Simplification and application of model F 129

EPFL Thesis 1617

6. SIMPLIFICATION AND APPLICATION OF MODEL F
The aim of Chapter 6 is to simplify and to demonstrate the possible applications of model F.
This chapter contains the following parts : firstly, the crack propagation rates calculated using
model F are related to those obtained using Paris equations (2.25) and (2.27), which leads to
new crack propagation rate equation. Next the model F is used to develop a simplified
fatigue crack propagation model - model SF. Finally, both model F and model SF are
applied to two example problems.
6.1 MODEL F VERSUS PARIS EQUATION
Both model F and the Paris equation can simulate fatigue crack propagation. Therefore, it
should be possible to relate crack propagation rates calculated using model F and the Paris
equations. The reason for this is to simplify the procedures used to determine constants C and
m of the Paris equations. A conventional experimental determination procedure is not always
possible. Some attempts have already been made to determine the constants C and m using
analytical rather than experimental methods ([6.1], [6.2]).
In addition to the constants C and m, the modified Paris equation (2.27) takes into account the
fatigue threshold. Further, two aspects related to the Paris equation are evaluated on the basis
of model F : constants C and m, and fatigue threshold. Finally, a crack propagation rate
equation similar to the Paris equation, is presented.
6.1.1 Paris Equation Constants
In this section, analytical formulas for the constants used in the Paris equation are developed.
This is done in following manner : a relationship between the effective stress intensity factor
range K
eff
and the crack propagation rate da/dN is established using the modeling principles
of model F (Chapter 3). Next, the constants C and m are extracted from established
relationships.
An average crack propagation rate da/dN over the length is a function of the fatigue life of
the first element at the crack tip, N
f,elem
:
da
dN N
f elem
=

,
(6.1)
N
f,elem
in Equation (6.1) can be determined by iterating the strain life relationship (3.11).
However, an analytical formula for the N
f,elem
is needed here. Assuming that the cyclic strains
at the fatigue crack tip are dominantly plastic, the fatigue life of element N
f,elem
can be
determined using the plastic part (2.19) of the strain-life relationship. Equation (2.19) leads to
the following formula for the N
f,elem
:
N
f elem
f
pl
c
,
'
'
=

!
"
#
#
$
%
&
&
1
2
1
2
1

(6.2)
Using the plastic part of the Ramberg-Osgood equation (3.20), it is possible to write the
plastic strain range
pl
in Equation (6.2) as a function of plastic stress range :

pl
pl
n
K
=

!
"
#
$
%
& 2
2
1
'
'
(6.3)
130 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
Including only the plastic component of Glinkas ESED range criterion (3.30) permits the
plastic stress range
pl
in Equation (6.3) to be expressed as a function of linear-elastic stress
range
le
:
( )
pl
n
n
le
n
n
K
n
E K
=
+

'
(
)
*
+
,

+

+
2
1
8
1
2
1
'
'
'
'
'
'
'
(6.4)
The linear-elastic stress range
le
in Equation (6.4) can be calculated using Equation (3.18).
Taking j=1 and substituting the
le
and K
eff
for the
le,j
and K
eff
in Equation (3.18) leads to :

le eff
K =

2

(6.5)
Making the substitutions as indicated in Figure 6.1 leads to the relationship between K
eff
and
da/dN, similar to the Paris equation :
da
dN
C K
pl eff
m
pl
= (6.6)
The constants C
pl
and m
pl
can be pre-calculated (Equations 6.7 and 6.8) and are functions of
material parameters used in model F. Equation (6.7) contains the element size , but its
influence on C
pl
is not significant (see Chapter 5 about the influence of ).
C
K E
n
pl f
n
c
=

+
!
"
#
$
%
&
!
"
#
#
$
%
&
&
+
2
4
1
1
1
1


'
'
'
'
'
(6.7)
m
c n
pl
=
+
2
1 ' ( ' )
(6.8)
K
eff
(6.5) le
(6.4) (6.3) pl
pl
(6.2) (6.1) N
f,elem da/dN

Figure 6.1 : Substitutions made in order to obtain Equation (6.6).
6.1.2 Fatigue Threshold
In this section, formulas which take into account the fatigue threshold in relationship to
da/dN-K
eff
, are developed using the modeling principles of model F. It is assumed, that
near the fatigue threshold where the K
eff
is close to the K
th
, the strain ranges at crack tip are
elastic and not plastic. This implies that the fatigue life of the element at the crack tip, N
f,elem

in Equation (6.9), can be calculated using the elastic part (2.18) of the strain-life relationship.
Equation (2.18) leads to the following formula for the N
f,elem
:
N
E
f elem
f m
el
b
,
'
( ' )
=


!
"
#
#
$
%
&
&
1
2 2
1

(6.9)
Simplification and application of model F 131
EPFL Thesis 1617
The elastic strain range
el
in Equation (6.9) can be calculated using the linear-elastic stress
range
le
and the elastic modulus E (Equation 6.10). The
le
in Equation (6.9) can be
calculated using Equation (6.5).

el
le
E
= (6.10)
Making the substitutions as indicated in Figure 6.2 leads to the following relationship between
K
eff
and da/dN :
da
dN
C K
el eff
m
el
= (6.11)
The constants C
el
and m
el
in Equation (6.11) can be calculated using Equations (6.12) and
(6.13) and are functions of material parameters used in model F. Equation (6.11) can be used
to calculate crack propagation rates at small K
eff
. A general crack propagation law, based on
Equations (6.6) and (6.11) is presented in following Clause.
( )
C
el f m
b
= 2 2
1
( ' )
'
(6.12)
m
b
el
=
1
'
(6.13)
K
eff
(6.5) le
(6.10) (6.9) el
(6.1) N
f,elem da/dN

