Nuclear Engineering and Design 311 (2017) 16–27

Contents lists available at ScienceDirect

Nuclear Engineering and Design

journal homepage: www.elsevier.com/locate/nucengdes

Impacts of weld residual stresses and fatigue crack growth threshold on

crack arrest under high-cycle thermal fluctuations
Said Taheri a,⇑, Emricka Julan b, Xuan-Van Tran c, Nicolas Robert d
a
EDF-LAB, IMSIA, 7 Boulevard Gaspard Monge, 91120 Palaiseau Cedex, France
b
EDF-LAB, AMA, 7 Boulevard Gaspard Monge, 91120 Palaiseau Cedex, France
c
EDF Energy R&D UK Centre/School of Mechanical, Aerospace and Civil Engineering, The University of Manchester, Manchester M13 9PL, UK
d
EDF-DPN, UNIE, Strategic Center, Saint Denis, France

h i g h l i g h t s

For crack growth analysis, weld residual stress field must be considered through its SIF in presence of a crack.

Presence of cracks of same depth proves their arrest, where equal depth is because mean stress acts only on crack opening.

Not considering amplitudes under a fatigue crack growth threshold (FCGT) does not compensate the lack of FGCT in Paris law.

Propagation rates are close for axisymmetric and circumferential semi-elliptical cracks.

a r t i c l e i n f o a b s t r a c t

Article history: High cycle thermal crazing has been observed in some residual heat removal (RHR) systems made of 304
Received 2 August 2016 stainless steel in PWR nuclear plants. This paper deals with two types of analyses including logical argu-
Received in revised form 10 November 2016 mentation and simulation. Crack arrest in networks is demonstrated due to the presence of two cracks of
Accepted 16 November 2016
the same depth in the network. This identical depth may be proved assuming that mean stress acts only
on crack opening and that cracks are fully open during the load cycle before arrest.
Weld residual stresses (WRS) are obtained by an axisymmetric simulation of welding on a tube with a
Keywords:
chamfer. Axisymmetric and 3D parametric studies of crack growth on: representative sequences for vari-
Variable amplitude
XFEM
able amplitude thermal loading, fatigue crack growth threshold (FCGT), permanent mean stress, cyclic
Weld counting methods and WRS, are performed with Code_Aster software using XFEM methodology. The fol-
Threshold lowing results are obtained on crack depth versus time: the effect of WRS on crack growth cannot be
Counting method determined by the initial WRS field in absence of crack, but by the associated stress intensity factor.
Crack arrest Moreover the relation between crack arrest depth and WRS is analyzed.
In the absence of FCGT Paris’s law may give a significant over-estimation of crack depth even if ampli-
tudes of loading smaller than FCGT have not been considered. Appropriate depth versus time may be
obtained using different values of FCGT, but axisymmetric simulations do not really show a possibility
of arrest for shallow cracks in contrast with logical argumentations.
Ó 2016 Elsevier B.V. All rights reserved.

1. Introduction RHR systems. Fig. 1b gives 10 s of temperature variation in the fluid

Thermomechanical crazing (shallow and dense unidirectional
or multidirectional crack networks), was discovered (Molinie
et al., 2002; Robert, 2004) in some nuclear power plants in differ-
ent areas of RHR systems made of 304 ASS (Fig. 1a). The crack
nucleation has been attributed to high cycle variable amplitude
thermal load in the mixing zones of cold and hot water flows of
⇑ Corresponding author.
E-mail address: Said.taheri@edf.fr (S. Taheri).
on a location on Father mock-up of a RHR systems (Le Duff et al.,
2011; Courtin, 2013). This is a broad band thermal loading where
significative values of DSP are obtained between 0.1 Hz and about
6 Hz in the fluid (Vincent et al., 2009). Several detrimental factors
impact crack nucleation in RHR systems such as stress singularities
at welding toe, surface slope change, and surface roughness. But
pre-hardening (Taheri et al., 2015a) (due to surface treatment
under high compression) and weld tension residual stresses are
also detrimental factors on which less attention has been paid.
The effects of these factors are amplified due to the dispersion on

http://dx.doi.org/10.1016/j.nucengdes.2016.11.008
0029-5493/Ó 2016 Elsevier B.V. All rights reserved.
 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27 17

Nomenclature

AC all cycles WRS weld residual stress

ASS austenitic stainless steel 2D axisymmetric XFEM simulation
EDF Electricite De France 3D tridimensional XFEM simulation
FCGT fatigue crack growth threshold DK stress intensity factor variation in mode I
FCGR fatigue crack growth rate DK th fatigue crack growth threshold for long cracks in mode I
FEM finite element method rm mean stress
XFEM extended finite element method KI SIF in mode I
RSE_M French nuclear maintenance code for nuclear plants DT temperature variation
NO-FCGT Paris law parameter identification without FCGT K mean mean value of SIF
RHR reactor heat removal
SIF stress intensity factor
VA variable amplitude

