Engineering Fracture Mechanics xxx (2018) xxx–xxx

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

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

Modelling of creep rupture of ferritic/austenitic dissimilar weld

interfaces under mode I fracture
Jia-nan Hu a,b,⇑, Takuya Fukahori c, Toshihide Igari c, Yasuharu Chuman c, Alan C.F. Cocks a
a
Department of Engineering Science, University of Oxford, Parks Road, OX1 3PJ, UK
b
Sente Software Ltd, Surrey Technology Centre, Guildford GU2 7YG, UK
c
Research & Innovation Centre, Mitsubishi Heavy Industry. Ltd, 717-1, Fukahori-machi, 5-chome, Nagasaki 851-0392, Japan

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

Article history: Dissimilar metal welded structures (DMWs) have been used extensively in conventional
Received 24 May 2017 and nuclear power generation plants. Evaluation of creep rupture properties of DMWs is
Received in revised form 29 October 2017 critical to the structural integrity assessment. Failure of DMWs can occur in the base metal,
Accepted 4 January 2018
the heat-affected zone (HAZ), or the dissimilar interface between the two welded materi-
Available online xxxx
als, depending on the operating stress and temperature. The primary focus of this work is
on interface failure in systems consisting of a ferritic steel (P91 or P22) and an Inconel filler
Keywords:
material, which has an austenitic structure. A planar damage zone is introduced within a
Dissimilar metal weld
Interface creep rupture
finite element (FE) framework to model the response of the interface. A traction-
Damagezone model separation constitutive law with a Kachanov-type damage accumulation relationship is
P91 employed to describe the interface response, with the material parameters calibrated
P22 against available creep rupture data in which failure occurred at the dissimilar weld inter-
Inconel face. It is found that the difference in damage accumulation along the interface of different
DMW systems can be attributed to the mismatch in creep properties of the continuum
materials either side of the interface. Diversion of the crack path into the HAZ is also cap-
tured as a result of damage accumulation in the heat affected zone (HAZ). The relationship
between the empirical damage accumulation model and the major microstructural fea-
tures that are responsible for interface failure is also discussed.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction

Dissimilar metal welded joints (DMWs) are used extensively in many high temperature engineering applications in which
different materials are used in different regions of a plant to satisfy the mechanical, environmental and/or economical
requirements of the design [1]. In conventional or nuclear power generating plant, DMWs between ferritic low alloy steels
and austenitic stainless steels or between different grades of ferritic steels are commonly used [2–4]. Due to the thermo-
mechanical mismatch between the two materials, filler metals (such as Inconel) are often used and buttered on one material
before being joined to the other. This, ensures the mutual solubility between the two welded materials, and also helps mit-
igate against the residual stresses that can be produced when the DMWs are heated to elevated temperatures in service [3,4].
Evaluation of the creep rupture properties of DMWs is critical to the structural integrity assessment. A variety of failure
modes associated with DMWs have been identified during monotonic uniaxial creep rupture tests of planar components or

⇑ Corresponding author at: Sente Software Ltd, Surrey Technology Centre, Guildford GU2 7YG, UK.
E-mail address: [email protected] (J.-n. Hu).

https://doi.org/10.1016/j.engfracmech.2018.01.001
0013-7944/Ó 2018 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
2 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Nomenclature

a coefficient of thermal expansion (CTE)

eeij elastic strain of continuum material
ecrij creep strain of continuum material
ethij thermal strain of continuum material
e_ 0 ðe_ 0H Þ strain rate in the base metal (heat affected zone) at a reference stress r0
n power-law creep stress exponent of continuum material
h softening constant for heat affected zone
b nucleation constant for heat affected zone
N number of nucleated cavities in heat affected zone
X triaxiality factor
dn (dt) normal (tangential) separation of interface zone
Tn (Tt) normal (tangential) traction of interface zone
a elastic constant of interface zone
d_ 0 creep separation rate of interface zone at a reference traction T0
dc critical crack opening separation
dcrn normal separation of interface zone resulted from creep
x damage variable of interface zone
m power-law creep constant of interface zone

solid pipes [3,5,6]. There are three major categories of failure (Fig. 1), namely: base metal failure (Type V); heat affected zone
(HAZ) failure (Type IV); and failure at the interface between one metal and the filler metal (Type VI or VII), whereby Type VII
refers to failure along the interface, and Type VI failure initiates where the interface meets the free surface, but eventually
diverts into the HAZ or even the base metal. Similar crack path changes have also been observed in similar metal welded
(SMW) components [7,8] which have been simulated by examining creep crack growth (CCG) in SMW models containing
an initial crack at different positions [7–12]. These studies demonstrate that, the effect of mismatch in creep properties sur-
rounding the crack, applied load level, initial crack position, specimen geometric constraint and even the width of the HAZ,
etc, can all contribute to the crack path change by promoting the initiation and propagation of a secondary crack in the soft
material to the side of the initial crack [7,8]. These factors may help explain the Type VI failure mode shown in Fig. 1 for
DMW components, but studies on intact DMW components without an initial defect have not been extensively conducted.
Moreover, changes in environmental conditions such as the operating temperature and applied stress level may lead to a
change of the dominant failure mode. A comprehensive understanding of the DMW failure mechanism and when a given
mechanism dominates is yet to be achieved.
In this work we are interested in the creep failure at dissimilar welded interfaces when one metal is ferritic 9Cr-1MoVNb
(designated P91) steel or 2.25Cr-1Mo (designated P22) and the filler metal is an austenitic Inconel alloy (designated Inco).
Monotonic uniaxial creep rupture tests of different DMWs have been carried out by different research groups at 550 °C
[3,5,6]. Laha et al. [3] adopted Inco182 as the filler metal, while other researchers [5,6] employed Inco82. Fig. 2 summarizes
the results of experiments on P91/Inco182/Alloy 800 and P22/Inco182/Alloy 800 DMWs conducted by Laha et al. [3] and
P91/Inco82/SUS304 conducted by Yamazaki and Yamashita et al. [5,6]. Results of creep tests on the pure base metals of
P91 or P22 and the trend lines for all the data obtained by Laha et al. [3] are also shown in the figure. At most applied stres-
ses, DMWs are found to be weaker than the counterpart base metals. Also P22 DMWs are found to be weaker than
P91 DMWs in all these uniaxial tests. In the double logarithmic plot of Fig. 2, a transition of slope of the trend lines can
be clearly seen when the mode changes to interface failure, which dominates at relatively low stresses and long times to
failure, and has occurred predominantly at the ferritic/austenitic interface, despite the use of different Inconel fillers. Note
that the interface failure life of the P91 DMWs tested by the different investigators is observed to be similar within the tested
stress range. Further, metallurgical observations of the tested samples have revealed a sharp transition in microstructure,
physical properties and chemical composition across the interface, which leads to discontinuous mechanical properties, such
as hardness, and the generation of residual stresses [3]. Elongated precipitates and cavities have also been observed along the

Fig. 1. Typical failure modes associated with dissimilar metal welded systems.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 3

Fig. 2. Comparison of creep rupture lives of P91 (P22) base metals and ferritic/austenitic dissimilar metal welded (DMW) joints P91 (P22)/Inconel at 550 °C
[3,5,6]. The solid lines are the trend lines of the tested data in [3].