Figure 6.2 : Substitutions made in order to obtain Equation (6.11).
Equation (6.12) contains the term
m
, which can be used to take into account the effect of the
local mean stress on the fatigue threshold. The
m
can be expressed as a function of the
effective stress intensity factor range K
eff
, and the maximum effective stress intensity factor
K
max,eff
. :

m m eff
m eff
m
C K
K
m
m
=
!
"
#
$
%
&
'
(
)
)
*
+
,
,
,max

2
(6.14)
Constants C
m
and m
m
in Equation (6.14) can be calculated using Equations (6.15) and (6.16)
correspondingly. Derivation of Equations (6.14), (6.15) and (6.16) is explained in
Annex A.4.2.
C K
n
E K
m
n
n
=
+

!
"
#
$
%
&
+
'
'
'
'
'
1
1

(6.15)
m
n
n
m
=

+
2
1
'
'
(6.16)
6.1.3 Crack Propagation Rate Equation
The objective of this section is to propose a crack propagation rate equation. Further, the
background of this equation is explained and the equation itself analyzed. The effective stress
intensity factor range at the intersection point of the two curves given by Equations (6.6) and
(6.11), K
eff
*, can be calculated using Equation (6.17) :
132 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
K
C
C
eff
el
pl
m m
pl el

=
!
"
#
#
$
%
&
&
1
(6.17)
The K
eff
* is a partition point of two regions of K
eff
: fatigue threshold region with
K
eff
<K
eff
*, and stable growth region with K
eff
>K
eff
*. Crack propagation rates da/dN
within the fatigue threshold region can be calculated using Equation (6.11). Values for da/dN
can be calculated within the stable growth region using Equation (6.6). Together, these
equations lead to the general crack propagation rate equation (6.18), which can be used to
calculate da/dN for any value of K
eff
. The C
el
, m
el
, C
pl
and m
pl
in Equation (6.18) can be
calculated using Equations (6.12), (6.13), (6.7) and (6.8), correspondingly.
da
dN
C K C K
el eff
m
pl eff
m
el pl
=

1
1 1

(6.18)
Three da/dN-K
eff
curves given by Equations (6.6), (6.11) and (6.18), are shown in Figure 6.3.
It can be seen that Equation (6.18) ensures a smooth transition from the fatigue threshold
region given by Equation (6.6), to the stable growth region given by Equation (6.11). The
proposed law has a bi-slope form if plotted on log-log scale. Measured da/dN curves, if
plotted versus K, also have two slopes : Herzberg [6.3], for example, shows the bi-slope
da/dN-K curves of 7075-T6. The same can be said about the da/dN-K curves in [6.4], [6.5],
[6.6].
da/dN [mm/cycle]
K
eff
*
(6.11)
model F
K
eff
[N/mm
1.5
]
m
pl
1
m
el
1
100 1000
10
-8
10
-7
10
-6
10
-5
10
-4
(6.6) (6.18)
ft sg
ft
sg
fatigue threshold
stable growth

Figure 6.3 : Crack propagation rate curves for St50 [6.7].
The crack propagation rates, calculated using model F, are also given in Figure 6.3. It can be
seen that Equation (6.18) leads to more conservative da/dN-K
eff
curves than model F. The
conservatism of Equation (6.18) arises due to the following reasons: Equation (6.11) was
developed using only the elastic term, and Equation (6.6) using only the plastic term of the
strain-life relationship and of the Ramberg-Osgood equation. In calculations of the
da/dN-K
eff
curve by model F, both the elastic and plastic terms of the strain-life
relationship and Ramberg-Osgood equation, are used. It is impossible to get an analytical
formula for the da/dN-K
eff
curve using both elastic and plastic terms of the strain-life
relationship and the Ramberg-Osgood equation. In model F, the calculation of the
da/dN-K
eff
curve is made by iteration.
Equation (6.18) leads to a non-fixed fatigue threshold concept : if K
eff
decreases then the
corresponding crack propagation rates calculated using Equation (6.18) decrease exponentially
Simplification and application of model F 133
EPFL Thesis 1617
and crack propagation practically stops. On the contrary to the modified Paris law (2.27),
Equation (6.18) does not require a special parameter to take account for the effect of the
fatigue threshold.
The term C
el
in Equation (6.18) is function of the local mean stress
m
. The
m
can be
calculated using Equation (6.14) and is a function of the maximum effective stress intensity
factor K
eff,max
. An increase of K
eff,max
results in an increase of C
el
and a corresponding decrease
of the fatigue threshold, which agrees with the test observations [6.3], [6.8].
Parameters C
el
, m
el
, C
pl
and m
pl
in Equation (6.18) can be calculated as functions of the
constants of the strain-life relationship and Ramberg-Osgood equation. These constants are
published for a great number of ductile materials [6.7]. Thus it is possible to rapidly compare
the fatigue properties of different materials. An example of this is given in Figure 6.4. The
comparison shows that StE790 steel has better fatigue properties than the St50 steel (i.e., both
the fatigue threshold and the stable growth region of the StE790 are below the corresponding
regions of the St50 steel.) Figure 6.4 also demonstrates that the proposed crack propagation
equation can account for the influence of the nominal stress ratio R. It is well known that
increasing R leads to decrease in fatigue threshold, which is well represented by curves in
Figure 6.4. Published data also show that an increase in yield stress leads to increase in the
fatigue threshold [6.9]. The comparison of StE790 and St50 steels shows the same tendency :
the steel with a greater yield stress (StE790) has a greater fatigue threshold than the steel with
lower yield stress (St50).
K
eff
[N/mm
1.5
]
da/dN [mm/cycle]
100 1000
10
-8
10
-7
10
-6
10
-5
10
-4
StE 790
St 50
R=0.5, 0, -0.5, -1
R=0.5, 0, -0.5, -1

