VerityTM Weld Fatigue Method in Fe-SafeTM Using ANSYS

Dr. Pingsha Dong

Center for Welded Structures Research
BATTELLE
Columbus, OH
dongp@battelle.org

Abstract

It is well known that stress concentration in welded joints (and notched structures) dominates
fatigue behavior of welded structures. However, traditional finite element methods are not
capable of consistently capturing the stress concentration effects on fatigue behavior due to their
mesh-sensitivity in stress determination at welds resulted from notch stress singularity. Any use
of an artificial radius is too arbitrary for the results to be reliable in fatigue design in practice.

In this presentation, a robust stress analysis procedure recently developed at Battelle and
extensively validated by various industries will be presented. The method is called VerityTM
mesh-insensitive structural stress method which serves as a FE post-processing procedure to
commercial FE packages such as ANSYS. The VerityTM method has been integrated into fe-
safeTM and available from Safe Technology Ltd. The method is based on the mapping of the
balanced nodal forces/moments along an arbitrary weld line available from a typical finite
element run into the work-equivalent tractions (or line forces/moments). In doing so, a complex
stress state due to notch effects can then be represented in the form of a simple stress state in
structural mechanics in terms of through-thickness membrane and bending components at each
nodal location. The resulting structural stress calculations are mesh-insensitive, regardless of
element size, element type, integration order used, as long as the overall geometry of a
component is reasonably represented in a finite element model.

A series of simple and complex examples will be represented to demonstrate the mesh-
insensitivity of the structural stress method, covering MIG seam welds, laser welds, resistance
spot welds, etc. In addition to its mesh-insensitivity, the effectiveness of the structural stress
parameter has been further validated by collapsing several thousands of fatigue tests available
from literature into a single curve, referred to as the master S-N curve. Additional applications
of the structural stress method will may also be touched upon. These include:

• Treatment of low cycle fatigue

• Treatment of multi-axial fatigue
• Solder fatigue in electronic packaging
Introduction
Fatigue design and evaluation of welded joints are typically carried out by weld classification approach in
which a family (theoretically infinite) of parallel nominal stress based S-N curves are used according to
joint types and loading modes [1]. Extrapolation-based hot spot stress methods offer the potential to
reduce the number of the S-N curves as required in weld classification approach, which has gained an
increasing popularity in offshore and marine applications [2-4]. Although extrapolation-based hot spot
stress procedures have been used for tubular structures for many years, their applications in plate joints
such as ship structures were only investigated during the recent past, as recently summarized Fricke [4]. As
shown in Fig. 1 for a plate to I-beam joint, the hot spot stress based SCF using three extrapolation
techniques demonstrate the variability showed a wide scatter band [4].

SCF
Base Plate
3.2 3.2 Experiment
3.0 3.0 Shell4
Shell4
2.8 2.8
Shell8
2.6 2.6 Shell8
Extrapolated s tres s es

s tres s es at ROP's

2.4 2.4 Shell4(css)

Shell4
2.2 2.2
Shell8w
2.0 2.0 Attachment 1Solid20w
1.8 1.8 Solidpw
2Solid20w
1.6 1.6
4Solid8w
1.4 1.4 4Solid8w
1.2 1.2 2Solid20w(f)

1.0 1.0

0.8 0.8
0 5 10 15 20 25 30
.5t/1.5t
.4t/1.0t
0.5t

Distance from Weld Toe

Extrapolation
Procedures
Fig. 1: Comparison of FEA surface stress distributions using various modeling procedures
and extrapolation-based hot stress SCF results at the weld toe on attachment plate [4]

One of the unique issues in using any extrapolation-based hot spot stress procedures in plate structures is
that the surface stress gradients on which any extrapolation techniques are based upon are that the stress
gradients are more localized in plate structures than in tubular structures, as illustrated by Dong and Hong
[5], as shown in Fig. 2. In this figure, the surface stresses normal to weld (indicated by arrows) are
normalized by the respective nominal bending stresses. The surface stress gradients shown become
increasingly localized as the joint type changes from tube-to-tube, tube-to-plate, and plate-to-plate joints.
As a result, extrapolations using 0.5t/1t, 0.5t/1.5t, or nodal value at 0.5t [1-4] yield a unity in SCF, i.e., the
nominal stress. If the finite element (FE) mesh is not refined enough, or not converged yet as referred by
Healy [6], the extrapolated hot spot stresses tend vary, depending on the element sizes, types, joint types,
and loading mode, as illustrated in Fig. 1 [4].
 F F

(a) Tube-to tube (b) Tube-to-

T joint [9] plate T joint

t = 20mm

6
F Tube to Tube
Tube-to-tube

N o r m a liz e d S u r fa c e S tr e s s
5 Plate-to-tube
Tube-to-plate
4 (d) Surface Stress Cruciform - Tension
Plate-to-plate
t Distributions
3
t
2

(c) Plate to plate T- Joint 1

0
0.0t 1.0t 2.0t 3.0t 4.0t
Distance from Weld Toe

Fig.2: Comparison of normalized surface stress distributions between tubular joint and plate joints:
(a) tubular joint under brace tension; (b) tube-to-plate joint; (c) plate-to-plate joint; (d) normalized
surface stress distribution (with respect to bending stress in chord for tubular joint and in base plate
for tube-to-plate joint) normal to weld toe.