interface [3,4]. However, despite the above experimental findings, very few theoretical and numerical simulation studies have
been conducted of damage accumulation and creep rupture along the interface. In addition, steam power generating plants
have seen an extensive use of P91/Inco/P22 DMWs, i.e. P91 welded with P22 using Inconel filler metal, whereby in these sys-
tems the primary interface failure can occur at either of the ferritic/austenitic interfaces under complex multi-axial and/or cyc-
lic thermo-mechanical loading experienced in service. Detailed theoretical and computational studies are needed to develop a
more in-depth understanding of the failure mechanisms, and when a given mechanism dominates, in order for engineers to
accurately predict the failure location and assess the creep rupture life of different DMW components.
In this study, we investigate the interface failure of two ferritic/austenitic DMW systems at 550 °C, namely P91/Inco and
P22/Inco system, through an analytical and computational damage zone approach. Such an approach has proved to be robust
in capturing the development of damage, describing crack initiation and subsequent propagation in a range of nonlinear frac-
ture processes [13]. We consider the situation where the load is applied perpendicular to the interface, to simulate remote
mode I fracture (although mix-mode conditions prevail at the interface due to the mismatch in mechanical and thermal
properties). Fracture under remote mixed-mode conditions will be considered in detail elsewhere [14]. An appropriate inter-
face zone model incorporating a Kachanov-type ductility exhaustion damage accumulation law [7–9,15] is implemented in
ABAQUS to simulate and evaluate interface failure of each DMW system. Such interface zone models have been adopted
in our previous work [16] on the interface failure of two DMW components with a simple two-dimensional plate geometry,
in which the effect of the HAZ was ignored. In this work, the model is improved to include the following aspects: (1) A more
practical three-dimensional axisymmetric cylinder structure is used to simulate the geometry of actual test specimens; (2)
The softening effect of HAZ is incorporated, with different layers possessing different creep properties, as obtained from the
insight given by Laha et al. [3] and Li et al. [17]; (3) In total four detailed DMW components are considered here. We first take
the two DMW components (P91/Inco182, P22/Inco182) tested by Laha et al. [3] and recalibrate the parameters against their
available interface failure data shown in Fig. 2 for both components. These parameters are subsequently used to evaluate
another two DMW components (P91/Inco82, P22/Inco82) tested by Yamazaki and Yamashita et al. [5,6]. Parametric studies
such as sensitivity to creep properties and geometric constraint are also examined, taking the P92/Inco82 DMW system as a
typical example. Results for different DMW systems from different sources are compared and explained in terms of the rel-
ative timescales for crack initiation and propagation. The aim is to probe the intrinsic interface failure mechanism and how it
is affected by the creep properties of the surrounding materials. Finally, the relationship between the calibrated parameters
and the microstructural features is also discussed.

2. Modelling of DMW interface creep rupture

A complete DMW system contains base or parent metal(s), heat-affected zone (HAZ), filler metal (or weld metal), as well
as the interface region between the two dissimilar materials. Different parts may possess different creep properties and thus
need to be modelled in different ways [3,18]. Here we are interested in the range of loading conditions identified in Fig. 2 that
result in failure along the interface, and we choose constitutive relationships that are appropriate for these situations.

2.1. Base metal and filler metal

Both the base and filler metals are considered to be homogeneous and isotropic, with their creep behaviour described by a
simple empirical power-law constitutive model, i.e. a steady-state Norton creep law. The transition in slope observed in

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
4 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Fig. 2 indicates that over the range of stress in which interface failure dominates, the base metal would fail at much longer
times to failure, and over the time scales of interest has not yet entered the tertiary stage of creep, thus damage development
and softening of the base metal can be ignored. The complete multi-axial form of the constitutive response for the total strain
rate for the base or filler metals can be written as the sum of the elastic, creep and thermal strain rate tensors:


n1
1þt t 3 re sij
e_ ij ¼ e_ eij þ e_ crij þ e_ thij ¼ r_ ij  r_ kk dij þ e_ 0 þ aT_ temp ð1Þ
E 1þt 2 r0 r0
where re and sij are true effective stress and true deviatoric stress, E and t are Young’s modulus and Poisson’s ratio, dij is the
Kronecker delta, e_ 0 and r0 are reference creep strain rate and reference stress respectively, a is the linear coefficient of ther-
mal expansion (CTE, assumed to be isothermal), T_ temp is the rate of change of temperature and n is the steady state creep
exponent. The thermo-elastic constants for the base and filler metals at 550 °C are listed in Table 1, while the creep constants
are shown and compared in Fig. 3 in the plots of creep rates at different loads. Note that the variation of creep properties can
be seen between measurements by different researchers even on the same ferritic steel.

2.2. Heat-affected zone (HAZ)

The HAZ can be sub-divided into a coarse-grained heat affected zone (CGHAZ), fine-grained heat affected zone (FGHAZ)
and inter-critical heat affected zone (ICHAZ) [19] with increasing distance from the filler metal. Type IV failure occurs pre-
dominantly in the FGHAZ as a result of strain softening and cavity nucleation and growth under the high local triaxial stress
states arising from the local constraint imposed on the deformation of the soft FGHAZ material by the adjacent more creep
resistant sections of the HAZ [17,20–22]. Over the stress range of interest type IV failure competes with interface failure and
even under situations where interface failure dominates the crack path can deviate into the HAZ towards the end of life (Type
VI failure). The effect of strain softening and cavitation on the constitutive response of the HAZ must now be taken into
account. The thermo-elastic part of the model in Eq. (1) is the same as for the ferritic base metal, while the creep part is mod-
ified to take into account the effect of strain softening [23,24]:

n1
3 re sij
e_ crij ¼ e_ 0H exp hecr ð2Þ
2 r0 r0
where h is a softening constant and ecr is the total effective accumulated creep strain. e_ 0H is a different reference creep strain
rate for the HAZ, not to be confused with that for the base metals in Eq. (1). Eq. (2) demonstrates that the creep rate increases
with increasing creep strain. Note that the creep parameters may vary in different zones. In practice, we simply indepen-
dently model the CGHAZ and FGHAZ in this study. A simple damage accumulation model is employed for each region of
the HAZ to characterize the formation of cavities of the order of the grain size developed by Gonzalez et al. [25] based on
the experimental observations of Masuyama and Yamaguchi [26]. They observed the following sequence of processes lead-
ing to failure: initially, small sub-micron size cavities initiate at triple points or precipitates at the grain boundaries; these
cavities grow and link to completely surround a grain, isolating it from the surrounding matrix, effectively forming a large
void of the order of the grain size; these cavities experience limited growth after formation, but as more and more form they
eventually coalesce leading to failure. It is found experimentally that failure occurs when the number density of these grain-
sized cavities reaches a critical value. The model assumes that failure is nucleation controlled and that the rate of nucleation
of the large grain-sized pores is proportional to average rate of nucleation of the small pores that form around the perimeter
of a grain; these pores in turn are assumed to nucleate according to a Dyson/Greenwood [27–29] criterion in which the rate
of nucleation is proportional to the macroscopic effective strain rate and the square of the stress normal to the grain-
boundary. By projecting the stress state onto a unit sphere, Gonzalez et al. [25] obtain the following relationship for the rate
of nucleation of the grain-sized pores


4 20 2
N_ ¼ be_ cr
e þ X ð3Þ
9 9

Table 1
Thermo-elastic constants for base and HAZ materials at 550 °C.

Parameter P91 P22 Inco82/182

Base HAZ Base HAZ
E (GPa) 174 169 182
t 0.3 0.3 0.3
a 106 /°C 12.5 14.6 15.0
h 6.0 [25] 6.0
b 1500 [21,30] 1500
Nc 40,000 [21,30] 10,000 [21,30]

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 5

Fig. 3. Comparison of creep properties (creep rate at different loads) of base metals, HAZ and filler metals at 550 °C measured by Laha et al. [3],
(a) P91/Inco182, (b) P22/Inco182, and by Kawashima et al. [31] and Fukahori et al. [30], (c) P91/Inco82, (d) P22/Inco82. All the figures are plotted in the
same scale. Note that reference creep rate of CGHAZ is assumed to be identical to that of base metal in (c) and (d).