amplitude thermal loading on the internal face have shown that

Fig. 1a. Crack networks near a weld seam in a RHR system.
Fig. 1b. A portion of the thermal loading sequence revealed on the mock-up Father
of a RHR system (500 points).
fatigue life as the thermal load amplitude is close to the fatigue
limit of a 304 stainless. All affected RHR systems were changed
in French power plants by improving the inner surface, e.g. by pol-
ishing and eliminating stress concentration zones such as weld
toes. Research work has been undertaken to study the conditions
of nucleation and crack arrest in high-cycle variable amplitude
thermal fatigue in RHR systems.
From the operational feedback on different areas of RHR sys-
tems of 900, 1300 and 1450 MWa plants, it was deduced
(Molinie et al., 2002; Robert, 2004) that maximum crack depth
after 3000 h of operation is smaller than about 2.5 mm for all
cracks in networks (Fig. 1c). This has suggested the possibility of
crack arrest when cracks have a maximal depth about 2.5 mm
(Kamaya and Taheri, 2008; Sbitti and Taheri, 2010). Axisymmetric
and 3D simulations (Taheri, 2007) for a tube under constant
for fully circumferential cracks the stress intensity factor variation
(DK) as a function of crack depth may become decreasing at a cer-
tain depth. Parametric studies (Kasahara et al., 2002; Taheri, 2007)
on thermal load amplitude, thermal exchange coefficient, and the
frequency of thermal loading show the particular importance of
frequency of thermal loading on decreasing of the DK. A decreasing
DK is obtained for high frequencies of loading as the stress ampli-
tude vanishes quickly in the depth. Due to the existence of a FCGT
for a 304 ASS (Lesur, 2005; Le-Roux and Akamatsu, 2002) the crack
may stop propagating when DK becomes equal to FCGT. As cracks
stop propagating, but nucleation at different locations continues,
crack density increases and thermal crazing appears. This analysis
on the possibility of crack arrest is a proposition and not a proof of
crack arrest before replacing RHR systems. A proof of crack arrest
will be given in Section 2.
It is moreover shown that the deepest cracks of the same depth
in a network have not been submitted to a mutual shielding effect
(Sbitti and Taheri, 2010), so only single cracks may be considered
for FCGR study.
To determine FCGR under elasticity hypotheses the French
maintenance code for nuclear power plants RSE_M [RSE_M] uses
Paris’s law without any FCGT where amplitudes of loading under
FCGT obtained by cycling counting are not considered in the
methodology. However, as will be presented in this paper, simula-
tions performed in the absence of FCGT in Paris’s law with a com-
monly used mean stress of 50 MPa show that for Father loading,
too conservative results on crack depth are obtained (predicted
depths are much larger than crack depths observed on compo-
nents). But the simulation of DK under a constant amplitude load-
ing DT ¼ 100  C which is higher than amplitudes obtained from
Father experiment, with a zero mean stress may give a good order
of crack depth at arrest about 2 mm (Sbitti and Taheri, 2010). This
is why in this paper extensive analyses on FCGT and mean stresses
and WRS are performed.
An important difference has to be noticed between the study of
crack arrest and obtaining a validated FCGR by numerical mod-
elling. For the first one (in axisymmetric case) only comparison
between variation of SIF and DK th of long cracks is needed, as max-
imal crack depths are sufficiently large in both RHR systems and
Father mock-up. For the second one validation on a structure is
more difficult, as for example in Father experience, 300 h of exper-
imentation includes nucleation time, short crack propagation and
long crack propagation time. In this case only qualitative analysis
of simulations results on cracks depths may be proposed.
The impact of WRS is another outcome of this work where its
effect on crack nucleation has been proved on RHR systems
(Taheri, 2007; Lei et al., 2014; Taheri et al., 2015b). Moreover max-
imum crack depths in networks are larger in the vicinity of welds
18 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27
Fig. 1c. Identical crack depth in crack networks in a RHR system and in Father mock-up.
(Robert, 2004) under tension WRS, compared to crack depths in
networks far from welds under a high compressive stress (due to
the surface treatment).
Numerical simulations are carried out with EDF’s open-source
Code_Aster software [Code_Aster]. Simulations for axisymmetric
circumferential cracks, and initially semi elliptical circumferential
cracks by XFEM methodology (Moës et al., 1999) were performed
(Taheri et al., 2015a). X-FEM’s major advantage as compared to
remeshing is that it represents the crack without explicitly
meshing it. The 3D simulations are carried out in the absence of
WRS and FCGT and are used for validation of axisymmetric
simulations.
The principal objectives of this paper are:

To show that the existence of deepest crack of the same depth
in a network proves that these crack are at arrest.

To analyze the relation between mean applied stress and WRS
on crack arrest depth.

Regarding simulation, the final objective of this work has been
to propose a valid numerical modelling of FCGR under variable
amplitude loading with a validation performed on crack arrest
depth data.
This work follows that presented in (Sbitti and Taheri, 2010).
Compared to the previous work, in this work 3 more ingredients
are considered: FCGT, WRS, and variable amplitude thermal
loading.
2. The proof of crack arrest and consequences
A proof of crack arrest in RHR systems network at the moment
they have been replaced or at the end of Father mock-up experi-
ment may be given under two following hypothesis:

The deepest cracks in a network have the same depth (at least
two cracks). Fig. 1c shows that this hypothesis may be consid-
ered as valid for a RHR system and for Father-Mock-up. Other
results also confirm this hypothesis (Taheri, 2014). It must be
noticed that the equality of depths concerns only cracks in
one network (in one location) as the loading is not the same
from one location to another, for example due to the distance
difference from TE junction.

Nucleation is not simultaneous at different points of a RHR sys-
tem: In high cycle fatigue, crack nucleation on the surface of a
component may not be simultaneous at different locations even
under the same loading. In fact for a 304 ASS, even for a labora-
tory test where the surface of the sample is well polished and
the stress field is homogenous, the dispersion is substantial
after 100 000 cycles (Taheri et al., 2015a). It is worth noting that
simulations under constant amplitude loading show that to
have an arrest at about 2.5 mm depth thermal loading requires
a frequency higher than 0.7 Hz (Sbitti and Taheri, 2010) which
gives a fatigue life upon 100 000 cycles due to an experimental
crack nucleation time higher than 100 h [private communica-
tion]. In addition, roughness on an industrially treated surface
(grounding or flap wheel) is more important than on laboratory
samples. This makes non simultaneous nucleation more proba-
ble. Non-simultaneous nucleation is shown by tests on a struc-
ture device (Maillot et al., 2005). The specimen geometry is a
thin plate and is polished. It is continuously heated by joule
effect and it is cyclically subjected to thermal shocks by water
spray on the two opposite sides. It is shown on some locations
that, between cycle numbers N and N þ dN the existing cracks
have not propagated whereas new cracks appear in the center
of the same area. This confirms continuous and non-
simultaneous nucleation.
Thus when one crack nucleates and propagates, because of the
previously explained non-simultaneous nucleation, another crack
nucleates later. If the first and second cracks are propagating, there
is no reason for these two cracks to have the same depth, as one is
nucleated after the other. One may argue that a difference between
loadings, for example on 4 cracks in Fig. 1c (RHR system) may
explain it. However, the probability that the difference in loadings
on 4 cracks are such as to compensate exactly the difference in
nucleation times must be zero.
Regarding the previous paragraph, a key point is to understand
why the deepest cracks in a network could have the same depth at
 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27 19

arrest. It is important here to notice differences of definitions

between: DK effectiv e used in the literature and DK efficient used in this
paper. We have
DK efficient ¼ K max  K min if K min > 0 and
DK efficient ¼ K max if K min < 0 ð1aÞ

DK effectiv e ¼ K max  K op ð1bÞ

Kop (Lawson et al., 1999) is the nominal SIF for which crack really
opens and depend in some cases significantly on the plasticity at
the crack tip through closure effect.
The analysis may start by the hypothesis that DK applied-load and
K applied-load
mean due to applied thermal and mechanical stress fields must
be identical on 4 cracks, if not the probability to have identical
depth must be zero.
However, as may be noticed in Fig. 1a (Fig. 1c is a cut on the line
given in Fig. 1a) the WRS has not a constant value and depends on
the distance from the weld. As the linear superposition of SIFs is
supposed (K total applied total
mean ¼ K mean þ K WRS ) thus K mean are different on 4
cracks because of the difference of WRS (positive). The question
is why these cracks have the same depth.
Authors propose a possible explanation by supposing that the
mean stress K totalmean acts only for crack opening within the definition
of DK efficient (Fig. 2). Thus, as we are in presence of small amplitude
loading, a high positive WRS helps the crack to stay fully open dur-
ing the cycle and the arrest depth is independent to mean stress. In
this case, cracks under negative WRS not being fully open at arrest
have smaller depths. This is in agreement with the fact that on the
part of component under compression (created by surface treat-
ment) in networks, cracks are less deep than the ones near the
weld.
However, the assumption that the mean stress (K totalmean ) acts only
for crack opening seems apparently in contradiction with existence
of K op which is different for different mean stresses (Tanaka, 1989).
Authors proposal to resolve this apparent contradiction is:

K WRS
OP ¼ 0: As has been noticed in (Lawson et al., 1999) to find a
relation between closure effect and residual stresses is a very
difficult task as crack closure is conventionally defined as taking
place behind the crack tip while residual stress affects the
material in front of it. Nevertheless as may be seen for axisym-
metric simulations (Figs. 9a and 9b, Section 6), when a crack is
created under tension WRS (crack is introduced in the body and
Fig. 3. Paris’s law identifications with and without FCGT compared to the RSE_M
law, and dependence of FCGT on temperature.
equilibrium solution is obtained under elasticity hypothesis) the
crack tip stays in tension. Therefore one may suppose K WRS
OP ¼ 0.
Obviously this may not be true if WRS relaxation through crack
propagation takes place under elastic-plastic behavior.

K applied-load ffi 0 Applied loadings are small, and in addition crack
op
propagation in a tube is far from plane stress situation where
closure effect is important (Lawson et al., 1999).
Having the same depth also proves, by the same argument as
before, that there is no interaction between the deepest cracks of
a network with equal depths. In fact the crack which nucleates first
would shield the one which nucleates later and so they may not
have the same depth. Obviously there exists a shielding effect of
the deepest cracks of the network on many shallower cracks
observed in networks, as simulations can show in a case study
(Seyedi et al., 2009). However if the shallowest cracks have been
subjected to a shieling effect by the deepest cracks there is a
mutual interaction so the shallowest also must impact the deepest
crack propagation. But the fact that the deepest cracks have the
same depth proves that the effect of the shallowest cracks on the
deepest cracks have been negligible. This negligible effect may be
justified by the fact that these cracks are too shallow to affect
the deepest cracks, probably due to an important retard in
nucleation related to high dispersion on fatigue life near to the

Fig. 2. Efficient DK versus crack depth as a function of KMEAN.

20 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27

Fig. 4. Axial WRS after 14 passes of welding, (red color tension, and blue color compression) and locations of cracks used for simulations. (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 5a. WRS, axisymmetric finite element simulation, stress fields on internal
surface of a tube with a chamfer.

Fig. 6. Organization of numerical simulations.

Fig. 5b. Crack networks configurations on a RHR system.

Fig. 7a. Representative signal in the absence of FCGT, sequential counting,

endurance limit. In the absence of shielding effects, only FCGR of sin-
gle cracks are considered.
With regard to thermal transients considered as an overloading
they have not been taken into account in this paper. In fact the
thermal transients have high values of DK and low values of fre-
quencies. Under low frequencies (0.1 Hz) for a tube under axisym-
metric loading (Seyedi et al., 2003) DK may stay always increasing
rm = 50 MPa.
and so prevent crack arrest, but as cracks are at arrest this proves
that the effect of the thermal transients are negligible. This is in
agreement with operational feedbacks which show that thermal
transient have been rare during RHR systems lives (private
communication).
 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27 21

Fig. 9b. SIF of WRS for 4 cracks at different heights.

Fig. 7b. Representative signal with & without FCGT (FCGT = 3.34), rainflow
counting, rm = 50 MPa.
3. Basic data for finite element simulations

It is a difficult task to obtain an improved evolution of temper-

ature field on the metallic structure of an RHR mixing tee through
thermo-hydraulic simulations (Kamide et al., 2009; Fontes et al.,
2010; Kamaya, 2014; Qian et al., 2015; Utanohara et al., 2016).
Simulations for codified applications are therefore carried out on
the main tube (straight part after mixing zone, named principal
area) of a RHR system. Moreover, thermal loading on the internal
face of this tube is assumed to be a constant function of spatial
coordinates but dependent on time. The dependence on time is
obtained through the temperature revealed on Father Mock-up at
a location where the loading is assumed to be most damaging
based on fatigue crack nucleation analysis. The Father mock-up
of a 1300 MWa RHR system produced a thermal variable loading
that is accurately measured and which is assumed to be close to
the thermal loading of a RHR system. In this study, the initial signal
is the signal described in (Vincent et al., 2009). Different parts of
this signal will be used to carry out simulations. A small part of
the signal is given in Fig. 1b.
To replace mean values of stress due to stratification or other
Fig. 8. The effect of different values of FCGT on FCGR for the sequence: 0–200 s.
bending moments or pressure, parametric studies are performed

MPa
200
100
0
-100
-200

Fig. 9a. Axial WRS versus depth at different heights from the middle of the pipe.
22 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27
with different mean stress values which are applied as a constant
tension on the top of the tube in axial direction. A constant value
of applied stresses through the thickness due to internal pressure
and bending is assumed as we are in presence of thin tubes.
Tube geometry: several thicknesses are considered

12.8 mm (thickness of the principal area of a 900 MWa RHR sys-
tem) with WRS.

7.5 mm (thickness of the principal area of the mockup Father)
without WRS.
The height of tubes is taken high enough to consider that the
middle of the tube is far from boundary conditions and so strain
or stress fields created by the thermal load may be assumed uni-
form on the surface.
Thermomechanical data for thermo-mechanical simulation (elastic
material)

Young’s modulus : 198; 500 MPa.

Poisson ratio: 0.3.

Thermal exchange coefficient: 50 W=mm2 C.

Thermal expansion coefficient: a ¼ 1:60  105 ð1=CÞ.

Thermal diffusion coefficient : k ¼ 15:38  103 W mm1 C1 .

Density ⁄specific heat : qC p ¼ 3:6  103 J mm3 C1 .
Landings and boundary conditions

Variable thermal loading: A representative part of Father-mock-
up signal.

Constant permanent mean loading: different values from 0 to
200 MPa will be used, however rm ¼ 50 MPa is the most used
in this paper.

Thermal boundary conditions: zero flux on the outer face of the
tube, previous variable loading on the inner face

Mechanical boundary conditions: clamped or simply supported
on the bottom of the tube.
Crack geometry:

Axisymmetric crack: fully circumferential cracks are consid-
ered. For a 304 ASS, long cracks (in contrast to short cracks)
have a length higher than a value between 0.5 and 0.8 mm
(Lesur, 2005). For simulations, an initial crack length of
0.6 mm is usually assumed to be the limit case for long cracks.

