International Journal of Fatigue 59 (2014) 129–136

Contents lists available at ScienceDirect

International Journal of Fatigue

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

Two lifetime estimation models for steam turbine components

under thermomechanical creep–fatigue loading
L. Cui a,b,⇑, P. Wang b
a
School of Mechanical Engineering, Xi’an Shiyou University, Dianzi Erlu 18#, 710065 Xi’an, Shaanxi, China
b
Institut für Werkstoffkunde (IfW), Technische Universität Darmstadt, Grafenstrasse 2, 64283 Darmstadt, Germany

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

Article history: The flexibility of steam turbine components is currently a key issue in terms of the fluctuations in the
Received 16 May 2013 power supply due to regenerative energy. Conventional steam power plants must run at varying utiliza-
Received in revised form 11 September 2013 tion levels. Life estimation methods according to standards, e.g. ASME Code N47 and TR, assess the influ-
Accepted 17 September 2013
ences of creep and fatigue separately under the assumption of isothermal conditions at the maximum
Available online 24 September 2013
operating temperature. The influence of thermomechanical fatigue (TMF) loading still requires a signifi-
cant number of experimental studies. Further, the interaction of creep and fatigue is not adequately taken
Keywords:
into account. Thus, new lifetime estimation methods are required for the monitoring, re-engineering and
Thermomechanical fatigue (TMF)
Chromium steel
new design of power plant components. In this paper, both a phenomenological and a constitutive crack
Lifetime estimation initiation lifetime estimation model for steam turbine components are introduced. The effectiveness of
Creep–fatigue each method is shown by recalculation of uniaxial as well as multiaxial service-type creep–fatigue exper-
Damage accumulation iments on high-chromium 10%Cr stainless rotor steel. Finally, the two models are compared with respect
to different aspects, such as the type and number of necessary experiments to determine model param-
eters, the prerequisite for the application and the limitations of each model.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction Steam pressure and centrifugal forces at rotors lead to quasi-static

In the foreseeable future, the coal-fired power supply will con-
tinue to play an irreplaceable role in the energy mix [1]. There ex-
ists a considerable energy saving potential by the existing power
plants. The improvement of efficiency of coal-fired power plants
is to be achieved by increasing the temperature and pressure of
steam turbines. On the other hand, the increasing share of electric-
ity produced by renewable energy will lead to more fluctuations in
the power supply. Power plants are increasingly forced to run at
varying utilization levels, which can shift the critical load to fatigue
domain by superimposed creep on the heated surface of compo-
nents. As a result, a very interesting point from an economic per-
spective will be the remaining life of these plants, as well as the
optimization of start-up and shut-down processes in order to
achieve maximum economic and ecological benefits. All these as-
pects require simple and effective methods for plant life analysis
of high temperature steam turbine components.
Temperature transients, constant or variable pressure in pres-
surized systems and constant or variable speed of rotors produce
a large variety of combined static and variable loading situations.
⇑ Corresponding author at: School of Mechanical Engineering, Xi’an Shiyou
University, Dianzi Erlu 18#, 710065 Xi’an, Shaanxi, China.
E-mail address: cuiluxa@hotmail.com (L. Cui).
creep damage. Load cycling due to start-up and shut-down causes
fatigue damage, especially on the heated surface of turbine compo-
nents. As a consequence, superimposed creep–fatigue can be con-
sidered to be the critical loading condition [2]. Traditionally,
thermal fatigue cracking has been assessed using the results of iso-
thermal tests conducted at (or close to) the peak operating temper-
ature. For life estimation or assessment under creep–fatigue
loading, the life fraction rule introduced by Robinson [3] and Taira
[4] is used in standards, e.g. ASME Code N47 [5]. The rule is deter-
mined by the summation of fatigue damage as a cyclic fraction and
creep damage as a time fraction up to a material dependent critical
creep–fatigue value. Instead of time fraction, a strain fraction rule,
the so called ductility-exhaustion method, is also widely applied
[6]. A resulting problem is the determination of long duration rup-
ture ductility properties [7]. Another well-known life assessment
method is the strain range partitioning method [8]. The method
distinguishes between damage due to time independent plastic
and time dependent plastic (creep) deformation. In this context,
the lifetime is estimated with the help of up to 4 S–N curves deter-
mined by experiments with 4 different cycle forms. Therefore, the
application of the life fraction rule is relatively simple with respect
to the preparation of experimental data, and the primary creep
load due to steam pressure and centrifugal force can be superim-
posed easily [9]. Farragher et al. [10] presented a multiaxial,

