Nuclear Engineering and Design 246 (2012) 220–232

Contents lists available at SciVerse ScienceDirect

Nuclear Engineering and Design

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

Creep behaviour and failure modelling of modified 9Cr1Mo steel

T. Massé ∗ , Y. Lejeail
CEA, DEN, DER/SESI/LE2S, F-13108 Saint-Paul-Les-Durance, France

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

Article history: A finite element modelling (FEM) software is used to model the creep behaviour and damage mechanisms
Received 24 October 2011 of modified 9Cr1Mo steel, CAsT3M 2008® , between 450 ◦ C and 650 ◦ C. Experimental creep results, col-
Received in revised form 31 January 2012 lected from a large literature review, are used to identify parameters for a strain rate viscoplastic model
Accepted 6 February 2012
that includes the effect of damage. Numerical calculations show good agreement with experimental
data over the whole range of temperature considered. In particular, the drop off in time to rupture in
the creep rupture curves is well predicted, and the change of failure mechanism from viscoplasticity-
assisted ductile rupture to diffusion-assisted creep cavitation damage is highlighted. The identification
of temperature-dependent hardening coefficients means the modelled strain results are also in a good
agreement with experiments whatever the applied stress. The model is seen to have some limitations,
but improvements are proposed to account for strain localizations and their geometrical effects.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction Because of the time required and cost of an experimental campaign

of long term creep tests in this work an alternative method has been
9%Cr steels are extensively used in industrial power plants used.
because of their good mechanical and thermal properties. They are An FEM approach, based on continuum mechanics, is adopted
potential candidates for structural material in the primary circuit in this study. A viscoplastic model that includes damage effects has
of the Very High Temperature Reactor (VHTR) and in the Steam been chosen from literature. This model was first developed for
Generator of the Sodium Fast Reactor (SFR). Compared to 316LN low content chromium steel base metal and weldments by Hall and
austenitic stainless steel they have a higher thermal conductivity Hayhurst (1991), Hayhurst et al. (2005a,b) and Perrin and Hayhurst
and a lower thermal expansion coefficient. One could thus expect (1996) and has been recently applied to P92 grade by Pétry and
weight gain by reducing the thickness of the structures, leading to Lindet (2009) (a 9%Cr steel grade very close to P91 steel). This work
lower prices for the high temperature components. proposes a set of parameters for this model, identified from experi-
In this context, the understanding of damage mechanisms and mental data taken between 450 and 650 ◦ C, and then simulates the
creep behaviour of 9%Cr steels has to be improved, especially at creep and tension behaviour of modified 9Cr1Mo steel.
the weld junctions. This paper focuses on the modified 9Cr1Mo This paper is organised as follows. Firstly, the material and
steel grade (also called T/P91 or X10CrMoVNb9-1 according to the experimental creep and tension results are detailed before the
European Standard EN 10216-2 (Technical Committee ECISS/TC29, model and the fracture criterion are introduced. Validation of the
2006)). The commercial P91 steel grade is one of the best known model implementation in the FEM code, using the P92 steel data,
9%Cr steels and has been already widely studied, and a large creep is briefly presented. A new set of parameters for P91 steel are then
and tension experimental database is available for a wide temper- identified using the method developed in Pétry and Lindet (2009),
ature range (Gaffard et al., 2005; Haney et al., 2009; Kimura, 2006; which is slightly modified here for better correlation with experi-
Kimura et al., 2007; Masuyama, 2006; Raj et al., 2004; Tabuchi and mental results. Numerical results of the creep and tension tests are
Takahashi, 2006; Tabuchi et al., 2009; Vivier, 2009; Watanabe et al., presented before, finally, numerical aspects of creep modelling are
2006). Most of the experimental results concern short term creep discussed to better understand the limits of the model and ideas
tests performed at high stress level. However, in case of nuclear for possible improvements are proposed.
reactor industrial components, long term creep results are required.

2. Experimental tests

∗ Corresponding author at: Laboratoire d’Étude et de Simulation des Systèmes,

2.1. Material
CEA, DEN, DER/SESI/LE2S, Bât. 212-P27, F-13108 Saintt Paul-Les-Durance, France.
Tel.: +33 4 42 25 43 46; fax: +33 4 42 25 71 87.
E-mail addresses: [email protected] (T. Massé), The X10CrMoVNb9-1 steel grade has a chemical composition
[email protected] (Y. Lejeail). range given by the European Standard EN 10216-2 (Technical

0029-5493/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.nucengdes.2012.02.006
 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232 221

Table 1
Specification of chemical composition of the modified 9Cr1Mo steel (wt.%).

C Mn P S Si Ni Cr Mo Nb V Al N

Min 0.08 0.3 0 0 0.2 0 8 0.85 0.06 0.18 0 0.03

Max 0.12 0.6 0.02 0.01 0.5 0.4 9.5 1.05 0.1 0.25 0.04 0.07

Committee ECISS/TC29, 2006) and is detailed in Table 1. The mean

Cr content of 9% gives good resistance to corrosion and lowers
the temperature of martensite formation. Mo contributes to solid
solution hardening. Nb and V enable fine MX precipitation (where
M = Nb or V and X = C and/or N) during tempering, which is con-
ducive to strengthening by pinning free dislocations in the matrix
as demonstrated by Czyrska-Filemonowicz et al. (2006), and they
also restrict the growth of austenitic grains during austenitization.
See Maruyama et al. (2001) for a complete explanation about the
role of each element.