In all the global based stress analysis procedures (such as nominal stress, extrapolation based hot spot
stresses, etc.) for fatigue evaluation purposes [1-6], the ultimate goal is to identify an appropriate stress
parameter which , being able to be consistently calculated in practice, can be used to effectively correlate S-
N data from various joint types and loading modes. This can be restated as both the necessary and
sufficient conditions for seeking a global stress-based fatigue correlation parameter as follows:

(a) A global stress parameter must be able to be calculated consistently with a minimum mesh-sensitivity
(mesh sizes, element shapes, element types, etc.) at a fatigue prone location such as at weld toe;
(b) Such a stress parameter must be demonstrated to be capable of correlating different fatigue behaviors
(such as S-N data) observed in various joint types, loading modes, etc.

Obviously, nominal stress definition, if applicable for some joint configurations, satisfies the necessary
conditions (a), since it can be calculated by simple formulae, i.e., without mesh-sensitivity. However, the
nominal stress definition, as it is well known, does not satisfy the sufficient conditions (b), since it cannot
be used to correlate S-N data from various joint types and loading modes. This is why a family of infinite
number of essentially parallel S-N curves has been used with respect to the nominal stress parameter as
shown in Fig. 3 [1].

The very fact that those S-N curves (Fig. 3) are essentially parallel to one another, as observed from a large
mount of fatigue data, suggests the existence of a master S-N curve. A scaling parameter that correctly
measures the stress concentration in various welded joint types and load modes should be able to collapse
all the parallel S-N curves in Fig. 3 into a single master S-N curve. It is the purpose of this paper to present
such an approach by formulating an effective global stress parameter which can be used as a basis to
establish such a master S-N curve. In this context, the nodal force (always implying moments in this paper)
based mesh-insensitive structural stress method (5-9) will be briefly highlighted for its consistency in stress
concentration characterization as required by the necessary conditions stated above. Then, the nodal force
based (referred as structural stress method throughout this paper) structural stresses are shown to posses a
unique property which can be used for a rapid estimation of the stress intensity factors (K) in an arbitrary
joint within fracture mechanics context. As a result, a two-stage crack growth model has been proposed
and validated by a large amount of experimental data. The two-stage growth laws unifies the
conventionally “short crack” anomalous crack growth with long cracks. By integrating the two-stage crack
growth model, a unique scaling parameter encompassing the structural stress based stress concentration
effects, loading mode effects, and thickness effects is then formulated and validated by a massive amount
of historical weld fatigue S-N data from 1947 to present.

Fig. 3: A family of infinite number of fatigue S-N (or “FAT”) curves

recommended by IIW for welded joints using nominal stress parameter [1]

THE STRUCTURAL STRESS METHOD

The essence of the new structural stress method was based on the following considerations for fatigue
evaluations of welded joints:

(a) It was postulated that stress concentration at a fatigue prone location, such as a weld toe as shown in
Fig. 4a, can be represented by an equilibrium-equivalent simple stress state (as shown in Fig. 4b) and
self-equilibrium stress state (as shown in Fig. 1c). The former describes a stress state corresponding to
an equivalent far field stress state in fracture mechanics context [4,6], or simply, a generalized nominal
stress state at the same location, while the latter can be estimated by introducing a characteristic depth
t1 as shown in Fig. 1 (dashed lines), as discussed in detail in [8];

(b) Within the context of displacement-based finite element methods, the balanced nodal forces and
moments within each element automatically satisfy the equilibrium conditions at every nodal position.
Therefore, the equilibrium-equivalent structural stress state in the form of membrane and bending can
be calculated by using the nodal forces/moments at a location of concern.

Weld Weld

(b)
(a)

t t τm
τ (y) σx (y)

σ mm σσbb
Fig. 4: Through-thickness structural
stresses definition: (a) local stresses from Weld
(c)
FE model; (b) structural stress or far-field
stress ; (c) self-equilibrating stress and t1
structural stress based estimation with
respect to t1 (dashed lines)
t
 Shell/Plate Element Procedures
However, in order to calculate the structural stresses in terms of membrane and bending components, line
forces and moments must be properly formulated by introducing work-equivalent arguments as discussed
in [8-9]. As an example of such formulation for a closed weld line (i.e., two ends of an arbitrarily curve
weld overlap each other, such as in a tubular joint), the nodal forces can be related to line forces along an
arbitrarily curved weld as:

 (l1 + ln −1 ) l1 ln −1 
0
 F1   3 6 6   f1 
F   l1 (l1 + l2 ) l2  
 2   0  f2  (1)
 F3     f 3 
6 3 6
 = l2 (l2 + l3 ) l3  
0
.   6 3 6  . 
.   0
0 ... ...
 . 
    