0142-1123/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijfatigue.2013.09.007
130 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136
critical-plane approach for life prediction. The lifetime model
works as a post processing step of a transient heat transfer FEA
and a sequential anisothermal viscoplastic FEA and was calibrated
with a simplified representative TMF cycle of a P91 steam header.
An extension of this methodology is required if loading histories
with variable strain ranges are to be assessed.
In this study, two lifetime prediction models for high-tempera-
ture components under TMF loading in steam turbines are intro-
duced. The first one is a phenomenological lifetime estimation
model (Model 1) developed from ‘‘engineering’’ approaches as an
extension of the generalized damage accumulation rule [4,11].
The second method (Model 2) is derived from a unified viscoplastic
model [12,13] with incorporated isotropic damage. This approach
is based on the concept of effective stress combined with the prin-
ciple of generalized energy equivalence [14]. The model parame-
ters of both methods were determined using uniaxial creep
experiments and low cycle fatigue (LCF) experiments. As valida-
tion, a lifetime prediction for the conducted uniaxial as well as
multiaxial service-type [9,15–17] experiments was carried out
with both methods. Finally, the two methods are compared with
respect to different aspects, such as the type and number of neces-
sary experiments, the prerequisite for the application, advantages,
disadvantages and limitations.
2. Experiment
The TMF experiments applied in this paper were carried out un-
der a ‘‘service-type’’ loading cycle according to [9,15–17] (Fig. 1a
and b). It is noted that the concept of ‘‘service-type’’ loading cycles
is rather similar to that of Holdsworth et al. [18] and was called
therein ‘‘service-like’’ cycle or service cycle. A similar idea of ‘‘sim-
plified cycle’’ and ‘‘realistic cycle’’ is presented in [10]. In any case,
the terms ‘‘service-type’’ and ‘‘service-like’’ are too similar and
should be renamed consistently in the future. The authors suggest
the terms ‘‘simplified service cycle’’ and ‘‘realistic service cycle’’
[19]. Accordingly, the strain controlled loading cycle illustrated in
Fig. 1 is referred to hereafter as a ‘‘simplified service cycle’’. It is
characterized by a compressive strain hold phase 1 simulating
start-up conditions, a zero strain hold phase 2 approximating tem-
perature equi-balance during constant loading, a tensile strain hold
phase 3 simulating shut-down conditions and an additional zero
strain hold phase 4 which characterizes a zero loading condition
[16]. The experiments were performed in accordance with the
thermal strain compensation method of the current TMF Standard
[20].
Fig. 1. Simplified service cycle: (a) temperature cycle and (b) strain cycle.
The tested material is a modern ferritic–martensitic
stainless steel of type 10Cr–1Mo–1W–V–Nb (German grade
X12CrMoWVNbN10-1-1). Uniaxial and biaxial loading conditions
were simulated with cylindrical and cruciform specimens. The de-
tails of performed experiments were introduced in [16]. An over-
view of the testing parameters of the applied isothermal and
TMF experiments are shown in Table 1.
3. Lifetime models
3.1. Phenomenological lifetime model
Model 1 proposed in this paper is a phenomenological lifetime
model which adopts a simplified service cycle as input. It is based
on the synthesis of stress–strain hysteresis loops and damage
assessment under consideration of creep–fatigue interaction. The
strain cycle begins with a strain of zero and reaches the maximum
compressive strain first. This process is described by a cyclic
stress–strain curve in accordance with the Ramberg–Osgood equa-
tion [21]:
r rn10
e ¼ eel þ epl ¼ þ ; ð1Þ
E K
where r is the stress, eel is the elastic part, and epl is the plastic part
of the total strain e. E is the temperature-dependent Young’s mod-
ulus. The material parameters K and n0 are temperature-dependent
and determined with stress–strain loops of LCF (low cycle fatigue)
experiments. To describe cyclic hardening and softening behavior
of materials, eight sets of parameters K and n0 are determined at
0%, 5%, 10%, 20%, 40%, 60%, 80%, and 100% of lifetime for 6 reference
temperatures between 300 °C and 625 °C. For stress–strain relation-
ships under intermediate temperatures, the parameters are
interpolated.
The stress relaxation process rðtÞ in hold time 1 (Fig. 2, 1 to 10 )
is described iteratively in j increments (Fig. 3). An effective stress
concept is applied. The effective stress r ^ is defined as the differ-

ence between normal stress r and internal stress r:
r^ ðtÞ ¼ rðtÞ  re ðtÞ: ð2Þ

The internal stress r at the beginning and end of the four hold
times are given in Table 2. Values of re at any other time points in a
hold phase can be approximately determined by a linear
interpolation.
For the first iteration, the creep strain increment D^e1 during the
time increment Dt1 for effective stress r ^ 1 is determined with a
creep equation of the type Norton–Bailey [23], and the strain hard-
ening rule is applied by varying stress (Fig. 3). The creep strain
increment D^e1 is calculated with the creep equation:
^ n1 Dtm
^e1 ¼ D^e1 ¼ Ar 1 with r^ 1 ¼ r1 ð3Þ
in which the parameters A, n and m for the temperate T are temper-
ature-dependent and determined with creep curves for the 6 refer-
ence temperatures. The total creep strain ^e1 at the first iteration is
the creep strain increment D^e1 :
The effective stress r
^ 1 for the first iteration is the normal stress
r1.
The effective stress r^ j and the normal stress rj for the incre-
ment j (j = 2 to J) are calculated as follows:
r^ j ¼ rj  re j ð4Þ
rj ¼ rj1  ^ej1 =E ð5Þ
where E is the temperature-dependent Young’s modulus. The time
span tj (Fig. 3) which the material needs to reach the total creep
strain ^ej1 under the stress r
^ j is to be determined as follows:
 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136 131

Table 1
Summary of the test results under service-type loading [16].