3D plane crack: initial semi elliptical circumferential cracks
with an aspect ratio c/a = 2 are used. Initial maximal depth for
the semi-elliptical crack is a = 0.6 mm.
4. Paris’s laws and parameter identifications
In this study, we are interested only in mode I propagation. In
the absence of mean stress, FCGR in Paris’s law is given respec-
tively by formula (1) in the absence of FCGT and by formula (2)
in presence of FCGT:
da=dn ¼ CðDKÞm
1 in the RSE_M code.
da=dn ¼ C 1 ðDK  DK th Þm
Several reasons may explain the absence of FCGT in RSE_M or
ASME codifications. The first may be that for high values of DK,
for example for thermal transients (thermal shocks of high ampli-
pffiffiffiffiffi
tude) the presence of DK th with values around 5 MPa m for ASS
(Lesur, 2005) does not significantly modify FCGR through formula
(2). Moreover the results of simulation remain conservative
compared to the case where FCGT is incorporated in Paris’s law.
A second reason may be that in the absence of FCGT a semi-
analytical method (Kasahara et al., 2002; Musi and Beaud, 2003)
using influence functions enables SIF to be obtained for some
structures such as a straight tube via a transfer function from a
temperature field which is given analytically. This makes simula-
tions very fast even for 3D cracks. Introducing FCGT makes the
transformation nonlinear and makes it difficult to obtain a transfer
function.
4.1. Mean stress effect
Mean stress effect in Paris’s Law in RSE_M code: In RSE_M code,
the effect of mean stress has been taken into account through for-
mula (3) for positive R ratio.
da=dn ¼ C½DK=ð1  R=2Þ4 where R ¼ K min =K max > 0 ð3Þ
Different modelling are given in the literature to combine mean
stress and FCGT effects on FCGR (Bathias and Pineau, 2008). How-
ever in this paper mean stress effect is taken into account only
through crack opening by DK efficient (Fig. 2). This choice is related to
the three following points: (a) to eliminate contact problem in sim-
ulations. Thus for all simulations we take K min ¼ 0 when K min < 0,
(b) absence of an improved modelling (only conservative modelling
are given in codifications) combing R ratio FCGR, in high tempera-
ture and under PWR environment, (c) presence of cracks of the same
depth under different WRS as explained in previous sections, related
to the effect of mean stress acting only for crack opening.
4.2. Identification of Paris’s law parameters
The most conservative values of parameters of Paris’s law in
RSE_M for ASS and for PWR environment (FCGR is given in mm/cy-
pffiffiffiffiffi
cle and SIF in MPa m) are:
C ¼ 1:8  E  9; m¼4 ð4Þ
Paris’s law parameters for an EDF-304 ASS have been provided
for various R ratios and temperatures (20–300 °C) in (Le-Roux and
Akamatsu, 2002). We identified Paris’s law parameters at room
temperature for a 304 ASS, using a data obtained for R = 0.7, to
be close to the previous law (formula (4)). The results of identifica-
tion with FCGT and without FCGT (named NO-FCGT), are plotted in
Fig. 3.
4.3. Cycle counting method and various propagation laws
In variable amplitude loading, SIF is obtained as a function of
time. A counting method is applied to SIF and different amplitudes
DK i (1 < i < NÞ are obtained where N is the number of cycles. Dif-
ferent combinations of rainflow counting method and Paris’s law is
used which are defined below:
1. Paris law with threshold = counting method + introduction of a
FCGT in Paris law.
2. Quasi-RSE_M = Paris’s law without FCGT + counting method
ð1Þ where no DK i < DK th is considered. The name quasi-RSE_M is
due to the fact that R ratio has not been taken into account as
ð2Þ
3. All cycles = Paris’s law without FCGT + counting method, where
all DK i are accounted for.
In (Taheri et al., 2015c) all cycles and Quasi-RSE_M methods are
compared under 50 MPa mean stress. Results show that for the
Father signal negligible difference exists between FCGR obtained
by two methods (about 20%) in spite of the fact that 70% of
 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27
amplitudes are under FCGT. In consequence all simulations without
FCGT in this paper use the all cycles method which is then very close
to the RSE_M method.
Another counting method which is designed by sequential
counting is also used in this paper. In this sequential counting,
two half successive unequal variations of loading are replaced by
two other half variations equal to the highest value of the two
unequal variations. The reason for this use is that rain-flow count-
ing destroys the chronology of the sequence while this is not the
case for sequential counting. The aim however is not to take into
account any effect of retard related to overloading.
5. Weld residual stress simulation
Welding simulation is a difficult task in terms of validation of the
simulation. So we carried out the same simulation as the one pre-
sented in (Razakanaivo et al., 2000). This is the axisymmetric
numerical simulation of 13 passes on a straight thin pipe with a
chamfer. The pipe thickness is h = 12.8 mm. For the requirements
of crack growth analysis a 14th pass on the outer face of the tube
has been added to completely fill in the chamfer. Fig. 4 shows ampli-
fied deformation of the pipe after welding simulation and axial
stress field which affects circumferential crack propagation in mode
I.
5.1. WRS local & structure effect, standard & tapered (weld) area
The aim of this paragraph is to justify why crack propagation is
studied only for cracks in the (weld) tapered area.
For the engineering issues, usually a straight pipe with a circum-
ferential weld is separated into two parts: tapered (weld) area and
the complementary part which is designed by standard area. The
tapered area (on internal face of the tube) is usually considered to
be an area with more damage than the standard area because of
geometrical singularities but also due to thickness reduction
achieved for welding. However, there may sometimes be a confu-
sion between the detrimental effect of tapered area and local
WRS effect under tension. Fig. 5a shows the axial and circumferen-
tial simulated WRS at the internal surface after 13 weld passes plot-
ted versus the axial coordinate which is in the direction of the tube
axis. These stress fields may be decomposed into 2 parts through an
analogy with stress fields under an axial thermal shock (Taheri,
1998): a local 3D effect, under tension on a length of the order of
h from the weld (h thickness of the tube) and a structure effect,
under compression where the maximum is located on axial coordi-
pffiffiffiffiffiffiffiffiffiffiffiffiffi
nate at a distance about h  r m (r m mean radius of the tube) from
pffiffiffiffiffiffiffiffiffiffiffiffiffi
the weld. The quantity h  r m is the characteristic length of thin
shell theory and has a value about 50 mm (h ¼ 12:8; r  20 mm).
Thus on a tube with a circumferential weld, areas may be separated
into three parts (not two: tapered area, standard area):

Weld local effect area in tension (superposed to tapered area).

Weld structure effect under compression.