 Fn −1   ln −1 ln − 2 (ln − 2 + ln −1 )   f n −1 
 0 
6 6 3 

In the above equation, a closed weld line (The first node at the weld start is the same node at the weld end)
is assumed, such as a tubular joint, i.e., Fn = F1 and f n = f1 . The lowercase f1 , f 2 ,..., f n −1 are line forces
along y’. In the matrix on the left hand of Eq. (1), li (i =1, 2, …, n-1) represents the element edge length
projected onto the weld toe line from ith element The corresponding line moments can be calculated in an
identical manner by replacing balanced nodal forces F1 , F2 ,..., Fn −1 in local y ' direction with balanced
nodal moments M 1 , M 2 ,..., M n −1 with respect to x ' in Eq (1) above, as depicted in Fig. 5. Note that nodal
force Fi in Eq. (1) represent the summation of the nodal forces at node i from the adjoining weld toe
elements situated on the positive side of y ' axis, as shown in Fig. 5. Before Eq. (1) can be constructed,
coordinate transformation for the nodal forces and nodal moments from the global x-y-z to local x’-y’-z’
system must be performed, with x’ traveling along the weld line and y’ being perpendicular to the weld
line. All these calculations have been automated as a structural stress post-processor. The linear system of
equations described by Eq. (1) can be solved simultaneously to obtain line forces for all nodes along the
line connecting all weld toe nodes. Substituting the corresponding nodal moments into Eq. (1), one obtains
line moments in the same manner. Then, the structural stress shown in Fig. 4b at each node along the weld
(such as weld toe) can be calculated as:
f y' 6m x'
σ s = σ m +σ b = + (2)
t t2
For parabolic plate or shell elements, Eq. (1) can be formulated in an identical fashion with the
relationships provided in [8]. In-plane shear can be treated in an identical manner [8].

N1
Node at Weld
Weld
Toe of Interest N2
E1
N3
Ni E2

E3
Ei
x’
y’

x
z

Fig. 5: The structural stress calculation procedures for an

arbitrarily curved weld using shell/plate element models
CALCULATION EXAMPLES

A tubular T-joint according to a recent round robin study on fracture assessment [5,9,10] is shown in Fig.
6a, where a detailed strain gauge measurements were also collected for deriving hot spot stress based stress
concentration at the saddle positions as shown. To demonstrate the effectiveness of the present structural
stress procedures, four shell element models with drastically different element sizes near the tube-to-tube
weld are shown in Fig. 6b, varying approximately from 0.25tx0.25t, 0.5tx0.5t, 1tx1t, to 2tx2t. Note that the
weld was not modeled at the tube-to-tube intersection in simplifying mesh generation efforts in the present
mesh-sensitivity study.
Fig. 6c summarizes the structural stresses along the weld toe on the chord side obtained from the four shell
models shown in Fig. 6b. Since the structural stresses along the weld possess the quarter symmetry, Fig. 6c
shows only the results for a quarter of the weld length measured from the saddle point shown in Fig. 6b.
The maximum structural stress concentration occurs at the saddle position. Within the angular span of 90o
along the 3D curved weld from saddle to crown positions, the 2tx2t mesh represents the weld line with only
three nodal positions (or about two and half linear elements) as shown by the triangle symbols in Fig. 6c.
Therefore, the difference in the structural stress calculations from the 2tx2t mesh is mainly due to the
geometric changes at the weld line (tube to tube intersection) resulting from the large linear element sizes
used. However, the structural stress based SCF at the saddle position is still within about 5% of the fine
mesh case (.25tx.25t). Excluding the 2tx2t mesh, the SCF variations in the other three models are all
within the 2% of each other.