Figure 6.4 : A comparison in between the crack propagation rate curves of two materials.
Therefore, the proposed crack rate propagation equation (6.18) is developed on the basis of
the modeling approach, presented in Chapter 3. There are some differences between
Equation (6.18) and the conventional Paris law. Specifically, Equation (6.18) leads to the non-
fixed fatigue threshold concept, it allows to account for the effect of maximum stress intensity
factor K
eff,max
, and the constants of this equation can be determined analytically.
Equation (6.18) has a wide range of application possibilities. Some of these possibilities are
reviewed later.
The summary of Section 6.1 is presented in Figure 6.5. On the basis of the plastic portion
(2.19) of strain-life relationship, Equation (6.6) was developed in Clause 6.1.1. Equation (6.6)
is similar to the Paris equation (2.25), and Equations (6.12) and (6.13) can be used to calculate
the constants of the classical Paris equation. On the basis of elastic part (2.19) of strain-life
relationship, Equation (6.11) was developed in Clause 6.1.2 in order to take into account the
fatigue threshold. Finally, based on Equations (6.6) and (6.11), the crack propagation rate
equation (6.18) was proposed in Clause 6.1.3.
134 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617

N
f
da/dN
K
eff
(6.11)
(6.6)
(6.18)
(2.19)
(2.18)
Section 6.1.1
Section 6.1.2
Section 6.1.3

Figure 6.5 : Summary of Section 6.1.
6.2 SIMPLIFIED MODEL F - MODEL SF
The aim of this section is to introduce a simplified version of model F. This simplified
version is called model SF, where SF stands for Simplified Fatigue. The main
requirement of the simplified fatigue crack propagation model is that it should be easily
programmable for a personal computer yet remain general and precise enough for use in
fatigue analysis. model SF is based on the crack propagation rate Equation (6.18).
6.2.1 Calculation of Fatigue Life
Crack Initiation
The objective of this section is to demonstrate how Equation (6.18) can be used to evaluate
the duration of the crack initiation stage, N
CI
. Further, something uncommon will be done - a
crack initiation period N
CI
will be calculated on the basis of stress intensity factor. The only
reason for this is to keep the calculation procedure simple. (i.e., to use similar formulas for
both : crack initiation stage and the stable crack growth stage.)
In order to use Equation (6.18), the effective stress intensity factor range K
eff
must be known.
Since there is no fatigue crack present in the detail during the crack initiation stage, K
eff
=0 if
it is calculated conventionally. In order to obtain an effective stress intensity factor range
greater than zero, the K
eff
is replaced by a fictitious stress intensity factor range K
fict
. The
fictitious stress intensity factor range is chosen so that it results in the same linear-elastic
loading of the element k=j=1 that is calculated using Equation (3.15). This condition can be
expressed by Equation (6.19), where the right side of Equation (6.19) equals the right side of
Equation (3.15), and its left side equals the right side of Equation (3.18) :

K SCF x dx
fict

=
-
2
0
0

( ) (6.19)
Solving Equation (6.19) for K
fict
gives :
K SCF
fict
=
*
0
2

(6.20)
where SCF
*
in Equation (6.20) stands for an average stress concentration factor over the
length of the element k=j=1. Very often the notation K
t
is used instead of SCF
*
in the
literature. Herein SCF
*
is used because K already stands for the stress intensity factor and
the monotonic strain hardening coefficient. Introducing K
fict
into Equation (6.18) and
Simplification and application of model F 135
EPFL Thesis 1617
integrating Equation (6.18) over the first element at the crack initiator leads to the expression
for the crack initiation life N
CI
:
N
C K C K
da
C K C K
CI
el fict
m
pl fict
m
el fict
m
pl fict
m
el pl el pl
=

!
"
#
#
$
%
&
&
=

-
1 1
0

(6.21)
in Equation (6.21) can be seen as the initial crack length a
0
and it can be taken :
a
0
= = 0.1 [mm] (6.22)
C
el
, in Equation (6.21) can be calculated using Equation (6.12). The local mean stress,
m
in
Equation (6.12), can be calculated using Equation (6.14), where K
eff,max
and K
eff
in
Equation (6.14) must be replaced by K
max,fict
and K
eff
, correspondingly. The fictitious
maximum stress intensity factor K
max,fict
can be calculated using Equation (6.23) which is
similar to Equation (6.20).
K SCF
fict max,
*
,max
=

0
2
(6.23)
0
[N/mm
2
]
N
CI
[cycles]
50
100
400
10
3
10
4
10
5
10
6
10
7
10
8
N
CI
*
SCF
*
=
6
5
4
3
2
Steel St50

Figure 6.6 : The influence of an average stress concentration factor SCF* on the N
CI
.
Figure 6.6 presents an example of a crack initiation period, calculated as a function of an
average stress concentration factor SCF
*
. The material parameters of St50 steel were chosen
to calculate the parameters of Equation (6.21). The local mean stress is taken as
m
=0. It can
be seen that the mean stress concentration factor has a very large influence on the crack
initiation period N
CI
. Figure 6.6 clearly illustrates that because the SCF
*
of the welded details
is very high, these details demonstrate a fast crack initiation period. The N
CI
also depends on
the nominal stress range
0
. If
0
is small, the crack initiation period increases
significantly, regardless if the SCF
*
remains high or not. Curves in Figure 6.6 have two
slopes. This is because Equation (6.21) is a function of two terms. The knee-points N
CI
* of the

0
-N
CI
curves can be calculated using Equation (6.24)
136 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
N
C
C
C
C
C
C
CI
el
el
pl
m
m m
pl
el
pl
m
m m
el
pl el
pl
pl el

!
"
#
#
$
%
&
&
=

!
"
#
#
$
%
&
&

(6.24)
It can be concluded that the use of the fictitious stress intensity factor range K
fict
makes it
possible to calculate the crack initiation period using da/dN-K
eff
curves.
Total Fatigue Life
The total fatigue life N
f
can be calculated using Equation (6.25). Parameters C
el
, m
el
, C
pl
and
m
pl
in Equation (6.25) can be calculated using Equations (6.12), (6.13), (6.7) and (6.8),
respectively. Equation (6.25) does not take into account the influence of simultaneous
damaging. How to appropriately account for the effect of simultaneous damaging will be
discussed in Clause 6.2.4.
N N
C K a
C K a
da
f CI
el eff
m
pl eff
m
a
a
el pl
cr
= +

