SAFE STRUCTURAL DESIGN FOR FATIGUE AND CREEP USING

CYCLIC YIELD STRENGTH

Yevgen Gorash* and Donald MacKenzie

Department of Mechanical and Aerospace Engineering, University of Strathclyde,
Glasgow, Scotland, UK

ABSTRACT

This study proposes cyclic yield strength (CYS, σyc ) as a potential characteristic of safe design for
structures operating under fatigue and creep conditions. CYS is defined on a cyclic stress-strain
curve (SSC), while monotonic yield strength (MYS, σym ) is defined on a monotonic SSC. Both
values of σyc and σym are identified using a 2-step fitting procedure of the experimental SSCs using
Ramberg-Osgood and Chaboche material models. A typical S-N curve in stress-life approach for
fatigue analysis has a distinctive minimum stress lower bound, the fatigue endurance limit (FEL,
f
σlim ). Comparison of σyc and σlimf
reveals that they are approximately equal. Thus, safe fatigue
design is guaranteed in the purely elastic domain defined by the σyc . A typical long-term strength
(LTS) curve in time-to-failure approach for creep analysis has 2 inflections corresponding to the σyc
and σym . These inflections separate 3 sections on a LTS curve, which are characterised by different
creep fracture modes and creep deformation mechanisms. Thus, safe creep design is guaranteed
in the linear creep domain with brittle failure mode defined by the σyc . These assumptions are
confirmed using 3 structural steels for normal and high-temperature applications. The advantage
of using σyc for characterisation of fatigue and creep strength is a relatively quick experimental
identification. The total duration of cyclic tests for a cyclic SSC identification is much less than the
typical durations of fatigue and creep rupture tests at the stress levels around the σyc .
Keywords: Creep, Fatigue, Failure, Plasticity, Softening, Steel, Yield Strength.

INTRODUCTION

Characterisation of long-term strength of structural materials is an important engineering task for

prevention of potential catastrophic failures of critical equipment. However, studies of this type are
usually very long-lasting, technically challenging and involve expensive experimental work. Thus,
the main scope of this study is the formulation of a simple way to predict characteristics of the long-
term material behaviour (creep and fatigue, in the first instance) using basic material properties.
Based upon the extensive availability of experimental material data, a significant progress to-
ward this challenge has been achieved so far and may be observed in the literature. Comparative
study by Kim et al. [1] evaluated seven basic methods for estimating uniaxial fatigue properties
f
(including σlim ) from tensile properties or hardness. This study was based upon the fatigue test data
for eight ductile steels under axial and torsional loading. Three of the evaluated methods were able
to predict over 93% of test cases within a factor of 3 compared with observed lives. The formulas
f
for σlim prediction included mechanical properties such as elasticity modulus E, ultimate tensile
f
strength σu and true fracture ductility εf . Among the variety of empirical formulations for σlim
prediction with different combinations of aforementioned mechanical properties, the simplest are
f f
based on σu : σlim = 1.9018 σu (Universal slopes method); σlim = 1.5 σu (Uniform material law);
and σlim = σu + 345 MPa (Mitchell’s method), which shown an accuracy of R2 = 0.88. Another
f