Brace
Hot Spot

0.25tx0.25t
(a) Tubular T-Joint
Chord

0.5tx5t 2tx2t
1tx1t

Saddle

(b) Four FE models with different elements sizes at weld location (t=20mm)

12
(c) Structural stress SCF results
10

8
2tx2t
SCF

6 1tx1t
0.5tx0.5t
4 0.25tx0.25t Crown
Saddle
2
0 30 60 90
Angle from Saddle Point (Deg.)
Fig. 6: Structural stress calculations for a tubular T joint investigated by Zerbst et al [10]: (a) T-
joint geometry and loading conditions; (b) Four FE models with different element sizes; (c)
Comparison of the current structural stress results along weld toe at chord
As another example, the plate to I-beam joint shown in Fig. 1 from Fricke [1] is analyzed here using the
new structural stress procedures discussed in the above. In the models shown in Fig. 7a, the box fillet weld
was modeled as simple nodal connections between attachment plate edge and I beam. The weld line in this
instance is considered as being open-ended. The virtual node method as discussed in [9] is automatically
activated in constructing Eq. (1). Four drastically different element sizes ranging from 0.5tx0.5t to 4tx4t are
used in the mesh designs, as shown in Fig. 7a. The structural stress distributions (normalized by the
nominal bending stress) calculated along the weld toe on the attachment plate side are shown in Fig. 7b. It
can be seen that the variation in the structural stress calculated at the weld end positions is within about 1%
for all four cases. The validity of the SCF was demonstrated using models with the fillet weld being
properly represented by a row of inclined shell elements [9]. Note that in all four models shown in Fig. 7a,
the weld is not modeled for simplicity. As the element sizes change at the beam to attachment intersection,
the geometric representation remains the same even if a 4tx4t mesh is used. This is not the case for the
tubular T-joint shown in Fig. 6 discussed earlier.

0.5tx0.5t 1tx1t

Base I Beam (a) Four FE meshes used

W/o weld
representation
S S calculatio n
4tx4t
F = 12 2 .95 N

2tx2t
Attachment plate
2
0.5tx0.5t
1.5
Top Weld Toe 1tx1t
2tx2t
Normalized Structural Stres

1
4tx4t
0.5

0
0 30 60 90 120 150
-0.5 Distance from Top W eld Toe
on Attachement, mm
-1

Bottom Weld Toe

-1.5
(b) Comparison of structural stress distributions
-2

Fig. 7: Mesh-size insensitivity demonstration for a plate to I-beam box joint used in [1]
(also see Fig. 1): (a) FE models with drastically different element sizes; (b) comparison
of structural stress distributions along weld toe on attachment plate.

MASTER S-N CURVE FORMULATION

In seeking a stress-based scaling parameter to correlate the multiple S-N curves as shown in Fig. 3, it may
be assumed that fracture mechanics principles are applicable, implying that crack propagation dominates
fatigue lives in welded joints. The validity of such an assumption must be demonstrated by correlating a
large amount of S-N test data.
 Structural Stress Based K Estimation
Naturally, the simple candidate fracture mechanics parameter which can be considered for the current
purpose is the stress intensity factor K. However, generalized K solutions are not available for welded
joints. Fortunately, the structural stress definition (Fig. 3b) is consistent with the far-field stress definition
( σ ) in fracture mechanics. Therefore, the structural stress calculation process can be viewed as a stress
∞

transformation process from an actual complex joint in a structure under arbitrary loading to a simple
fracture specimen, in which the complex loading and geometry effects are captured in the form of
membrane and bending, as shown in Fig. 8. As a result, K for any crack size along the weld can be
estimated by using the existing K solution for a simple plate fracture mechanics specimen subjected to both
membrane tension and bending, by considering either an edge crack or a surface elliptical crack.

A general 3D Joint Geometry

and Loading Mode

F
2c

a t
F

σb
σm

a
t

A Simple 2D Crack Problem

Fig. 8: Structural stress based transformation and K calculation

using a simple fracture mechanics specimen

The detailed derivations and validations can be found in [8]. For demonstration purposes, Fig. 9 shows the
validation for considering an edge crack in a T fillet weld. In Fig. 9, the case corresponds to “W/O notch
stress” was obtained by directly plugging the structural stress components (membrane and bending)
calculated using the present structural stress method into the existing K solution for an edge notch
specimen under remote tension and bending, respectively. The case “W/ notch stress” refers to the use of
the self-equilibrating part of the stress state (Fig. 3c) which is analytically estimated, as discussed in [8].
t1/2 t1 /t=1 Remote Loading:
h/t=1 Pure Tension

Sym. h

K K
σ0 πa σL0 πa

6 6

Far-Field Stress(Eq. 9)
(b)Current solution W/O notch stress
Far-Field Stress (Eq. 9)
5 (a) Current solution W/O notch stress 5
Weight
Glinka’s weight Function[14,15]
function
Glinka’s weight Weight
function
Function[14,15]

4 Current solutionNotch
W/ notch stress
Stress (Eqs. 8&11) 4 Current solution
NotchW/ notch
Stress stress
(Eqs. 8&11)

3 3

(a) (b)
2 2

1 1

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
a/t a/t

Fig. 9: Validations of the structural stress K estimation for T-fillet joint using Glinka’s
weight function method [8]: (a) Remote bending; (b) remote tension
It can be seen that without considering the notch stress (or self-equilibrating part of the stress state), the
current solution provides an accurate K estimation for crack size a/t larger than about 0.1. With the use of
the notch stress effects, K can be calculated for any given infinitesimally small a/t. The current solution in
Fig. 9 is higher for small a/t than the weight function solution from Glinka (see [8] for detail). This is due
to the fact that Glinka introduced a small weld toe radius in performing the finite element stress calculation
to avoid the mesh-sensitivity. In the present calculations, the weld toe radius was assumed to be zero, i.e.,
simulating a sharp notch at the weld toe.