2.2. Creep tests

A total of 224 creep tests, between 450 ◦ C and 650 ◦ C, have been
identified: 63 from CEA experimental campaign, and 161 from liter-
Fig. 2. Creep curves at 600 ◦ C.
ature. All results are presented in Fig. 1, which shows a comparison
between stress/time-to-rupture data from the literature (unfilled
symbols) and from joint EDF/AREVA/CEA experiments (filled sym- agreement with the results from the literature for the temperature
bols). range 450–650 ◦ C. The influence of temperature and stress is easily
In case of the tripartite collaboration: observable, from Fig. 1, particularly at high temperatures where a
steeper gradient is seen at longer rupture times.
• 19 creep tests have been performed by Gaffard (2009) (identified Creep curves from CEA experimental campaign at 600 ◦ C are
VG in Fig. 1) plotted in Fig. 2. We observe usual features of the creep curves as
• In the HTR-M and RAPHAEL European project by Hegeman (2009), also described by Gaffard et al. (2005) and Orlová et al. (1998) such
(8) creep tests have been done. They are identified as JRC in Fig. 1 as:
• 36 creep tests have been completed by CEA (Haney et al., 2009;
Tavassoli et al., 1992). 1. A short primary creep corresponding to a decrease of the creep
strain rate (Fig. 3)
And from the literature: 2. A secondary creep only present for low stresses with a charac-
teristic steady strain rate (Fig. 3)
• 16 creep tests from Watanabe et al. (2006) at 550–600–650 ◦ C 3. A relatively long tertiary creep where the strain rate increases
• 9 creep tests from Tabuchi et al. (2009) at 550–600–650 ◦ C typically over 75% of strain (Fig. 3).
• 14 creep tests from Raj et al. (2004) at 550–600–650 ◦ C
• 12 creep tests from Masuyama (2006) at 550–600–650 ◦ C Additionally, Fig. 3 shows that, at 600 ◦ C, the minimum creep
• 22 creep tests from Kimura (2006) between 500 and 700 ◦ C, 88 strain rate ε̇s is reached around a constant strain value of 3%.
creep tests in Kimura (2006).
3. Modelling
Both data sets lie within the same scatter band, and so the creep
lifetimes from the tripartite cooperation are considered to be in Recently, a continuum mechanics based model – first devel-
oped to study low %Cr steels creep behaviour (Hayhurst et al., 2004,
2005a; Mustata and Hayhurst, 2005; Perrin and Hayhurst, 1996),

Fig. 1. Stress vs. time-to-rupture: experimental results between 450 and 650 ◦ C. Fig. 3. Creep strain rates at 600 ◦ C.
222 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232

and which accounts for three-dimensional creep deformation and

damage mechanisms – has been modified and successively applied
to 9%Cr steels (P92) by Pétry and Lindet (2009). In this paper,
the identification of a new set of model parameters for modified
9Cr1Mo steel is described. Numerical results from this model are
compared to the experimental data previously presented.

3.1. Model equations

Hayhurst et al. (2005a,b) developed a creep strain rate model

that includes the effect of hardening; this model was also able to
represent the precipitation coarsening phenomenon as well as the
evolution of intergranular damage. Indeed, in Hayhurst’s model,
the state variable H only represents the creep primary phase, i.e.
the increase of the hardening until the maximum H* which also
corresponds to the minimum steady-state creep strain rate. The
damage is decomposed in two state variables: one for the decrease
of hardening (the loss of the creep resistance) which represents the
coarsening of the precipitates. The second variable represents the
inter-granular cavitation damage. Pétry and Lindet (2009) slightly
modified the internal variables of the model and only two state
variables are used. The increase and the decrease of the hard-
ening are included in the state variable H with H1 and H2 (Eqs.
(3.11)–(3.13)). The second state variable deals with the descrip-
tion of damage which is introduced thanks to a phenomenological
approach. This method is derived from the pioneering works of
Kachanov (1958) on the continuum damage theory, which describe
the progressive degradation of metallic materials during creep by
using a power law. In the modified model of Pétry and Lindet
(2009), the power law has been replaced by a hyperbolic sinusoidal
function in order to better describe the change of exponent which
Fig. 4. (a) Damage characterization of creep specimens tested under vacuum and
usually occurs during the change of failure mechanisms of 9%Cr
(b) ductile failure with dimples (scanning electron microscope (SEM) picture with
steels. This empiricism also enables to avoid the use of advanced back scattered electrons) at 625 ◦ C and 120 MPa of modified 9Cr1Mo steel.
optimization techniques in the identification process. From Gaffard et al. (2005)
Detailed equations of the creep model are described in Pétry and
Lindet (2009) and are re-written below with further explanation.
The total strain is decomposed into an elastic and a viscoplastic The coefficient K also depends on temperature:
part:
K = ak T + bk (3.10)
ε̄ = ε̄e + ε̄vp (3.1)
Hardening kinetic is divided in two parts: H1 for the increase of
¯ = C · ε̄e (3.2) hardening and H2 representing the softening for high strains:

where C is the Hooke’s matrix and Eq. (3.2) represents the elastic H = H1 + H2 (3.11)
behaviour. with:
The viscoplastic strain rate tensor is normal to the yield surface:
h1
ε̄˙ vp = ṗn̄ (3.3) Ḣ1 = (H ∗ − H1 )ṗ (3.12)
J 2 () 1
3 s̄ h2
with : n̄ = the normal to the yield surface (3.4) Ḣ2 = ṗ (3.13)
2 J2 () J 2 ()
1
s̄ = ¯ − 3
tr()
¯ Ī the stress deviator (3.5) In Pétry and Lindet (2009), h1 , H1∗ and h2 are considered constant
 whatever the considered temperature for reasons of simplicity.
3 However, our later parameter identification (cf. Section 4.1) shows
J2 () = s̄ : s̄ the equivalent von Mises’ stress (3.6)
2 a quasi-linear evolution of these three parameters with tempera-
ṗ = ε̇0 sinh(fv ) the viscoplastic strain rate module (3.7) ture. Thus, h1 , H1∗ and h2 are written as follows:
⎧
In this last Eq. (3.7), the thermal-activated aspect of the vis- ⎨ h1 = ah1 T + bh1
coplasticity is taken into account by mean of an Arrhenius law in H ∗ = aH ∗ T + bH ∗ (3.14)
the ε̇0 coefficient: ⎩ h 1 = a 1T + b 1
2 h2 h2