!
"
#
#
$
%
&
&
-
1 1
0

( )
( )
(6.25)
Integration of the total fatigue life using Equation (6.25) can result in very long computing
time, especially if small nominal loads and/or complex loading histories are present. The
increase in computing time is directly related to the fact that the effective stress intensity
factor range K
eff
, must be re-calculated every time the crack length a changes. The algorithm
of the simplified model presented in Clause 6.2.4 will include some techniques to save
computing time.
6.2.2 Effective Stress Intensity Factor Range
A parametric study carried out in Chapter 5 showed that the nominal load range is one of the
most important fatigue parameters. In addition, the influence of the nominal mean load was
shown to be important as well. According to the crack closure concept, only the effective part
of the nominal load range causes the growth of the fatigue crack. If the reduction of the
nominal load range due to the crack closure effect is not taken into account, the results of the
fatigue analysis will be conservative.
The crack closure effect can be taken into account using the effective stress intensity factor
range K
eff
instead of the stress intensity factor range K. The K
eff
is introduced because it is
used in the crack propagation law (6.18). The objective of this section is to present a
simplified method for calculating the K
eff
due to both constant-amplitude and variable-
amplitude loading.
Constant-Amplitude Loading
The K
eff
as a function of constant-amplitude loading can be calculated as follows :
Using Equation (3.45) to find the nominal crack opening stress at a crack length a,
op
(a).
Calculate the crack opening stress intensity factor K
op
using Equation (2.7) and substituting

op
for
0
in Equation (2.7).
Using Equation (2.7), calculate the stress intensity factors due to the maximum and
minimum nominal loads, K
max
and K
min
. Using Equation (2.8), calculate the stress intensity
factor due to fabrication-introduced residual stress, K
res
.
Using Equations (6.26), (6.27) and (6.28), calculate the effective stress intensity factor
range K
eff
.
K
eff
= K
max
- K
min
if K
max
> K
min
> (K
op
- K
res
) (6.26)
K
eff
= K
max
- K
op
if K
max
> (K
op
- K
res
) > K
min
(6.27)
Simplification and application of model F 137
EPFL Thesis 1617
K
eff
= 0 if (K
op
- K
res
) > K
max
> K
min
(6.28)
Variable-Amplitude Loading
K
i
, K
i*
, K
res
Define K
op
K
min,abs
= K
min
1
3
K
max
= Max(K
i
; K
i*
) + K
res
K
min
= Min(K
i
; K
i*
) + K
res
2
?
K
max
> K
max,abs
?
K
min
< K
min,abs
NO
K
max,abs
= K
max
K
min,abs
= K
max
YES
YES
4
NO
K
min,eff
= K
min
K
eff,i
= K
max,eff
- K
min,eff
?
K
max
K
op
NO
YES
?
K
min
K
op
K
min,eff
= K
op
K
max,eff
= K
max
NO
YES
K
max,eff
= K
op
5
6

Figure 6.7 : Calculation algorithm of the effective stress intensity factor range.
The calculation algorithm for the effective stress intensity factor range K
eff,i
corresponding to
variable-amplitude load reversal, is presented in Figure 6.7. The following steps are needed :
1. Calculation of the stress intensity factors K
i
, K
i*
and K
res
, as a function of the
0,i
,
0,i*
and
fabrication induced residual stresses
res
. The index* in K
i*
corresponds to the nominal
load peak that forms a closed stress-strain hysteresis loop with the load peak (see Clause
3.4.2). The K
i
and K
i*
can be calculated using Equation (2.7), the K
res
using Equation (2.8).
2. Determination of the K
max
and K
min
as functions of K
i
, K
i*
and K
res
.
3. Re-compute, if necessary, the absolute maximum and minimum stress intensity factors
K
max,abs
and K
min,abs
.
4. Calculation of the opening stress intensity factor K
op
using Equation (6.29). The stress
intensity factor K
*
op
in Equation (6.29) corresponds to the stabilized crack opening stress

op
, and can be calculated using Equation (3.47). Similar to Equation (3.45),
Equation (6.29) takes into account the change in crack opening stress due to the change in
138 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
crack length, where K
max,abs
(a-) is the absolute maximum stress intensity factor just before
the previous element breaks.

( )
K a K
K a
K
K K
op abs
abs
abs
op
abs
( )
( )
min,
max,
max,
*
min,
= +

(6.29)
5. Calculation of the effective stress intensity factors K
max,eff
and K
min,eff
.
6. Calculation of the effective stress intensity factor range K
eff,i
.
Advantages of the proposed calculation method with respect to the effective stress intensity
factor range are :
It's generality : this method can be used to calculate the effective stress intensity factor
ranges, within a large region, while still accounting for the influence of plate thickness,
load history and yield stress, on the crack closure phenomenon. In addition, Equation (6.29)
automatically takes into account increases in the crack opening stress intensity factor if the
crack length increases.
It's simplicity : the proposed formula are simple and the algorithm in Figure 6.7 is linear.
No iterations are needed in order to calculate K
eff,i
.
It's rapidity : the proposed method leads to a very fast calculation of the effective stress
intensity factor ranges. A big advantage, when compared to existing crack closure
approaches, can be obtained. This is especially true when the influence of a complex load
history is analyzed.
6.2.3 Algorithm
The aim of this section is to present the overall algorithm of the simplified fatigue crack
propagation model - model SF. The algorithm of model SF is shown in Figure 6.8 and the
steps in computing it are reviewed later.
Data Input
Fatigue crack propagation simulation
Output of results
Element damaging and
failure
a = a +
Element
redistribution
Initialisation
?
a a
cr
NO
YES