where N_ is nucleation rate, e_ cr

e is effective creep rate, b is a material constant, rm is the hydrostatic component of stress and
X ¼ 3rm =2re is the triaxiality factor. Gonzalez et al. [25] demonstrate that this model provides a good correlation of the dis-
tribution of grain-sized cavities in crept notched components of simulated P91 FGHAZ material. Igari et al. [21] and Fukahori
et al. [21,30] have adopted this model for welded components and identified a critical number of cavities Nc = 40,000 for Type
IV failure in P91 ferritic steel, based on the measurement of the average grain size in P91 FGHAZ. For P22 ferritic steel, Nc =
10,000 due to a different average grain size from that of P91. In this paper, our main focus is on interface failure, and we do
not further model the detailed influence of cavity nucleation on the creep behaviour of HAZ. We simply track the parameter
N/Nc, which is stored as a state variable, to provide an indication of damage development in the HAZ as the interface damages
and a crack propagates along the interface. The values of the thermo-elastic, softening and cavity nucleation material param-
eters in Eqs. (1)–(3) are listed in Table 1 for the different regions of the welded component. Note that for simplicity, the
thermo-elastic properties of HAZ are assumed to be the same as the corresponding base metal. Due to a lack of material
properties for P22, we use the same values of h and b as for P91. The creep properties of the different weld components
are shown and compared in Fig. 3. Note that both P91/Inco and P22/Inco considered here are creep-hard [8,11] or over-
matched [12] DMW components at moderate stress levels, where the weld metal Inconel is more creep resistant than other
components. The FGHAZ is further observed to be much softer than the CGHAZ. A variation of creep properties can be seen
between the measurements and simulations undertaken by Laha et al. [3], Kawashima et al. [31] and Fukahori et al. [30].
Laha et al. [3] conducted experiments on simulated HAZ material which had microstructures, prior austenitic grain size
and hardness equivalent to those of the HAZ of the actual welded joint [32]. While Kawashima et al. [31] and Fukahori
et al. [30] have regarded the CGHAZ to have the same creep properties as the base metal. Here we have also included the
effects of strain softening and cavitation.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
6 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

2.3. Interface region

The interface region of each DMW system is modelled as an ultra-thin layer of interface zone. Deformation and failure of
the interface can be characterized by the interface separation or opening behaviour using appropriate traction-separation
constitutive relationships. In two-dimensional problems, both tractions and separations have two components, tangential
(T t ; dt ) and normal (T n ; dn ). A detailed description of interface models of this type and how they are implemented can be
found in [13,33]. However, these models, either classified as non-potential-based or potential based models, largely adopt
rate-independent traction-separation laws that relate the normal and tangential displacements or displacement jumps (d)
to the corresponding tractions (T). Here in the current study, we employ a new phenomenological rate-dependent interface
zone model, i.e. we relate tractions to the displacement rates (or separation rates d) _ under creep loading. Further, a Kachanov
type damage variable x, is incorporated to reflect the gradual accumulation of damage at the interface zone, as motivated by
the classical continuum damage mechanics approach originally proposed by Kachanov [34]

m
aT_ n Tn
d_ n ¼ d_ en þ d_ cr
n ¼ þ d_ 0 ð4aÞ
1x T 0 ð1  xÞ

m
aT_ t Tt
d_ t ¼ d_ et þ d_ cr ¼ þ _0
d ð4bÞ
t
1x T 0 ð1  xÞ

where all the subscripts ‘‘n” and ‘‘t” refer to normal and tangential directions and all the superscripts ‘‘e” and ‘‘cr” refer to
elastic and creep contributions. a and m are material constants, where m may be different to the exponent n in the
power-law creep models of the base and HAZ materials (Eqs. (1) and (2)). d_ 0 ; T 0 are reference separation rate and reference
traction. In order to limit the number of material parameters we assume that the reference separation rate is the same in the
normal and tangential directions.
We assume that damage accumulation in the interface zone is only related to the normal creep separation dcr n , and that
failure occurs when dcr n reaches a critical separation dc. We further assume that

x ¼ dcrn =dc ð5Þ

The use of a displacement-based damage criterion of this type resembles the strain-based creep ductility exhaustion model
frequently used for creep crack growth [7–10,12,15]. This type of model parallels the conventional continuum damage
mechanics stress-based damage law proposed by Kachanov-Rabotnov [34,35] but is simpler and contains fewer fitting
parameters [15]. In addition, use of the damage variable x in both the elastic and creep parts in Eqs. (4a) and (4b) provides
a characterization of the degradation of interface stiffness and rate dependent properties as the damage gradually accumu-
lates at this interface zone. The effect of using different types of damage model with the structure of Eq. (4) on creep crack
growth is evaluated in [36]. We also explore the coupling between the normal and tangential response using a physically
based interface cavitation model [14].

3. Finite element analysis framework

3.1. Dissimilar metal welded solid cylinder (axisymmetric problem)

Here we explore how the proposed empirical Kachanov-type damage model described above can be used to capture and
simulate interface failure in DMW systems through numerical simulation using the commercial finite element code ABAQUS.
Only fracture under remote mode I loading at 550 °C at a dissimilar interface is considered in this section (note: due to mate-
rial mismatch, mixed mode conditions exist at the interface – we do not evaluate the degree of mode mixity here, but simply
examine the effect of material combinations on the global macroscopic response). Interface creep rupture data obtained by
Laha et al. [3] (Fig. 2) is used to calibrate the model. Accordingly, an axisymmetric solid cylinder model is established as
shown in Fig. 4(a) to simulate the geometry of the actual specimen, with the specimen gauge length and diameter as well
as the dimensions of FG(CG)HAZ following the experiment conducted by Laha et al. [3]. The model is comprised of two mate-
rials (ferritic steel and Inconel) with one ferritic-austenitic dissimilar interface, as simulated by a thin layer of interface ele-
ments (described in Section 3.2) inserted between the CGHAZ of the ferritic steel and Inconel. Interfaces between P91/
Inconel and between P22/Inconel are investigated independently. The applied uniaxial load direction is perpendicular to
the HAZ and interface. Boundary conditions are also shown in the figure. A mesh dependence analysis has been carried
out by Hu et al. [16], where different numbers of interface elements have been adopted, which are equally distributed along
the interface zone. The results show that the creep crack initiation time and creep crack growth rate both decrease with
decreasing mesh size along the dissimilar interface, but the sensitivity becomes negligible as the mesh size decreases to
30 mm. Therefore, 100 interface elements were used with interface elements of size 25 mm. Meshes in the HAZ and Inconel
are refined to provide the finest elements adjacent to the interface zone (Fig. 4b).

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 7

Fig. 4. Schematic of axisymmetric solid cylinder rod models of (a) P91(P22)/Inco182(82) DMW system subject to uniaxial creep loading at 550 °C. FGHAZ
and CGHAZ are created as parallel to the interface and perpendicular to the direction of the applied load. The interface is modelled as a zone filled with four-
node linear interface elements, each node having two degrees of freedom (normal and tangential). (b) Refined meshes around and towards the interface
(e.g. in CGHAZ).

3.2. Linear interface element

The interface element consists of two lines and four nodes with a linear displacement formulation (Fig. 4a). The two lines
connect the faces of adjacent bulk elements, i.e. they share the same surface and nodes. In the current computation, the order
of the nodes of an interface element is counterclockwise. Each node has two degrees of freedom (normal and tangential) and
each interface element has a total of eight nodal displacements (u) and corresponding displacement increments (Du). The
element is coded as an ABAQUS User-defined Element Library Subroutine (UEL), as described in Appendix A. At the beginning
of a simulation, the thickness of interface layer is zero, but the two surfaces of each interface element separate as the adja-
cent bulk elements deform in different rates. Transformation of local coordinates or rotation of interface elements is taken
into account in the simulation [13,37].

3.3. Computational scheme

In the simulation, for the continuum elements of the base and filler metals, we employ the in-built thermo-mechanical
model (Eq. (1)) provided by ABAQUS. For the continuum elements of the FG(CG)HAZ, while the thermo-elastic properties are
given by ABAQUS, the power-law creep model with softening (Eq. (2)) and the cavity nucleation model (Eq. (3)) are imple-
mented using a CREEP subroutine. The number ratio of cavities (N/Nc) is stored as a solution dependent state variable in the
subroutine. For the interface elements, the rate-dependent traction-separation law (Eq. (4)) and Kachanov-type damage vari-
able is implemented in the ABAQUS User-defined Element Library Subroutine (UEL). Axisymmetric quadrilateral elements
(CPX4) are used for all the continuum elements, which are compatible with the shape functions employed in the newly
developed interface elements. The integration area of each individual axisymmetric interface element is calculated by
R
2prdrdh, where r is the distance from the center of the interface element to the axisymmetric axis.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
8 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Two steps are used in the simulation, with the first one being STATIC elastic loading and heating (550 °C), and the last one
being VISCO creep deformation. Loads are selected based on the creep rupture data shown in Fig. 2. In the VISCO creep step,
the FE analysis requires global convergence in the solution of stress and strain in the continuum elements as well as local
convergence in the solution of tractions and separations in the interface elements. At the beginning of each time increment,
ABAQUS provides UEL with the current time increment (Dt), nodal displacement and displacement increment, which deter-
mines the normal and tangential separation increments (Ddn, Ddt). The normal and tangential tractions (Tn, Tt) from the last
increment are stored as state variables in the UEL, and specification of the evolution relationships oT/od are derived from Eq.
(4) (see Appendix A). Based on these, the change of normal and tangential tractions (DTn, DTt) over the increment and the
values at the end of the increment (Tn + DTn, Tt + DTt) are calculated through a Newton-Raphson iteration scheme. Details of
the Newton-Raphson scheme can be found in Appendix A.