ε̇0 = fε e−Qε /RT (3.8)

The damage evolution kinetic is a linear combination of the max-
f is decomposed into two parts: one for the hardening variable imal principle stress  1 and J2 (), associated with the triaxiality
H and another for intergranular creep damage D: parameter ˛ (Hayhurst, 1972):

J2 ()(1 − H)
 ˛ + (1 − ˛)J () 
1 2
fv = (3.9) Ḋ = A0 sinh (3.15)
K(1 − D) 0
 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232 223

Fig. 5. Electron Back Scattered Diffraction (EBSD) map near the rupture area of (a) a creep test at 90 MPa (tR = 7000 h) illustrating a diffusion-assisted creep cavitation damage
and (b) EBSD map of a creep test at 120 MPa (tR = 900 h) illustrating a viscoplasticity-assisted ductile damage; black lines delineate high angle boundaries (>15◦ ), grey or red
lines delineate low angle boundaries (5–15◦ ), cavities are in black colour. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of the article.)
From Gaffard et al. (2005).

Temperature variations are also represented by Arrhenius laws, phenomenon, is triggered by lath martensite microstructure recov-
since creep is a thermally activated phenomenon: ery as illustrated in Fig. 5b. The considered time-to-rupture for
this mechanism has been identified from available experimental
A0 = fA e−QA /RT (3.16) creep curves and creep times to failure. Because of numerical con-
0 = f0 e−Q0 /RT (3.17) vergence time, calculations cannot be performed to a creep strain
higher than 12%. If creep time at 10% of creep strain is considered,
the remaining life is always lower than 1% of the elapsed creep time.
3.2. Failure criterion
Therefore, the time at a 10% strain was taken as the time-to-rupture
tR . It was also verified numerically that the remaining service life
It is widely acknowledged in the scientific community that
could be estimated at less than 1% of the elapsed time
two temperature- and stress-dependent failure mechanisms are
For low stress levels, the main failure mechanism is diffusion-
observed during 9%Cr steel creep. Indeed, at a given temperature, a
assisted creep cavitation damage. This damage starts with void
change in the mechanism which governs failure takes place, accord-
nucleation at grain boundaries, then with void growth assisted by
ing to the applied creep stress level, as highlighted (Cocks and
diffusion along the boundaries of small equiaxed, highly disori-
Ashby, 1982; Eggeler, 1989; Frost and Ashby, 1982) and leads to
ented, grains – as emphasized by the EBSD map in Fig. 5a. Finally,
a fall in time-to-rupture (Gaffard et al., 2005; Kimura et al., 2007;
void coalescence leads to micro and macro cracks. Since diffu-
Kimura, 2006). Formal treatment of constitutive models for mixed
sion phenomenon is thermally activated, it has been observed that
high–low stress levels has been done by Hayhurst et al. (2008).
the higher the temperature, the earlier the appearance of creep
However, a dual criterion, developed by Pétry and Lindet (2009)
damage. The considered time-to-rupture for this mechanism is the
and which enables to differentiate the two failure mechanisms at
damage threshold Dc corresponding to the local void coalescence
high and low stresses, has been applied here.
and microcrack initiation. It is also assumed that D is zero at the
The principle mechanism at high stress levels is the
initial state.
viscoplasticity-assisted ductile rupture mechanism. The mech-
In this model, Pétry and Lindet (2009) applied the weakest link
anism begins with void nucleation at the preferential stress
approach, leading to a dual criterion that takes into account both
concentration areas of grain boundaries, such as second phase par-
mechanisms:
ticles or triple junctions. Then void growth is assisted by grain
deformation and followed by coalescence of small intergranu-
lar cavities (Fig. 4a). The ductile fracture (illustrated in Fig. 4b 
tR = Min t(ε = εc = 10%), t(D = Dc ) (3.18)
with low void volume fraction specimen), coming from necking
224 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232

Fig. 7. Cast3m numerical results (dashed lines), numerical (lines) and experimental
results (symbols) from Pétry and Gariboldi (2009) of P92 tensile test at 600 ◦ C.
Fig. 6. Numerical (lines) and experimental (symbols, extracted from Pétry and
Lindet, 2009) time to rupture for the P92 steel. The red dashed line represents the
limit between the area of diffusion-assisted creep cavitation damage (low stresses 4.1. Hardening parameters h1 , H1∗ and h2
and high temperature) and the one corresponding to viscoplasticity-assisted ductile
damage (high stresses and low temperature). (For interpretation of the references to
As a first order approximation, the stress can be considered as
colour in this figure legend, the reader is referred to the web version of the article.)
constant during a creep test. The integration of Eqs. (3.11)–(3.13)
with  constant, neglecting elasticity and considering H1 (t = 0) = H2
(t = 0) = 0 can be rewritten:
with Dc typically taken between 0.1 and 0.3, as reported by
H = H1 + H2 (4.1)
Tveergaard (1984) and Van der Giessen et al. (1995). The exact value
of Dc will be identified in Section 3.3. with:
  h 
H1 = H1∗ 1 − exp −
1
ε (4.2)