simple method in this comparison, proposed by Roessle & Fatemi [2], used a Brinnell hardness
* Corresponding author. Tel.: +44 141 5484851. E-mail: [email protected]
 f
HB for prediction as σlim = 4.25 HB + 225 MPa. This approach showed a reasonable accuracy
of R2 = 0.86 for experimental data fit.
f
The study by Casagrande et al. [3] investigated a relationship between σlim and Vickers hardness
f
HV in steels and developed a method to predict σlim . A good correlation was observed between HV
f
and σlim for for four kinds of steels in different metallurgical states. However, the proposed empirical
method is not straightforward and involves a number of parameters and equations to achieve a
f
reasonable of accuracy of σlim predictions. Recently, Bandara et al. [4] proposed a formula for
f
predicting σlim of steels in the gigacycle regime. It uses a combination of σu and HV as material
parameters and was verified using the experimental results for 45 steels.
A different approach was developed by Li et al. [5], who estimated theoretically σyc and σlim f
c
using test data for 27 alloy steels. One formula expresses σy by two conventional mechanical per-
formance parameters – σu and the reduction in area ψ. The other formula expresses the FEL by
the CYS with a reasonable accuracy of R2 = 0.883 as σlim f
= 1.13 (σyc )0.9 . Despite the relative
simplicity, the proposed relation can’t be considered as mathematically elegant, most probably be-
cause of the conventional assumption of 0.2% plastic strain offset for σyc and σym . Nevertheless, this
f
formula by Li et al. [5] demonstrated the tendency that σlim is not too much different from σyc .
Less progress has been achieved in methods for creep rupture strength evaluation, but recently
an important observation was discovered by Kimura [6]. The creep strength of ferritic and austenitic
steels has been investigated in [6] through the correlation between creep rupture curve, presenting
stress vs. creep rupture life, and 50% of 0.2% offset yield stress (half yield) at a wide range of
temperatures. The inflection of the creep rupture curve at half yield was recognised for ferritic
creep resistant steels with martensitic or bainitic microstructure, e.g. T91, T92 and T122. This was
explained in terms of different mechanisms of microstructural evolution during creep at high- and
low-stress regimes. The purpose of this study was to point out a significant risk of overestimation
of long-term creep rupture strength by extrapolating the data for martensitic and bainitic steels (e.g.
ASTM T91/P91) in high-stress regime to low-stress regime, which are separated by half yield.
A similar problem with particular application to ASTM P91 steel was investigated and discussed
by Gorash et al. [7, 8] for the purpose of a creep constitutive model development. In these works,
apart from inflection of creep rupture curve, the simultaneous inflection of the minimum creep rate
curve, presenting minimum creep rate vs. stress, was recognised. Alternation of minimum creep
rate slope was explained in terms of different creep deformation mechanism (linear creep for low
stress and power-law for high stress), while alternation of creep rupture life slope was explained in
terms of different damage accumulation modes (brittle fracture for low stress and ductile for high
stress). The inflection of both curves was characterised by the same value σ0 called transition stress,
which had the meaning of material parameter in the developed “double-power-law” creep model.
However, σ0 was identified in [7, 8] using minimum creep rate data, and no relation of σ0 to basic
mechanical properties of ASTM P91 steel was recognised.
The principal aim of the present study is to investigate a link in characterisation of long-term
strength of structural steel by finding a similar quantative feature in available experimental data.
This establishes a straight relation between characteristics of creep and fatigue behaviour on one
hand and yield strength as a basic material property and characteristic of plasticity on other hand.

CONCEPT OF THE SAFE STRUCTURAL DESIGN

Definition of the yield strength

Dowling [9] discusses several methods to characterise the yield strength σy . The first is the propor-
tional limit σyp , which is the stress where the first departure from linearity occurs. The second is the
elastic limit σyel , which is the highest stress that does not cause plastic deformation. The third is the
offset yield strength σy0.2% , which is the stress in the point on stress-strain curve typically defined by
the plastic strain offset of 0.2% from elastic line. This value is generally the most practical means
 ultimate monotonic S-S curve

normal temperature design

fatigue S-N curve
strength, su
cyclic S-S curve

high
plastic behaviour

monotonic yield strength, s ym

moderate
stress, s

mixed behaviour
caused by softening

cyclic yield strength, s yc fatigue endurance limit

elastic behaviour

low
safe design

total strain, e number of cycles to failure, log(N * )

ultimate minimum creep creep rupture curve

high temperature design

strength, su strain rate curve

high
power-law
monotonic yield breakdown ductile fracture
stress, log(s)

strength, s ym

moderate
mixed mode
power-law creep
fracture
cyclic yield
strength, s yc transition stress
linear creep

low
safe design brittle fracture

min. creet rate, log(e& c ) time to failure, log(t * )

Figure 1: Concept of the safe structural design for fatigue and creep using cyclic yield strength

of defining the yielding event for engineering metals. Therefore, σy0.2% is usually meant to define
the yield strength σy in the literature. However, here the elastic limit σyel , defined in the scope of
unified Chaboche model [10, 11], is used as the yield strength σy .
This study proposes σyc as a key characteristic for the definition of safe design for engineering
structures operating under fatigue and creep conditions, as illustrated in Fig. 1. It is conventionally
defined in context of a cyclic stress-strain curve (SSC), which is obtained from results of cyclic tests
for a number of different strain ranges. Each cyclic test produces a stabilised stress response, which
is effected either by hardening or by softening depending on the type of steel. In the case of steels
with a cyclic softening effect, σyc separates the low stress range of purely elastic behaviour from
moderate stress range of mixed elasto-plastic behaviour. Monotonic yield strength σym , which is
conventionally defined in context of a monotonic SSC, separates the moderate stress range of mixed
elasto-plastic behaviour from the high stress range of purely plastic behaviour. Both values of σym
and σyc are identified using a 2-steps fitting procedure of the experimental S-S curves. The first step
applies the Ramberg-Osgood material model, which produces basic smoothing and extrapolation,
to the both monotonic and cyclic SSCs separately. The second step of fitting involves a typical
rate-independent form of the Chaboche material model with 3 kinematic backstresses. Fitting the
Chaboche model with two separate sets of material constants sequentially to the both SSCs provides
the values of σym and σyc with minimum offset from the elastic line as elastic limits.

