimensional 3D definition of
SCF5. A two-dimensional 2D definition applying only x,exp,
y,exp and xy,exp can be used if the stress state is primarily perpendicular to the weld axis. If the two alternative stress parameters are compared to the stress parameter in, e.g., EC3, then it is
seen that they both result in much lower standard deviations. The
standard deviation on the new stress measures 0.20.25 is equal
to the size of the standard deviations related to SN curves for
standard welded details. Relative to the results for the standard
Eq. 1 fatigue-stress measure plotted in Fig. 10, it is seen that
the new fatigue-stress measure gather both the low and high bending test results around the same mean curve.
It is noted that for the present simple and planar geometry with
only one load case, there is almost no difference between the
accuracy of SCF4 and SCF5. However, for more complicated
nonplanar geometries and combination of different load cases, differences between the two approaches will occur. SCF5 is the theoretically best stress measure.
Selection of Design SN Curve. For the test conditions, the
applicable design SN curve can be taken directly from Tables 11
and 12. However, some care needs to be taken. Looking carefully
into the tests and analysis results, one will note that the best fit
inverse log. slope m changes between the different test series
varying DOB. The explanation for this cannot be found in different residual stress, etc., as the same fillet weld size is present in
all specimens. The reason is believed to be load shedding. For

Fig. 13 Test results and characteristic SN curve with alternative stress parameter SCF4 maximum-likelihood estimates

Journal of Offshore Mechanics and Arctic Engineering

Fig. 14 Test results and characteristic SN curve with alternative stress parameter SCF4 maximum-likelihood estimates;
fixed m = 3

both the cruciform and the doubler-plate specimen, two fillet


welds are present in the critical section total failure occurs at the
time two welds separate. The fillet welds are part of a statically
indeterminate system, see Fig. 18 for a doubler-plate specimen.
The bending load in the fillet welds is dependent on the rotational support stiffness of the fillet welds Fig. 18. A crack growing through the throat section reduces the bending moment in the
fillet throat. Failure occurs at the time the uncracked weld cannot
carry the loading any longer. This load-shedding effect will slow
down the crack growth for specimen with a high bending load in
the fillet welds. The effect will be most pronounced in specimens
with a high degree of bending and relatively low loading i.e.,
only a small fraction of the weld area is required to carry the
membrane load. The test results can be expected to have a larger
fatigue life relative to the specimen with lesser bending and higher
load level. In the SN plot, the load-shedding effect will be seen as
a higher value of the inverse log slope m. This is also what the test
results show. From Table 4, it is seen that m is 3.01 for test series
1 cruciform joints. For the cruciform joints, the load shedding is
marginal due to the low bending. Including all test data, the best
fit inverse log slope increases to almost 4 Table 12. Thus, for the
specimen with a high degree of bending, a non-negligible part of
the fatigue life will relate to a situation where a crack has propagated a distance through the throat section. The tests are performed for constant amplitude force loading. In case of variable
amplitude loading, failure may occur at a much earlier stage because a single large load peak may rupture the weld. Thus, in case
of variable amplitude loading, the best fit free SN curve from
Table 12 may not be applied because it will lead to a nonconservative fatigue life estimation for the low-stress high-cycle part of
the curve. In this case, an SN curve with m = 3 should be applied.
The curve should be selected to match the high stress low cycle
test data to ensure the load shedding effect is not included. For
variable amplitude loading, the SN curve from Table 4 with fixed
m = 3 should be applied or alternatively as the standard deviations
are almost equal, a standard ISO 19902 F2 classification can be
applied logA = 11.633.
The present tests give no information about the SN curve
around the endurance limit. However, there is no reason not to
believe that the approach generally applied for other welded components should not be applicable for fillet welds i.e., in lack of
further information, the stress cut off for constant amplitude
tests, or transition to a higher inverse log slope m at approx.
107 cycles for variable amplitude loading, as specified in common analysis codes, is recommended applied e.g., use the entire
F2 curve.
Requirements for FE Analysis. The new stress definitions are
based on FE analysis. The present section sets forth minimum
requirements for the FE analysis in order to obtain reliable and
FEBRUARY 2006, Vol. 128 / 71