3.3. Implementation and validation on P92 steel grade
h2
H2 = ε (4.3)
The creep behaviour model has been implemented into 
Cast3m2008® . The numerical method used to solve the differen- Furthermore, without damage and neglecting elasticity, Eq.
tial equations is the second order Runge–Kutta technique which (3.11) can also be reformulated as follows:
gives good accuracy and stability. This implementation has been
K
 ε̇ 
validated by comparing the results on the P92 steel to data taken H =1− a sinh (4.4)
from Pétry and Lindet (2009). In terms of time-to-rupture, a maxi-  ε̇0
mal deviation of 4.5% is observed, as shown in Table 2. A comparison At any time, pseudo-experimental values of H of Eq. (4.4) can
with experimental P92 creep rupture times is presented in Fig. 6, then be easily determined, since K and ε0 are already known by the
where good agreement is observed. Fig. 6 also plots the limit curve previous identification step of viscosity parameters (see Pétry and
(red dashed line) which divides the chart into two sub-areas which Gariboldi (2009) for more details),  is also known and ε can be cal-
correspond to the two different fracture mechanisms. This numeri- culated as the instantaneous gradient of the creep curve. Using Eqs.
cal limit curve is identified by the determination of particular points (4.1)–(4.3), the best hardening parameters can be identified by fit-
x(t (h);  (MPa)) at every temperature where both failure criteria ting to experimental values of H. In Pétry and Gariboldi (2009), the
are fulfilled: it corresponds to the intersection of the creep time to authors averaged each hardening parameter over the temperature
failure (obtained with the criterion ε = εc ) – stress curve with the range. However in our case, a quasi-linear evolution of H, depend-
same curve but considering the creep time obtained with the cri- ent on temperature, has been observed, as shown in Fig. 8. For this
terion D = Dc . The resulting limit red curve is nearly the same as the reason a linear regression based on temperature is applied for each
one given by Pétry and Gariboldi (2009). hardening parameter as written in Eq. (3.14).
Pure tension tests of P92 at 600 ◦ C for different strain rates have
been also simulated using Cast3m (dashed lines) and are seen to be 4.2. Triaxiality influence with ˛ parameter
very close to those obtained by Pétry and Gariboldi (2009) as shown
in Fig. 7, but the correlation with experiments is not accurate for This section of the work takes advantage of available creep tests
the lower strain rates. This point will be discussed in Section 4.3, on notched specimens (NC) with different notched radii (Hayhurst
where constant strain rate tension tests have also been performed and Henderson, 1977; Hayhurst et al., 1984). This enables one to
on P91 steels. investigate the influence of the stress triaxiality ratio via the ˛
parameter and its effect on the maximum principal stress  1 and on
the von Mises equivalent stress J2 () (Hayhurst, 1972) – defined by
4. Identification procedure the peak rupture stress (the numerator term in Eq. (3.15)). The coef-
ficient ˛ has been taken equal to 0.43, which is the value identified
A specific procedure for the identification of parameters has by Gaffard (2009) for T/P91 thanks to experimental creep tests on
been developed, based on a progressive interpolation of variables NC specimens. Experimental creep curves and creep time to failure
over a large temperature range as detailed in Pétry and Gariboldi are respectively presented in Figs. 11 and 12.
(2009). The same procedure is adopted here for modified 9Cr1Mo The mesh and boundary conditions of the round V-notched half-
steel. The modifications in the model that have been presented specimen (NC0.25) are presented in Figs. 9 and 10. This specimen
earlier (see Eq. (3.14)) are mainly detailed in the next sections. has a maximum diameter of 23 mm, a minimum diameter of 14 mm
 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232 225

Table 2
Comparison of P92 creep time-to-rupture between Cast3m simulations and numerical results of Pétry and Gariboldi (2009).

T(◦ C)  R (MPa) tR (h) from Pétry and Lindet (2009) tR (h) from Cast3m Deviation (%)

550 200 31622 31482 0.44

600 100 146780 148344 1.07
625 100 25645 25363 1.10
650 100 4410 4308 2.31
675 100 631 636 0.79
750 60 90 94 4.51

and a 45◦ V-notch with a radius of 0.25 mm. Elongation of the spec- numerical model does not correctly predict the notch opening dur-
imens was measured between points located on both sides of the ing the tertiary creep, in both cases, i.e. with and without ˛. The
notched area with a gauge length of 6 mm. use of the ˛ parameter in the description of the creep behaviour
The results of numerical creep tests on notched specimens at has nearly no influence on the creep curves. However, the model
625 ◦ C with and without ˛ are compared with experimental data is able to correctly predict the time to rupture (after applying the
(extracted from Gaffard, 2009) in terms of creep curves and time- dual criterion), if the parameter ˛ is used, as shown by the results
to-rupture measurements in Figs. 11 and 12. The figures show the with and without ˛ in Fig. 12 (note that SC means smooth creep

Fig. 8. Identification of hardening parameter (a) H1∗ (b) h1 and (c) h2 according to
the temperature. Fig. 9. Mesh and boundary conditions of a notched creep half-specimen.
226 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232

Fig. 10. Detailed view of the meshed notch.

Fig. 13. Comparison between model predictions and experimental observations

from Gaffard (2009) of the location of the most damaged area at rupture.

The effect of a notch on the stress, as already studied by Al-

Faddagh et al. (1982), is also highlighted by Fig. 12 which compares
the time to rupture of smooth creep (SC) specimens (see Fig. 23
for the geometry of a SC specimen) and notched specimens. As a
result of the notch geometry, the mean stress triaxiality ratio is
increased, which is why for the same equivalent strain rate, the
notched specimen can sustain a higher nominal stress.
Finally, the model’s prediction of the area of most damaged is
compared to the observed locations in Fig. 13 for the NC0.25 spec-
Fig. 11. Experimental (symbols) creep curves (from Gaffard, 2009) and numerical
imen. Both show the occurrence of damage near the notch root,
(lines) creep curves with and without ˛ at 625 ◦ C.
which results from the high severity of the notch. The numeri-
cal modelling also displays a slight translation of the damage area
towards the specimen axis with the increase of the stress.
specimens and corresponds to the results of the CEA experimen-
tal campaign at 625 ◦ C reported in Fig. 1). This may come from the 4.3. Damage parameters
introduction of ˛ only in the damage kinetic equation (Eq. (3.15))
and not directly on the hardening. As described in Section 2.2, creep damage only influences the
While at high stresses the time to rupture is the same with or failure mechanisms of long-term creep tests, i.e. at low stresses.
without the use of ˛ in the model, at low stresses ˛ is essential For the modified 9Cr1Mo steel, we assume that damage parameters
to predict the correct time-to-rupture. This difference is because described in Eqs. (3.16) and (3.17) are the same as those identified
at high stresses the damage mechanism is viscoplasticity-assisted by Pétry and Lindet (2009). This choice is guided by the very similar
ductile rupture, and the time to rupture is determined by the chemical composition of P91 and P92, as well as their macroscopic
strain in the failure dual criterion and thus the parameter ˛ has behaviour. Indeed, even if P91 steel is supposed to be more duc-
no influence. For low stresses, the major damage mechanism is tile, the creep time to failure is lower for P91 compared to P92,
diffusion-assisted cavitation damage and thus damage itself is the as depicted by Fig. 14, but the total evolution seems very close
criterion for the time to rupture, and where ˛ has an influence via
the peak rupture stress defined in Eq. (3.15).