Test piece ID Tmax (°C) Rth (h) Stage e_ r (1/s) DTC/DTW/DTH (°C) DeC/DeW/DeH (°C) Nia (–)

Uniaxial loading
uA16b1 600 3.2 1 103 –/–/– e/–/– 811
uA16ba2 600 3.2 1 103 300/–/– e/–/– 583
uA16b31 600 1.0 1 105 –/–/– e/–/– 787
uA16ba26 600 1.0 1 105 300/–/– e/–/– 443
uA16ba27 600 1.0 3 105 300/100/50 e/0.77e/0.59e 1239
Biaxial loading
uA16df7 600 1.0 1 105 –/–/– –/–/0.62e 1268
uA16dfba1 600 1.0 1 105 –/–/50 –/–/0.62e 1058b
uA16df6 600 1.0 1 105 –/–/– e/–/– 655
uA16dfba3 600 1.0 1 105 300/–/– e/–/– 356
uA16df4 600 1.0 3 105 –/–/– e/0.77e/0.59e 1400
uA16dfba2 600 1.0 3 105 300/100/50 e/0.77e/0.59e 1060
a
Crack initiation, load drop 1.5%.
b
Dismount.

Table 2
The internal stress at the beginning and end of the four hold times [22]. The symbol
r0001 denotes e.g. the stress at the point 100 in Fig. 2.
Beginning End
Hold phase 1 (1–10 ) r0003 þr1 r003 þr0001
2 2
Hold phase 2 (2–20 ) r0001 þr2 r0001 þr002
2 2
Hold phase 3 (3–30 ) r0001 þr3 r001 þr0003
2 2
Hold phase 4 (4–40 ) r0003 þr4 r0003 þr004
2 2

sffiffiffiffiffiffiffiffiffi
m ej1
^
t j ¼ : ð6Þ
Ar^ nj

Then the creep strain increment D^ej (Fig. 3) is calculated as

follows:
Fig. 2. Synthesis of stress–strain hysteresis loop.
^ nj ðt j þ Dt j Þm  ^ej1 :
D^ej ¼ Ar ð7Þ

After every increment, the creep strain increment is

accumulated:
^ej ¼ ^ej1 þ D^ej : ð8Þ
The creep strain increment D^e at an intermediate temperature T
is interpolated between the values D^eðT ¼ T u Þ and D^eðT ¼ T l Þ,
which are calculated with the Norton–Bailey-parameters of the
upper and lower temperatures Tu and Tl.
The synthesis of a stress–strain hysteresis loop under simplified
service cycle is based on the following principles. First of all, the
loop is supposed to be enveloped in a hypothetical loop consisting
of two flank curves (Fig. 2, 1000 –300 and 100 –3000 , also f(Tmax) and
f(Tmin)). The upper flank curve f(Tmax) is derived from a cyclic
stress–strain curve under the maximum temperature Tmax by mul-
tiplying by a factor of two, while the lower flank curve f(Tmin) is de-
rived from a cyclic stress–strain curve under the minimum
temperature Tmin by multiplying by a factor of two. The starting
points of the stress relaxation in the four hold phases 1, 2, 3 and
4 (Fig. 2, points 1, 2, 3 and 4) are all located at the flank curves,
Fig. 3. Calculation of stress relaxation (a) during hold times with creep data (b).
and the end points of stress relaxation in hold phases 1 and 3
(Fig. 2, points 10 and 30 ) are also located at the flank curves. The
stress relaxation in hold phases 2 and 4 can be described as anal-
The life fraction rule introduced by Robinson/Taira with modifi-
ogous to hold phase 1. Point 200 is the intersection point of the
cation is used for life estimation under TMF loading. This rule pre-
upper flank curve and the straight line which starts from the point
sents a summation of creep damage Dc and fatigue damage Df:
20 with a slope equal to Young’s modulus’ at Tmax. Point 400 is the
Ni
!
intersection point of the lower flank curve and the straight line X X
K X
J
Dt l;k;j 1
which starts from the point 40 with a slope equal to Young’s mod- Dc þ Df ¼ þ : ð9Þ
l¼1 k¼1 j¼1
tu;l;k;j Nio;l
ulus’ at Tmin.
132 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136