A Two-Stage Growth Model

The non-monotonic K behavior as a function of a/t shown in Fig. 9 is characteristic among all joint types
investigated [8]. As the crack size a/t becomes smaller than about a/t ~ 0.1, the elevated K is attributed to
the dominance of the notch stresses at weld toe. It can then be postulated that both the short crack and long
crack growth processes may be characterized by the two distinct stages of the K behavior as a crack
propagates from a/t < 0.1 to a/t > 0.1. Along this line, it can be argued that two stages of stress intensity
solutions in the form of the notch-stress dominated ∆ K a / t < 0 .1 and far-field stress dominated ∆K a / t > 0.1 can be
separated to characterize the full range crack growth behavior from 0 < a / t < 0.1 (“small crack”) to
0.1 ≤ a/t ≤ 1 (“long crack”). Here, the term ∆K refers to the stress intensity factor range corresponding to
remote stress range. It then follows:

da
= C[ f1 ( ∆K )a / t ≤ 0.1 × f 2 ( ∆K )a / t > 0.1 ] (3)
dN

By introducing a stress intensity magnification factor Mkn in dimensionless form and assuming a power-law
form of the two stage crack growths corresponding to f1 ( ∆K ) a / t ≤ 0.1 and f 2 ( ∆K ) a / t > 0.1 , respectively, Eq. (3)
can be re-written as:

da = C ( M ) n ( ∆K ) m (4)
dN kn n

The terms Mkn and Kn are defined below:

K ( with local notch effects) (5)

M kn =
Kn (based on through thickness σ t m and σ t b )

signifying the notch-induced magnification of the stress intensity factors as a/t approaches zero. The
constants n represents the crack growth exponent for the first stage of the crack growth and m the
conventional Paris law exponent, both of which are to be determined by experimental crack growth rate
data.
The validation of the two-stage growth model is shown in Fig. 10. The crack growth data were taken from
well-known short crack growth data by Tanaka and Nakai [11] and Shin and Smith [12]. Without relying
on any crack closure arguments, all the so called anomalous crack growth data in Fig. 10a are collapsed
into single straight data band in Fig. 10b, with a unified slope of m=3.6. Note that the first exponent in Eq.
(4) was empirically determined as n=2. More detailed discussions on the notch stress formulation and the
two stage growth law can be found in [13].
 SEN (a)
1.E-06 CN

(Shin, 1988)

DEN

da/dN, m/cycle
1.E-08

SEN-0.4mm-Steel
DEN-0.4mm-SS
(Tanaka, 1983)
DEN-0.71mm-SS
1.E-10 CN-s=60-Steel
CN-s=76-Steel
CN-s=160-Steel

1.E-12
1.E+01 1.E+02 1.E+0
delta K, MPa*m^1/2
1.E-06
(b)Structural
(b) Notch Notch Structural Stress
(b) KBased
Stress Based and
1.E-07 Two StageTwo StageModel
Growth Growth Model
(Eq. 19)

1.E-08
da/dN*(1/Mkn2), m/cycle

1.E-09
m
1.E-10
1
1.E-11
SEN - 0.4mm Steel
CN-s=60-Steel
1.E-12 CN-s=76-Steel
CN-s=160-Steel
1.E-13 DEN-0.71mm-SS
DEN-0.4mm-SS
1.E-14
1.E-01 1.E+00 Kn, MPa*(m)^.5 1.E+01 1.E+02
delta Kn MPa*m**1/2
Fig. 10: Consolidation of short crack growth data from various specimen types and
notch geometries [11,12] using the current two stage growth model with n = 2: (a)
da/dN versus nominal ∆K range referred as anomalous crack growth in [11,12]; (b)
current two-stage growth model

Equivalent Structural Stress Parameter

The two-stage crack (Eq. 4) with two stage growth exponents being n = 2 and m = 3.6 ) can be
integrated as

a=a f
da
N= ∫
a →0
C ( M kn ) ( ∆K )
n m
(6)