Fig. 14. Comparison of experimental creep time to failure between P91 (dashed
Fig. 12. Experimental creep time-to-rupture for NC and SC (filled symbols, from line: logarithmic interpolation of experimental results presented in Fig. 1) and P92
Gaffard, 2009) and numerical (unfilled symbols) specimen. (symbol) steel.
 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232 227

Fig. 15. Experimental creep time to failure (symbol) compared to numerical creep
time to failure without damage (dashed line) of P91 steel.

on the whole range of stresses and temperatures. Moreover, when

comparing numerical prediction of failure times without damage
(A0 = 0) with experimental results, shown in Fig. 15, only few creep
tests take place in the cavitation creep damage area for temper-
atures above 550 ◦ C, compared to the Fig. 7 in Pétry and Lindet
(2009) where a high number of creep tests are involved in the Fig. 17. Boundary conditions of the REV.
cavitation creep damage, which makes the identification of P91
complex. Therefore, in a first approximation, damage parameters
A0 and  0 have been taken as the same as in Pétry and Lindet
(2009). In case of inaccurate final results, in terms of creep time
for 9–10%Cr steels (Kimura et al., 2007; Kloc and Sklenička, 1997,
to failure in the cavitation creep damage area, adjustments will be
2004; Sklenička et al., 2003) and characterizes the change of dam-
performed. In the model parameter identification procedure, dam-
age and failure mechanism. In order to get Dc , the boundary stress
age is neglected at high stresses, but is taken as significant with both
 lim corresponding to the intersection of each of the high and low
decreasing stress and increasing temperature. This leaves the crit-
stress curves is identified from Fig. 16, then creep test simulations
ical damage coefficient Dc defined in Eq. (3.18), to be determined.
of these two cases are performed. In the simulations the value
This coefficient represents the change in the stress vs. time curve
of Dc is adjusted in order to achieve tR (ε = εC = 10%) = tR (D = Dc ) at
for low stresses and high temperature; although, unfortunately, no
 =  lim , and returns a mean value of the critical damage coefficient
long-term experimental data are available to confirm this effect.
of Dc = 0.132.
The minimal creep strain rate ε however, may used to identify Dc .
The evolution of εs for different applied stresses and two differ-
ent temperatures (600 ◦ C and 650 ◦ C) is plotted in Fig. 16, which
highlights the different gradient at high and low stresses and also 4.4. Summary
the dependence on temperature. This change in gradient, with an
associated change in the stress exponent of the power function All of the modelling parameters identified previously are
(instead of the hyperbolic sinusoidal function in the formulation of reported in Table 3.
the damage kinetic in Eq. (3.15)), has been already demonstrated

5. Numerical simulation of the modified 9Cr1Mo steel

behaviour

5.1. Mesh and boundary conditions

In order to reduce computation time, calculations have been

performed on a Representative Elementary Volume (REV). We also
take advantage of the cylindrical geometry of the creep specimens,
and an axisymmetric model can be used (cylindrical coordinates
r, , z). Quadrangular elements at eight nodes with two degrees of
freedom are used to mesh the REV. In the axisymmetric calculation,
the nodal unknowns are the two displacements UR and UZ . The
loading is applied under constant force assumption as in a real creep
test, which is numerically represented by a negative pressure well
distributed on the upper face of the REV, as depicted in Fig. 17.
The simulation proceeds with the progressive application of a load
which is then maintained at a constant level. A large displacements
option is used and results in a geometry reactualization at every
Fig. 16. Identification of the boundary stresses separating the two failure mecha-
nisms for 600 and 650 ◦ C.
time step.
228 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232

Table 3
Identified parameters on the base metal of the modified 9Cr1Mo steel.

