Вы находитесь на странице: 1из 12

International Journal of Fatigue 21 (1999) 10511062

www.elsevier.com/locate/ijfatigue

Parametric equations for T-butt weld toe stress intensity factors


F.P. Brennan a, W.D. Dover a, R.F. Kare b, A.K. Hellier

c,*

NDE Centre, Department of Mechanical Engineering, University College London, Torrington Place, London, WC1E 7JE, UK
b
Technical Software Consultants Ltd, 6 Mill Square, Featherstone Road, Wolverton Mill, Milton Keynes, MK12 1RB, UK
Materials Division, Australian Nuclear Science and Technology Organisation, Private Mail Bag 1, Menai, NSW 2234, Australia
Received 10 March 1999; received in revised form 5 July 1999; accepted 5 July 1999

Abstract
This paper describes the generation of parametric equations for weld toe stress intensity factors. The methodology employed
used a two-dimensional finite element analysis to evaluate the crack opening stress distribution in the uncracked plane of T-butt
geometries. This was then used as input into a dedicated weight function solution for the determination of stress intensity factors.
The final parametric equations describe the stress intensity factor distributions for tension and bending as a function of plate
thickness, weld attachment width, weld angle, weld root radius, crack length and crack shape. The equations are compared and
validated against a wide spectrum of published values and appear by comparison accurate and wide ranging. The validation exercise
uncovered situations where present design guidance is unconservative. 1999 Elsevier Science Ltd. All rights reserved.
Keywords: T-butt; Weld toe; Crack; Stress intensity factor; Weight function; Parametric equation; Fatigue life; Engineering design

1. Introduction
A set of parametric equations for the calculation of
stress intensity factors (SIFs) known as Yw parametric
equations for cracks at weld toes have been produced,
based on the NiuGlinka weight function method [1].
These new equations are available for both membrane
and bending loadings and include the influence of crack
size and shape, weld toe radius, weld attachment width
and weld angle. A previous study [2] showed that the
NiuGlinka weight function [1] could be successfully
used to produce accurate stress intensity factors for surface cracks emanating from weldments. That study,
however, did not take into account the attachment size,
and the equations are valid for limited ranges of geometric parameters. Another set of equations for the weld
correction factor, known as the MK factors [3] are also
available for the calculation of SIF for cracks at weld
toes. The equations for MK factors, however, do not consider the effect of weld angle or weld toe radius but do
include the effect of attachment size. MK factors are

given in PD6493 [3] as SIF solutions suitable for plate


welds. The availability of these two sets of equations is
an important step forward in the application of fracture
mechanics in design, and it would be very valuable to
determine the accuracy and limits of the two sets of
equations.
This study extends the previous study of Yw parametric equations to include the effect of attachment size
and compares the SIF solutions calculated using the Yw
parametric equations against known finite element solutions, in order to provide information on the accuracy
and limits of application and recommendation for their
use. The study provides confirmation of guidance on the
determination of weld toe stress intensity factors for
plate and tubular welds.

2. Parametric equations
The general expression for the stress intensity factor,
K, is as follows:

KYs pa
* Corresponding author. Tel: +61-(0)2-9717-3482; fax: +61-(0)29543-7179.
E-mail address: akh@ansto.gov.au (A.K. Hellier)

(1)

where s nominal stress; a crack depth; Y stress intensity


calibration factor; Y, in general, is the unknown in this

0142-1123/99/$ - see front matter. 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 2 - 1 1 2 3 ( 9 9 ) 0 0 0 9 0 - 0

1052

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

Nomenclature
a
a/c
c
FE
K
L
MK
SCF
SIF
t
T
W
Y
Yb
Yt
Yw
a
r
s
snom
sx

Crack depth
Crack aspect ratio
Half surface crack length
Finite element
Stress intensity factor
Overall width of the welded attachment
Weld correction factor
Stress concentration factor
Stress intensity factor
Attachment thickness
Plate thickness
Plate width
Stress intensity calibration factor
Stress intensity calibration factor in bending
Stress intensity calibration factor in tension
Stress intensity calibration factor based on weight function method
Weld angle (degrees or radians)
Weld toe radius
Nominal stress
Nominal stress
Stress at depth x in plate

equation and is the term calculated from SIF parametric equations.