Outside the effect of the weld.
The principal reason for restricting crack propagation analysis
below to the local effect area of WRS is that the deepest cracks
in the networks are in this area and not outside the WRS effect
(Robert, 2004).
5.2. Comparison between simulated WRS and crack networks
configurations in a RHR system
Fig. 5b gives crack configuration on a RHR system and validate
simulations. As may be noted on the area of structure effect under
23
compression, there are no cracks. Close to the weld, the presence of
uni-directional axial cracks (outside the weld toe) may be related
to the largest WRS component which is a circumferential stress.
In the standard zone where the effect of WRS disappears, a
multi-directional crack network is observed. For these areas one
may relate crack initiation to the detrimental effect of roughness
as usually assumed, but also to the effect of pre-hardening due
to a high compressive stress (Taheri, 2007) obtained by surface
treatment.
6. Simulations of fatigue crack growth
As far as the high cycle fatigue regime is concerned, simulations
may not be achieved cycle by cycle. Moreover, the elasticity
hypothesis enables propagation to be controlled by increment of
crack length. Fig. 6 gives the organization of the simulation in
axisymmetric and in 3D. As the mean stress effect in this paper is
taken into account only through crack opening, for all simulations
we take K min ¼ 0 when K min < 0 (closed crack). This eliminates sim-
ulation with contact which is virtually impossible in HCF because of
CPU time. It may be noticed that replacing the contact problem by
K min ¼ 0 according to some work (Pierard et al., 2016) may intro-
duce some error margins. Moreover it is worth noting that due to
thermal diffusion some parts of the crack may be closed while the
tip of the crack remains open. This means contact on some areas
of cracks when other areas are not in contact. This problem will
be eliminated in many cases by taking mean stresses sufficiently
high to keep cracks fully open all the time, this is the case for
rm ¼ 50 MPa. In fact for mean stresses of 50 and 100 MPa for any
crack depth larger than 0.6 mm, crack depths versus time are super-
posed (Taheri et al., 2015b) which shows that cracks are fully open
during cycling. Moreover rm ¼ 50 MPa is the limit case value,
usually used for engineering assessments on RHR systems.
6.1. A representative loading signal
A numerical simulation of crack growth with totality of Father
signal requires considerable CPU time. In fact the CPU time is pro-
portional to the duration of the signal due to SIF simulation which
must be performed at each instant where the stress field is deter-
mined. Thus a part of the signal is selected for simulations and the
increment of crack for this sequence is obtained. However, this
portion of signal must be representative. The representative load-
ing sequence is defined by a portion of initial signal so that when
repeated enough to get the same duration as the initial signal,
the same crack geometry must be obtained. As no simulation
may be achieved with the total signal only, convergence through
increase of sequence duration may be considered at different load-
ing starting instants. However, one difficulty could arise from rain-
flow counting; the amplitudes extracted from the signal T (T0 to T0
+T) are not all included in the signal T+T (T0 to T0+2T). For exam-
ple, if just after T signal there is a high amplitude loading this
would modify some amplitudes previously extracted for T
sequence, as the highest value is brought into the beginning of
the sequence due to standard ANFOR (Anfor, 1993). Thus the con-
vergence as noted before may not exist. This is why sequential
counting is also used for comparison, where theoretically this con-
vergence must exist.
Fig. 7a shows that in absence of FCGT for a shallow crack, for an
axisymmetric solution, 10 s of Father signal taken from the starting
point T0 = 500 s may be considered as more representative than
10 s taken from the starting point T0 = 100 s. However, due to high
CPU demands, no more simulations are performed with T0 = 500.
In Fig. 7b results obtained for different sequences are compared:
0–100 s, 100–200 s, and 0–200 s. Results are perfectly superposed
24 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27

for shallow cracks less deep than 2.5 mm. From now all simulations residual stresses and external loading. A linear superposition is then
will be performed with the sequence 0–200 s. Getting a representa- supposed and programmed in free software Code_Aster. Thus we
tive sequence in presence of deep cracks needs new investigations have:
with longer signals and so higher CPU times, but this is not within
K total ¼ K WRS þ K rm þ K Fatherload ð6Þ
the scope of this paper.
Fig. 7b shows that sequences: 0–100 s, 0–200 s or 100–200 s The loading is composed of Father signal, axial WRS, and a per-
may be considered as less representative in presence of a FCGT, manent mean stress rm = 50.
compared to the case without FCGT. However it may be considered
as representative for shallow cracks up to 2.5 mm. The reason for 6.3.1. Crack growth simulation in absence of FCGT
which the presence of FCGT makes necessary to have longer signals SIF due to WRS (axial stress): WRS are self-equilibrated, it
to be representative is that as cycles with an amplitude smaller means that the integration of axial stress through the thickness
than FCGT are not considered, the number of cycles in the must be equal to zero (y represent axial direction and r radial
sequence would not be statistically enough anymore to be repre- direction in the tube). We have:
sentative. Results in (Taheri et al., 2009) show that almost 20 Z Rext Z Rext
cycles are necessary for the Father loading after removing the r ryy dr ffi r m ryy ¼ 0 ð7Þ
cycles under FCGT to be statically representative. For higher values Rint Rint