4. Simulation of creep rupture at dissimilar interface

4.1. Calibration of creep properties of interface zone

Following the procedure described in our previous work [16], here we first calibrate the creep properties of the interface
zone based on the interface creep rupture data in Fig. 2. We divide the calibration process into two steps (analytical and
numerical). In step 1, it is assumed that the normal traction Tn along the interface is uniform and equal to the applied uni-
axial stress. The damage accumulation rate x _ is then uniform for each interface element and all the elements fail at the same
time. Considering Eqs. (4a) and (5), x_ in each interface element can be written as

m
d_ cr d_ 0 Tn
x_ ¼ n
¼  ð6Þ
dc dc T 0 ð1  xÞ
The time to failure at a given applied load can then be obtained by directly rearranging and integrating Eq. (6), noting that x
= 0 at time t = 0 and 1 when t = tf. Then
Z Z
m
m
1 tf
d_ 0 T n 1 d_ 0 T n
ð1  xÞm dx ¼  dt ! ¼  tf ð7Þ
0 0 dc T0 m þ 1 dc T0
Using a double logarithmic scale, Eq. (7) can be rewritten as



Tn 1 1 dc 1
log ¼  log t f þ  log  ð8Þ
T0 m m d_ 0 m þ 1
Assuming the reference traction lies within the applied stress range of the experimental data (e.g. T0 = 200 MPa for P91/
Inconel DMW and T0 = 100 MPa for P22/Inconel DMW), the two independent parameters, dc =d_ 0 and m, can be directly
obtained by fitting the interface failure data for each DMW system (Fig. 2). The fitted result is summarized in Table 2.
In practice the normal traction at the interface is not uniform due to elastic, thermo- and creep mismatch between the
adjacent continuum materials surrounding the interface. Accordingly, stress can be elevated and concentrated where the
interface meets the free surface. Thus damage grows preferentially and cracks initiate in these regions and propagate grad-
ually along the interface until the specimen fails. Therefore, in step 2, we simulate this process in ABAQUS and systematically
adjust dc =d_ 0 , keeping m constant, until the simulations match the experimental data. The DMW systems adopted by Laha
et al. [3] (P91/Inco182 and P22/Inco182) are used for the calibration, with the creep properties extracted from Fig. 3 (a)
and (b). Then we predict and evaluate the creep rupture life of the DMW systems adopted by Yamazaki et al. and Yamashita
et al. [5,6] using the same calibrated parameters. For simplicity, elastic constants in Eq. (4) remain the same as used in our
previous study (a = 1106 mm/MPa) [16]. d_ 0 is chosen such that the material length scale d_ 0 =e_ 0H is finite, enabling the sin-
gularity of tractions where the interface meets the free surface be eliminated, and so that the separation at the interface is
dominated by creep rather than elasticity. It is further found that, the creep rupture life depends mainly on the ratio of dc =d_ 0
rather than the detailed values of dc and d_ 0 , thus the magnitude of dc has been simply chosen to be 0.01 mm.
Fig. 5 shows the simulated and predicted total creep rupture life of each DMW system. The newly calibrated interface
zone parameters are also shown in the figure. The calibrated value of dc =d_ 0 at the end of step 2 is around 50% higher than
that computed in step 1 for the P91/Inco182 DMW system and 13% lower for the P22/Inco182 DMW system. Comparing data
from different sources it is further noticed that, the predicted creep rupture life of P91/Inco82 is close to that of P91/Inco182,

Table 2
Creep properties directly fitted from the interface creep rupture data in Fig. 2.

DMW system m dc =d_ 0 (h)

P91/Inconel 4.48 2.2  104
P22/Inconel 3.93 4.2  104

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 9

Fig. 5. Experimental and simulated (axisymmetric model) creep rupture life of each DMW system (P91(P22)/Inco182(82)) under interface failure mode at
550 °C. Hollow points are data and solid lines are model calibrated or predicted results. The fitted value of the ratio dc =d_ 0 is shown in the figure.

consistent with the tested data, while the predicted life of P22/Inco82 is almost a half that of P22/Inco182. Such a trend can
be linked to the creep properties mismatch between the two sides of the interface, as illustrated below.
Taking one tested load from each of the P91 DMWs (160 MPa) and P22 DMWs (130 MPa) as an example, the detailed
creep damage evolution along the interface of each system is quantitatively illustrated in Fig. 6. Cracks are observed to ini-
tiate from the free edge where the interface meets the free surface and gradually propagate towards the center, which is a
direct reflection of the stress concentration or enhancement of normal traction at the free edge as a result of physical prop-
erty mismatch between the materials on the two sides of the interface. Once the crack has propagated across about a quarter
of the width of the interface, the rate of propagation increases substantially and extremely small time steps are required for
the solution to converge at each increment and to capture the full details of the crack growth process up to final separation.
The overall time to failure, is however, well approximated by the time at which the simulations are allowed to terminate,
which corresponds to a time increment of 1  108h to achieve convergence. It is further noted that, most of the life is spent
initiating a crack in most systems (>95% of total life) and the remaining life for crack propagation is very short (cracks prop-
agate along a heavily damaged interface). However, an exception is observed for the P22/Inco82 DMWs (Fig. 6d), where a
crack initiates at a relatively earlier stage (at 75% of total life) and the crack propagates slower along a less heavily damaged
interface. Comparing between P91/Inco182 and P91/Inco82, the crack is found to initiate slightly later in P91/Inco82 (at
14,400 h) than in P91/Inco182 (at 13,900 h), but both systems have a similar crack propagation period. Similarities
between the two P91 DMW systems tested by different researchers can be attributed to a similar mismatch in creep prop-
erties (difference in creep rates at a given stress) between P91 CGHAZ and Inconel within the range of the applied stresses
(Fig. 3 a and c, e.g. 108/s at 160 MPa). However, it is evident that the creep mismatch between P22 CGHAZ and Inconel
measured by Kawashima et al. and Fukahori et al. [30,31] is more pronounced (e.g. 106 at 130 MPa) than that measured
by Laha et al. [3] (e.g. 108 at 130 MPa) within the range of the applied stresses (Fig. 3b and d). The relatively earlier crack
initiation (or faster damage accumulation rate where the interface meets the free surface) in P22/Inco82 DMWs is thus
caused by a higher stress (traction) concentration at the free edge as a result of the larger creep mismatch.
The much longer life for crack initiation than that for crack propagation in all the tested systems at each applied stress
seems to indicate that the crack growth assessment before a DMW component completely fails is unreasonable for practical
applications. However, care needs to be taken when interpreting the simulated creep rupture life from the engineering point
of view. While the detailed ratio of crack initiation and propagation depends on the damage accumulation and localized
stress state (Eq. (6)), one important geometric feature that can contribute to the late crack initiation is the topology of
the interface. Here we have adopted the idealization that the interface is flat as shown in Fig. 4. The stress only concentrates

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
10 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Fig. 6. Comparison of simulated damage evolution along the interface of (a) P91/Inco182 and (b) P22/Inco182 DMW systems; (c) P91/Inco82 and
(d) P22/Inco82 systems. One example stress is chosen for each DMW system.