2.1. Yw parametric equations
The Yw parametric equations from the study [2] completed in March 1990 provide Y factors for a crack emanating from the weld toe of a T-butt welded plate. The
equations include the effect of weld toe radius, weld
angle, crack depth and crack aspect ratio, and have been
produced using data derived from the use of the weight
function approach.
This study has extended the original equations to
include attachment size. The parametric equations were
produced from studies of welded connections with the
following parameters (see also Fig. 1):
Weld Angle a=30, 45, 60
Range of Crack Aspect Ratios 0a/c1.0
Range of Crack Depths 0.01a/T1.0
Range of Weld Toe Radii 0.01r/T0.066
Range of Attachment Widths 0.3L/T4.0

3. The NiuGlinka weight function


Stress intensity factors (SIFs) are functions of component geometry, crack shape and loading. For convenience SIFs may be expressed in terms of a weight

Fig. 1. (a) Local weld geometry studied (b) Geometry and loading
used to derive stress intensity factors (c) Crack geometry (semi-elliptical crack).

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

function (representing geometrical influences) and the


crack opening stress distribution in the uncracked
plane.
Niu and Glinka [1] developed a closed form parametric representation for the weight function for cracks
in T-butt welded plates using geometric mapping techniques. Their original solution was developed for a twodimensional system in terms of weld angle a, weld toe
radius r, plate thickness T and crack depth a. This
was then modified using Newman and Rajus flat plate
SIF solutions [4] for cracks emanating from threedimensional weldments.
The NiuGlinka weight function solutions are applicable to any symmetric stress system and valid for a
wide range of geometries. In order to derive the SIF from
the weight function solution, the stress distribution in the
uncracked plane in the path of an expected Mode I crack
must first be known. A major part of the work reported
here was to define these stress distributions for the
required loading systems.

4. Stress analysis
The stress distribution in the uncracked plane for a
range of T-butt geometries was determined using twodimensional finite element analysis. The entire analysis
was carried out using the ideas [5] package. A twodimensional idealisation was considered adequate as the
NiuGlinka weight function solution is primarily for an
edge crack and corrected for a plane stress three-dimensional weldment. Eighty-six geometries varying weld
angle, weld toe radius, weld attachment size and plate
thickness were considered. Eight-noded shell elements
were used in the mesh definition, an example of which
is shown in Figs. 2 and 3. Fig. 2 shows the weldment

Fig. 2.

Finite element mesh.

Fig. 3.

1053

Finite element fillet radius detail.

loaded in axial tension, the boundary restraint opposite


simulating symmetry about the xz plane achieved by
applying full restraint in the ydirection. No restraints
were applied in the x or zdirections. Extensive convergence studies, mesh optimisation and investigation of the
effect of non-variable geometric parameters (i.e. attachment length and distance of critical region of interest to
the applied boundary conditions) were conducted. Fig.
3 shows the fillet detail and mesh displaying minimal
element distortion and stretch.
Free-end boundary conditions were applied to simulate a uniformly distributed axial force over thickness
T for tension and a pure bending moment over the edge
T for the bending cases. The maximum component of
stress in the xdirection (Mode I crack opening) was
found to vary in position from geometry to geometry. In
each case the stress distribution was plotted along the
fillet radius to identify the position and magnitude of the
stress concentration factor (SCF). Subsequent maximum
stress values were chosen in the ydirection through the
thickness T. These values were not curve fitted but
used directly as input to the NiuGlinka weight function.
The study initially set out to consider the parameter
r/T=0; however, after observation of those results it was
concluded that, using finite elements, the resultant SCF
was due to a numerical singularity instead of real structural behaviour. In practical terms r/T=0.01 represents a
realistic extreme. Therefore the six geometries with
r/T=0 were not used in this study.
Some additional cases (to those first planned) were
modelled for completeness and contributed to better SIF
curve definition. In all, eighty meshes were generated
and run in tension and bending, producing on average
sixty points of crack opening stress component over the
range 0x/T1.0 for each case.