of FCGT obviously longer signals are needed to be representative. Fig. 9a seems to validate this assertion. But when the crack
propagates, WRS is modified to take into account the balance given
6.2. Comparison of crack depth with and without FCGT and the effect by formula (6). So equilibrium must stay valid through staying
of temperature ligament. This is for example the case at the height 7 mm. In spite
of a region of high compressive stress near to the external surface,
At room temperature literature gives different values of FCGT the associated SIF stays positive (Fig. 9b). In contrast with heights
pffiffiffiffiffi
for a 304 ASS. These values vary between DK th ¼ 3 MPa m and 7, 14, 20 mm, at height 26 mm, in spite of a region of tension near
pffiffiffiffiffi
DK th ¼ 5 MPa m (Le-Roux and Akamatsu, 2002), while our identi- the associated external surface, SIF stays negative. A negative SIF
pffiffiffiffiffi
fication here gives the value DK th ¼ 3:34 MPa m. Moreover FCGT has no meaning but the following constraint: K min ¼ 0 when
depend also on temperature. Fig. 3 shows that FCGT at 150 °C is K min < 0 has been removed to have previous representation of a
pffiffiffiffiffi
even higher than DK th ¼ 5 MPa m (maximum temperature in negative SIF. These results show that it is not possible to analyze
the mixing zone of a RHR system is given equal to 160 °C). For the effect of WRS on crack growth through the initial WRS field,
the need of simulation and to take into account high values of but only through SIFs associated with the WRS fields.
FCGT, Paris’s curve with FCGT identified at 20 °C is translated along SIF due to the sum (rm ¼ 50 MPa)+ WRS: Fig. 9c shows the SIF for
x-axis. After translation, for high values of DK we find the Paris’s rm ¼ 50 MPa and the SIF for (rm ¼ 50 MPa)+ WRS. Note that for the
law without FCGT (NO-FCGT). height of 26 mm the SIF associated with (rm ¼ 50 MPa)+ WRS is
Results in Fig. 8 compare crack depths versus time in the almost zero up to 5 mm. As indicated by the figure, this result is
pffiffiffiffiffi related to an initial compressive WRS on the internal face of the tube.
absence of FCGT and for DK th ¼ 5 MPa m and
pffiffiffiffiffi Crack depths versus time for the total loading (thermal loading
DK th ¼ 3:34 MPa m. The significant sensibitivity to the value of
FCGT is proved for Father loading where the loading amplitude is + rm + WRS) is given in Fig. 9d. The results can be explained as
small. Comparison of results with and without FCGT (close to follows:
RSE_M code) show that it is not sufficient to eliminate cycles under
FCGT (RSE_M methodology) but one must use Paris’s law with
A: The crack at the height of 7 mm and the crack at the height
FCGT for the remaining amplitudes (Taheri et al., 2015c). 14 mm have the same behavior despite the different values of
WRS, because rm + WRS keeps the crack fully open during load-
ing cycles for both heights from the length of 0.6 mm. In the
6.3. Axisymmetric solution for crack propagation in presence of WRS
absence of FCGT the quasi arrest at about 12.6 mm (near exter-
nal surface) may be explained by the fact that DK for high fre-
Fig. 4 shows amplified simulated deformations of the pipe
quency of thermal loading is nearly equal to zero at this
(h = 12.8 mm) after welding. WRS field is then transferred to a
depth even if the crack is fully open.
more refined mesh for crack growth study. The problem is sym-

B: The crack at the height 20 mm has a behavior different from
metrical with respect to the plane of the middle of the pipe, so only
the cracks at 7 or 14 mm. However, its arrest depth is the same
y > 0 (y is the axial direction of the pipe) is concerned. Cracks are
as for 7 and 14 mm. More precisely, at 3 mm depth they begin
located at different heights: 7 mm, 14 mm, 20 mm, and 26 mm rel-
ative to mid plane of the pipe. y  26 mm is almost the boundary
of WRS local effect with WRS structure effect (two times thickness
from the middle of the seam).
Axial stress versus depth for different heights (7, 14, 20, 26 mm)
are given in Fig. 9a. For the three first height WRS is in tension on
internal surface while it is under compression for the last one.
The introduction of residual stresses in fracture mechanics in
absence of plasticity is very complex (Bui, 2006), due to changing
the elasticity law which becomes inhomogeneous and anisotropic.
The linear relation (formula (4)) between stress and strain on
which is based linear superposition of SIF is valid only out of a
small region around the crack tip.

rij  r0ij ¼ ktracðeÞdij þ 2leij ð5Þ

Based on this linear relation through energy release rate, one may Fig. 9c. KI for mean stress Sm = 50 MPa and sum of SIFs of WRS and Sm, for cracks
prove validity of linear superposition of SIFs associated with at different heights.
 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27
Fig. 9d. Crack depths versus time, in presence of the WRS rm ¼ 50 MPa, for cracks
at different height.
to be different and at about 10 mm depth they have the same
FCGR. This means that the crack at 20 mm height is not fully
open on a cycle between 3 mm and 10 mm. After being partially
open it becomes fully open and finds the same FCGR and arrest
depth as the two others.