Figure 6.8 : The overall algorithm of model SF.
Simplification and application of model F 139
EPFL Thesis 1617
Data Input
Similar to model F, there are three groups of input data required : loading history, detail
geometry and material properties. The input data should be made as described in
Clause 3.6.1.
Initialization
Initialization of the model SF includes the following tasks :
Rainflow analysis of the load history in order to determine the tips of the stress-strain
hysteresis loops (see Clause 3.4.2).
Set up of elements. The elements are placed close to the stress concentrators and amount of
elements used at the same time in the calculations should not exceed 5.
Determination of average stress concentration factors SCF
*
k
, for elements k=1...5. The
SCF
*
is required for the crack initiation stage only.
Determination of the stress intensity correction factor Y
k
, for elements k=(1), 2...n
global
.
(Y
1
=0 if no initial crack is assumed to be in the detail).
Calculation of the material-related parameters m
el
, C
pl
and m
pl
used in the crack
propagation rate law (6.18). Parameter C
el
in Equation (6.18) is a function of K
eff
and
K
eff,max
and therefore can not be pre-computed.
Calculation of constants
el
and
pl
. These constants are used in order to take account for
the simultaneous damaging of the elements. Calculation of the
el
and
pl
differs whether in
the crack initiation stage or in the stable crack growth stage :
crack initiation stage :

el k
k
m
SCF
SCF
el
,
*
*
=
!
"
#
$
%
&
1
(6.30)

pl k
k
m
SCF
SCF
pl
,
*
*
=
!
"
#
$
%
&
1
(6.31)
stable crack growth stage :
( )

el j
m
j j
el
,
= 1 (6.32)
( )

pl j
m
j j
pl
,
= 1 (6.33)
where k=1...5 in Equations (6.30) and (6.31), and j=1...5 in Equations (6.32) and (6.33).
Element Damaging and Failure
The algorithm for element failure and simultaneous damaging is given in Figure 6.9 and
includes the following steps :
Initialization of the damage calculation loop which contains the initialization of the
reversal counter (n=0) and determination of the crack propagation stage. It is assumed that
the crack initiation stage changes into the stable crack growth stage when the element at
the crack initiator (k=1) is broken.
Calculation of the effective stress intensity factor range K
eff,i
. Calculation of the K
eff,i

depends on the crack propagation stage : if crack initiation takes place, K
eff,i
must be
calculated using Equation (6.20) at the stable crack growth stage using the algorithm
presented on Figure 6.7.
140 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
Element damaging and failure
next i
n = n + 1
K = K
eff,i
Damage increment of elements
Initialisation
?
Propagation
stage
CI
SCG
Calculate K
eff,i
K = K
fict,i
?
D
1
1
NO
YES
a=a+

Figure 6.9 : Algorithm of the element failure and simultaneous damaging.
Damage increment of elements
j = 1
?
j 5
NO
YES
j = j + 1
Damage of element j
D
j
= D
j
+ d
j,i
Calculate d
j,i
(Equation 6.34)
?
D
11

Figure 6.10 : Algorithm of the damage increment of elements.
Damage increment of elements. This calculation step is presented in Figure 6.10 and it
consists of looping, where within every loop, the damage increment of the element j, d
j,i
is
Simplification and application of model F 141
EPFL Thesis 1617
calculated
1
. This can be made using Equation (6.34). Then the d
j,i
is added to the total
damage of the element j, D
j
. Looping is continued until the damage increment of all 5
elements is found.
d
C K C K
j i
el j el eff i
m
pl j pl eff i
m
el pl
,
, , , ,
=

+

!
"
#
#
#
#
$
%
&
&
&
&
1
2
1
1 1

(6.34)
Check if the element j=1 has failed. If the damage of element j=1 is greater or equal to 1,
then it is assumed that the first element has failed. Otherwise the damage increment of the
elements is calculated using the next load reversal, i.
Increment of the Crack Length
After failure of the element j=1, the crack length a is increased by the element size . Then,
the new crack length a is compared to the critical crack length a
cr
. If aa
cr
, then the
calculation is finished, and the output is written, else the elements are redistributed.
Redistribution of Elements
Since the element j=1 has failed, it is removed and a new element is added. The redistribution
procedure is similar to the procedure described in Clause 3.6.2 : the intermediate results are
saved, the elements re-numerated and the new element is added at the crack propagation path.
The initial damage of the new element, D
5
, can be calculated using Equation (3.39). Finally,
the absolute maximum and the absolute minimum stress intensity factors K
max,abs
and K
min,abs
,
must be re-calculated as indicated in Clause 3.5.3.
Output of Results
In this step all the results are written in a result file. This can be made as described in
Clause 3.6.3.
6.2.4 Verification of model SF
In chapter 4 model F was compared to fatigue test results. In this clause, the same is done
using model SF. In addition to comparing it to the test results, model SF is compare to
model F.
Comparison to Test Results
As an example, one series of comparisons is shown. Fatigue test for these comparisons are
taken from Section 4.2.1 corresponding to the plate with a welded attachment. It should be
noted that welded details have relatively short crack initiation lives compared to their total
fatigue life. The needle peened specimen under constant-amplitude and spectrum A and B
loading is used for verification of model F. Specimen geometry and loading conditions are
given in Figure 4.2 and Figure 4.3, respectively.
Comparisons between model SF-simulated results and data from section 4.2.1 are presented
in Figure 6.11. Each comparison includes two types of curves : a-N curves and da/dN-a
curves. The variation of the simulated curves is obtained by changing the weld toe angle and
the distribution of the residual stress (see Figure 4.2).