around the free edge at the outside surface, which is also the site for crack initiation. However, the actual welded interface
would not be flat but wavy, and the surface might be irregular, in which case the stress would concentrate at a number of
positions along the interface, leading to earlier nucleation at multiple sites. Also, the crack path would become tortuous. The
time for crack propagation could then dominate the life of the component. While the actual wavy interface can be modelled
and simulated, the idealized flat interface can provide insight and fundamental understanding of the creep crack initiation
and propagation processes.
Further, crack initiation and propagation can be visualized by the evolution of axial stress (perpendicular to the interface)
at different stages of creep, as shown in Fig. 7, taking the P91/Inco82 DMW system at 160 MPa as an example. Patterns at
other applied stresses as well as other DMW systems are similar. Black crosses in these figures refer to the interface ele-
ments. Different stages of behaviour include: pure thermo-elastic loading before creep (Fig. 7a); crack initiation (Fig. 7b);
onset of fast crack propagation (Fig. 7c), and final stage of creep rupture (Fig. 7d). Stress is observed to become more and
more concentrated at the crack tip as the crack propagates.
Following the same stages in Fig. 7, the evolution of the normalized cavity density N/Nc (Eq. (3)) in the HAZ of P91/Inco82
DMW is presented in Fig. 8. For a comparison, equivalent result for the P22/Inco82 DMW is presented in Fig. 9. In both sys-
tems, the number ratio in the FGHAZ and CGHAZ increases with time as the creep strain accumulates, with the accumulation
rate much faster in the FGHAZ than the CGHAZ (Figs. 8b&c and 9b&c). This agrees with the findings that the FGHAZ is the
softest region and is the most favorable site for Type IV failure [20–22]. Note that N/Nc accumulates 1–2 orders of magni-
tude faster in the FGHAZ of P22/Inco82 compared with that in the FGHAZ of P91/Inco82, partly due to the lower critical cav-
ity density (Nc) in P22 (Table 1) and higher stress triaxiality in the softer P22 FGHAZ compared to the P91 FGHAZ (Fig. 3).
Thus P22/Inco82 is more prone to Type IV failure than P91/Inco82. Once the creep crack initiates where the interface meets
the free surface, the stress concentration leads to localized creep cavity accumulation around the crack tip in the CGHAZ

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 11

Fig. 7. Axial stress distribution around interface of P91/Inco82 system subjected to a remote applied stress of 160 MPa at different creep rupture stages: (a)
before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep rupture.

(Figs. 8c&d and 9 c&d). An important feature is noted that when the normalized cavity density exceeds unity the crack could
divert into the HAZ, i.e. indicating a transition from a Type VII to a Type IV crack path, giving an overall Type VI failure pro-
cess as discussed in the Introduction and illustrated in Fig. 1. Here we do not further explore the quantitative relationship
between Type VI failure and cavity concentration, but emphasize that taking into account the evolution of N/Nc, the diversion
of the crack path into the HAZ for both DMW systems would only occur at the very late stage of creep when the stress con-
centration is very high and takes only a minimal period that would not significantly affect the total rupture life. This is sim-
ilar to the observation in creep crack growth (CCG) studies of similar metal welds that the rupture life is mainly determined
by the initiation time of the secondary crack [8].
Finally, it is envisaged that, the simulated total creep rupture life is relatively independent of the dimension of the model
because of the predicted short period of crack propagation in these models. However, as mentioned before, the model has
employed an idealized geometry of an interface, and the geometric size effect could become more important in practical
irregular interfaces.

4.2. Effect of change of material creep properties surrounding the dissimilar interface

The comparison between P91/Inco and P22/Inco DMWs presented in Section 4.1 in terms of crack initiation and propa-
gation life highlights the significance of creep property mismatch between the two materials surrounding the dissimilar
interface (Inconel and CGHAZ of the ferritic steel). Here a number of sensitivity studies are carried out, both in terms of phys-
ical properties and geometrical constraint. Firstly, in this section, we take P91/Inco82 DMWs and the same axisymmetric
model in Fig. 4(b) as an example and further examine the effect of change of creep property mismatch on the creep rupture
life. In practice, we only change the reference creep rate (e_ 0H in Eq. (2) or e_ 0 in Eq. (1)) but do not change the nature of the
weldment configuration [8,11,12]. All other parameters remain unchanged. Four cases are conducted here:

CASE 1: increase e_ 0H by three orders of magnitude in P91 CGHAZ while keeping that of Inco82 constant;
CASE 2: decrease e_ 0H by three orders of magnitude in P91 CGHAZ while keeping that of Inco82 constant;
CASE 3: increase e_ 0 by three orders of magnitude in Inco82 while keeping that of P91 CGHAZ constant;
CASE 4: decrease e_ 0 by three orders of magnitude in Inco82 while keeping that of P91 CGHAZ constant.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
12 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Fig. 8. Evolution of the solution dependent variable N/Nc (number ratio of creep cavities Eq. (3)) in HAZ of P91/Inco82 DMW system, subjected to a remote
applied stress of 160 MPa, at different creep rupture stages: (a) before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep
rupture.

The simulation results are shown in Fig. 10. Both data measured by Laha et al. [3], Yamazaki et al. [5] and Yamashita et al.
[6] as well as the original predicted result for P91/Inco82 DMWs are presented in the figure for comparison. The total creep
rupture life is divided into crack initiation and propagation and is presented in Table 3 as ratios relative to the original values.
In general, it is observed that for all the cases the creep rupture life has not significantly changed, indicating that the total
rupture life is insensitive to the change of reference creep rates of the materials surrounding the interface within the range of
changes adopted in this study. There are however some important differences between the different sets of results. For
example, an increase in the reference strain rate (CASE 1 and CASE 3) gives rise to 20% increase in the total life while a
decrease (CASE 2 and CASE 4) results in a similar life to the original at each stress. Further, the detailed contribution to
the total life from crack initiation and propagation varies, as reflected in Table 3. The following two trends can be found
in Table 3:

(1) A longer crack initiation life than the original is found in most CASES, which still takes the dominant part in the total
rupture life compared with the propagation.
(2) As the applied stress decreases, a change of creep properties of the materials surrounding the interface tends to give a
longer propagation life than the original.

Taking the applied stress of 160 MPa as an example, the quantitative damage accumulation along the interface is pre-
sented in Fig. 11. A similar damage evolution process is found in CASE 2 (Fig. 11b) and CASE 4 (Fig. 11d) compared with
the original (Fig. 6c), with a smooth distribution of damage along the interface. The crack initiation life is subtly enhanced
in CASE 2 while that in CASE 4 remains almost unchanged. Interestingly, a different damage evolution process is found in
CASE 1 (Fig. 11a) and CASE 3 (Fig. 11c) compared with the original (Fig. 6c), with a rather uniform damage accumulation
in the interior region of the interface. CASE 1 even demonstrates a significantly sharper transition compared with other
CASES close to the free edge at the outside surface, which is observed to gradually decrease with time. Here we attempt
to explain these features in terms of two major effects that occur as a result of the change in creep properties i.e.: (a) change
in the creep mismatch between the two materials surrounding the interface as shown in Fig. 3(c); and (b) change in the
relaxation rate of the normal traction close to the free edge.
A decrease of the reference creep rate in the P91 CGHAZ (CASE 2) reduces both the relaxation rate of the normal traction
close to the free edge and the creep mismatch, with the former potentially increasing and the latter decreasing the damage

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 13

Fig. 9. Evolution of the solution dependent variable N/Nc (number ratio of creep cavities Eq. (3)) in HAZ of P22/Inco82 DMW system, subjected to a remote
applied stress of 130 MPa, at different creep rupture stages: (a) before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep
rupture.

Fig. 10. Effect of change of reference creep rate in P91 CGHAZ and Inco82 on the creep rupture life of P91/Inco82 DMW system. Both data measured by Laha
et al. [3], Yamazaki et al. [5] and Yamashita et al. [6] as well as the original predicted result of P91/Inco82 DMWs are shown in the figure for a comparison.

Table 3
Simulated crack initiation and propagation life from CASE 1 to CASE 4 (ratios are relative to the original value).