Stress-strain curves fitting procedure

As experimental SSCs usually demonstrate some level of scatter, the first step in data fitting for the
material parameters identification is basic curve smoothing. The conventional Ramberg-Osgood
(R-O) equation [15] is optimal for such curve smoothing since it was formulated to describe the
 450 900

400 800

350 700

300 elastic 600 elastic

mono. data
Stress (MPa)

Stress (MPa)
mono. data
250 500
mono. fit Ch. mono
200 cyclic data 400
σy mono

R-O cyclic cyclic data

150 300
Ch. cyclic R-O cyclic
100 200
σy cyclic Ch. mono

50 100 σy cyclic

0 0
0 0.0025 0.005 0.0075 0.01 0.0125 0.015 0.0175 0.02 0.0225 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Total strain Total strain

Figure 2: Fitting of monotonic and cyclic SSCs Figure 3: Fitting of monotonic and cyclic SSCs
of ASTM A36 steel from [12] at RT of AISI 4340 steel from [13] at RT
400 300

350
250

300

200 elastic
250 elastic
R-O mono
Stress (MPa)

Stress (MPa)

R-O mono
Ch. mono
200 Ch. mono 150
σy mono
σy mono
150 cyclic data
cyclic data 100
R-O cyclic
R-O cyclic
100 Ch. cyclic
Ch. cyclic
50 σy cyclic
50 σy cyclic

0 0
0 0.002 0.004 0.006 0.008 0.01 0 0.002 0.004 0.006 0.008 0.01
Total strain Total strain

Figure 4: Fitting of monotonic and cyclic SSCs Figure 5: Fitting of monotonic and cyclic SSCs
of ASTM P91 steel from [14] at 550◦ C of ASTM P91 steel from [14] at 600◦ C

non-linear relationship between stress and strain in materials near their yield point. It is particularly
useful for metals that harden or soften with plastic deformation showing a smooth elastic-plastic
transition. The equations for the monotonic and cyclic SSCs are:
( )1/β
σ ( σ )1/β ∆εtot ∆σ ∆σ
εtot = + and = + , (1)
E B 2 2E 2B

where ∆εtot is the total strain range and ∆σ is the total stress range (MPa) for each cyclic test
respectively; B and β are the R-O material parameters; and Young’s modulus E in MPa. Using
the value of E, the total strain εtot in the experimental curves is decomposed into elastic and plastic
strain. Then the plastic component εp of strain is fitted using the the least squares method by the
following power-law relations, which are derived from the Eq. (1):
( )β
β ∆σ ∆εp
σ = B (εp ) and =B . (2)
2 2
Table 1: Fitting parameters of the Ramberg-Osgood model (1) for different steels and temperatures

Type of plastic Elasto-plastic constants

material response E (MPa) B (MPa) β σy (MPa)
ASTM A36 RT cycl. 189606 1015.61 0.2362 –
AISI 4340 RT cycl.∗ 193053 1897.94 0.5175 320
ASTM P91 RT mono. 710 0.047 –
215000
ASTM P91 RT cycl. 1180 0.155 –
ASTM P91 500◦ C m. 594 0.066 –
180000
ASTM P91 500◦ C c. 763 0.15 –
ASTM P91 550◦ C m. 482 0.054 –
172000
ASTM P91 550◦ C c. 613 0.144 –
ASTM P91 600◦ C m. 330 0.042 –
158000
ASTM P91 600◦ C c. 446 0.123 –
ASTM P91 650◦ C m. 269 0.071 –
140000
ASTM P91 650◦ C c. 343 0.125 –
∗
Extended version of the R-O model (6) is used for data fitting.

The resultant R-O fits for monotonic and cyclic curves are then used to identify the parameters for
the Chaboche material model. The range of applicability for the R-O fit is usually quite narrow not
exceeding 1% of εtot depending on the grade of curvature grade for a SSC.
The basic variant of the rate-independent Chaboche model [10, 11] is presented as a combination
of nonlinear kinematic hardening and nonlinear isotropic hardening models. The model allows
the superposition of several independent backstress tensors and can be combined with any of the
available isotropic hardening models. Since in this study monotonic and cyclic SSCs are fitted
separately only for the identification of σy , only the kinematic hardening component is considered:

∑
N
X= Xi , with Ẋi = Ci ε̇p − γi Xi ṗ, (3)
i=1

where ε̇p is the plastic strain rate, and ṗ is its magnitude. The total backstress X in Eq. (3) is given
by the superposition of a number N of kinematic backstresses Xi with a corresponding evolution
equation initially proposed by Armstrong & Frederick [16] for Ẋi , where Ci and γi are kinematic
material constants. Chaboche et al. [10] recommended N = 3 in order to provide a good fit of
experimental SSCs, which include large strain areas. Therefore, three backstresses are considered
in this study providing an excellent match of the R-O fit (1) for a whole range of strains.
The kinematic hardening constants (Ci , γi ) and σy , which define the size of the yield surface,
are identified as recommended in [11]. The cyclic SSC is fitted by the following relation:

∑N ( )
∆σ c Ci ∆εp
= σy + tanh γi , (4)
2 γ
i=1 i
2

which is obtained in [11] by integrating Eq. (3) and considering εp ≈ const at the peak stresses for
strain-controlled cyclic loading. Relation (4) is valid for the cyclic curve after stabilisation of the
hardening or softening effects. Constants (Ci , γi and cyclic σyc ) are identified by automatic fitting
Eq. (4) to the R-O extrapolation (2) with “cyclic” values of constants B and β. The identification
procedure is implemented in Microsoft Excel using an add-in Solver [17]. The Solver searches for
an optimal (minimum in this case) value for a formula in one cell – called the objective cell – subject
to constraints, or limits, on the values of other formula cells on a worksheet. The Solver works with
a group of cells, called decision variables or simply variable cells, that participate in computing
the formulas in the objective and constraint cells. In this case, the Solver adjusts the values in
 Table 2: Fitting parameters of the Chaboche model (3)-(5) for different steels and temperatures

Type of plastic Three kinematic hardening backstresses Yield σ

material response C1 (MPa) γ1 C2 (MPa) γ2 C3 (MPa) γ3 σy (MPa)
ASTM A36 RT cycl. 87345.7 984.7 14013.4 111.78 3918.32 13.477 115.792
AISI 4340 RT mono. 205524.6 535.8 8966.94 92.268 782.893 1.0739 341.153
AISI 4340 RT cycl. 35912.1 650.7 6972.29 53.297 4221.72 5.7356 330.727
ASTM P91 RT mono. 1120466 23911 125301.9 2539.9 17295.23 227.86 406.098
ASTM P91 RT cycl. 1030320 11608 136282.4 1254.6 29535.03 148.08 197.493
ASTM P91 500◦ C m. 1059420 23359 122317.7 2469.7 17631.89 219.49 270.687
ASTM P91 500◦ C c. 659430 11229 87028.5 1248.7 19146.80 149.22 134.541
ASTM P91 550◦ C m. 1059420 23359 122317.7 2469.7 17631.89 219.49 270.687
ASTM P91 550◦ C c. 659430 11229 87028.5 1248.7 19146.80 149.22 134.541
ASTM P91 600◦ C m. 511703 24975 56536.0 2630.3 7588.97 232.90 199.970
ASTM P91 600◦ C c. 444752 12216 11344.6 160.13 56238.9 1347.6 107.731
ASTM P91 650◦ C m. 498277 23543 56252.6 2433.8 8263.19 217.10 115.346
ASTM P91 650◦ C c. 353928 12801 44816.6 1396.6 8916.41 162.14 80.6307

the decision variable cells containing material constants (Ci , γi and σyc ) in order to minimise the
value in the objective cell. This cell contains an average value of the absolute difference between
columns containing ∆σ 2 calculated by Eq. (2) and Eq. (4) correspondingly in a particular range of
∆εp . Applying this approach, an excellent match of Eqs (2) and (4) is achieved.
The monotonic SSC is fitted by the different relation in the following form [11]:

∑
N
Ci
σ = σym + [1 − exp(−γi εp )] , (5)
i=1
γi

which contains the monotonic σym and different values of kinematic hardening constants (Ci , γi ).
These constants are identified by fitting Eq. (5) to the R-O extrapolation (2) with “monotonic” values
of the R-O parameters B and β. The identification procedure is implemented in Microsoft Excel
using an add-in Solver [17] in the same way as for cyclic SSC. An advanced step-by-step guideline
for the estimation of the Chaboche viscoplasticity model parameters with their further optimisation
was developed by Hyde et al. [18].