As discussed in [8,13], an extensive investigation of Mkn for various joint types showed that it can be
approximated by a single curve as a function of a/t for all joint types once the denominator in Eq. (5) is
formulated using the mesh-insensitive structural stress.
Note that the integral in Eq. (6) is not very sensitive to the final crack size af, and therefore, can be written
in a relative crack length form as:
 a / t =1 m
td (a / t ) 1 1− (7)
N= ∫
ai / t →0 C ( M kn ) ( ∆K )
n m
= ⋅ t 2 ⋅ (∆σ s ) − m I (r )
C

where I (r ) is a dimensionless function of bending r ( r = ∆σ b / ∆σ s ) after performing the following

integration for a given m:

a / t =1
d (a / t )
I (r ) = ∫  a 
m
ai / t →0 a  a
( M kn )  f m ( ) − r  f m ( ) − f b ( ) 
n

 t  t t 

Then, Eq. (7) can be expressed in terms of N once the dimensionless I(r) function is known:

1 2−m 1 1
− −
∆σ s = C m
⋅t 2m
⋅ I (r ) m ⋅ N m (8)

Eq. (8) uniquely describes a family of an infinite number of structural stress based S-N curves ( ∆σ s −N)
as a function of thickness effects (t), and bending ratio effects r. If Eq. (8) provides a good representation
of the fatigue behavior of welded joints, an equivalent structural stress parameter can be defined by
normalizing the structural stress range ∆σ s with the two variables expressed in terms of t and r on the right
hand side of Eq. (8):

∆σ s
∆S s = 2− m 1
(9)
t 2m
⋅ I (r ) m

( 2− m ) / 2 m
where the thickness term t becomes unity for t =1 (unit thickness) and therefore, the thickness t
can be interpreted a ratio of actual thickness t to a unit thickness, rendering the term dimensionless. With
this interpretation, the equivalent ∆S s retains a stress unit. It is worth noting that the equivalent structural
stress parameter described by Eq. (9) captures the stress concentration effects ( ∆σ s ) , thickness effects (t),
and loading mode effects (r) on fatigue behavior.

Initial Crack Size Effects

Before Eq. (9) can be used to construct a single master S-N curve for welded joints, assumptions in
performing the integration and the effects of Mkn on I (r ) must be quantified. It is well known that initial
crack size ( ai / t ) can make a significant difference in the final life prediction based on fracture mechanics
as described in Eqs. (6-8).
The effects of a series of assumed initial crack sizes are shown in Fig. 11 by using the edge crack based K
1/ m
solution. Note that I ( r ) is presented as I ( r ) after considering the exponent 1/m in Eq. (11) to
facilitate the comparison between Figs. 11a and 11b. Without considering the local notch effects, i.e., Mkn,
different initial crack sizes ai / t produce significantly different I (r )1 / m curves as a function of r. Once Mkn
is considered, the dependency of the I (r ) on initial crack size ai / t becomes insignificant, particularly for
the two cases with small initial crack size ( ai / t ), as shown in Fig.11b. The increase in I ( r )1/ m is about
8.5% as r increases from r = 0 (pure membrane) to r = 1 (pure bending) under load controlled conditions.
This implies that with a strong notch effects in typical welded joints characterized by Mkn, the usual initial
crack size effects on life predictions observed in typical fracture mechanics specimens without stress riser
are significantly diminished in welded joints. Based on Fig. 11b, ai / t =0.001 will be used in the rest of
this paper.
5
(a) w/o considering Mkn

4
ai/t=0.0001
ai/t=0.001
I(r)^(1/m)

ai/t-0.01
3

1
0 0.2 0.4 0.6 0.8 1
r
4
(b) with considering Mkn ai/t=0.0001
ai/t=0.001
3 ai/t=0.01
I(r)^ (1/m )

0
0 0.2 0.4 0.6 0.8 1
r
Fig. 11: Comparisons of I(r) functions with and without Mkn and effects of initial ai/t
(edge crack solution): (a) without considering Mkn; (b) with consideration Mkn

Joint Gb (t=20mm) Joint B(t=12.7 mm), Joint B(Kihl)(6.35 mm),

Joint G’ (t=12.7mm) 13/10/8AW (13mm), 50/50/16AW (50 mm),
50/50/16AW (DW )(50mm),100/50/16AW (100mm),
100/50/16AW (QT Steel)(100 mm)

t
t

Joint C(t=12.7mm)
Joint D(t=12.7mm)
Joint-Cb(Booth)(t=38mm),
Joint-Cb(Pook)(38mm) t
t

t Bell (t=16mm)
Joint F (t=12.7mm),
t = 5-80mm
Joint F(Rorup)(12.5mm)
t
Joint E (t=12.7mm)

t t Double Edge Gusset (90mm)

Fig. 12: Illustration of some representative joint types analyzed in this investigation for the
development of the master S-N curve
 S-N Data Correlation
The ultimate test for proving if the equivalent stress parameter (Eq. 9) is valid is to demonstrate if a large
amount of experimental S-N data can be collapsed into a single narrow band. In doing so, a massive
amount of S-N data (over 800 fatigue tests) from drastically different joint geometries, plate thicknesses,
and loading modes were collected from literature and published reports, as highlighted in Fig. 12.