Applied stresses (MPa) Crack initiation Crack propagation

CASE 1 CASE 2 CASE 3 CASE 4 CASE 1 CASE 2 CASE 3 CASE 4
200 1.23 1.37 1.58 1.23 0.90 0.14 0.05 0.18
175 1.07 1.20 1.31 1.12 1.68 0.20 0.40 0.22
160 1.08 1.05 1.14 0.99 4.00 0.55 1.00 1.50

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
14 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Fig. 11. Comparison of damage accumulation along the interface of P91/Inco82 DMW system at 160 MPa in (a) CASE 1 (increase of reference creep rate in
P91 CGHAZ); (b) CASE 2 (decrease of reference creep rate in P91 CGHAZ); (c) CASE 3 (increase of reference creep rate in Inco82) and (d) CASE 4 (decrease of
reference creep rate in Inco82).

accumulation rate at the free edge, in terms of Eq. (6). The later crack initiation in CASE 2 (Fig. 11b) than the original is thus a
combined effect but is dominated by the reduced creep mismatch. A decrease of the reference creep rate of Inco82 (CASE 4)
gives rise to very limited change in the traction relaxation and the creep mismatch (as the P91 CGHAZ creeps faster than
Inco82 at the considered stresses). As a result, the damage evolution resembles that of the original (Fig. 11d). An increase
in the reference creep rate of the P91 CGHAZ (CASE 1) enhances both the relaxation rate of the normal traction and the creep
mismatch, which is the opposite of CASE 2. The faster damage growth along the interior region of the interface (Fig. 11a) is a
direct result of the two effects, where the normal traction at the interior region increases by the enhanced traction relaxation
around the free edge (the average normal traction must be in equilibrium with the remote applied stress). Also due to the
enhanced relaxation of the traction in the region with greater damage around the free edge, the region does not grow into
the interior, which allows the damage to grow uniformly. The even later crack initiation in CASE 1 than that in CASE 2 is
again a combined effect, but is dominated more by the enhanced traction relaxation. Finally, an increase in reference creep
rate in Inco82 (CASE 3) reduces the creep mismatch but enhances the traction relaxation rate, both of which can reduce the
damage growth rate, as reflected in the longest crack initiation life in Fig. 11(c). The damage evolution in the interior region
is faster than CASEs 2 and 4 and also the original situation, as a result of the enhanced traction relaxation (equilibrium effect
as mentioned above), but is slower than CASE 1 because of a reduced creep mismatch.

4.3. Effect of different geometric constraints (plane stress & plane strain)

In this section, we further explore how the change of geometric constraint influences the interface creep rupture life, tak-
ing P91/Inco82 DMW system as an example. Here we establish a two-dimensional plate model in ABAQUS (Fig. 12) with the

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 15

Fig. 12. Schematic of two-dimensional plate models of P91/Inco82 DMW system subject to uniaxial creep loading.

Fig. 13. Simulated (plane model) creep rupture life of P91/Inco82 DMW system. Data of P91/Inco182 tested by Laha et al. [3], Yamazaki et al. [5] and
Yamashita et al. [6] are shown, together with the original predicted result using the axisymmetric model.

gauge length and thickness of HAZ exactly the same as that of the axisymmetric solid cylinder model shown in Fig. 4(b),
similar to the structure in our previous work [16] but with the HAZ incorporated. Note that the width of the plate equals
the diameter of the solid cylinder rod. The mesh of the plate model follows the criteria evaluated before in Section 4.1. Plane
stress (CPS4) and plane strain (CPE4) elements in ABAQUS are chosen for the continuum materials for a comparison. All
parameters remain the same, except that the numerical integration area of each interface element is modified in the UEL
R R
from axisymmetric problems to plane problems, i.e. 2prdrdh ! dxdy [38].
The predicted results of total creep rupture life for the plate models at the aforementioned same stress levels are shown in
Fig. 13. Results using the data measured by Laha et al. [3], Yamazaki et al. [5] and Yamashita et al. [6] are compared with the
original predicted results for P91/Inco82 DMWs using an axisymmetric solid cylinder model. The predicted total creep rup-
ture life for both plane stress and plane strain models is found to be shorter than the axisymmetric model, but still within the
scatter of the experimental data. The plane strain model is further observed to be more creep-resistant than the plane stress
model and agrees better with the actual axisymmetric model at the applied stress levels used in these simulations. The crack
initiation and propagation life at each applied stress normalized by the life for the original axisymmetric solid cylinder model
are tabulated in Table 4. Both plane problems exhibit a majority of their rupture life spent initiating a crack, and tend to give
rise to a longer propagation life than the original as the applied stress level decreases, which are consistent with the previous
findings in the axisymmetric problem, but the ratio between crack initiation and propagation varies. At each applied stress,
plane stress is found to exhibit a shorter period of crack initiation but longer propagation life than plane strain.
Taking the applied stress 160 MPa as an example, the distribution of axial stress around the interface and the evolution of
the state variable N/Nc in the HAZ at different creep rupture stages are shown in Figs. 14 and 15 (plane stress) and Figs. 16
and 17 (plane strain), illustrating the crack initiation and propagation stages. The stress legend scales in Figs. 14 and 16 have
been unified for comparison. The main observations are listed below.

(1) Cracks initiate at both edges when the interface meets the free surface in the plane problems, which is consistent with
the axisymmetric solid cylinder problem where the crack initiates from the outer surface around the entire
circumference.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
16 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Table 4
Simulated crack initiation and propagation life of plane stress/strain problems (ratios are relative to the original value).

Applied stresses (MPa) Crack initiation Crack propagation

Plane stress Plane strain Plane stress Plane strain
200 0.90 0.99 0.90 0.90
175 0.73 0.93 1.60 1.44
160 0.56 0.85 20.0 11.0

Fig. 14. Plane stress model: Axial stress distribution around interface of P91/Inco82 DMW system subjected to a remote applied stress of 160 MPa at
different creep rupture stages: (a) before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep rupture.

(2) Asymmetric crack propagation is observed, which results from numerical perturbation in the computation. A small
perturbation may accumulate and influence subsequent computations. This would result in asymmetric evolution
of damage in the interface elements and ultimately leads to a single predominant crack propagation path.
(3) Before initiation of any crack, a stress singularity where the interface meets the free surface is similar in both plane
problems (Figs. 14a & 16a). However, the traction concentration relaxes faster in plane strain at the interface than that
in plane stress, leading to a much delayed crack initiation in plane strain. Once a crack initiates, the more constraint in
plane strain leads to higher triaxiality [8] (twice that in plane stress, not shown here) at the crack tip surrounding the
interface, thus resulting in higher stresses at the crack tip than for plane stress (Figs. 14c&d and 16c&d). As a result, the
creep crack growth rate is faster in plane strain along an already heavily damaged interface, giving rise to a much
shorter crack propagation life than plane stress.
(4) Localized cavity nucleation can be seen at the crack tip as the crack propagates towards the center. A potential for
diversion of the crack path into the HAZ due to the formation of a large number of grain-sized cavities can also be seen,
although this is less pronounced than in the axisymmetric model (Fig. 8).

These results and comparisons demonstrate that the change of geometric constraints of the problem can affect the
detailed combination of crack initiation and propagation life at the dissimilar interface, but would not change significantly
the total creep rupture life and nature of the creep rupture process.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 17

Fig. 15. Plane stress model: Evolution of the state variable N/Nc in HAZ of P91/Inco82 DMW system, subjected to a remote applied stress of 160 MPa, at
different creep rupture stages: (a) before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep rupture.

Fig. 16. Plane strain model: Axial stress distribution around interface of P91/Inco82 DMW system subjected to a remote applied stress of 160 MPa at
different creep rupture stages: (a) before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep rupture.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
18 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

Fig. 17. Plane strain model: Evolution of the state variable N/Nc in HAZ of P91/Inco82 DMW system, subjected to a remote applied stress of 160 MPa, at
different creep rupture stages: (a) before creep; (b) at crack initiation; (c) crack propagation and (d) final stage of creep rupture.