1054

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

5. Generation of parametric equations


Normalised, through-thickness stress distributions
(sx/snom vs x/T) were obtained from the 2-D FE analyses
of T-butt welded joints, possessing various values of the
geometric input parameters (a, r/T, L/T), under first tension and then bending loading. The NiuGlinka weight
function method [1] was used for each stress distribution
in turn (with a/c values of successively 0.0, 0.04, 0.1,
0.2, 0.4, 0.6, 0.8 and 1.0, and with c/W fixed at 0.1) to
derive the stress intensity calibration factor Y for 100
values of a/T ranging from 0.01 to 1.00 in increments
of 0.01. This resulted in a total of 64,000 Y factors
80 geometries8 (a/c) ratios100 (a/T) ratios as a
function of various discrete values of the input parameters (a, r/T, L/T, a/c, a/T) for each of the tension and
bending cases.
Existing parametric equations for Yt (tension) and Yb
(bending) due to Hall, Topp and Dover [2] were available, based on a much smaller number of 2-D FE results,
and for L/T=2.8 only. As a starting point, the predictions
of these equations were evaluated for all of the cases in
the present database, and the percentage errors calculated
for each. Representative distributions of Y, as obtained
from the NiuGlinka weight function calculation using
the present FE stress distributions, and as predicted by
the existing parametric equations, were plotted as a function of a/T, and then examined to see how closely
they corresponded.
5.1. Tension
The existing parametric equation for Yt was found to
provide a surprisingly reasonable (but nevertheless
unacceptable) fit to the present results even for low L/T,
although it systematically overpredicted, particularly for
a/T values less than about 0.2, and gave very large overpredictions as a/T approached zero, in many instances.
It was therefore decided to utilise this equation as a starting point for the present derivation. Accordingly, the first
step taken was to modify the existing equation by the
addition of a negative term C3 of the functional form:
k1(0.2a/T)k2(1.1a/c)k3(L/T)k4, applicable where
a/T0.2. The parameters a and r/T were omitted for
simplicity because, unlike the other variables, there was
no obvious correlation between their magnitudes and
that of the error. Initial values for the parameters k1,
k2, k3 and k4 were obtained by linear regression of the
logarithms of the differences between predicted and
actual values of Yt, against the corresponding logarithms
of (0.2a/T), (1.1a/c) and (L/T), at selected values of
a/T spanning the range over which the correction
applied. All regression analyses were conducted using
MINITAB statistical software [6]. The incorporation of
this term resulted in a marked improvement in the accuracy of the parametric equation, as evidenced by a histo-

gram plot of percentage errors. However, for some


ranges of the geometric input parameters a kink was
apparent at the point where C3 took effect. This problem
was removed by multiplying C3 with a negative
exponential smoothing function, so that its effect was
introduced more gradually. The parameters k1k4 were
then, successively, iteratively adjusted to optimise the
distribution of percentage errors.
Upon re-examining the accuracy of the equation for
various values of the geometric parameters, as a function
of a/T, it was still found to often overpredict considerably for very low a/T. Accordingly, a further negative
corrective function C4 was overlaid on the equation, for
a/T0.03, of the form: k1(0.03a/T)k2 (1.1a/c)k3
ak4 (r/T)k5 (L/T)k6. In a similar manner as for C3,
the values of k1k6 were obtained by linear regression,
C4 was multiplied by an appropriate smoothing function,
and k1k6 were then readjusted. The smoothing function
for C4 had to be modified slightly for low L/T under
certain conditions to avoid a pronounced kink in the Yt
versus a/T curve.
Finally, a multiplicative factor of just over unity was
applied to the original component of the parametric equation, to correct a small systematic underprediction
which had resulted from the changes made. The equation
was then optimised by iteratively tweaking all of the
newly introduced constants, so as to produce a percentage error distribution with the highest possible peak and
the smallest possible spread. The final form of the parametric equation is given in the next section and the
resulting histogram plot of percentage errors is shown
in Fig. 4. The largest underprediction was by 9.0%;
the largest overprediction was by +32.6%. It is clear that
this equation is a very good fit to the data.