 fε (/h) Qε (J) ˛k (MPa/K) bk (MPa) ˛H1 (MPa/K) ˇH1 (MPa) ˛H ∗ (/K) ˇH ∗

1 1

0.3 5.20E + 07 3.07E + 05 −4.45E − 02 45.4 −165 1.78E + 05 1.20E − 03 −0.693

˛H2 (MPa) ˇH2 (MPa) fA (/h) QA (J) f0 (MPa) Q0 (J) R (J/mol/K) ˛

0.374 −399 9.00E + 04 2.00E + 05 3.50E − 02 −4.85E + 04 8.31 0.43

method already described in Section 2.3. The resulting boundary

curve between the two damage mechanisms described in 2.2, is
plotted in Fig. 18, and which highlights the small number of exper-
imental tests lying in the creep damage domain. It is clear that creep
damage is responsible for the sharp drop in time-to-rupture and as
already discussed in Sections 2.2 and 3.3, this drop is due to the dif-
ferent damage and failure mechanisms for high and low stresses.
The results from the numerical model show that the drop in time
to rupture happens earlier at high temperature as the change from
ductile damage to creep damage also takes place earlier (this is
due to the diffusion mechanism which assists creep damage and is
thermally activated).
To further validate the Cast3m model, creep simulations were
performed at 550 ◦ C and 600 ◦ C for different stress levels and
Fig. 18. Simulated (lines) and experimental (symbols) creep time-to-rupture for the were compared to experimental results of creep strain. The use
modified 9Cr1Mo steel. of temperature-dependent hardening coefficients (see Section 3.1)
means the model response is globally in accordance with experi-
5.2. Creep tests results mental results whatever the applied stress level, as illustrated in
Fig. 19.
The creep times to failure from the simulations are plotted
in Fig. 18 for a range of temperatures (experimental results are 5.3. Constant strain rate tension tests results
depicted by symbols and numerical results by lines). Good agree-
ment with the experimental data is observed for short lifetimes, For the sake of completeness, pure tension tests at 450 ◦ C, 550 ◦ C
and the numerical results also predict a sharp drop of the time and 625 ◦ C were also performed with at least three different dis-
to rupture with a decrease of stress and an increase of temper- placement rates applied to the upper face of the specimens. The
ature, corresponding to long-term creep time, which cannot be resulting engineering strain–stress curves are plotted in Fig. 20
correlated to experimental results since only a few data are avail- alongside calculated values, using the REV.
able at very low stresses. However, this drop of stress vs. time to At 450 ◦ C and 550 ◦ C, stress levels are correctly represented, but
failure curves has been already highlighted (Haney et al., 2009; at 625 ◦ C, these stresses are underestimated. Moreover, at this tem-
Kimura, 2006; Panait et al., 2010) and is associated with a change in perature, the viscosity effect, clearly visible when different strain
creep mechanisms from power-law creep to viscous creep (Haney rates are tested, is not accurately described. Finally, the mate-
et al., 2009; Kloc and Sklenička, 1997, 2004). The long-term diffu- rial softening, represented by the decreasing part of the tensile
sion mechanisms which occur at very low stresses strongly modify curve (after the maximum of strain) is not well captured. Indeed,
the microstructure (Panait et al., 2010) and may explain the loss a linear softening is observed whatever the temperature and the
of creep strength. Based on the identification of the critical dam- applied strain rate, whereas in the experiments, the material soft-
age threshold Dc (Section 3.3), boundary stresses are calculated ening increases with the progressive occurrence of the necking
for other temperatures than those identified in Fig. 16, with the phenomenon. This point will be discussed in the following section.

Fig. 19. Numerical (lines) and experimental (symbols) creep curves at (a) 550 ◦ C and (b) 600 ◦ C.
 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232 229

Fig. 20. Experimental (symbols) and numerical (lines) tensile curves at (a) 450 ◦ C, (b) 550 ◦ c and (c) 625 ◦ C.

6. Discussions constant hardening coefficients. The improvement of the model’s

response is clearly seen with a better accuracy in both primary
6.1. Model efficiency and tertiary creep stage, especially for high stresses. Indeed, with
constant hardening parameters, an increase in the numerical error
6.1.1. Creep curves is observed when the stress level increases, whereas with tem-
As described in Section 3.1, hardening parameters have been perature dependent hardening parameters the numerical error is
taken to be dependent on temperature. To highlight the benefit approximately constant or lower whatever the stress level.
of this, calculations with constant hardening coefficients that do
not vary with temperature have also been performed (as those 6.1.2. REV and reduction of the computation time
done by Pétry and Lindet (2009)). Creep curves at 600 ◦ C using In order to validate numerical results obtained with the REV,
both approaches are plotted in Figs. 21 and 22 alongside experi- some creep tests have been also simulated on an axisymmetric
mental values, with lines representing numerical creep curves for model of a creep half-specimen. The mesh and boundary condi-
temperature dependent hardening parameters and dashed lines for tions of this specimen are illustrated in Fig. 23. Here, the loading

Fig. 21. Numerical (lines and dashed lines) and experimental (symbols) primary Fig. 22. Numerical (lines and dashed lines) and experimental (symbols) tertiary
creep stage at 600 ◦ C. creep stage at 600 ◦ C.
230 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232

Fig. 24. Comparison of creep time-to-rupture for the modified 9Cr1Mo steel at
650 ◦ C between simulations on specimen (black dashed line), simulations on an
REV (red line) and experimental (symbols). (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of the article.)

6.2. Strain localization and its geometrical effect

This model works under the assumption of homogeneous strain,

i.e. the final stage of a creep or tension test (including necking,
crack initiation and propagation and final failure of the specimen)
is not taken into account. The chosen approach consists of run-
ning the calculation until a 10% macroscopic strain occurs, which
is estimated by Pétry and Lindet (2009) to be an accurate predic-
tor of the time to rupture. As shown in Fig. 25 for the modelled
response of a creep test, no necking and no strain/stress/damage
localization are actually observed, even at the considered time to
rupture tR (the same observations can be made with the modelled
tension test). The major reason this assumption was made is the dif-
ficulty in describing the necking phenomena with finite elements
with a strong mesh-dependency. In order to overcome this prob-
lem, either non-local variables or crystal plasticity could be used.
Another solution would be to use automatic remeshing during com-
putation, as used in specific large strain finite element software.
However, in terms of final time-to-rupture prediction, which is
the main issue of this work, this simplification has no major influ-
ence as shown by the numerical results of the rupture times in
Fig. 18. Some modifications of the model can even be proposed in
order to improve its response and they are formulated in the next
section.