C: The crack at the height 26 mm is almost at arrest at 2 mm
depth without any FCGT. The reason is that the K mean (rm +
WRS) is very small up to 5 mm depth where DK has to be signif-
icant. But when K mean increases, DK is too small to have a signif-
icant impact on crack depth. The smallness of K mean is related to
the fact that WRS is compressive at internal surface at the
height of 26 mm. It is worth noting that with an exponent of
3–4 in Paris’s law, FCGR in the presence of a mean stress which
fully opens the crack is up to 9–16 times greater than that with-
out mean stress.
6.4. Tridimensional solution (3D) in absence of WRS and FCGT
3D simulations for plane crack propagation with XFEM have
been performed. The mesh is given in (Taheri et al., 2015b). Local
energy release rate along a semi-elliptical crack frontier obtained
by XFEM and by remeshing method are compared. Perfect super-
position of results is obtained. SIF in mode I at each point on the
crack frontier is then obtained through Irwin’s relation under plane
strain conditions G ¼ K 2 =ð1  mÞE.
The first objective of 3D simulation of FCGR has been validation
of axisymmetric simulations which requires much less CPU time
for deep cracks. That is why simulations of 3D circumferential
crack are preferred to axial one. Moreover a representative
sequence for the 3D case must have at least the same length as
for the axisymmetric case but here we suppose that the represen-
tative sequences for the axisymmetric case are also valid for the 3D
case. So in the absence of FCGT, 20 s of signal is assumed to be rep-
resentative for shallow cracks (see Fig. 7b).
Regarding mean stress, in the axisymmetric case, for cracks
with a depth larger than a = 0.6 mm, rm ¼ 50 MPa is sufficient
(limit case) to keep the crack fully open during a cycle in the
absence of compressive WRS. For the 3D case, a larger value is
obtained rm ¼ 60 MPa for an aspect ratio of c/a = 2 and a = 6 mm.
However, a mean stress of 200 MPa is used to ensure a fully open
crack during cycling for crack depths even smaller than 0.6 mm.
Fig. 10a shows the evolution of a 3D crack under Father loading
with more extension in circumferential direction than in depth.
6.4.1. Comparison of two cycling counting methods
A comparison between the two counting methods including
rain-flow and sequential is presented in (Taheri et al., 2015b) for
sequences less than 12 s for a crack with c/a = 2. Results are very
25
close for the two methods. Fig. 10b shows some larger differences
between two methods with a sequence of 20 s.
6.4.2. Comparison between axisymmetric and 3D circumferential
cracks
In (Taheri et al., 2015a) FCGR for axisymmetric and 3D initially
semi-elliptic cracks (with aspect rations, c/a = 2, and c/a = 10) are
compared for a sequence of two seconds, close results are obtained
(the aspect ratio in the case of the RHR systems may reach c/
a = 20). Fig. 10b compares FCGR of a axisymmetric crack with an
initially semi-elliptical crack with c/a = 2 for a sequence of 20 s.
Close results are obtained for shallow cracks.
6.5. Discussion
Analysis of the results of crack depth versus time shows impacts
of cycle counting methodologies and the choice of axisymmetric or
3D simulations. Some error margins may come also from 3D sim-
ulation due to the error margins in evaluation (Vu et al., 2015) of
the SIF on the free surface when the crack is not orthogonal to
the free surface. However, the effect of previous parameters seems
negligible compared to the difference of crack depths obtained in
the absence and in the presence of FCGT in Paris’s law (Fig. 8)
due to the small amplitude loading (Father loading).
Validation of modelling of FCGR based on Father test is difficult
as 300 h of experimentation includes the nucleation time, short
crack propagation and long crack propagation. Nevertheless as
the deepest cracks in networks are long cracks and are assumed
to be at arrest so modelling of crack arrest depth may be carried
out using FCGT. However, it is not easy to choose the value of FCGT
as the temperature varies. In fact temperature elevation increases
FCGT in air, but in PWR environment beneficial effect of tempera-
ture elevation on FCGT is reduced, which may cancel partially ben-
eficial effect of high temperature on FCGT. Moreover the
dependence on frequency of FCGT is difficult to analyze as for
RHR systems we are in presence of broad band loading where sig-
nificative values of DSP are obtained between 0.1 Hz and about
6 Hz in the fluid (Vincent et al., 2009; Qian et al., 2015). It is so dif-
ficult in a quantitative way to decide which conservative value of
FCGT must be used in simulations. Maximal cracks depths
obtained for crack networks on RHR systems may be interesting
data in future to determine a medium value of FCGT in presence
of variable temperature loading, PWR environment, and frequency
variation. However, the main qualitative shortcoming of axisym-
metric simulation seems to be the convexity of curves giving crack
depth versus time. In fact this curvature makes crack arrest at
small depth not possible (Fig. 8). 3D simulation gives a concave
function which corresponds better to an arrest. This may be related
to a quicker circumferential extension than depth extension at
high frequency of thermal loading. A 3D simulation in presence FCGT
may help to confirm the importance of 3D simulation on this aspect.
A question may be raised if a possible retard due to variable
amplitude loading or overloading may have an impact on the
Fig. 10a. 3D crack propagation: circumferential extension greater than radial
extension.
26 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27
Fig. 10b. Simulations for a sequence of 20 s, rm = 200 MPa (Seq: sequential, RF:
rainflow).
curvature discussed before. In (Okazaki et al., 2014) the interaction
effects between HCF (thermal fluctuation) and LCF (start up or shut
down of plant) in crack propagation are considered. Retard effect
has not been taken into account in this paper because this is very
difficult in the presence of FCGT. However, several reasons may
justify the absence of a sufficient retard effect in the RHR system:
(a) high cycle thermal fluctuation is a broad band random loading
for which the effect of cycle interaction is considered not to be
important (Skorupa, 1998). Concerning overloading, for example
for thermal transients, we showed that crack arrest also imposes
scarcity of thermal transient.
7. Conclusions
The effect of WRS on crack propagation cannot be determined
by the initial WRS field in absence of crack, but through associated
SIF as WRS relaxes during crack propagation.
Conclusions 1 and 2, 3 below are the result of logical argumen-
tation and are valid for axisymmetric cases. They are based on the
hypothesis that mean stress only has an effect on crack opening
which has been discussed in the paper.
1. Expertise on damaged RHR systems which have been removed
show that in a localized network the deepest cracks have the
same depth. Using two hypotheses which seem likely: (a)
nucleation is not simultaneous at different locations of a com-
ponent due to HCF, (b) mean stress acts only on crack opening:
it is proved that: (a) These cracks are at arrest, b) they have not
be submitted to a mutual shielding effect, (c) the effect of tran-
sients (thermal shocks) has been negligible.
2. Crack is fully open at arrest: This case gives the maximum depth
at arrest and explains the identical depth for the deepest cracks
of a network. This case may occur for both cases of cracks fully
or partially open during the cycle up to just before arrest. As a
final and important conclusion, maximum crack depth is deter-
mined only by the variation of SIF of the thermal loading and
not by the mean value of SIF (permanent mean stress or WRS).
3. Crack is not fully open at arrest: In this case, the crack depth at
arrest depends on mean stress (rm and WRS). This case may be
produced for a sufficiently large compressive stress issued by
WRS, or by surface treatment.
For the principal area of a RHR system, 2 and 3 may explain the
presence of networks of shallower cracks in the standard area out
of weld influence compared to the tapered (weld) area which is the
zone of WRS tension.
The following results are also obtained in relation with axisym-
metric and 3D simulations:

Parameters of Paris’s law with and without threshold have been
identified for a 304-EDF stainless steel.

For a high cycle thermal variable amplitude load, the use of Par-
is’s law + Rainflow counting with removing amplitudes under
FCGT gives an over-estimation of crack depth. It is necessary to
use a FCGT in Paris’s law because applying DK or DK  DKth on
the remaining amplitudes may give significantly different
results, while the impact of amplitudes under FCGT is negligible.

For the variable loading used here for simulations, the impact of
the choice of counting method on FCGR seems negligible.

Axisymmetric and 3D solutions in the absence of FCGT give
close results for shallow cracks.