1.E+03

Nominal Stress Range, MPa

1.E+02

(a)

AT122 AT140 AT180 AT222 AT240

AT280 Bell Joint G' Joint D Detail_3(Fricke)
Joint F joint F(rorup) Joint-Cb(Booth) Joint-Cb(Pook) AC110
AC122 AC140W AC140N AC180 AC210
AC222 AC240 AC280 AC310 AC340
AC380 AC422 AC440 Joint C Joint B
13/10/8 AW 50/50/16 AW 50/50/16 AW (DW) 100/50/16 AW 100/50/16 AW (QT)
Joint E Gurney -LW2 HHI_3 9mm-w25 9mm-w50
9mm-w100 9mm-w160 20mm-w25 20mm-w50 20mm-w100
20mm-w160 40mm-w25 40mm-w50 40mm-w100
1.E+01
1.E+04 1.E+05 1.E+06 1.E+07
Life
1.E+04
AT122 AT140 AT180 AT222
∆σ s AT240 AT280 Bell Joint G'
∆S s = 2− m 1 Joint D Detail_3(Fricke) Joint F joint F(rorup)
Joint-Cb(Booth) Joint-Cb(Pook) AC110 AC122
t 2m
⋅ I (r ) m
AC140W AC140N AC180 AC210
AC222 AC240 AC280 AC310
Equivalent Structural Stress Range, MPa

AC340 AC380 AC422 AC440

Joint C Joint B 13/10/8 AW 50/50/16 AW
50/50/16 AW (DW) 100/50/16 AW 100/50/16 AW (QT) Joint E
Gurney -LW2 HHI_3 9mm-w25 9mm-w50
9mm-w100 9mm-w160 20mm-w25 20mm-w50
20mm-w100 20mm-w160 40mm-w25 40mm-w50
40mm-w100

1.E+03 (b)

1.E+02
1.E+04 1.E+05 1.E+06 1.E+07
Life

Fig. 13: Correlation of existing S-N data for various joint types, loading modes,
and plate thicknesses: (a) nominal stress range versus life; (b) equivalent
structural stress range versus life.

For each set of the specimens and fatigue tests, the structural stress calculations were performed under
given loading conditions and failure criteria. The results are summarized in Fig. 13 in terms of both
nominal stress range and equivalent structural stress range as given in Eq. (9). A wide scatter can be seen
in Fig. 13a, which is expected. Once the equivalent structural stress range is used according to Eq. (9), the
all the S-N data are collapsed into a narrower band, regardless of the diverse joint types, plate thicknesses,
and load modes under which the fatigue tests were conducted over about 40 years. The same methodology
has been proven with the same effectiveness in correlating tubular and pipe/vessel joints [8].

By regression analysis, the mean line of Fig. 13b can be represented in the form of:

−1
∆σ s
1
−
= CN m'
= 16308 × N 3.333 (10)
2− m 1

t 2m
I (r ) m
with the stress unit in MPa and thickness in mm and m = 3.6. In Eq. (10), 1 / m' represents the negative
slope of the master S-N curve in Fig. 13b. Note that although in various publications, m = m ' (=3 for
steel welds) is often assumed. In this investigation, it is found that are close, but not necessarily the same.
For simple fatigue test specimens, nominal stresses are often well-defined, then,

∆σ s = SCFss × ∆S n

in which SCFss signifies the structural stress based SCF and ∆S n the typical nominal stress range
definition.

1000
(a) Predicted nominal stress range versus N for Data by Maddox[13]
N om inal Stress Range, MPa

13mm

100mm
100

13/10/8 AW 50/50/16 AW
50mm
50/50/16 AW (DW) 100/50/16 AW
100/50/16 AW (QT) 13mm - Prediction
50mm -Prediction 100mm -Prediction
10
1.E+04 1.E+05 1.E+06 1.E+07
Life

1000
(a) Predicted nominal stress range versus N for Data from SR202[14]

22mm
Nominal Stress Range, MPa

40mm
80mm
100

AT122 AT140
AT180 AT222
AT240 AT280
#REF! AT140-Prediction
AT180-Prediction AT222-Prediction
AT240-Prediction AT280-Prediction
10
1.E+04 1.E+05 1.E+06 1.E+07
Life

Fig. 14: The use of the master S-N curve (mean line) for the prediction of nominal stress
range versus N curves generated by Maddox [13] and SR202 [14].