1
Factor 2 in denominator of Equation (6.34) counts for the damage due to load reversal i. If damage due to
load cycle is calculated, then factor 1 must be used instead.
142 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
0.000001
0.00001
0.0001
0.001
0 2 4 6 8 10
N [cycles] a [mm]
a [mm] da/dN [mm/cycles]
= 40
o
res
: min
measured
simulated
= 60
o
res
: max
a) CA loading
0
2
4
6
8
10
0 200000 400000

N [cycles] a [mm]
a [mm] da/dN [mm/cycles]
= 40
o
res
: min
measured
simulated
= 60
o
res
: max
b) Spectrum A
0
2
4
6
8
10
0 1000000 2000000
0.0000001
0.000001
0.00001
0.0001
0 2 4 6 8 10

N [cycles] a [mm]
a [mm] da/dN [mm/cycles]
= 40
o
res
: min
measured
simulated
= 60
o
res
: max
c) Spectrum B
0
2
4
6
8
10
0 4000000 8000000 12000000
0.0000001
0.000001
0.00001
0 2 4 6 8 10

Figure 6.11 : Comparison of model SF to the test data [4.1] : a) constant-amplitude
loading ; b) spectrum A ; c) spectrum B.
Comparisons show that model SF is able to simulate fatigue behavior of specimens of
complicated geometry, taking into account the influence of residual stresses and varying
loading conditions. This conclusion is valid for welded details having relatively short crack
initiation life compared to their total fatigue life.
Simplification and application of model F 143
EPFL Thesis 1617
Comparison to model F
The second series of comparisons are made between model SF and model F. Two types of
crack propagation is simulated by the models; 1) retardation effect of the overload and;
2) influence of mean stress.
da/dN [mm/cycle]
N [cycles] a [mm]
2a
120 mm
0
2
18 mm
0
a [mm]
1
2
pcf = 2.0
'
ys
= 311 N/mm
2
cycles
0
[N/mm
2
]
150
0
200
9
11
13
15
17
19
0 600000 1200000 1800000
1
0.0000001
0.000001
0.00001
0.0001
0.001
9 11 13 15 17 19
1 - model F
2 - model SF
small crack
behaviour

Figure 6.12 : Comparison between model SF and model F : retardation effect due to
tensile overload.
Figure 6.12 presents the effect of tensile overload on fatigue behavior, simulated using
model SF and model F. Figure 6.13 presents the influence of changing the nominal mean
stress. The results are given as a-N curves and da/dN-a curves. These curves reveal the
following :
it appears that similar to model F, model SF can simulate small crack behavior, but
crack propagation rates of small cracks predicted by model SF are smaller than those
predicted by model F.
The crack initiation life simulated by model SF is much shorter than the crack initiation
life predicted by model F. This is because model SF is based on the crack propagation
rate equation (6.18) which always leads to greater crack propagation rates than the strain
life relationship (3.11) which used as the basis of model F (see also Figure 6.3).
The stable crack growth life predicted by model SF is longer than that predicted by
model F. This is because the effective stress intensity ranges calculated by model F are
smaller than those found by model SF. Correspondingly, model SF predicts shorter
retardation of crack propagation due to tensile overload than model F (Figure 6.12). In
addition, the influence of the change in nominal mean stress on crack propagation is also
less important if calculated using model SF rather than using model F (Figure 6.13).
144 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
It appears that model SF is not able to consider a short acceleration in crack propagation
just after application of the tensile overload. Since this acceleration is very short, it is not
important to take it into account.
da/dN [mm/cycle]
N [cycles] a [mm]
2a
120 mm

0
2
18 mm

0
a [mm]
1
2
pcf = 2.0
'
ys
= 311 N/mm
2
cycles

0
[N/mm
2
]
150
0
1
1 - model F
2 - model SF
9
11
13
15
17
19
0 600000 1200000 1800000
0.0000001
0.000001
0.00001
0.0001
0.001
9 11 13 15 17 19
50
200
125
-25

Figure 6.13 : Comparison of model SF to model F :change in mean stress and small crack
behavior.
It can be concluded that model SF and model F result in approximately the same results in
cases when the crack initiation life is small compared to total fatigue life. On the other hand,
in cases where crack initiation life constitutes a large portion of the total fatigue life,
model F leads to shorter crack propagation than model F.
6.2.5 Discussion of model SF
This section reviews the basic assumptions, outlines the advantages and disadvantages, and
establishes the application limits and application field of simplified model - model SF.
Assumptions
Model SF is based on a crack propagation rate equation (6.18) which was developed on the
basis of strain-life relationships (2.18) and (2.19), Glinkas ESED criterion and Ramberg-
Osgood equations (3.19) and (3.20). The above mentioned equations and relationships are
related to each other according to the principles used in model F. These principles are
reviewed and analyzed in Chapter 3, including use of the SCF(x) and Y(a) in order to
characterize the linear-elastic stress field at a stress concentrator using the average value of the
same linear-elastic stress as element loading, etc.. The basic assumptions considering
model SF, are :
Simplification and application of model F 145
EPFL Thesis 1617
Elements represent the average material fatigue behavior along their length. Elements are
placed along crack propagation path, where the latter is situated perpendicular to the
maximum principal tensile stresses occurring in the specimen due to nominal loading.
Fatigue life of elements can be calculated using strain-life relationships obtained by smooth
specimen fatigue testing (see Clause 3.3.2). For that, it is assumed that if elements, placed
on the surface of smooth specimen, at a crack initiator or, at a fatigue crack tip, are loaded
by equal strain ranges, then the fatigue lives of these elements are equal.
The crack initiation stage finishes after the element at the crack initiator fails. This
assumption was largely confirmed by fatigue analysis carried out using model F.
Advantages and Disadvantages
The advantages of the model SF are :
The constants of the crack propagation rate law can be evaluated using material data
obtained from smooth specimen testing. A large range of these data are available in the
literature. In addition, the proposed crack propagation rate equation does not require any
special considerations to take into account the fatigue threshold.
The crack initiation period can be evaluated, using the same approach and the same
material constants used to calculate the duration of the stable crack growth period.
A simplified crack closure model accounts for the increase in fatigue life due to the crack
closure effect. In addition, the evolution of the opening stress intensity factor, as function
of crack length, is considered.
Model SF accounts for the influence of fabrication-introduced residual stresses and the
effect of the simultaneous damaging. This gives the model the ability to simulate small
crack behavior, without adding any special features.
The disadvantages of model F are :
Greater initial efforts are needed to put the proposed approach into computer code than if
the conventional fracture mechanics approach was used. However, the increase in the range
of application is many times greater than the loss in time invested in programming.
Since the fatigue analysis is carried out using average material data, safety factors should
then be applied to the results if used for fatigue design. The calculation of the appropriate
safety factors is out of the scope of this study.
Application Field and Application Limits
Based on the modeling assumptions, as well as on the comparisons of model SF to test data
and to model F, it can be said that model SF can be used in fatigue analysis for specimens
of any geometry and containing any distribution of residual stress. Covered materials include
structural steels of ductile stress-strain behavior. Since features to consider crack closure
effect are included, model SF is able to account for the influence of variable-amplitude
loading on fatigue behavior.
However, the maximum applied loads should not cause large scale yielding of the net section.
This condition must be guaranteed in order to ensure that small scale yielding conditions are
present at the crack tip (see condition (5.20)). The limits on element size are established by
condition (5.19). If model SF is applied on the analysis of details made of aluminum, the
limits of the element size should be reviewed. This is due to fact that micro-structure of
aluminum differs from that of steel.
Finally, it must be said that like model F, model SF is a deterministic model and therefore
does not take into account variation of material properties. Due to its wide application range
and relative simplicity, model SF can, however, be supplemented by including statistical
aspects of the material properties into the model.