5. Discussions and conclusion

In this paper damage development along a dissimilar metal weld interface and the associated degradation of elastic-
creep properties of the interface has been described using a simple Kachanov type damage variable x, empirically defined
as the ratio between the normal creep separation and a prescribed critical separation (Eq. (5)). Many factors can influence
the growth of damage at the dissimilar interface. These include specimen geometrical constraint, and also, from the
mechanical perspective, the load level at a given temperature, as well as the effect of mismatch in creep properties of
the surrounding materials. These have been examined and discussed in Section 4 and also in the creep crack growth lit-
erature in similar metal welded components containing an initial crack at different positions [7,8,10,12]. Physically, the
separation at the interface leading up to failure is caused by the nucleation, growth and coalescence of cavities, thus x
is related to the area fraction of cavities in the plane of the interface. The development of micromechanical interface mod-
els for this process is explored elsewhere [14,39]. It is further noted that, for dissimilar metal welds, the preferential sites
of these cavities have been found to be largely provided by the elongated carbides that nucleate and grow at the interface
throughout the life [3]. Cocks and Ashby [40] and Cocks and Ponter [41] have explored the relationship between microme-
chanical and empirical Kachanov models and demonstrate that if a single process dominates the rupture process, it is pos-
sible to map from one description to the other. Under these conditions we would expect the results presented here to
provide a good description of the material response and an accurate representation of the development of damage in a
welded component. In situations where multiple processes determine the rupture life, the material description may need
to be further developed to provide a more elaborate multi-state variable micromechanical model in order to adequately
capture the appropriate constitutive response.
To conclude, in this paper, we have adopted an empirical interface zone model with a Kachanov-type ductility exhaustion
damage variable to simulate and evaluate the uniaxial creep rupture at 550 °C due to failure at dissimilar weld interfaces of
different P91/Inco and P22/Inco dissimilar metal weld (DMW) systems, as tested independently by Laha et al. [3] and Yama-
zaki et al. [5] and Yamashita et al. [6]. An axisymmetric solid cylinder model is established in ABAQUS. Creep parameters
have been calibrated against the available experimental interface failure data obtained by Laha et al. [3]. The calibrated
model has been further used in a number of sensitivity studies taking P91/Inco82 as an example, such as the effect of change
in creep property mismatch and geometric constraint. The main findings are summarized below:

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 19

(1) Similarities and differences in the damage accumulation along the interface of different DMW systems can be largely
attributed to the mismatch in creep properties of the continuum materials surrounding the dissimilar interface and
the associated traction relaxation close to the free edge. A proper design for creep mismatch between weldment con-
stituents can possibly enhance the creep rupture life.
(2) A major proportion of the creep failure life at a flat interface is taken by crack initiation. A wavy interface would be
expected to enhance the ratio between crack propagation/initiation, as it can provide a more significant enhancement
of stress where the interface meets the free surface and more positions along the irregular interface where the stress is
enhanced, promoting the nucleation of internal cracks, and also a tortuous crack path, which could slow down the
crack propagation.
(3) Diversion of the crack path into HAZ, i.e. Type VI failure mode, can be qualitatively captured by monitoring the damage
(cavity) accumulation in the HAZ. However, such a diversion would not affect the total life, since it only occurs at the
very late stage of creep rupture when the crack propagates very rapidly.
(4) A change in the geometric constraint may change the relationship between crack initiation and propagation, but does
not affect significantly the total creep rupture life and general features of the creep rupture process in the DMWs
described above.
The model presented here can be used to assess actual high-temperature DMW components at lower stress levels with
longer creep life, as the failure at lower stress levels is still dominated by the interface failure mode (Fig. 2). The structure can
be further extended to incorporate the detailed physics of damage accumulation by characterizing the evolution of carbides
and cavities at the interface within the framework of a micromechanical description.

Acknowledgement

Jianan Hu and Alan Cocks would like to acknowledge the financial support of Mitsubishi Heavy Industries (MHI).

Appendix A. Computational scheme for cohesive zone models

For the nonlinear constitutive traction-separation rate laws (Eqs. (4a) and (4b)) in cohesive zone models as described in
the context for a creep rupture problem, update of normal and tangential tractions (Tn, Tt) and damage variable (x ¼ dcr n =dc )
at each time increment is computed in UEL by Newton-Raphson iteration method. Expression of oT/od is also needed in order
to comply with the global integration scheme for adjacent bulk elements. This appendix shows the detailed update process
and how to compute oT/od from Eqs. (4a) and (4b).
For either normal or tangential direction, two unknowns (Ddcr and DT) are to be solved, thus two equations are required.
At the beginning of each time increment, the initial values of Tn, Tt, dcr cr
n ; dt are given from the last time increment, while the
total separation increments Ddn, Ddt are given from ABAQUS. The first equation for each direction can be derived from the
creep component in Eqs. (4a) and (4b)
 m  m
T n þ hDT n T t þ hDT t
Gn ¼ Ddcr
n  b n Dt ¼ 0; Gt ¼ Ddcr
t  b t Dt ¼0 0<h61 ðA1Þ
n þ hDdn Þ=dc
1  ðdcr n þ hDdn Þ=dc
cr
1  ðdcr cr

n over the time increment Dt are determined at an instant hDt. h is an integration constant. In
where the rates of Tn, Tt and dcr
the simulations presented here a value of h = 0.5 was chosen to ensure that the solution is unconditionally stable. The second
equation for each direction can be derived from the elastic component in Eqs. (4a) and (4b), noting that
1  dcr
n =dc n þ Ddn Þ=dc
1  ðdcr cr  
n Þ; T n þ DT n ¼
ðdn  dcr ðdn  dcrn Þ þ ðDdn  Ddn Þ ;
cr
Tn ¼
an an
1  dcr
n =dc n þ Ddn Þ=dc
1  ðdcr cr  
t Þ; T t þ DT t ¼
ðdt  dcr t Þ þ ðDdt  Ddt Þ
ðdt  dcr cr
Tt ¼ ðA2Þ
at at
Therefore

n þ Ddn Þ=dc
1  ðdcr cr   1  dcr n =dc
n Þ þ ðDdn  Ddn Þ 
ðdn  dcr cr
Hn ¼ ðdn  dcr
nÞ
an an
1  ðdn þ Ddn Þ=dc
cr cr
T n Dd ncr
¼ DT n  ðDdn  DdcrnÞþ ¼ 0;
an dc  dcr n

n þ Ddn Þ=dc
1  ðdcr cr
  1  dcr n =dc
t Þ þ ðDdt  Ddt Þ 
ðdt  dcr cr
Ht ¼ ðdt  dcr
t Þ
at at
n þ Ddn Þ=dc
1  ðdcr T t Ddcr
cr
¼ DT t  ðDdt  Ddcr
t Þþ
n
¼0 ðA3Þ
at dc  dcr n

For each direction, the two equations can be solved using Newton-Raphson iteration.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
20 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx

@Gix cr i @Gix @Hix cr i @Hix

Giþ1 ¼ Gix þ cr ðdDdx Þ þ ðdDT x Þi ¼ 0; Hiþ1 ¼ Hix þ cr ðdDdx Þ þ ðdDT x Þi ¼ 0 ðA4Þ
x
@ Ddx @ DT x x
@ Dd x @ DT x

The subscript ‘‘x” is ‘‘n” for normal direction and ‘‘t” for tangential direction. After i + 1 iterations within the increment the
iþ1
values of ðDdcr
x ; DT x Þ are given by
2 31 ( ) ( )
 iþ1  i @Gix @Gix  i
Ddcr Ddcr @ Ddcr @ DT x Gix Ddcr Gix
x
¼ x
4 x 5 ¼ x
 ½J x  ðA5Þ
DT x DT x @Hix @Hix Hix DT x Hix
@ Ddcr
x @ DT x

where [Jx] is the Jacobian matrix. The iteration process continues until a prescribed tolerance is reached for both Ddcr
x and DT x .
As for oT/od, it can be determined using a procedure as described below. Note that

@T x @ DT x
¼ ðA6Þ
@dx @ Ddx
Differentiating Eqs. (B1) and (B3) with respect to Ddx then gives
" @Gx @Gx
# ( @ Ddcr ) ( @ Ddcr ) " #
@ Ddcr @ DT x
x
@ Ddx
x
@ D dx
0
@Hx
x
@Hx
 @ DT x
¼ ½J x  @ DT x
¼ 1 ðA7Þ
@ Ddcr @ DT x @ Ddx @ D dx ax
x

Given the value of Jacobian matrix determined in the final iteration described above, oT/od can be then computed by Eq. (B7).