The creep damage of a cycle is calculated by accumulating the 3.2. Constitutive model
creep damage of all time increments Dtl,k,j. in terms of Dtl,k,j/tu,l,k,j,
where index k (for this case k = 1–4) denotes the serial number The constitutive damage model is incorporated into a unified
of hold phases in cycle l, and index j means the j-th time increment viscoplastic material model [13] in the form of an isotropic damage
in hold phase k. The rupture time tu,l,k,j is carried out according to variable D according to the principle of generalized energy equiv-
the instantaneous acting stress rl,k,j from rupture characteristic alence [14].
curves, which were generated by data of static creep tests. The For small deformations, the total strain E is additively decom-
modeling of creep rupture time tu is done with help of the Larson posed into an elastic and a plastic part:
Miller parameter [24].
E ¼ Ee þ Ep : ð15Þ
PLM ðT; t u Þ ¼ T  ðC þ lg t u Þ ð10Þ The relationship between stress T and elastic strain Ee follows
The fatigue damage for cycle l presented in terms of 1/Nio,l is cal- Hooke’s law with C as the stiffness tensor for isotropic materials,
culated with reference number of cycles to crack initiation Nio,l, where the damage D leads to a smaller stress in damaged material
which is generated with data of standard LCF tests under consider- than in undamaged material under the same elastic strain.
ation of creep–fatigue interaction and the mean stress effect on T ¼ ð1  DÞ  C½Ee  with C½A ¼ C ijkl Akl ð16Þ
lifetime.
An equivalent stress f
Nio ¼ N ith  mr ð11Þ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 ðT  nÞD ðT  nÞD
The number of cycles to crack initiation Nith for the TMF loading f ¼ pffiffiffiffiffiffiffiffiffiffiffiffi : pffiffiffiffiffiffiffiffiffiffiffiffi ð17Þ
2 1D 1D
is determined with the equation:
is defined as the deviator of the difference between the stress tensor
1 1 1 T and the back stress tensor of kinematic hardening n. The operator
¼ þ : ð12Þ
Nith Nith ðT ¼ T min Þ Nith ðT ¼ T max Þ (. . .)D denotes deviator of a tensor and (:) indicates the inner prod-
uct of second order tensors. The overstress F for plastic deformation
The number of cycles to crack initiation Nith as a function of
is defined as
strain ranges for a certain temperature is described using the Man-
son–Coffin [25,26] equation for maximum temperature Tmax and F ¼ f  k0 ð18Þ
minimum temperature Tmin of a hysteresis loop.
where the material parameter k0 represents the yield strength.
Furthermore, the mean stress effect is described by a mean
The plastic strain increment is given as
stress factor:
3 ðT  nÞD
N ðr –0; Deeff Þ E_ p ¼ pffiffiffiffiffiffiffiffiffiffiffiffi s_ ð19Þ
mr ¼ ith m ð13Þ 2 1Df
Nith ðrm ¼ 0; Deeff Þ
with s as accumulated plastic strain. Its evolution is generally de-
which characterizes the relationship of fatigue lifetime under load- scribed by a power law s_ ¼ hFim =g. As noted in [28], a single power
ing with mean stress (rm –0) and without mean stress (rm ¼ 0). The law exponent is not sufficient to model the complete stress regime a
fatigue lifetime under loading with mean stress (N ith ðrm –0; Deeff Þ) is modern 9–12%Cr steel experiences. A modification of the power law
derived from the Smith-Watson-Topper parameter [27] as part of further development in comparison to [15,29] with the
d
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi term eaF is introduced for better conformance to a wide stress
 