Fig. 4.

Deviation of curve fit from NiuGlinka Y values tension.

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

1055

5.2. Bending
The procedure used to derive the parametric equation
for Yb was very similar to that already described for the
tension case. The existing parametric equation for bending due to Hall, Topp and Dover was again found to be
a reasonable starting point for producing new equations.
The Yb values from Ref. [2] tend to be low, frequently
less than 1, and go negative just before a/T reaches 1.
This leads to a considerable number of very large positive and negative percentage errors, although the FE
results and predicted values may actually lie quite close
together in absolute terms. For this reason, and because
Yb values are most likely to be used for a/T0.5, the
histogram of percentage errors used to develop the equation was compiled from this region only. It was apparent that the existing equation systematically overpredicted over most of this region, with the largest
overpredictions occurring for very low a/T (although
even in this region there were also some large
underpredictions).
The same approach used for the tension case, of
adding functional forms involving products of the input
parameters to various powers, was applied to the bending equation. In this case, all of the input parameters
were utilised in each corrective function, in order to gain
the maximum possible improvement in accuracy. A
negative corrective function C3 was first added for
a/T0.25, and its parameters optimised. Secondly, a
function C4 was superimposed upon this where
a/T0.05, which was negative for L/T0.455 and positive for L/T0.455. Thirdly, another negative function
C5 was incorporated where a/T exceeded 0.35. Lastly, a
multiplicative factor of just less than unity was applied
to the natural logarithm component of the original equation, in order to remove the systematic overprediction
which was still apparent from the histogram of percentage errors. All of the coefficients obtained by regression
were then tweaked to optimise the overall accuracy of
the final parametric equation, which is given in the following section. The largest underprediction was by
19.3% and the largest overprediction by +55.8%. A
histogram plot of percentage errors is shown in Fig. 5
(for a/T0.5), from which it is apparent that the equation
fits the data well.
6. Parametric equations for stress intensity factors

Fig. 5.

Deviation of curve fit from NiuGlinka Y values bending.

NOTE: exp denotes exponential function.


where







0.5

a
c

P0.3650.207

a
c

C00.9631.102

a
Mp
c

0.144

0.5

1.430

a
M0
c

C13.0846.542

a
c

0.5

a
a
M1 0.156
c
c

C12.9133.245

a
c

0.5

a
a
M1 0.156
c
c

C122.62710.767

9.023

a
c

C120.06250.557

1.761

0.5

a
c

9.553

a
c

0.5

0.156

a
0.2
c

a
c

a
0.2
c

if C120.914 then C120.914


C2C12M2

Mp=0.1720.1550 a0.0016 (T/r)


M0=0.2840.1780 a0.0046 (T/r)
M1=0.317+0.0115 a+0.0099 (T/r)
M2=0.0045+0.2060 a0.0054 (T/r)

6.1. Tension
The parametric equation for tension is:
Yt1.03

a P
a
a
exp C0C1
C2
T
T
T

C3C4

(2)