Application to three structural steels

The above described fitting procedure is applied to SSCs of three structural steels for the purpose
of σym and σyc identification. The first is ASTM A36 steel, with mechanical properties reported
in [19, 12], which is a standard low carbon steel, without advanced alloying and is a principal
carbon steel employed for bridges, buildings, and many other structural uses. The monotonic
SSC for this steel shown in Fig. 2 exhibits perfectly plastic behaviour when reaching the stress
of 36 ksi = 248.211 MPa in average, which is considered as σym . The perfectly plastic yielding lasts
for approximately of εp = 1 (%) of strain plateau, which is followed by the strain hardening area,
then gradually approaching failure at εtot = 30 (%). The cyclic SSC for this steel shown in Fig. 2
from [12] is fitted by the 2-step procedure, and the obtained material parameters for the R-O (1) and
Chaboche (3)-(5) models are listed in Tables 1 and 2 correspondingly.
The second material is AISI 4340 steel [13], a high-strength alloy steel, which has good machin-
ability features and used for a wide range of applications including aircraft landing gears, shafts or
axels for power transmission, gears, high pressure pump housings, etc. Both monotonic and cyclic
SSCs shown in Fig. 3 and mechanical properties are taken from [13]. Since it is available explicitly,
the monotonic SSC is fitted by the Chaboche model (5) directly, and the material parameters are
listed in Table 2. The cyclic SSC for this steel shown in Fig. 3 from [13] is available at ten times
wider strain range than for the ASTM A36 steel. Therefore, the R-O model (1) is not able to provide
an accurate fit of the cyclic SSC. In this case, the following modification of the R-O equation (1) is
used for fitting analysis:
( )1/β ( )1/β
σ σ − σy ∆εtot ∆σ ∆σ − σy
εtot = + and = + , (6)
E B 2 2E 2B

Compared to Eq. (1), this notation contains an additional parameter of the yield strength σy in the
meaning of σyel , and can be applied for an accurate fitting of much wider strain range than Eq. (1).
Thus, the cyclic SSC is fitted by the 2-step procedure. The obtained material parameters for the
modified R-O (6) and Chaboche (3)-(5) models are listed in Tables 1 and 2 correspondingly.
The third material is ASTM P91 (modified 9Cr-1Mo) steel [20, 14], an advanced ferritic steel
with martensitic microstructure, which has already been widely used over the last 2 decades as
tubes/pipes for heat exchangers, plates for pressure vessels, and other forged, rolled and cast com-
ponents for high temperature services. Both monotonic and cyclic SSCs shown in Figs 4 and 5
and mechanical properties at room temperature (RT), 500◦ C, 550◦ C, 600◦ C and 650◦ C are taken
from [14]. Firstly, the monotonic SSCs are presented in [14] by the material parameters for the R-O
model (1) listed in Table 1. The cyclic SSCs are presented in [14] by raw data, which is fitted by the
R-O model (1) with material parameters listed in Table 1. Secondly, both monotonic and cyclic R-O
extrapolations are fitted by the Chaboche model (3)-(5) with material parameters listed in Table 2.

RELATION IN MECHANICAL CHARACTERISTICS

The next step is a check for possible correlations between the obtained yield strength values (σym
and σyc ) for ASTM A36, AISI 4340 and ASTM P91 steels and their fatigue and creep behaviour.
This identifies a clear similarity for characteristic transition stresses in S-N fatigue, minimum creep
strain rate and creep rupture curves, as explained below.

Fatigue behaviour at normal temperature

Engineering structures operating under cyclic loading conditions at normal temperature are usually
designed against fatigue failure using the conventional stress-life approach. This approach involves
experimental fatigue S-N curves with number of cycles to failure N ∗ vs. stress. A typical S-N curve
is a straight line in double logarithmic coordinates with a distinctive minimum stress lower bound,
f f
which is called a fatigue endurance limit (FEL, σlim ). Referring to [9, 18], σlim is observed for a
number of structural steels in benign environmental conditions and represents a stress level below
which the material does not fail and can be cycled infinitely without fatigue damage. Comparison
of σyc defined as material constant and experimentally observed σlim
f
reveals that they are close. This
assumption is confirmed by high-cycle fatigue (HCF) experimental data for ASTM A36 [21] and
f
AISI 4340 [22, 23, 24] steels shown in Fig. 6. Comparison of σlim with σyc summarised in Table 3 for
ASTM A36 steel gives 27.6% accuracy and 5.5% accuracy for AISI 4340 steel. These observations
indicate that safe fatigue design is guaranteed in the purely elastic domain defined by σyc .

Creep behaviour at elevated temperature

Engineering structures operating under constant loading conditions at high temperature are usu-
ally designed against creep failure using the conventional time-to-failure approach. This approach
involves experimental creep rupture curves with stress vs. time to failure t∗ . A typical creep rup-
ture curve is a trilinear smoothed curve in double logarithmic coordinates, with two inflections
corresponding to σyc and σym . These inflections separate three sections on the creep rupture curve,
which are characterised by three different creep damage accumulation modes – brittle, ductile and
mixed. Three sections with different creep deformations mechanisms can be typically observed
 650 exp., Atlas of Fatigue Curves
exp., Dowling (2004)
600
exp., Ragab et al. (1989)
exp., did not fail
550
4340 fatigue fit (σ_lim = 350MPa)