Dreff Deeff range as follows:
Pswt ¼ þ rm  E  ; ð14Þ
2 2
hFim d
s_ ¼  eaF : ð20Þ
g
where Dreff is the effective stress range, and Deeff is the effective
strain range. The creep–fatigue interaction was investigated in [9] Four material parameters m, n, a and d are used. As illustrated in
d
according to metallographic observations. In viscoelastic stress– Fig. 4, the term eaF covers the range of power law breakdown and
strain loops with small strain amplitudes, there exists no significant permits a smooth transition from low stress to high stress. With
plastic deformation. The creep damage dominates and the damage the help of this modification, only one evolution equation for kine-
is represented by intergranular cracks. On the other hand, the matic hardening is required, which means fewer material parame-
creep–fatigue damage dominates in viscoplastic loops with large ters are to be determined. Alternatively, the power law can be
strain ranges and transgranular cracks are expected. For the first replaced with a hyperbolic sine function. This approach was cali-
case, the creep damage is derived from the 4 hold times of the sim- brated and validated in [28] for P91 steel across a range of cyclic
plified service cycle (Fig. 2). The fatigue damage can be estimated
with the help of fatigue life curves generated with LCF data without
hold times. For the second case, hold time 1 and a part of hold time
3, depending on the strain range, are assigned to the fatigue dam-
age. The creep–fatigue damage is calculated on the basis of life
curves generated by creep–fatigue experiments with symmetric
hold times. Correspondingly, the creep damage results from the
complete hold phases 2 and 4 and the remaining time of hold phase
3. The evaluations of creep damage for the two cases remain the
same as described above.
Accumulating the creep damage and fatigue damage until a pre-
defined critical value of damage Dcrit, the counted cycle number is
defined as the number of crack initiation Ni under the loading.
A computer program SARA-TMF has been developed to carry out
the above introduced phenomenological approach. Fig. 4. Comparison of simulated strain rate of uniaxial creep tests with test data.
 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136
and noncyclic high-temperature loading conditions. An internal
variable Y for kinematic hardening strain is applied and its evolu-
tion follows the Armstrong–Frederick hardening rules with an
additional static recovery term pY for high-temperature creep
behavior. The equation can be written as
pffiffiffiffiffiffiffiffiffiffiffiffi
Y_ ¼ E_ p  B  b 1  DYs_  pY
where b and p are material parameters. In Eq. (21) a scalar function
B, which is given by
BðsÞ ¼ B1 þ ð1  B1 ÞeB2 s
is applied to the term of dynamic recovery in a multiplicative man-
ner to describe the cyclic softening or hardening. B1 and B2 are
parameters, whereby B1 > 1 represents cyclic softening and B1 < 1
cyclic hardening.
Furthermore the kinematic hardening back stress tensor n is de-
fined as a scalar multiplication of Y by the parameter c:
n ¼ ð1  DÞ  cY
The viscosity function in the form of a power law or its im-
proved version (Eq. (20)) is to be distinguished from the power
law creep equation (Eq. (24)), which describes, as is generally
known, only the secondary creep.
^e_ ¼ Arn
Take, for example, the case of uniaxial creep tests: other than
the stress r (r is equal to the applied constant load stress at the
specimen) in a creep equation, the over stress F in Eq. (21) is de-
fined as the applied load stress less back stress. It is not constant
but steadily being reduced by increasing back stress. Therefore,
the creep rate slows until the back stress reaches the maximum
(Y_ ¼ 0). This process correlates to the primary creep. Subsequently,
the strain rate reaches a minimum and becomes near constant due
to the constant over stress. This stage is known as secondary creep.
In tertiary creep, the strain rate exponentially increases as a conse-
quence of the damage accumulation. In summary, Model 2 consi-
derates primary, secondary and tertiary creep deformation
response.
As an assumption, the total damage D represents the superposi-
tion of fatigue damage Df and creep damage Dc. The fatigue damage
is addressed to crack initiation from the surface, which is assumed
to correlate with the accumulated plastic strain s. The isotropic
damage rule of Kachanov-Rabotnov [30] with the parameters Ac,
rc and kc is applied to the creep damage evolution. The following
equations describe total damage D, its components and evolution
equations:
D_ ¼ D_ f þ D_ c
D_ c ¼ ðr =Ac Þkc  ð1  DÞrc
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !2
2 3 D D 1
Rm ¼ ð1 þ mÞ þ 3ð1  2mÞ T : T = traceðTD Þ
3 2 3 of the cruciform specimen was conducted for the first loading cycle
1 þ ac  ðRm  1Þ
Y  ¼ ðEe : TÞ  þ ðn : YÞ
Rm
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ ð1  DÞ  2  E  Y 
1 þ af  ðRm  1Þ
D_ f ¼ Af   s_
Rm
In the evolution equation of the creep damage, the fictitious
stress r* is used, which can be derived from the modified strain en-
ergy release rate Y*. The strain energy release rate was modified by
133
an additional parameter ac which controls the effect of multiaxial
stress states on the creep damage evolution. For 0 6 ac < 1, the neg-
ative impact of multiaxial stress states on the development of
creep with respect to the uniaxial creep test (Rm = 1) at the same
equivalent stress is reduced. Otherwise the impact of multiaxiality
is increased. Analogous to ac, the parameter af controls the effect of
ð21Þ multiaxial stress states on the fatigue damage evolution. At uniax-
ial stress state (Rm = 1), the fatigue damage evolution is assumed to
be proportional (by a factor Af) to the increment of accumulated
plastic strain s.
ð22Þ It is noted that the damage variable D correlates to isotropic
damage. Due to the fact that fatigue cracking tends not to be isotro-
pic, the presented damage model is suitable for the prediction of
lifetime to crack initiation.
The material parameters were determined in two steps. In the
first step, the deformation parameters for deformation are deter-
mined by using uniaxial creep tests and push–pull LCF tests. The
damage parameters are identified in the second step. The parame-
ð23Þ ters Ac, rc and kc are determined with uniaxial tertiary creep data.
As the parameter describing the sensitivity of the creep damage
due to multiaxial stress states, ac can be estimated by comparing
the rupture time in uniaxial and biaxial creep tests under the same
equivalent stress. The parameter af can be determined with biaxial
push–pull tests. It is assumed that the fatigue damage is responsi-
ð24Þ ble for the linear load drop, while the non-linear load drop can be
described as cyclic softening with function B(s) (Fig. 5). TEM inves-
tigations of the steel X12CrMoWVNbN10-1-1 indicate that the cyc-
lic softening correlates with the subgrain coarsening [31]. The
subgrain size decreases at the beginning and reaches a steady
state. This process can be described very well by the function B(s).
4. Results
As validation of the two proposed lifetime models, the experi-
ments listed in Table 1 are calculated with both models. The calcu-
lation with Model 1 was carried out with the program SARA-TMF. A
two-dimensional loading, which is the case in the biaxial experi-
ments, cannot be calculated directly with the phenomenological
approach, since the model requires uniaxial input data. Thus, an
equivalent strain cycle was composed for the first cycle. Generally,
the equivalent strain and stress can be composed in accordance
with different hypotheses, e.g. von Mises, or based on crack open-
ing displacement (COD). A comparison of these hypotheses by
application for lifetime estimation was made in [32], and the meth-
od using von Mises equivalent strain and stress shows the smallest
scatter band. Thus, in this paper, a uniaxial equivalent strain eeq in
accordance with von Mises is determined as follows:
ð25Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
ð26Þ eeq ¼ signðe11 Þ 2ððe11  e22 Þ2 þ ðe22  e33 Þ2 þ ðe33  e11 Þ2 Þ ð31Þ
3
The principal strains e11 and e22 for the loading direction x and y
are given in the form of simplified service cycles. An inelastic FEA
ð27Þ
in order to determine the strain for the third direction e33. Further,
a user subroutine ‘‘UAMP’’ was developed to realize the so called
ð28Þ ‘‘strain control mode’’. The amplitudes of the forces on the speci-
men arms are determined by UAMP and applied in every time
increment of the FEA in order to reach the specified simplified ser-
ð29Þ
vice cycle in the test zone. Data of the node under extensometer
heads are evaluated to determine the equivalent strain.
ð30Þ Model 2 was applied for an FEA of the whole specimen. For a
better comparison with Model 1, the calculation with Model 2
was carried out with an FEA model with one element. This corre-
sponds approximately to the point below the extensometer. It is
assumed that this point is also the point with the maximum
134 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136