C30.45 0.2

a
T

0.409

1.1

a
c

0.3

L
T

0.549

1.0

1056

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062


a
T
0.15

0.2

exp

a
0.2
T

a
0.2
T

C30

a
C42.5 0.03
T

1.1


a
c

0.7

r
T

a1.5


0.03

a
T

0.285

L
T

29.768

0.5

a
c

C21.6480.926

0.5

a
a
M1 0.1
c
c

a
a
M1 0.1
c
c

2.876

a
c

C28.75841.156

1.28

22.264

0.5

a
c

a
a
M0 0.1
c
c

0.648

0.5

a
c

C12.66422.408
C13.8606.128

0.5

a
c

C00.6451.111

a
a
M2 0.1
c
c

0.00393

a
a
M2 0.1
c
c

0.394

1.0 exp
MA=0.5970.649 a0.0028 (T/r)
M0=1.2821.325 a0.0077 (T/r)
M1=2.222+2.154 a+0.0170 (T/r)
M2=0.7890.621 a0.0097 (T/r)

a
0.03
T

S4


0.5

a
T

C30.25 0.25

1.1

where S4=0.006
r
L
but S4=0.018 if a0.6109 and 0.035 and 0.35
T
T
C40

a
0.03
T

a
c

0.16

r
T

a2.0

0.16

0.37

L
T

1.0 exp

NOTE: a is in radians.
6.2. Bending


a
T
0.15

0.25

The parametric equation for bending is:

a
a
a 2
Yb0.96A ln
C0C1
C2
C3C4
T
T
T
C5

a
0.25
T

C30

NOTE: ln denotes natural logarithm.


where

A0.3880.958

a
c

0.5

a
c

0.5

A0.6860.310
C00.5444.125

(3)

a
c

0.5

a
MA
c

1.111

0.0622

4.018

C44.0 0.05
a
0.1
c

a
a
MA 0.1
c
c

a
a
M0 0.1
c
c

a
0.25
T

a
T

0.565

1.1

a
c

0.3

a1.35

r
T

0.3

0.455

L
T

0.204

1.0 exp

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062


0.05

a
T

1057

L
a
0.05 and 0.455
T
T

S4

where S4=0.05
L
r
but S4=0.06 if a0.6109 and 0.04 and 0.35
T
T

C40.5 0.05

0.455

1.1

a
T

1.1

0.0384

a
c

0.486

a2.66

r
T

0.11

L
T


a
T
0.015

0.05

1.0 exp

a
L
0.05 and 0.455
T
T
a
0.05
T

C40

C50.14


a
c


a
0.35
T

a0.675

C50

a
0.35
T
0.2

1.1

0.862

0.098

r
T

0.077

L
T

0.148

1.0 exp

Fig. 6. Comparison of the weld toe SIF parametric equation (tension)


with NewmanRaju flat plate solution [4].

published data for validation purposes. Although relevant available data are sparse, sufficient cases were
compared with the SIF parametric equations to establish
confidence in their behaviour.
The equations were firstly compared with the NewmanRaju flat plate SIF solutions [4] for surface cracks.
These are shown in Figs. 6 and 7 for tension and bending
cases respectively. These show the decaying effect of
the weldment as a/T increases from zero. As expected
there is some influence of the weldment through to the
end; however, behaviour of the SIF parametric equations
is consistent with an anticipated variation based on NewmanRaju.

a
0.35
T

a
0.35
T

NOTE: a is in radians.

7. Assessment of parametric equations


The weld toe SIF parametric equations described in
the previous sections were compared with a variety of

Fig. 7. Comparison of the weld toe SIF parametric equation


(bending) with NewmanRaju flat plate solution [4].

1058

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

Fig. 8. Comparison of the weld toe SIF parametric equation (tension)


with PD6493 solution [3].

Fig. 10. Comparison of the weld toe SIF parametric equation


(tension) with PD6493 solution [3] and Bell data [7].