References

[1] Mvola B, Kah P, Martikainen J. Dissimilar ferrous metal welding using advanced gas metal arc welding processes. Rev Adv Mater Sci 2014;38:125–37.
[2] Clark JWG, McCartney DG, Saghafifar H, Shipway PH. Modelling chemical and microstructural evolution across dissimilar interfaces in power plants.
Proc ASME Power Conf 2014;1.
[3] Laha K, Chandravathi KS, Parameswaran P, Goyal S, Mathew MD. A comparison of creep rupture strength of ferritic/austenitic dissimilar weld joints of
different grades of Cr-Mo ferritic steels. Metall Mater Trans A – Phys Metall Mater Sci 2012;43A:1174–86.
[4] DuPont JN. Microstructural evolution and high temperature failure of ferritic to austenitic dissimilar welds. Int Mater Rev 2012;57:208–34.
[5] Yamazaki M, Watanabe T, Hongo H, Tabuchi M. Creep rupture properties of welded joints of heat resistant steels. J Power Energy Syst 2008;2:1140–9.
[6] Yamashita T, Takaya S, Nagae Y, Date S, Tomobe M. Factor analysis of interface failure mechanism for dissimilar weld joints. Jpn Atom Energy Agency
(JAEA) 2014. Internal Report.
[7] Chen G, Wang GZ, Zhang JW, Xuan FZ, Tu ST. Effects of initial crack positions and load levels on creep failure behavior in P92 steel welded joint. Eng Fail
Anal 2015;47:56–66.
[8] Chen G, Wang GZ, Xuan FZ, Tu ST. Mismatch effect in creep properties on creep crack growth behavior in welded joints. Mater Des 2014;63:600–8.
[9] Chen G, Wang GZ, Xuan FZ, Tu ST. Effects of HAZ widths on creep crack growth properties of welded joints. Weld World 2015;59:851–60.
[10] Segle P, Andersson P, Samuelson A. Numerical investigation of creep crack growth in cross-weld CT specimens Part I influence of mismatch in creep
deformation properties and notch tip location. Fatigue Fract Eng Mater 2000;23:521–31.
[11] Tu ST, Yoon KB. The influence of material mismatch an the evaluation of time-dependent fracture mechanics parameters. Eng Fract Mech
1999;64:765–80.
[12] Lee KH, Kim YJ, Yoon KB, Nikbin K, Dean D. Quantification of stress redistribution due to mismatch in creep properties in welded branch pipes. Fatigue
Fract Eng Mater 2010;33:238–51.
[13] Park K, Paulino GH. Computational implementation of the PPR potential-based cohesive model in ABAQUS: educational perspective. Eng Fract Mech
2012;93:239–62.
[14] Hu J, Elmukashfi E, Fukahori T, Igari T, Chuman Y, Cocks ACF. Effect of weld angle on the interface creep rupture life of dissimilar metal welded
components. In preparation; 2017.
[15] Wen JF, Tu ST, Gao XL, Reddy JN. Simulations of creep crack growth in 316 stainless steel using a novel creep-damage model. Eng Fract Mech
2013;98:169–84.
[16] Hu J, Fukahori T, Igari T, Chuman Y, Cocks ACF. An evaluation of creep rupture strength of ferritic/austenitic dissimilar weld interfaces using cohesive
zone modelling. Procedia Struct Integrity 2016;2:934–41.
[17] Li M, Barrett RA, Scully S, Harrison NM, Leen SB, O’Donoghue PE. Cyclic plasticity of welded P91 material for simple and complex power plant
connections. Int J Fatigue 2016;87:391–404.
[18] Kumar Y, Venugopal S, Sasikala G, Albert SK, Bhaduri AK. Study of creep crack growth in a modified 9Cr-1Mo steel weld metal and heat affected zone.
Mater Sci Eng A – Struct Mater Propert Microstruct Process 2016;655:300–9.
[19] Mannan SL, Laha K. Creep behaviour of Cr-Mo steel weldments. Trans Indian Inst Met 1996;49:303–20.
[20] Francis JA, Mazur W, Bhadeshia HKDH. Type IV cracking in ferritic power plant steels. Mater Sci Technol 2006;22:1387–95.
[21] Igari T, Fukahori T, Kawashima F, Tokiyoshi T, Chuman Y, Komai N, et al. Micro-macro creep damage simulation for welded joints. Mater High Temp
2011;28:181–7.
[22] Laha K, Chandravathi KS, Parameswaran P, Rao KBS, Mannan SL. Characterization of microstructures across the heat-affected zone of the modified 9Cr-
1 Mo weld joint to understand its role in promoting type IV cracking. Metall Mater Trans A – Phys Metall Mater Sci 2007;38A:58–68.
[23] Eggeler G, Ramteke A, Coleman M, Chew B, Peter G, Burblies A, et al. Analysis of creep in a welded P91 pressure-vessel. Int J Press Vessels Pip
1994;60:237–57.
[24] Lim R, Sauzay M, Dalle F, Tournie I, Bonnaillie P, Gourgues-Lorenzon AF. Modelling and experimental study of the tertiary creep stage of Grade 91 steel.
Int J Fract 2011;169:213–28.
[25] Gonzalez D, Cocks ACF, Fukahori T, Igari T, Chuman Y. Creep failure of a P91 simulated heat affected zone material under multiaxial state of stress. In:
Proceedings of 3rd international ECCC conference on creep & fracture; 2014.
[26] Masuyama F, Yamaguchi T. Creep damage initiation mechanism and life of martensitic structures. In: Proceedings of 51st symposium on high-
temperature strength of materials; 2013. p. 110–4.
[27] Dyson BF. Continuous cavity nucleation and creep fracture. Scr Metall 1983;17:31–7.
[28] Kelly DA. Creep cavitation failure in copper subject to biaxial stress systems at 250 degreesc. Acta Metall 1975;23:1267–75.
[29] Vandergiessen E, Tvergaard V. Interaction of cavitating grain-boundary facets in creeping polycrystals. Mech Mater 1994;17:47–69.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001
 J.-n. Hu et al. / Engineering Fracture Mechanics xxx (2018) xxx–xxx 21

[30] Fukahori T, Tokiyoshi T, Igari T, Chuman Y, Komai N. Prediction of Type IV creep failure of a seam-welded mod. 9Cr-1Mo elbow based on microscopic
damage simulation. Mater High Temp 2017;34:194–207.
[31] Kawashima F, Igari T, Tokiyoshi T, Shiibashi A, Tada N. Micro-macro combined simulation of the damage progress in low-alloy steel welds subject to
type IV creep failure. JSME Int J Series A – Solid Mech Mater Eng 2004;47:410–8.
[32] Chandravathi KS, Laha K, Rao KBS, Mannan SL. Microstructure and tensile properties of modified 9Cr-1Mo steel (grade 91). Mater Sci Technol
2001;17:559–65.
[33] Park K, Paulino GH. Cohesive zone models a critical review of traction-separation relationships across fracture surfaces. Appl Mech Rev 2011;64.
[34] Kachanov LM. Rupture time under creep conditions. Int J Fract. 1999;97:Xi–Xviii.
[35] Rabotnov YN. Creep problems in structural members. Amsterdam: North-Holland; 1969.
[36] Elmukashfi E, Cocks ACF. A theoretical and computational framework for studying creep crack growth. Int J Fracture. 2017 [IUTAM Baltimore].
[37] Yu CH, Huang CW, Chen CS, Gao YF, Hsueh CH. Effects of grain boundary heterogeneities on creep fracture studied by rate-dependent cohesive model.
Eng Fract Mech 2012;93:48–64.
[38] Chester SA, Di Leo CV, Anand L. A finite element implementation of a coupled diffusion-deformation theory for elastomeric gels. Int J Solids Struct
2015;52:1–18.
[39] Yalcinkaya T, Cocks ACF. Micromechanically motivated cohesive zone formulation for ductile fracture. ENUMATH 2015.
[40] Cocks ACF, Ashby MF. Creep fracture by coupled power-law creep and diffusion under multiaxial stress. Metal Sci 1982;16:465–74.
[41] Cocks ACF, Ponter ARS. Creep deformation and failure under cyclic thermal loading. Nucl Eng Des 1989;116:363–87.

Please cite this article in press as: Hu J-n et al. Modelling of creep rupture of ferritic/austenitic dissimilar weld interfaces under mode I
fracture. Engng Fract Mech (2018), https://doi.org/10.1016/j.engfracmech.2018.01.001