500 exp., Wang et al. (2010) traditional

Alternating stress (MPa)

exp., Wang et al. (2010) energy

450 A36 fatigue fit (σ_lim = 160MPa)

400

350

300

250

200

150
10000 100000 1000000 10000000
Number of cycles to failure

Figure 6: S-N curve fits of ASTM A36 steel based on HCF data by Wang et al. [21] and AISI 4340
steel based on HCF data from Atlas of Fatigue Curves [22], Dowling [23] and Ragab et al. [24]

on the minimum creep rate curve, presenting minimum creep strain rate vs. stress, which is also a
trilinear smoothed curve in double logarithmic coordinates. The deformations mechanism (linear
creep, power-law creep and power-law breakdown) are separated by the same two inflections.
This assumption is confirmed by experimental observations for ASTM P91 steel at elevated
temperatures. Data for creep rupture shown in Fig. 7 is all taken from the recent study by Kimura [6].
The inflections of corresponding curves were well observed at 600 and 650◦ C and explained in terms
of half monotonic yield (σy0.2% /2). In contrast to [6], in current study, σym and σyc from Table 2 are
used in combination with test data [6] to provide a basic polylinear fitting. Data for min. creep strain
rate shown in Fig. 8 is taken from studies by Sklenička et al. [25], Kloc & Fiala [26] and Kimura
[20]. The inflections of corresponding curves were observed at 550, 600 and 650◦ C and explained in
terms of transition between different creep deformation mechanisms. As in the case of creep rupture,
here the same σym and σyc from Table 2 are used in combination with test data [20, 26, 25] to provide
a basic polylinear fitting. Since the inflections are captured reasonably well on both types of data in
Figs 7 and 8, the correspondence of transition stresses on creep rupture and min. creep rate curves
proposed by Gorash et al. [7, 8] is proved by relating them to σym and σyc . It should be noted that
Dimmler et al. [27] associated these inflections with microstructurally determined threshold stresses
(back-stress concept). The applicability of this concept was shown using the experimental minimum
creep rate and creep rupture curves for several 9-12%Cr heat resistant steels (P91, GX12, NF616,
X20 and B2). Dimmler et al. [27] emphasised that the knowledge of these threshold stresses limits
the range of experimentally based predictions, thus preventing from overestimation of long-term
creep rate and creep strength from extrapolated short-term creep data. Therefore, these observations
arise a consideration that the most safe creep design is guaranteed in linear creep domain with brittle
failure mode, which is also defined by the σyc .
Finally, the fatigue performance of ASTM P91 steel is analysed using the HCF experimental data
by Matsumori et al. [28] at three different temperatures (RT, 400 and 550◦ C) illustrated in Fig. 9.
Stress (MPa)

100

exp., 500°C, Kimura (2013)

stress-based data fit, 500°C
exp., 550°C, Kimura (2013)
stress-based data fit, 550°C
exp., 600°C, Kimura (2013)
stress-based data fit, 600°C
exp., 650°C, Kimura (2013)
stress-based data fit, 650°C
cyclic yield σyc, all temperatures
monotonic yield σym, all temp.
10
100 1000 10000 100000
Time to failure (h)

Figure 7: Stress vs. creep rupture life of ASTM P91 steel based on the data by Kimura [6]

1.0E+01
exp., 550°C, Kimura et al. (2009)
1.0E+00
exp., 550°C, Sklenicka et al. (1994)

1.0E-01 stress-based data fit, 550°C

Minimum creep strain rate (1/h)

exp., 600°C, Kloc & Fiala (2005)

1.0E-02
stress-based data fit, 600°C
1.0E-03 exp., 650°C, Kloc & Fiala (2005)

stress-based data fit, 650°C

1.0E-04
cyclic yield σyc, all temperatures
1.0E-05 monotonic yield σym, all temperatures

1.0E-06

1.0E-07

1.0E-08

1.0E-09
1 10 100
Stress (MPa)

Figure 8: Min. creep rate vs. stress of ASTM P91 steel based on several sets of data [20, 26, 25]
 550

500

450
Alternating stress (MPa)

400

exp., RT
350
exp., 400°C
exp., 550°C
300 exp., did not fail

RT fit (σ_lim = 418MPa)

250 400°C fit (σ_lim = 350MPa)
550°C fit (no σ_lim exists)
200
10000 100000 1000000 10000000 100000000
Number of cycles to failure

Figure 9: S-N curve fits of ASTM P91 steel based on HCF data by Matsumori et al. [28]