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Fig. 15 Alternative stress parameter 2-SCF5

accurate results complying with the basis for the SN curve development. Requirements to the FE analysis are particularly important due to the stress singularity at the root of the fillet weld.
The analysis requirements are developed on the basis of 35
different analysis series 28 analyses in each series, investigating
the sensitivity to different modeling and analysis parameters.
Based on the base case, the following parameter variations have
been evaluated:

element type first- and second-order brick and tet

element size within fillet weld, along fillet weld and in adjacent plates close to the fillet
surface profile concave and convex investigated varying
the fillet leg length with 2, 1, 0, 1, and 2 mm
fillet weld mesh pattern 1T or 2T, see later
stress mapping to extrapolation points. Two cases investigated: Mapping from all elements or mapping only from
elements within the fillet weld. In addition, differences in
mapping to extrapolation points located on element bound-

Table 12 Maximum-likelihood estimates-SCF5


m free
Mean
SN curve
Standard deviation
Characteristic SN curve

logN=
14.20 3.93 log
0.19
logN=
13.86 3.93 log
c : + static uncertainty MPa 83
c : static uncertainty MPa 85

m=3
logN=
12.08 3 log
0.24
logN=
11.67 3 log
61
62

Fig. 17 Test results and characteristic SN curve with alternative stress parameter SCF5 maximum-likelihood estimates;
fixed m = 3

Fig. 16 Test results and characteristic SN curve with alternative stress parameter SCF5 maximum-likelihood estimates

72 / Vol. 128, FEBRUARY 2006

Fig. 18 Bending moment in doubler-plate specimen as function of support stiffness: a stiff rotational support and b soft
rotational support

Transactions of the ASME

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

Fig. 19 Mesh size requirements in FEM analysis

ary lines relative to extrapolation points in the interior of


elements have been investigated.
For each series, the best fit S-N curve has been determined.
Sensitivity of parameter selection is performed by comparison of
the SN curves log A values and standard deviation. Based on this
study, the following minimum requirements for the FE analysis
are determined:

Elements: Element types should be second-order brick or


tet elements. Linear brick elements may be applied, but it
will require an extensive number of elements. Linear tet
elements are not acceptable.
Element size requirements: See Fig. 19.
Stress mapping: Extrapolated stresses from integration
point to nodes using element shape functions, average
stresses at common nodes, extrapolate stresses to extrapolation points using the shape functions.
Weld-surface profile: If the nominal specification for the
fillet weld is with a straight surface profile and the fabrication tolerances comply with a recognized standard, e.g., ISO
5817, the weld profile may be modeled as straight in the FE
model.
Contact: In case of a doubler-plate geometry, no contact
between the main tube and doubler plate may be assumed
due to fabrication tolerances.

Effects of Scale and Environment. The work performed is


based on specimens of equal scale anominal = 5 mm, T = 10,
20 mm and constant stress amplitude tests at laboratory conditions in air. To allow a general application of the results, the
influence of the scale the thickness effect and the environment
corrosive effects are discussed in the following.
All test specimens in the present work are of a relatively small
Journal of Offshore Mechanics and Arctic Engineering

scale. If the critical location of cracking was at the toe of a weld,


one would expect a scale effect giving rise to decreasing fatigue
capacity with increasing thickness. In this case, application of
small-scale results would require careful correction of results before applying these for larger-scale structures. However, in the
case of fillet welds failing through the throat section, crack initiation is at the weld root. Doing an investigation of the scale effect
based on the 400 test results in 1, which include a range of
different plate and fillet weld thickness, one finds a slight increase
in fatigue strength with increasing thickness and/or a measure.
The likely reason being that the welding process and rotations
constraint created by two welds create residual compressive
stresses at the weld root. These residual compressive stresses
counterbalance other scale factors. Based on the above and the
fact that fillet welds are rarely very large, no scale effect is to be
applied for fatigue assessment of fillet welds failing through the
throat section.
The present tests are performed in air. The test data will cover
most fillet welded structures, even structures located below water.
This is because codes generally do not allow single fillet welds,
but require the fillet welds being continued on the whole circumference of a detail, i.e., in most cases the weld root will not be
exposed to seawater. If the weld root is exposed to seawater,
generally applicable fatigue life knockdown factors can be applied
e.g., a factor 2 for a case with adequate cathodic protection and
3.0 for a case without.