Fig. 5. Determination of function B(s) and parameter Af for fatigue damage Df with ‘‘load drop’’ data of LCF tests.

damage. This assumption is valid only under the condition that
there exists one homogeneous strain and stress field in the test
area of cruciform specimens as well as in the gauge section of
the cylindrical specimens. However, this is not always true, espe-
cially in TMF loading at high temperature rates [16]. In this case,
the whole specimen should be calculated.
4.1. Description of deformation behavior
Both models describe the deformation behavior of the investi-
gated material fairly well. Fig. 6 shows the obtained results of a
uniaxial TMF experiment (uA16ba26) for the cycles N = 1 and
N = Ni/2. In the first cycle, it can be seen that the relaxation behav-
ior in hold time 2 is not sufficiently simulated with the constitutive
model. This is due to the fact that a simple kinematic hardening
rule with only one back stress is applied. Furthermore, large devi-
ations exist for both methods in the description of material behav-
ior at low temperature in the half lifetime cycle. The reason for this
is that both models are calibrated using isothermal experiments.
However, analyses in [16,18] show a significant shift of axial
strains in the TMF specimens. The stress field in gauge length of
the TMF specimens is not homogeneous and the local axial mean
strain in the symmetry plane at axis is higher than that at the sur-
face [16]. Nevertheless, Model 2 yields significantly larger errors in
all cases, which occur up to 50% relative to the test data at maxi-
mum compressive stress (Fig 6c).
4.2. Lifetime estimation
The determination of Dcrit is a challenging subject. In this work
for Model 1, a material-specific mean creep–fatigue damage value
of 0.68 which is determined by diverse uniaxial service-type tests

Fig. 6. Comparison between two models and experiments under thermomechanical loading according to uA16ba26 in Table 2, (a) (b) data by cycle N = 1 and (c) (d) by N = Ni/
2.
 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136 135

Fig. 7. Comparison of number of cycles to crack initiation of service-type experiments to estimated numbers, (a) phenomenological model (Model 1) N i and (b) constitutive
model (Model 2) N i .

is chosen for the material X12CrMoWVNBN10-1-1 and was applied Table 4

Number and type of experiments used to identify the parameters of Model 2.
as the critical damage value Dcrit for recalculation of the experi-
ments shown in Table 1. For Model 2, a critical damage value for fa- T (°C) 300 400 500 550 600 625
tigue Df,crit (T) and a critical damage value for creep Dc,crit (T) are LCF (hold time 0/0, strain rate 10 3
1/s) 1 1 1 1 1 1
determined respectively in dependence on the temperature. An Creep 3 3 3 3 7 3
internal variable Dac is applied to accumulate the normalized dam-
age increments.
Above all, the advantage of Model 1 is its relatively short com-
D_ f D_ c puting time, because only explicit equations are applied in the
D_ ac ¼ þ ð32Þ
Df ;crit ðTÞ Dc;crit ðTÞ model. Hence it is especially suitable for real time monitoring
and optimization of start-up and shut-down processes. On the con-
The critical damage value for fatigue Df,crit (T) is determined by
trary, evolution equations, which can only be accomplished with
evaluating the accumulated plastic strain of push–pull experi-
high computational complexity, are used in Model 2. Its applicabil-
ments at different temperatures. The critical damage value for
ity in industrial practice for lifetime assessment depends on the
creep Dc,crit (T) is defined as the damage value at which tertiary
speed of calculation of large finite element models, particularly un-
creep begins. A calculation will be terminated by a critical damage
der cyclic loading. In this context, a sequential extrapolation meth-
value Dac = 1.
od [17], which accelerates the calculation of deformation and
The estimated numbers of cycles to crack initiation N i with
lifetime, can be of crucial importance for the application of a con-
both models for X12CrMoWVNbN10-1-1 under simplified service
stitutive approach such as Model 2.
cycles are plotted against the experimental values in Fig. 7. The
In the presented phenomenological method, the influence of
predictions associated with Model 1 are significantly better than
creep–fatigue interaction on lifetime is considered explicitly.
Model 2 and lie on the conservative side. This correlates to the fact
Depending on strain range, hold times 1 and 3 are assigned either
that Model 2 yields significantly larger errors in the description of
to creep damage or to fatigue damage and evaluated respectively.
deformation behaviors, especially at half lifetime. However, both
The constitutive method takes creep–fatigue interaction into con-
models predict the creep–fatigue lives within a scatter band of fac-
sideration implicitly by using the damage variable D, which in-
tor two.
cludes both creep damage and fatigue damage. The damage
variable D is integrated in the evolution equations of plastic defor-
mation and back stress, which in turn determine the evolution of
5. Discussions
damage.
The parameters of both models are temperature dependent, and
Both methods provide satisfactory results regarding the lifetime
one set of parameters is identified for each of the 7 selected tem-
estimation and have advantages and disadvantages. The following
peratures in Model 1 and 6 selected temperatures in Model 2. The
differences can be observed:

Table 3
Number and type of experiments used to identify the parameters of Model 1.

T (°C) 300 400 450 500 550 600 625

LCF (hold time 0/0, strain rate 103 1/s) 2 – 2 2 4 7 4
LCF (3/3, 103 1/s) 2 – 2 1 2 3 3
LCF (20/20, 103 1/s) – – 2 – 2 2 –
LCF (0/0, 105 1/s) – – 2 – 2 2 –
LCF (0/0, ½  106 1/s) – – 2 – 2 2 –
Creep – 2 2 2 8 10 8
136 L. Cui, P. Wang / International Journal of Fatigue 59 (2014) 129–136
type and number of applied experiments for the parameter identi-
fication are given in Tables 3 and 4. Significantly more experiments
are needed to generate the parameters of Model 1. But the determi-
nation of parameters of the phenomenological method is easier,
because the number of parameters, which are to be determined
at the same time, is much smaller than in the constitutive model.
In the lifetime prediction of steam turbine components, the
phenomenological method (Model 1) is applied as a post processing
step of a thermal FEA, because the FEA provides local equivalent
strain cycles as input. In the calculation of components, a thermal
FEA should be conducted to determine the strain cycle, and the
integration point with the maximum equivalent strain range is
chosen. The model adopts a uniaxial loading cycle as input, and
multiaxial strain states should be converted to a uniaxial equiva-
lent strain according to a hypothesis, e.g. von Mises. As the next
step, an abstraction of the calculated cycle into a simplified service
cycle is usually necessary, so that the deformation and damage
analysis can be conducted with the phenomenological approach.
Furthermore, the phenomenological method (Model 1) considers
the deformation and damage behavior of an isolated local material
point. However, local plastic deformations, such as flattening of
notches, can shift the point of maximum. On the contrary, Model
2 can be easily combined with an FEA of complex structures under
multiaxial loading conditions and thus can be applied for design
and re-engineering of important power plant components.
6. Conclusions
In this paper, two lifetime estimation models for the develop-
ment of new, more efficient turbines as well as for the optimization
of start-up and shut-down processes of the existing plants under
complex loadings are introduced. They each describe material
deformation and damage in a different manner and both models
show satisfactory results using the recalculation of uniaxial and
biaxial experiments under simplified service cycle.
For a solid understanding of the approaches in their application,
the two models are compared in the following aspects: generation
of the model parameters, accuracy of prediction and practical
applicability. A phenomenological approach has its strengths in
an engineering-based application, which already provides a practi-
cal use in the design and manufacture of turbines with acceptable
conservative results (Fig. 7a). Another advantage is the coverage of
a wider range of materials. According to the current state of the
technique, the application of a constitutive model is relatively
expensive in terms of the parameter identification, but behind this
approach is more potential for describing complex loading condi-
tions, since the model can describe multiaxial deformation better
and allows for immediate use in finite element programs. The
recalculation of the two approaches yielded acceptable predictions
(Fig. 7) with the potential for improvement, taking into account an
improved data base.
Acknowledgements
Thanks are due to the ‘‘Forschungsvereinigung der Arbeits-
gemeinschaft der Eisen und Metall verarbeitenden Industrie e.V.’’
(AVIF No. A 232 and A 242), and the ‘‘Forschungsvereinigung Verb-
rennungskraftmaschinen e.V.’’ (FVV No. 608 951 and 609 250) for
financial support, and to the working group W10 of the German
power plant industry for their accompaniment.
References
[1] International Energy Agency, World energy, outlook 2010.
[2] Berger C, Scholz A, Müller F, Schwienheer M. Creep fatigue behavior and crack
growth of steels. In: Abe F, Kern TU, Viswanathan R, editors. Creep-resistant
steels. Cambridge: Woodhead Publishing Limited; 2008. p. 446–71.
[3] Robinson E. Effect of temperature variation on the long-time rupture strength
of steels. Trans ASME 1952;74(5):777–80.
[4] Taira S. Lifetime of structures subjected to varying load and temperature. In:
Hoff NJ, editor. Creep in structures. New York: Academic Press; 1960. p.
96–124.
[5] ASME Code Boiler and Pressure Vessels. Code Case N 47-17, Class I components
in elevated temperature service, Section III, Division 1, 1996.
[6] Priest R, Ellison E. A combined deformation map ductility exhausion approach
to creep–fatigue analysis. Mater Sci Eng 1981;49:7–17.
[7] Holdsworth SR. The ECCC approach to creep data assessment. ASME J Press
Vessel Technol 2008;130(May). 024001/1-6.
[8] Manson SS, Halford GR. Relation of cyclic loading pattern to microstructural
fracture in creep–fatigue, Proc. Of Fatigue 1984, The sec. Int. Conf. on Fatigue
and Fatigue Thresholds, Vol. III, 1984. p. 1237–55.
[9] Scholz A, Berger C. Deformation and life assessment of high temperature
materials under creep fatigue loading. Mater.-wiss. und Werkst.-tech
2005;36:722–30.
[10] Farragher TP, Scully S, O’Dowd NP, Leen SB. Development of life assessment
procedures for power plant headers operated under flexible loading scenarios.
Int J Fatigue 2013;49:50–61.
[11] Manson SS, Halford GR. A method of estimating high-temperature. In: Hult J,
editor. Creep in structures. New York: Academic Press; 1960.
[12] Chaboche JL. Constitutive equations for cyclic plasticity and cyclic
viscoplasticity. Int J Plast 1989;5:247–302.
[13] Chaboche JL. A review of some plasticity and viscoplasticity constitutive
theories. Int J Plast 2008;24:1642–93.
[14] Tsakmakis C, Reckwerth D. The principle of generalized energy equivalence in
continuum damage mechanics. In: Hutter K, Baaser H, editors. Deformation
and failure in metallic materials. Berlin: Springer; 2003. p. 381–406.
[15] Samir A, Simon A, Scholz A, Berger C. Service-type creep–fatigue experiments
with cruciform specimens and modeling of deformation. Int J Fatigue
2006;28:643–51.
[16] Cui L, Wang P, Scholz A, Oechsner M, Berger C. Influence of temperature
transients and multiaxial loading on crack initiation behavior of modern high
chromium rotor steel under service type loading. Mater Sci Eng A 2013;560:
767–80.
[17] Wang P, Cui L, Lyschik M, Scholz A, Berger C, Oechsner M. A local extrapolation
based calculation reduction method for the application of constitutive
material models for creep fatigue assessment. Int J Fatigue 2012;44:253–9.
[18] Holdsworth SR, Mazza E, Binda L, Ripamonti L. Development of thermal fatigue
damage in 1CrMoV rotor steel. Nucl Eng Des 2007;237:2292–301.
[19] Kostenko Y, Almstedt H, Naumenko K, Linn S, Scholz A. Robust Methods for
creep fatigue analysis of power plant components under cyclic transient
thermal loading. Proceedings of the ASME Turbo Expo 2013, June 3–7, 2013,
Texas, USA.
[20] TMF Standard ISO 12111, ISO/TC 14/SC 5, 2009, Metallic materials – Fatigue
testing – Strain-controlled thermomechanical fatigue testing method, 2011.
[21] Ramberg W, Osgood WR. Technical Report No. 902, NACA, 1943.
[22] Scholz A. Beschreibung des zyklischen Werkstoffverhaltens bei betriebs
ähnlicher Landzeithochtemperaturdehnwechselbeanspruchung, Dr-Ing.
Thesis, TU- darmstadt, D17, 1988.
[23] Norton FH. The creep of steel at high temperature. New York: McGraw Hill;
1929.
[24] Larson FR, Miller J. A time-temperature relationship for rupture and creep
stresses. Trans ASME 1952;74:765–75.
[25] Coffin Jr LF. A Study of the effects of cyclic thermal stresses on a ductile metal.
ASME Trans 1954;76:931–50.
[26] Manson SS. Behavior of materials under conditions of thermal stress. Technical
Report NACA-TR-1170, National Advisory Committee for Aeronautics, 1954.
[27] Smith KN, Watson P, Topper TH. A stress–strain function for the fatigue of
metals. J Mater 1970;4:767–78.
[28] Barrett RA, O’Donoghue PE, Leen SB. An improved unified viscoplastic
constitutive model for strain-rate sensitivity in high temperature fatigue. Int
J Fatigue 2013;48:192–204.
[29] Simon A, Scholz A, Berger C. Validation of a constitutive material model with
anisothermal uniaxial and biaxial experiments. MP Mater Test 2009;9:532–41.
[30] Rabotnov YN. Creep problems in structural members, North Holland,
Amsterdam; 1969.
[31] Dubey JS, Chilukuru H, Chakravartty JL, Schwienheer M, Scholz A, Blum W.
Effects of cyclic deformation on subgrain evolution and creep in 9–12% Cr-
steels. Mater Sci Eng A 2005;406:152–9.
[32] Zhang S, Sakane M. Multiaxial creep–fatigue life prediction for cruciform
specimen. Int J Fatigue 2007;29:2191–9.