From these data, it can be concluded that at elevated temperatures heat-resistant steels don’t exhibit
f
σlim on S-N fatigue curves, which is usually observed at normal temperature. The reason for this is
the elimination of purely elastic behaviour at high temperature, since there is always some amount
of inelastic strain, which is caused by creep. Therefore, there is always a permanent accumulation of
creep damage, even at low stress levels and high-strain rate, which leads to inevitable failure. This
f
fact is confirmed by experimental observations [28], which demonstrated the extinction of σlim at
◦ 8 f m
550 C for over 10 loading cycles. However, a good match of σlim with σy with accuracy of 2.8% is
observed at RT for this steel as shown in Table 3, which makes advanced martensitic steels different
f
from simple ferritic steels is σlim prediction. This effect can be explained by the assumption of
f
Terent’ev [29], who recognised two types of the fatigue endurance limit σlim – standard in HCF range
(N = 10 -10 cycles) and ultrahigh in gigacycle fatigue (GCF) range (N = 107 -1011 cycles). The
2 7
f
existence of ultrahigh σlim was proved by the experimental data for high-strength steels (50CrV4,
54SiCrV6 and 54SiCr6), which demonstrated two inflections of the fatigue curves followed by
horizontal plateaus – first in HCF area (N ≈ 105 -106 ), second in GCF area (N ≈ 108 -109 ). The
correspondence of σyc with ultrahigh σlimf
for ASTM P91 steel is expected to be found at N > 108
cycles, but no experimental data is available for this range.

CONCLUSIONS

Kimura’s [6] assumption of half monotonic yield (σy0.2% /2) agrees very well with the outcomes of
the current study. According to Table 3, the relation σyc ≈ σym /2 is valid for all temperatures except
the highest 650◦ C. This assumption is not relevant to AISI 4340 steel, which exhibits σyc ≈ σym .
The principal advantage of the σyc application to the characterisation of fatigue and creep long-
term strength is the relatively fast experimental identification. The total duration of all cyclic tests,
which are required to reach the stabilised stress response for the construction of cyclic SSC is much
less than the typical durations of fatigue and creep rupture tests at stress levels around σyc .
 Table 3: Comparison of σym , σyc and σlim
f
for ASTM A36, AISI 4340 and ASTM P91 steels

Steel ASTM A36 AISI 4340 ASTM P91

Temp., ◦ C RT RT RT 400 500 550 600 650
σym , MPa 248.2 341.2 406.1 – 270.7 253.0 200.0 115.3
σyc , MPa 115.8 330.7 197.5 – 134.5 116.6 107.7 80.6
σym /σyc 2.1 1.0 2.1 – 2.0 2.2 1.9 1.4
f
σlim , MPa 160.0 350.0 418.0 350.0 – – – –
∆σ, % 27.6 5.5 2.8 – – – – –

The critical point in the work presented here is an application of the advanced material model
(i.e. Chaboche model [10, 11]) to the estimation of a single value of elastic limit σyel , which may
seem to be complected. However, this approach is effective in typical cases when experimental
SSCs are unavailable in explicit form, but available in the form of R-O [15] fittings (1). In other
cases, when all necessary experimental SSCs are available in form of raw data, the modified form
(6) of the R-O model may reduce the fitting procedure just to one step. Since Eq. (6) contains σy as
a material parameter, the application of Chaboche model equations (3)-(5) is no longer needed.

REFERENCES

[1] Kim, K. S. et al. “Estimation methods for fatigue properties of steels under axial and torsional
loading.” Int. J. Fatigue, vol. 24, no. 7, (2002), pp. 783–793.
[2] Roessle, M. L. and Fatemi, A. “Strain-controlled fatigue properties of steels and some simple
approximations.” Int. J. Fatigue, vol. 22, no. 6, (2000), pp. 495–511.
[3] Casagrande, A. et al. “Relationship between fatigue limit and Vickers hardness in steels.”
Mater. Sci. & Eng. A, vol. 528, no. 9, (2011), pp. 3468–3473.
[4] Bandara, C. S. et al. “Fatigue Strength Prediction Formulae for Steels and Alloys in the
Gigacycle Regime.” Int. J. Mater., Mech. & Manufacturing, vol. 1, no. 3, (2013), pp. 256–260.
[5] Li, J. et al. “Theoretical estimation to the cyclic yield strength and fatigue limit for alloy
steels.” Mechanics Research Communications, vol. 36, no. 3, (2009), pp. 316–321.
[6] Kimura, K. “Creep rupture strength evaluation with region splitting by half yield.” Proc.
ASME 2013 PVP Conf. PVP2013-97819, ASME, Paris, France (Jul. 14-18, 2013), pp. 1–8.
[7] Gorash, Y. Development of a creep-damage model for non-isothermal long-term strength
analysis of high-temperature components operating in a wide stress range. PhD thesis, Martin-
Luther-University Halle-Wittenberg, Halle (Saale), Germany (Jul. 21, 2008).
[8] Altenbach, H. et al. “Steady-state creep of a pressurized thick cylinder in both the linear and
the power law ranges.” Acta Mechanica, vol. 195, no. 1-4, (2008), pp. 263–274.
[9] Dowling, N. E. Mechanical Behavior of Materials: Engineering Methods for Deformation,
Fracture, and Fatigue. Pearson Education Limited, Harlow, UK, 4th ed. (2013).
[10] Chaboche, J.-L. et al. “Modelization of the strain memory effect on the cyclic hardening of
316 stainless steel.” Trans. 5th Int. Conf. on Structural Mechanics in Reactor Technology, no.
L11/3 in SMiRT5, IASMiRT, Berlin, Germany (Aug. 1979), pp. 1–10.
[11] Chaboche, J.-L. “A review of some plasticity and viscoplasticity constitutive theories.” Int.
J. Plasticity, vol. 24, no. 10, (2008), pp. 1642–1693.
[12] Higashida, Y. and Lawrence, F. V. Strain controlled fatigue behavior of weld metal and heat-
affected base metal in A36 and A514 steel welds. FCP Report No. 22, University of Illinois,
Urbana, Illinois, USA (Aug. 1976).