6.3. Model improvements

The modelling approach used could be improved in two minor

ways. First, as shown in the previous figure, it would have been
Fig. 23. Mesh and boundary conditions of a creep specimen. interesting to completely simulate creep and tension tests up to
final failure or at least until the strain localization appears, so as
to accurately represent the necking phenomenon at high stresses
and damage localization at low stresses. Secondly, by abandoning
the homogeneous strain assumption, so that the softening part of
the tensile curve is better predicted. Effective modelling of neck-
ing would improve the prediction of the decreasing stress part of
is applied on the threaded surface of the creep specimen. Fig. 24 the numerical tension test strain–stress curve. It is our view that
compares the time to creep rupture between both modelling this model can be improved without introducing much complexity,
approaches and with experimental results at 650 ◦ C. The numerical by representing both a plastic damage Dp , which predominantly
results are quite similar for the whole range of stresses, suggesting occurs at low temperature and high stresses and a preponderant
the REV approximation is valid. Moreover, the REV approach sig- creep damage Dc at high temperature and low stresses. Triaxiality
nificantly reduces the computation time from 1 h 30–1 min in case should be taken into account in both damage terms and the total
of high stresses and from 3 h to 3 min in case of low stresses. damage will be the summation of the two damages Dp and Dc .
 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232 231

Fig. 25. Results of a creep test at 650 ◦ C and 120 MPa at t = tR = 14.57 h in terms of (a) the deflected mesh (in red – the initial mesh is in black) (b) the total strain (c) the von
Mises equivalent stress and (d) the damage D. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article.)

7. Conclusions metal in order to have a complete description of modified 9Cr1Mo

weldments behaviour. Following this, we will employ an anisother-
This paper has been concerned with the modelling the creep mal elasto-viscoplastic behaviour model to allow the simulation of
behaviour of modified 9Cr1Mo steel over a large temperature range. actual complex structures.
A large review of experimental creep test data for this mate-
rial has allowed an elasto-viscoplastic creep behaviour model, Acknowledgements
which includes damage effects, to be fitted for a temperature range
between 450 ◦ C and 650 ◦ C. Modifications to the parameter identi- EDF/MMC Renardières and AREVA-NP are both gratefully
fication procedure have been introduced in order to better describe acknowledged for their financial support. The authors are also
the creep behaviour of modified 9Cr1Mo steel. After the imple- grateful to the DM2S and DMN – two departments of the CEA-, for
mentation of the model in the finite element code Cast3m, creep making available the experimental database of creep and tension
and tension tests have been simulated under different loading con- tests.
ditions, i.e. for different temperatures and stresses in creep and
different strain rate in tension.
The numerical results show good prediction of creep time to References
rupture for the whole range of temperature, and areas correspond-
Al-Faddagh, K.D., Fenner, R.T., Webster, G.A., 1982. Steady state distributions in
ing to the two damage mechanisms have been clearly identified. circumferentially notched bars subjected to creep. J. Strain Anal. 17 (3), 123–132.
Additionally, the simulated creep curves are in agreement with Cocks, A.C.F., Ashby, A.F., 1982. On creep fracture by void growth. Prog. Mater. Sci.
27, 189–244.
experimental results whatever the applied stress and particularly
Czyrska-Filemonowicz, A., Zielińska-Lipiec, A., Ennis, P.J., 2006. Modified 9% Cr steels
in the primary creep phase. The model’s response has been also for advance power generation: microstructure and properties. J. Achiev. Mater.
tested in tension, and although, the assumption of homogeneous Manuf. Eng. 2, 43–48.
deformation leads to some imprecision. Finally, the model’s effi- Eggeler, G., 1989. The effect of long-term creep on particle coarsening in tempered
martensite. Acta Metall. 37 (12), 3225–3234.
ciency has also been discussed and some improvements have been European Standard EN 10216-2:2002/prA2, 2006. Seamless Steel Tubes for Pressure
suggested to better account for strain/damage localization. Purposes – Technical Delivery Conditions. Part 2. Non Alloy and Alloy Steel Tubes
In the next stage of this study it is planned to identify two With Specified Elevated Temperature Properties.
Frost, H.J., Ashby, M., 1982. Deformation-Mechanism Maps: The Plasticity and Creep
new sets of parameters for the heat affected zone and the welded of Metals and Ceramics. Pergamon Press, Oxford.
232 T. Massé, Y. Lejeail / Nuclear Engineering and Design 246 (2012) 220–232