For a given simple joint specimen of interest, the structural stress based SCF can be calculated using the
present structural stress procedure. Then, the nominal stress based S-N behavior can be predicted by using
the master S-N curve from Eq. (10) as:

 22−mm 1
 1
 t I (r ) m  − m' (11)
∆S n = C  N
 SCFss 
 
Both C and m’ are given in Eq. (10) based on the master S-N database generated in this investigation. Two
examples are given in Fig. 14. The mean nominal stress S-N curves for various plate thicknesses are
predicted using Eq. (11) for the cruciform joint tests (under remote tension loading) by Maddox [13] and T-
fillet joint tests (under 3-point bending) reported in [14]. A good correlation between the test data and
those predicted by the master S-N (Eq. 10) curve is evident.

Conclusion
The mesh-insensitive structural stress parameter not only can be calculated consistently with the
demonstrated mesh insensitivity, but also has been shown to be an effective fatigue parameter to correlate
the fatigue behavior of welded joints regardless of joint geometries, loading modes and plate thicknesses.
Furthermore, the structural stress parameter can be directly related to the far-field stress in a fracture
mechanics context. As a result, a rapid K estimation scheme has been demonstrated to be effective for
analyzing arbitrary joints in plate structures, as well as other joint types such as tubular joints, pipe and
vessel welds. With the aid of fracture mechanics principles, a master S-N curve approach has been
developed by introducing an equivalent structural stress parameter, which captures three well-known
factors that contribute to fatigue behavior in welded joints: (a) the stress concentration due to joint
geometry; (b) the loading mode; and (c) the plate thickness. A large amount of S-N data has been collected
and correlated by the equivalent structural stress parameter. The present master S-N curve approach can
simplify fatigue evaluation procedures for ship structures and significantly reduce testing requirements,
since the S-N data transferability in the form of the equivalent structural stress parameter has been
established.

References
1. Hobbacher, A., “Fatigue Design of Welded Joints and Components: Recommendations of IIW Joint
Working Group XIII-XV, Abington Publishing, Abington, Cambridge, 1996.
2. “Fatigue strength Analysis of Offshore Steel Structures,” DNV RP-C203, May 2000.
3. “Guide for the Fatigue Assessment of Offshore Structures,” ABS, April, 2003.
4. Fricke W., “Recommended Hot-Spot Analysis Procedure for Structural Details of FPSO’s and Ships
Based on Round-Robin FE Analysis, ISOPE Proceedings, Stavanger, Noway, June 2001.
5. Dong, P. and Hong, J.K., “Analysis of Hot Spot Stress and Alternative Structural Stress Methods,”
Proceedings of 22nd International Conference on Offshore Mechanics and Arctic Engineering, June 8-
13, 2003, Cancun, Mexico.
6. Healy, B.E., “A Case Study Comparison of Surface Extrapolation and Battelle Structural Stress
Methodologies,” to appear in Proceedings of the 23rd International conference on Offshore Mechanics
and Arctic Engineering, June 20-25, 2004, Vancouver, British Columbia, Canada.
7. Dong, P., “A Structural Stress Definition and Numerical Implementation for Fatigue Analysis of
Welded Joints,” International Journal of Fatigue, 23, pp. 865-876, 2001.
8. Dong, P., Hong, J.K., Osage, D., and Prager, M., “Master S-N curve approach for welded
components,” Welding Research Council Bulletin, No. 474, December, 2002, New York, New York,
10016.
9. Dong, P., “A Robust Structural Stress Method for Fatigue Analysis of Ship Structures,” Proceedings of
the 22nd International Conference on Offshore Mechanics and Arctic Engineering, June 8-13, 2003,
Cancun, Mexico.
10. Zerbst, U., Heerens, J., and Schwalbe, K.-H., “The fracture behavior of a welded tubular joint – an
ESIS TC1.3 round-robin on failure assessment methods Part I: experimental data base and brief
summary of the results,” Engineering Fracture Mechanics, 69, 2002, pp. 1093-1100.
11. Tanaka, K., and Nakai, Y., "Propagation and Non-Propagation of Short Fatigue Cracks at a Sharp
Notch," Fatigue of Engineering Materials and Structures, Vol. 6, No.4, pp.315-327, 1983
12. Shin, C.S., and Smith, R.A., "Fatigue Crack Growth at Stress Concentrations- the Role of Notch
Plasticity and Crack Closure," Engineering Facture Mechanics, Vol. 29, No.3, pp.301-315, 1988.
13. Dong, P., Hong, J.K., and Cao, Z., “Stresses and Stress Intensities at Notches: ‘Anomalous Crack
Growth’ Revisited”, Int. J. of Fatigue, Vol. 25(9-11), pp. 811-825, 2003.
14. Maddox, S.J., The Effect of Plate Thickness on the Fatigue Strength of Fillet Welded Joints, The
Welding Institute, Abington Hall, Abington, Cambridge CB1 6AL, 1987.
15. SR202 of Shipbuilding Research Association of Japan, Fatigue Design and Quality Control for
Offshore Structures, 1991 (in Japanese).