[13] Smith, R. W. et al. Fatigue behavior of materials under strain cycling in low and intermediate
life range. Technical Note No. D-1574, NASA, Cleveland, Ohio, USA (Jan. 1963).

[14] Data sheets on elevated-temperature, time-dependent low-cycle fatigue properties of ASTM

A387 Grade 91 (9Cr-1Mo) steel plate for pressure vessels. NRIM Fatigue Data Sheet No. 78,
National Research Institute for Metals, Tokyo, Japan (Dec. 25, 1993).
[15] Ramberg, W. and Osgood, W. R. Description of stress-strain curves by three parameters.
Technical Note No. 902, NASA, Washington DC, USA (Jul. 1943).
[16] Armstrong, P. J. and Frederick, C. O. A mathematical representation of the multiaxial
Bauschinger effect. Report No. RD/B/N731, CEGB, Berkeley, UK (Dec. 1966).

[17] Microsoftr Office Professional Plus. Excel Help System // Analyzing data // What-if analysis
// Define and solve a problem by using Solver. Microsoft Corp., Release 2010 ed. (2009).

[18] Hyde, T. et al. Applied Creep Mechanics. McGraw-Hill Education, New York, USA (2004).
[19] ASTM Standard. Standard Specification for Carbon Structural Steel. A36/A36M – 08, West
Conshohocken, USA (2008).
[20] Kimura, K. et al. “Long-term creep deformation property of modified 9Cr-1Mo steel.” Mater.
Sci. & Eng. A, vol. 510-511, (2009), pp. 58–63.
[21] Wang, X. G. et al. “Quantitative Thermographic Methodology for Fatigue Assessment and
Stress Measurement.” Int. J. of Fatigue, vol. 32, no. 12, (2010), pp. 1970–1976.
[22] Boyer, H. E. Atlas of Fatigue Curves. ASM International, Materials Park, Ohio, USA (1986).

[23] Dowling, N. E. “Mean Stress Effects in Stress-Life and Strain-Life Fatigue.” SAE Technical
Paper, , no. 2004-01-2227, (2004), pp. 1–14.

[24] Ragab, A. et al. “Corrosion Fatigue of Steel in Various Aqueous Environments.” Fatigue
Fract. Engng Mater. & Struct., vol. 12, no. 6, (1989), pp. 469–479.

[25] Sklenička, V. et al. “Creep Behaviour and Microstructure of a 9Cr Steel.” D. Coutsouradis
et al., ed., Proc. Conf. Materials far Advanced Power Engineering 1994. Part I, Kluwer Aca-
demic Publishers, Liège, Belgium (Nov. 3-6, 1994), pp. 435–444.
[26] Kloc, L. and Fiala, J. “Viscous creep in metals at intermediate temperatures.” Kovové Mater-
iály, vol. 43, no. 2, (2005), pp. 105–112.
[27] Dimmler, G. et al. “Extrapolation of short-term creep rupture data – The potential risk of
over-estimation.” Int. J. of Pressure Vessels & Piping, vol. 85, (2008), pp. 55–62.
[28] Matsumori, Y. et al. “High Cycle Fatigue Properties of Modified 9Cr-1Mo Steel at Ele-
vated Temperatures.” Proc. ASME 2012 Int. Mechanical Engineering Congress & Exposition.
IMECE2012-87329, ASME, Houston, Texas, USA (Nov. 9-15, 2012), pp. 85–89.
[29] Terent’ev, V. F. “Endurance limit of metals and alloys.” Metal Sci. & Heat Treatment, vol. 50,
no. 1-2, (2008), pp. 88–96.