Figs. 8 and 9 compare the parametric equations with


the solution given in PD6493 [3]. This is applicable for
weld angles of 45 only and does not take r/T into
account. The case of a=45, r/T=0.01 and a/T=0.01 was
arbitrarily chosen for comparison purposes. Fig. 8 shows
PD6493 is quite conservative in its estimation of Y relative to the SIF parametric equation for both extremes of
a/c=0.2 and a/c=0.8. It should be noted that this is an
extreme case and that the general trend is comparable.
Fig. 9 shows the same case in bending. PD6493 guidance underestimates Y relative to the SIF parametric equation by up to 25% irrespective of crack aspect ratio.
Figs. 10 and 11 show the same data plotted against a/T,
L/T=2.3. Bells three-dimensional finite element data [7]
are also shown. It can be seen that for bending PD6493
Fig. 11. Comparison of the weld toe SIF parametric equation
(bending) with PD6493 solution [3] and Bell data [7].

Fig. 9. Comparison of the weld toe SIF parametric equation


(bending) with PD6493 solution [3].

values are low for a/T values below 0.1, whereas for the
parametric equations the predictions are low above a/T
values equal to 0.1.
In tension, PD6493 [3] is conservative; Bells data
correlates well with the SIF parametric equation. In
bending (Fig. 11), PD6493 again shows unconservatism
relative to both Bells data and the SIF parametric equation in the critical near surface region for a/T values
0.05. Bells data is for r/T=0 and is compared with
the parameter r/T=0.01 limit when calculating the SIF
parametric equation curves.
Figs. 12 and 13 show the parametric equation and
another set of Bells data [7] in tension and bending
respectively. Again Bells data shows Y marginally
higher than the SIF parametric equations, reflecting the
different values of r/T. The only significant mismatch

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

Fig. 12. Comparison of the weld toe SIF parametric equation


(tension) with Bell data [7].

Fig. 13. Comparison of the weld toe SIF parametric equation


(bending) with Bell data [7].

between the SIF parametric equations and Bells data [7]


is shown in Figs. 14 and 15. Although the equations are
valid only up to a=60, values were generated for the
test case a=70 and L/T=2.3 for comparison purposes
with Bells data [7]. These appear slightly higher than
a=60 curves but are inconsistent with Bells data [7].
Examination of the full set of data shows that, the Y
values presented for a=70 were lower than those for
a=45 and a=30. This is surprising and needs confirmation from the original authors.
The figures presented so far validate the surface
crack aspect of the SIF parametric equations (parameter
a/c) and the weld angle a. The following comparisons
aim to demonstrate the accuracy of the SIF equations
considering parameters r/T and L/T. For this purpose,

1059

Fig. 14. Comparison of the weld toe SIF parametric equation


(tension) with Bell data [7].

Fig. 15. Comparison of the weld toe SIF parametric equation


(bending) with Bell data [7].

the most extensive and reliable data available was that


of Dijkstra [8]. These are presented in terms of the twodimensional parameter MK as defined in PD6493 [3].
Figs. 16 and 17 compare the case a=70, L/T=1.319.
Dijkstras data [8] is for r/T=0.0071 which is marginally
beyond the validity limits of the SIF parametric equations and is therefore shown against curves generated for
r/T=0.01. Again agreement is excellent for both tension
and bending cases.
Similar comparisons are shown in Figs. 1823 for a
variety of cases, varying r/T, L/T and a for both tension
and bending. In all, a wide range of applicable test cases
have been used in this validation. The SIF parametric
equations would appear to be consistent over their validity range when compared with published data from
other sources.

1060

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

Fig. 16. Comparison of the weld toe SIF parametric equation


(tension) with Dijkstra data [8].

Fig. 17. Comparison of the weld toe SIF parametric equation


(bending) with Dijkstra data [8].

8. Limits of validity
The derived equations have been produced using the
NiuGlinka weight function equations. This solution
requires input of the through-thickness stress distribution
on the uncracked body and these values have been calculated using 2-D finite element methods. The NiuGlinka
routine from FACTS [9] which was used has been extensively verified and validated. The geometries investigated were T-butt welded joints with the attachment
unloaded, as shown in Fig. 1. The data has been pro-

Fig. 18. Comparison of the weld toe SIF parametric equation


(tension) with Dijkstra data [8].