Taking into account the FCGT, it is possible by simulation to
approach a desired shallow crack depth within an appropriate
time. However, for the axisymmetric simulation the shape of
the curves presenting crack depth versus time are not in accor-
dance with an arrest at small depth, which has been demon-
strated by argumentation. Better curvature is obtained by 3D
simulation. Thus a 3D crack growth simulation in the presence
of FCGT may help to a better understanding of this issue.
References
ANFOR Standard AFNOR A 03-406, novembre 1993. Fatigue sous sollicitations
d’amplitude variable: Méthode Rainflow de comptage des cycles.
Bathias, C., Pineau, A., 2008 Fatigue des matériaux et des structures 1, Ed. Lavoisier.
Bui, H.D., 2006. Fracture Mechanics, Inverse Problems and Solutions, Appendix A:
Residual Stresses in Fracture Mechanics. Springer.
Code aster <http://www.code_aster.org>.
Courtin, S., 2013. High cycle thermal fatigue damage prediction in mixing zones of
nuclear power plants: engineering issues illustrated on the FATHER case.
Procedia Eng 66, 240–249.
Fontes, J.P., Braillard, 0. Cartier, 0, Dupraz, S., 2010 High cycle thermal fatigue in
mixing zones: investigations on heat transfer coefficient and temperature fields
in PWR mixing configurations. In: The 18th international conference on nuclear
engineering (ICONE18), ICONE18-29179.
Kamaya, M., 2014. Assessment of thermal fatigue damage caused by local fluid
temperature fluctuation (part I; characteristics of constraint and stress caused
by thermal striation and stratification). Nucl. Eng. Des. 268, 121–138.
Kamaya, M., Taheri, S., 2008. A study on the evolution of crack networks under
thermal fatigue loading. Nucl. Eng. Des. 238, 2147–2154.
Kamide, H., Igarashi, M., Kawashima, S., Kimura, N., Hayashi, K., 2009. Study on
mixing behavior in a tee piping and numerical analyses for evaluation of
thermal striping. Nucl. Eng. Des. 239, 58–67.
Kasahara, N., Takasho, H., Yscumpai, A., 2002. Structural response function
approach for evaluation of thermal striping phenomena. Nucl. Eng. Des. 212,
281–292.
Lesur, N., 2005. Propagation fatigue des fissures physiquement courtes dans un
acier inoxydable 304L Ph.D thesis. ENSMA, France.
Lawson, L., Chen, E.Y., Meshii, M., 1999. Near-threshold fatigue: a review. Int. J.
Fatigue 21, 15–34.
Le Duff, A., Tacchini, B., Stephan, J.M., Tampigny, R., Fissolo, A., Vincent L., 2011. High
cycle thermal fatigue issues in RHRs mixing tees and thermal fatigue test on a
representative 304L mixing zone. PVP2011-57951, Baltimore, Maryland, USA.
Le-Roux, J.C., Akamatsu, M., 2002. EDF internal report HT-26/02/027/B.
Lei, B., Tran, V.X., Taheri, S., Wan, L., Zhou, Y., 2014. Effects of biaxial mean stress on
the critical plane orientation under biaxial tension/compression fatigue loading
conditions. Int. J. Fatigue 66, 194–206.
Maillot, V., Fissolo, A., Degallaix, G., Degallaix, S., 2005. Thermal fatigue crack
networks parameters stability and experimental study. In: Int. J. Solids Struct.
42 (1), 759–764.
Moës, N., Dolbow, J., Belytschko, T., 1999. A finite element method for crack growth
without remeshing. Int. J. Numer. Meth. Eng. 46, 135–150.
Molinie, E., Monteil, N., Delatouche, S., Roux, S., Robert, N., Pages, C., 2002.
Caractérisation des troncons RRA 900-1300MWe déposes: synthèse des
enseignements acquis. In: Proc. Fontevroud 5, France, pp. 883–895.
Musi, S., Beaud, F., 2003. An analytical model for thermal fatigue crack initiation and
propagation in mixing zones of piping systems. In: Proc. ASME-PVP, Cleveland,
USA.
Okazaki, M., Muzvidziwa, M., Iwasaki, A., Kasahara, N., 2014. Effect of thermo-
mechanical loading histories on fatigue crack growth behavior and the
threshold in SUS 316 and SCM 440 steels-for prevention of high cycle
thermal fatigue failures. E-J. Adv. Maint. 6, 24–32.
 S. Taheri et al. / Nuclear Engineering and Design 311 (2017) 16–27
Pierard, O., Berto, F.B., Campagnolo, A., 2016. Simulation of contact on crack lips and
its influence on fatigue life prediction. In: Proc. ICMFF11, June 1–3 Seville.
Qian, S., Kanamaru, S., Kasahara, N., 2015. High-accuracy CFD prediction methods
for fluid and structure temperature fluctuations at T-junction for thermal
fatigue evaluation. Nucl. Eng. Des. 288, 98–109.
Razakanaivo, A., Desroches, X., Bois, C., Lejail, Y., Kichenin, K., 2000. Tubular welding
numerical simulation: experimental validation. In: Proc. ICRS-6.
Robert, N., 2004 Internal Report, EDF/DPN D4550.02-04/4932:REA/RBT.02/0385.
RSE_M, 1997 Règles de Surveillance en exploitation des matériels Mécaniques des
îlots Nucléaires REP, tome II, édition.
Seyedi, M., Taheri, S., Stephan, J.M., Hild, F., 2003 Study of crack propagation in a
multicracked structure in thermal fatigue. SMIRT 17, Prague, paper D220.
Skorupa, M., 1998. Load interaction effects during fatigue crack growth under
variable amplitude loading – a literature review part I – empirical trends.
Fatigue & Fract. Eng. Mater. Struct. 21, 987–1006.
Sbitti, A., Taheri, S., 2010. Crack arrest in high cycle thermal fatigue. Nucl. Eng. Des.
240 (1), 30–38.
Seyedi, M., Taheri, S., Hild, F., 2006. Numerical modeling of crack propagation and
shielding effects in a striping network. Nucl. Eng. Des. 236, 954–964.
Taheri, S., 1998. Three-dimensional local effect and shell theories. Int. J. Press.
Vessels Pip. 36, 225–246.
Taheri, S., 2007. Some advances on understanding of high cycle thermal fatigue
crazing. ASME J. Press. Vessel Tech. 129, 400–410.
Taheri, S., 2014. Synthesis of studies on fatigue crack growth and their arrest in RHR
systems with XFEM method in presence of weld residual stresses. EDF internal
report H-B60-11451-EN.
27
Taheri, S., Sbitti, A., Galenne, E., 2009 Study of crack arrest under thermo-
mechanical variable amplitude loading for 3D plane crack by remeshing and
XFEM. Fatigue design, November 25 & 26, Senlis, France.
Taheri, S., Julan, E., Tran, X.V., 2015a. Fatigue crack growth and arrest under high-
cycle thermal loading using XFEM in presence of weld residual stresses. In:
Proc. of 5th Conference on Crack Path (CP-2015) September, Ferrara Italy.
Taheri, S., Vincent, L., Le-Roux, J.C., 2015b. A Classification of metallic alloys for
fatigue damage accumulation: a conservative model under strain control for
304 stainless steels. Int. J. Fatigue 70, 73–84.
Taheri, S., Tran, V.X., Julan, E., Robert N., 2015c Fatigue crack growth and arrest
under high-cycle thermal loading using XFEM in presence of weld residual
stresses. Proc. of SMIRT 23 paper 504, 14–18 August Manchester UK.
Tanaka, K., 1989. Mechanics and Micromechanics of Fatigue Crack Propagation.
ASTM STP 1020 Philadelphia: 151–183.
Utanohara, Y., Nakamura, A., Miyoshi, K., Kasahara, N., 2016. Numerical simulation
of long-period fluid temperature fluctuation at a mixing tee for the thermal
fatigue problem. Nucl. Eng. Des. 305, 639–652.
Vincent, L., Chaise, S., Braillard, O., Le-Roux, J.C., Taheri, S., Beaufils, R., 2009. On the
significant difference of lifetime prediction between strain and stress control
analysis. In: Proc. of 2th Int. Conf. on Material and Component Performance
Under Variable Amplitude Loading. Darmstaat, Germany.
Vu, M.N., Geniaut, S., Massin, P., Marigo, J.J., 2015. Numerical investigation on corner
singularities in cracked plates using the G-theta method with an adapted h field.
Theor. Appl. Fract. Mech. 77, 59–68. 788.