Discussion and Conclusion


The importance of including the influence of bending in the
weld throat section on the fatigue strength of load-carrying fillet
welds is demonstrated by considering doubler-plate connections.
Fatigue tests have been performed. The fatigue tests show that the
FEBRUARY 2006, Vol. 128 / 73

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

degree of bending DOB has a large influence on the fatigue


lifetime. The fatigue lifetime decreases significantly with increasing degree of bending, and SN curves in existing codes of practice
are unsafe if DOB 0.3 0.5.
No desk calculations are possible because the stress field is far
too complex. FEM analyses are required for reliable stress prediction. In order to take into account the effect of the degree of
bending, two new definitions of the fatigue stress are presented.
Using the test results, it is shown that the new definitions of the
fatigue stress can be used for a wide range of DOB with a reasonably low standard deviation of the resulting SN curve.
The new approach is applicable primarily for fillet welds loaded
in a plane transverse to the weld axis giving rise to mode I loading along the root of the fillet weld. The SN curve obtained using
a free m parameter can only be used for constant amplitude loading. The high value of m is mainly due to stress redistribution in
fillet welds when the crack grows through the throat section. For
variable amplitude loading, the allowable crack size will be less
than for constant amplitude loading because failure can occur for
a single high amplitude load peak. It is recommended to use m
= 3 for variable amplitude loading.
The stress parameter SCF5 is recommended to be used in practice since superposition of load cases is possible and ensures consistency in the calculation of the principal stresses. SN curves

74 / Vol. 128, FEBRUARY 2006

from existing codes can be used, e.g., the ISO 19902 F2 curves.

References
1 Gurney, T. R., and MacDonald, K., 1995, OTH 91 356: Literature Survey on
Fatigue Strengths of Load-Carrying Fillet Welded Joints Failing in the Weld,
The Welding Institute, Cambridge.
2 2003, EN 1993-1-9: Eurocode 3: Steel Structures, Part 19: Fatigue.
3 Andersen, J. U., and Brandstrup, R. D., 2003, Fatigue Strength of Fillet Welds
Subject to BendingIn Connection With Doubler Plate Joints, MSc. thesis,
Aalborg University.
4 2002, EN 1990: Eurocode 0: Basis of structural design.
5 Schittkowski, K., 1986, NLPQL: A FORTRAN Subroutine Solving NonLinear Programming Problems, Ann. Operat. Res., 5, pp. 485500.
6 Lindley, D. V., 1976, Introduction to Probability and Statistics from a Bayesian Viewpoint, Vol. 12. Cambridge University Press, Cambridge.
7 OTH: Comparison of fatigue provisions in codes and standards. HSE report
OTH 2001/083, 2001.
8 Madsen, H. O., Krenk, S., and Lind, N. C., 1986, Methods of Structural Safety,
Prentice-Hall, Englewood Cliffs, NJ.
9 Ditlevsen, O., and Madsen, H. O., 1996, Structural Reliability Methods, Wiley,
New York.
10 Brozzetti, J., Hirt, M., Sedlacek, G., Smith, I., and Ryan, I., Eurocode 3: Steel
Structures, Fatigue background document to Chap. 9 of EC3.
11 2004, ISO 19902: Petroleum and Natural Gas Industries-Fixed Steel Offshore
Structures draft.
12 Tychsen, J. P., 2004, Fatigue Assessment of Fillet Welds, Tech. Report,
Mrsk Oil and Gas confidential.
13 Tychsen, J. P., 2002, General Jacket Design and Reassessment Conditions,
Fatigue Analysis, Mrsk Oil and Gas confidential.

Transactions of the ASME

Downloaded From: http://offshoremechanics.asmedigitalcollection.asme.org/ on 07/08/2014 Terms of Use: http://asme.org/terms

You might also like