Fig. 19. Comparison of the weld toe SIF parametric equation


(bending) with Dijkstra data [8].

duced using the following geometries and ranges of the


input parameters:
a/T 100 values in the range 0.01a/T1.0
a/c 8 values in the range 0a/c1.0 (Semi-elliptical
surface crack of depth a and total length 2c).
a 30, 45, 60
r/T 0.01, 0.02, 0.04, 0.066
L/T 0.34.0
Consequently the parametric equations should only be
used in the ranges:

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

Fig. 20. Comparison of the weld toe SIF parametric equation


(tension) with Dijkstra data [8].

1061

Fig. 22. Comparison of the weld toe SIF parametric equation


(tension) with Dijkstra data [8].

Fig. 23. Comparison of the weld toe SIF parametric equation


(bending) with Dijkstra data [8].
Fig. 21. Comparison of the weld toe SIF parametric equation
(bending) with Dijkstra data [8].

0.01a/T1.0
0a/c1.0
30a60
0.01r/T0.066
0.3L/T4.0
Full details concerning the development of these equations are contained in Ref. [10].

9. Conclusions
SIF parametric equations have been developed for surface cracks emanating from the weld toe of T-butt
welded joints. These are more comprehensive than any
other similar study to date as parameters included are
weld angle, weld root radius, weld attachment length,
parent plate thickness and crack aspect ratio. Validity
ranges cover realistic engineering values.
The equations, predictions of which were compared
to published data, are consistent, stable and accurate over
the validity ranges.

1062

F.P. Brennan et al. / International Journal of Fatigue 21 (1999) 10511062

Acknowledgements
The authors are grateful to the Marine Technology
Support Unit of the Health and Safety Executive, UK
for sponsoring this work.
References
[1] Niu X, Glinka G. Theoretical and experimental analyses of surface fatigue cracks in weldments. In: Proceedings The Symposium on Surface Crack Growth: Model, Experiments and
Structures, ASTM, Reno (Nevada, USA), 25 April 1988.
[2] Hall MS, Topp DA, Dover WD. Parametric equations for stress
intensity factors. Technical Software Consultants Report No.
TSC/MSH/0244 for Dept. of Energy, UK, March 1990.
[3] BS PD6493 Guidance on methods for assessing the acceptability
of flaws in fusion welded structures, appendix E: stress intensity
factor solutions for cracks in welded joints. British Standards
Institution, 1991.

[4] Newman JC Jr., Raju IS. Eng Fract Mech 1981;15:185.


[5] ideas (Integrated Design Engineering Analysis Software), finite
element modelling users manual. Structural Dynamics Research
Corporation, USA, 1994.
[6] minitab Statistical Software, Minitab Inc, 3081 Enterprise Drive,
State College, PA 16801 (USA).
[7] Bell R. Determination of stress intensity factors for weld toe
defects. Final Report BSS 22ST.23440-2-1083/7, Faculty of
Engineering, Carlton University, Ottawa (Canada), October 1985.
[8] Van Straalen IJJ, Dijkstra OD, Saijder HH. Stress intensity factors and fatigue crack growth of semi-elliptical surface cracks at
weld toes. Paper 15 Weld Failures, November 1988.
[9] facts (FAtigue Crack growTh Software), Technical Software
Consultants Limited, 6 Mill Square, Featherstone Road, Wolverton Mill, Milton Keynes MK12 5RB (UK).
[10] Brennan FP, Dover WD, Kare RF, Hellier AK. Development of
parametric equations for stress intensity factors. Technical
Software Consultants Ltd/University College London Report to
Marine Technology Support Unit, Health and Safety Executive,
UK, Project No. 8707-HSEP3317, 1995.

Вам также может понравиться