146 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617

6.3 APPLICATION EXAMPLES
The objective of this section is to present two examples of how model SF and model F can
be applied o fatigue analysis. The examples include :
The study of fatigue properties of structural steels by means of crack propagation rate
equation (6.18).
Comparison of variable-amplitude fatigue analysis carried out by model F to similar
analysis made using S-N curves and linear damage accumulation rule.
6.3.1 Analysis of Material Parameters
In this section, two topics are presented : the comparison of fatigue properties of structural
steels with the aid of the crack propagation equation (6.18), leading to recommendations
concerning the choice of the material.
Influence of Steel Grade
Comparison is made between five structural steels : St37, St42, St46, St50 and St52. The
material properties needed to calculate the constants C
el
, m
el
, C
pl
and m
pl
used in
Equation (6.18), are taken from [6.7]. The local mean stress
m
in Equation (6.12) is taken 0.
Figure 6.14 represents 25 da/dN-K
eff
curves calculated using to 25 sets of the material data
from [6.7]. The raw material of the test specimens were treated by high temperature in some
cases.
K
eff
[N/mm
1.5
]
B
A
10
-4
10
-5
10
-6
10
-7
10
-8
heat treatment :
T 880
o
C
flame-cut
Dubois [4.1]
da/dN [mm/cycle]
100 1000

Figure 6.14 : Crack propagation rate curves for unalloyed steels St37, St42, St46, St50 and
St52.
A comparison of the curves show that :
the da/dN-K
eff
curves of the steels analyzed generally fall within a relatively narrow
scatter band. The boundaries of the scatter in crack propagation rate curves are marked
with A and B in Figure 6.14. A scatter band of crack propagation rate curves measured by
Dubois [4.1], are shown in same figure. It appears that da/dN-K
eff
curves predicted using
Equation (6.18) correspond quite well with results of Dubois [4.1]. Differences between
measured and predicted curves are observed in the region of small K
eff
. This can be
explained by the fact that in the analysis of test results, Dubois did not take into account the
Simplification and application of model F 147
EPFL Thesis 1617
effect of small crack behavior and therefore obtained relatively high crack propagation
rates at small values of K
eff
.
The effective threshold stress intensity factor range K
eff,th
at the crack propagation rate
da/dN
th
=10
-8
mm/cycle is :
K
eff,th
= 100...130 [N/mm
1.5
] (6.35)
An analysis shows the large influence of heat treatment on crack propagation rate - heating
the material to a temperature of T>880
o
C leads to catastrophic loss in fatigue properties.
The curves of heat-treated materials (dotted curves in Figure 6.14), are out of the region
that is bounded with curves A and B. Figure 6.14 also shows the negative influences of
flame cutting on fatigue properties
1
.
If heat treatment is made under a temperature of T<660
o
C, then the curves of fatigue crack
propagation rate fall within the region that is bounded by curves A and B.
Comparison showed that high temperature has a very large influence on the fatigue properties
of material. This implies that the material inside the heat affected zones around welds can
loose a considerable part of it fatigue resistance.
Constants C
el
, m
el
, C
pl
and m
pl
used in Equation (6.18) are sensitive to the constants of the
strain-life relationship. The constants of the strain-life relationship can be determined by
fatigue testing of a smooth specimen, but a large amount of these constants are available in the
literature [6.7].
Even though fatigue properties of structural steels are within a relatively narrow scatter band,
the fatigue life of two identical details made of different steel can, however, differ many
times. Fatigue design with the steel with good fatigue properties allows a remarkable gain in
fatigue life.
Choice of Material for Fatigue Design
Based on Figure 2.17 and on the modeling principles, the following recommendations can be
made, in order to select the material :
if the detail contains high local stress concentrations (welded details, for example) then a
ductile material should be chosen. The use of a ductile material leads to a slower stable
crack growth compared to tough or strong materials, the crack initiation stage of this type
of details is short regardless the type of material.
If the detail contains small or medium local stress concentrations (details with holes and
smooth notches, for example) then a strong material should be chosen. The use of a strong
material leads to a longer crack initiation stage compared to tough or ductile materials, the
stable crack growth stage of this type of detail is short compared to the crack initiation,
regardless the type of material.
If the detail is made of a composite material, then for the first layer, a strong material
should be used which leads to a long crack initiation life. Other layers should be made of a
ductile material that makes the stable crack growth stages slower than if tough or strong
materials were used.
6.3.2 model F versus S-N Curves
Even though S-N curves are normally used for fatigue design and not for prediction of fatigue
life, sometimes they are used in order to evaluate (remaining) fatigue life of existing
structures. It is interesting to show the difference in prediction of the fatigue life made using
model F and S-N curve. The aim of this section is to compare the fatigue lives N
f
(MF)
obtained using model F, to the fatigue lives N
f
(SN) obtained by means of the S-N curve.