Gaffard, V., Gourgues-Lorenzon, A.F., Besson, J., 2005. High temperature creep flow Maruyama, K., Sawada, K., Koike, J.I., 2001. Strengthening mechanisms of creep
and damage properties of 9Cr1MoNbV steels: base metal and weldment. Nucl. resistant tempered martensitic steel. ISIJ Int. Jpn., 641–653, 41–6.
Eng. Des. 235, 2547–2562. Masuyama, F., 2006. Creep degradation in welds of Mod, 9Cr–1Mo steel. Int. J. Pres.
Gaffard, V., 2009. Experimental study and modeling of high temperature creep flow Ves. Pip. 83, 819–825.
and damage behavior of 9Cr1Mo–NbV steel weldments. Ph.D. Thesis. ENSMP, Mustata, R., Hayhurst, D.R., 2005. Creep constitutive equations for a 0.5Cr 0.5Mo
France. 0.25V ferritic steel in the temperature range 565–675 ◦ C. Int. J. Pres. Ves. Pip. 82,
Hall, F.R., Hayhurst, D.R., 1991. Continuum damage mechanics modelling of high 363–372.
temperature deformation and failure in a pipe weldment. Proc. Roy. Soc. Lond. Orlová, A., Buršík, J., Kuchařová, K., Sklenička, V., 1998. Microstructural development
A 433, 383–403. during high temperature creep of 9% Cr steels. Mater. Sci. Eng. A245, 39–48.
Haney, E.M., et al., 2009. Macroscopic results of long-term creep on a modified Panait, C.G., Bendick, W., Fuchsmann, A., Gourgues-Lorenzon, A.-F., Besson, J., 2010.
9Cr–1Mo steel (T91). Mater. Sci. Eng. A 510–511, 99–103. Study of the microstructure of the Grade 91 steel after more than 100,000 h of
Hayhurst, D.R., 1972. Creep rupture under multiaxial states of stress. J. Mech. Phys. creep exposure at 600 ◦ C. Int. J. Pres. Ves. Pip. 87, 236–335.
Solids 20, 381–390. Perrin, I.J., Hayhurst, D.R., 1996. Creep constitutive equations for a
Hayhurst, D.R., Henderson, J.T., 1977. Creep stress redistribution in notched bars. 0.5Cr–0.5Mo–0.25V ferritic steel in the temperature range 600–675 ◦ C. J.
Int. J. Mech. Sci. 19, 133–146. Strain Anal. 31, 299–314.
Hayhurst, D.R., Dimmer, P.R., Morrison, C.J., 1984. Development of continuum dam- Pétry, C., Gariboldi, E., 2009. Experimental characterisation and modeling of the
age in the creep rupture of notched bars. Phil. Trans. Roy. Soc. Lond. A 311, creep properties of a P92 steel weldment. In: 4th EPRI Expert Workshop on
103–129. Creep–Fatigue Damage Interaction, Paris.
Hayhurst, R.J., Vakili-Tahami, F., Mustata, R., Hayhurst, D.R., 2004. Thickness and Pétry, C., Lindet, G., 2009. Modeling creep behavior and failure of
multi-axial stress creep rupture criteria of the type IV component of a ferritic 9Cr–0.5Mo–1.8W–VNb steel. Int. J. Pres. Ves. Pip. 86, 486–494.
steel weld. J. Strain Anal. 39 (6), 729–743. Raj, B., Choudhary, B.K., Singh Raman, R.K., 2004. Mechanical properties and non-
Hayhurst, R.J., Mustata, R., Hayhurst, D.R., 2005a. Creep constitutive equations for destructive evaluation of chromium–molybdenum ferritic steels for steam
parent type ZAT, CG-ZAT and weld material in the range 565–640 ◦ C for Cr–Mo–V generator application. Int. J. Pres. Ves. Pip. 81, 521–534.
weldments. Int. J. Pres. Ves. Pip. 82, 137–144. Sklenička, V., Kuchařová, K., Svoboda, M., Kloc, L., Buršík, J., Kroupa, A., 2003. Long-
Hayhurst, D.R., Hayhurst, R.J., Vakili-Tahami, F., 2005b. CMD predictions of creep term creep behavior of 9–12%Cr power plant steels. Mater. Charact. 51, 35–48.
damage initiation and growth in ferritic steel weldments in a medium bore Tabuchi, M., Takahashi, Y., 2006. Evaluation of creep strength reduction factors for
branched pipe under constant pressure at 590 ◦ C using a 5-material weld model. welded joints of modified 9Cr–1Mo steel (P91). In: Proc. PVP2006/ICPVT-11
Proc. Roy. Soc. Lond. A 461, 2303–2326. ASME Pressure Vessels and Piping Division Conference, PVP2006-ICPVT-11-
Hayhurst, D.R., Lin, J., Hayhurst, R.J., 2008. Failure in notched tension bars due 93350, 2006.
to high-temperature creep: interaction between nucleation controlled cavity Tabuchi, M., Hongo, H., Li, Y., Watanabe, T., Takahashi, Y., 2009. Evaluation of
growth and continuum cavity growth. I. J. Sol. Struct. 45, 2233–2250. microstructures and creep damages in HAZ of P91 steel weldment. J. Pres. Ves.
Hegeman, H., 2009. V/HTR vessel materials – Creep testing of pre-irradiated weld Technol. 131–132, 6.
specimens. Deliverable D-ML 1.4, RAPHAEL project. Tavassoli, A.A., Le Naour, F., Tournié, I., 1992. Mechanical properties of 9Cr ferritic
Kachanov, L.M., 1958. On the time to rupture under creep conditions. Otd. Tech. steels developed for steam generators – tensile, impact and creep properties of
Nauk. 8, 26–31. thick plates. In: Technical Note NT SRMA, pp. 92–1968.
Kimura, K., 2006. Assessment of long-term creep strength and review of allowable Tveergaard, V., 1984. Analysis of cup–cone fracture in a round tensile bar. J. Mech.
stress oh high Cr ferritic creep resistant steels. In: Proc. PVP2006/ICPVT-11 ASME Phys. Sol. 32 (5), 373–393.
Pressure Vessels and Piping Division Conference, PVP2006-ICPVT11-93294, Van der Giessen, E., Vanderburg, M.W.D., Needleman, A., Tveergaard, V., 1995. Void
2006. growth due to creep and grain boundary diffusion at high triaxialities. J. Mech.
Kimura, K., Sawada, K., Kushima, H., Toda, Y., 2007. Long-term creep strength of creep Phys. Sol. 43 (1), 123–165.
strength enhanced ferritic steels. Chall. Power Eng. Environ. 13, 1059–1065. Vivier, F., 2009. Fluage à 500 ◦ C d’un joint soudé d’un acier 9Cr–1Mo modifié. Évo-
Kloc, L., Sklenička, V., 1997. Transition from power-law to viscous creep behavior of lution de la microstructure & Comportement mécanique. Ph.D. Thesis. ENSMP,
P-91 type heat-resistant steel. Mater. Sci. Eng. A234–A236, 962–965. France (in French).
Kloc, L., Sklenička, V., 2004. Confirmation of low stress creep regime in 9% Watanabe, T., Tabuchi, M., Yamazaki, M., Hongo, H., Tanabe, T., 2006. Creep damage
chromium steel by stress change creep experiments. Mater. Sci. Eng., 633–638, evaluation of 9Cr–1Mo–V–Nb steel welded joints showing type IV fracture. Int.
A387–389. J. Pres. Ves. Pip. 83, 63–71.