1
The dotted curve noted flame-cut in Figure 6.14 corresponds to the point flame-cut in Figure 5.13.
148 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
The fatigue life N
f
(MF) is obtained by numerical simulations using model F. The detail
geometry used in simulations is a plate with hole. The dimensions of the plate are identical to
those used in the parametric study (Figure 5.1). The material properties used are given in
Table 5.2. Simulations with model F were carried out under variable-amplitude loading. The
64 load histories used were calculated in the way shown in Annex A.4.3. The fatigue life
N
f
(MF) is calculated in units of load history.
The fatigue life N
f
(SN) is calculated as follows :
firstly, the rainflow analysis of the 64 load histories (see Annex A.4.3), identical to those
used in numerical simulations, is carried out in order to obtain the load spectra.
Secondly, using Equation (6.36), the damage due to every load spectrum, is calculated.
Since the plate with hole corresponds to the detail category 90 in [6.10], the double-slope
fatigue resistance curve of the DC90 (Figure 5.18) is used to calculate N
f,i
in
Equation (6.36).
d
n
N
ls
i
f i i
=
.
,
(6.36)
Depending on the load range level in spectrum, the N
f,i
in Equation (6.36) is calculated
using Equations (6.37), (6.38) or (6.39) :

i D
: N
N
f i
D
m
D
i
m ,
=

1
1
(6.37)

CUT OFF i D
< : N
N
f i
D
m
D
i
m ,
=

2
2
(6.38)

i CUT OFF
<

: N
f i ,
(6.39)
where the constants in Equations (6.37), (6.38) and (6.39) correspond to the fatigue
resistance curve DC90, and they are :

D
= 66 [N/mm
2
]
m
1
= 3
N
D
= 510
6
[cycles]

CUT-OFF
= 36 [N/mm
2
]
m
2
= 5
The fatigue life N
f
(SN), is calculated in units of load spectra using Equation (6.40), where
the damage due to one load spectrum d
ls
in Equation (6.40) is calculated using
Equation (6.36).
N SN
d
f
ls
( ) =
1
[load spectra] (6.40)
In comparisons, all variable-amplitude load histories are normalized using the equivalent
constant-amplitude stress range
e
, which can be calculated using Equation (5.17). The
results of comparison are presented in Figure 6.15. The comparison shows that if

e
>100 N/mm
2
, use of S-N curve leads approximately the same fatigue life as model F. On
the other hand, the difference between fatigue lives predicted using model F and S-N curves
increases if
e
decreases. Within the range 36
e
80 N/mm
2
, this difference is at least
three times. If the equivalent constant-amplitude stress range is less than the cut-off limit, then
both model F and S-N curves predict fatigue threshold.
There are two reasons why the S-N curves lead to more conservative results than does
model F for use in prediction of fatigue life for details loaded by variable-amplitude loading.
Simplification and application of model F 149
EPFL Thesis 1617
First, use of the linear damage accumulation rule instead of non-linear rule leads to shorter
fatigue lives. Second, S-N curves correspond to fatigue lives reduced by a safety factor. In the
example given, fatigue lives predicted by S-N curves are not much shorter than those
calculated using model F. This is because the mean stress in the load histories used for the
simulations are high. The conservatism of the S-N curves would be larger if load histories
with lower mean stresses (with compressive nominal stresses), were used. It should be noted
that the comparison made is valid, however, for the given detail geometry and material
properties only.

e
[N/mm
2
]
model F and S-N
curves lead to fatigue
threshold
1
10
100
0 50 100 150 200 250
N
f
(MF)
N
f
(S-N)
model F leads to the
fatigue threshold

Figure 6.15 : The comparison of variable-amplitude fatigue analysis made using model F
and S-N curves.
6.4 SUMMARY
Model F versus Paris Equation
In the first part of Chapter 6, a crack propagation rate equation was developed. This equation
leads to results similar to the Paris equation. There are three advantages of the proposed
equation compared to Paris equation. First, the constants of Equation (6.18) can be determined
without testing (i.e., they can be calculated using the material constants of the strain-life
relationship and Ramberg-Osgood equation). Second, Equation (6.18) accounts for the fatigue
threshold without introducing an extra parameter into crack propagation rate equation. Third,
the proposed equation accounts for the effect of the maximum stress intensity factor on the
crack propagation rate.
Model SF
Computational procedures used in model F are relatively complex : they include a number of
iterations and rather involved algorithms. A simplified crack propagation model - model SF
was developed on the basis of model F. It leads to similar results as a fatigue analysis
conducted using model F, but uses much simpler computing procedures and algorithms.
150 Extended numerical modeling of fatigue behavior
EPFL Thesis 1617
model SF is based on the crack propagation rate equation (6.18). It allows calculation of the
fatigue life by taking into account all important aspects of fatigue crack propagation: the
influence of the crack initiation and stable crack growth stages, the crack closure effect, the
small crack behavior, the effect of residual stresses, etc. An original crack closure model,
presented in Chapter 3, is simplified and introduced into model SF. Working algorithms of
model SF are also given. Model SF is compared to fatigue test data and to model F, its
advantages and disadvantages outlined, and application limits and application field
established.
Application Examples
Two examples were introduced in order to show how the approaches developed can be used in
fatigue analysis. The first example showed how the crack propagation rate equation (6.18) can
be used to compare the fatigue properties of steels. Comparison of structural steels St37, St42,
St 46, St50 and St 52 showed that their fatigue properties are relatively similar and that heat
treatments at high temperatures (T>880
o
C) significantly reduce the fatigue resistance of these
materials. The recommendations concerning the choice of the material were also given.
In the second example, fatigue lives obtained using model F were compared to fatigue lives
obtained using the S-N curves. The comparison showed that use of the S-N curve for variable-
amplitude fatigue analysis of a plate with center hole predicts shorter fatigue lives than
model F, especially in the region of a small equivalent constant-amplitude stress range.