International Journal of Fatigue 146 (2021) 106138

Contents lists available at ScienceDirect

International Journal of Fatigue

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

Essential structure of S-N curve: Prediction of fatigue life and fatigue limit
of defective materials and nature of scatter
Yukitaka Murakami a, *, Toshio Takagi b, c, Kentaro Wada d, Hisao Matsunaga e
a
Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan
b
Takagi Corporation, 2-4-1 Ishida Minami, Kokura Minami-ku, Kitakyushu 802-8540, Japan
c
Graduate School of Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan
d
Department of Mechanical Engineering, Fukuoka University, 8-19-1 Nanakuma, Jyonan-ku, Fukuoka 814-0180, Japan
e
Department of Mechanical Engineering, Kyushu University, 744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan

A R T I C L E I N F O A B S T R A C T

Keywords: Historically, S-N curve has been expressed in the form of an exponential relationship between applied stress σ
S-N curve and number of cycles to failure Nf. This paper elucidates the essential structure of S-N curve based on the pre­
Fatigue limit diction method for fatigue life and limit of materials containing defects from the viewpoint of mechanics of small
Fatigue life
crack. The extended application of the proposed method to step loading and variable amplitude loading is dis­
Defects
cussed. The question about the applicability of Miner’s rule is also discussed from the viewpoint of continuous
Variable amplitude loading
decrease in fatigue limit with number of stress cycles and crack growth.

1. Introduction where, σ: stress, N: number of cycles to failure, C1, C2: material

constants.
It seems that the phenomenon of metal fatigue was recognized in the The concept of the fatigue limit was not recognized in the Basquin
beginning of the Industrial Revolution in Europe [1]. The terminology of model. Later in 1914, Stromeyer [6] analyzed Wöhler’s data and carried
“fatigue” was first used by Jean-Victor Poncelet, French mathematician and out bending and torsional fatigue tests on various materials to verify the
engineer, in his book in 1841 [2]. Later, F. Brathwaite, British engineer, first existence of definite fatigue limit and determined the relationship be­
used the English term “fatigue” in his paper 1850s. Circa 1850, the fatigue tween the applied stress and the number of cycles N as Eq. (2).
behavior of steels was systematically investigated for the first time by
Wöhler [3,4]. Since the concept of metal fatigue itself was not clear at that σ = σw + C(106 /N)1/4 (2)
time, the mechanics and microscopic aspects of S-N data were not well
understood. It was natural that the exact understanding of metal fatigue was where, σ: stress, σw: fatigue limit, N: number of cycles to failure, C:
very difficult in the days of Wöhler. However, regarding the interpretation material constant.
and prediction of S-N data, what substantial progress have we made since Namely, the Stromeyer law expresses S-N data by truncating the curve of
the days of Wöhler? The objective of Wöhler’s study was the acquisition of Basquin model at fatigue limit σw (Fig. 1). Stromeyer plotted the applied
the basic fatigue data for railroad axles in which the study on the effect of stress σ on the ordinate and C(106/N)1/4 on the abscissa and extrapolated
defects contained in materials on fatigue was not explicitly considered. In the data point to C(106/N)1/4 = 0, i.e. the abscissa = 0, and the ordinate of
the days of Wöhler, the confirmation and determination of endurance cy­ applied stress σ was defined as fatigue limit σw. However, the knee point [7],
cles was the first priority for the safe design of railroad axles. This situation Nknee, was not defined explicitly. It is interesting that Stromeyer noted the
continues to the present day. Thus, currently we have to still continue fa­ relationship between fatigue limit and heat evolution from a specimen
tigue tests and acquire S-N data for safe fatigue design. during fatigue testing [6].
In 1910s, Basquin [5] expressed S-N data obtained by Wöhler et al. in Palmgren [8] proposed a model similar to the Basquin model and the
the form of the following exponential equation as in Fig. 1. Equation (1) Stromeyer law as Eq. (3) based on the endurance test of ball bearings.
is called the Basquin model.
σ = C(aN + e)x + σ w (3)
σ = C1 N C2
(1)
where, C, a, and e are material constants and x is a negative number.
* Corresponding author.
E-mail address: murakami.yukitaka.600@m.kyushu-u.ac.jp (Y. Murakami).

https://doi.org/10.1016/j.ijfatigue.2020.106138
Received 5 November 2020; Received in revised form 30 December 2020; Accepted 30 December 2020
Available online 5 January 2021
0142-1123/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

Fig. 1. Curve fitting models of S-N data. Fig. 2. Composition of S-N curve: Line of finite life regime, knee point Nknee
and fatigue limit σw.
Weibull [9] proposed a model in which the statistical scatter near fa­
tigue limit is considered. Weibull had the principal objection against the (a) The inclined line of finite life region (Nf in logarithmic scale)
usually accepted model of the S-N curve as two straight lines which may which shows longer fatigue life Nf with decreasing stress ampli­
give the impression of a knee without any physical significance. However, tude σ (Fig. 2).
physical or mechanistic factors that cause the scatter have not been dis­ (b) The minimum number of fatigue cycles Nknee (knee point) at
cussed in these models. According to the discussion with Socie [10] in terms which fatigue failure does not occur with additional stress cycles
of his experience that he was requested by an industry engineer who had (Fig. 2).
never conducted a fatigue test or analysis to propose a method of fatigue life (c) The horizontal line of fatigue limit σw below which no fatigue
prediction for a component within ±10% accuracy, he explained to the failure occurs.
industry engineer that it is impossible. Industry engineers who are not
familiar with fatigue phenomena are likely to misunderstand that fatigue The Basquin model [5] as Eq. (1), Stromeyer model [6] as Eq. (2) and the
life prediction should be easier than other technical problems. Weibull model with adding failure probability analysis [9], are made to
Later, even nowadays other modified versions similar to Eqs. (1)–(3) express S-N data such as Fig. 2 by a curve fitting technique. Although these
have been repeatedly proposed [11]. However, these models, though models have a name of model and gave some contributions to the fatigue
having the name of a model, are basically made by curve fitting procedures research and fatigue design in the early stage of fatigue study, they were not
for fatigue test data and there has been no discussion based on mechanics of a model made on the basis of the physical meaning of the data points. Even if
cracks or defects which should critically influence the form of an S-N curve. we manipulate the equations of these models, we cannot derive a mecha­
The equations obtained by curve fitting procedure are naturally dependent nistic understanding of fatigue phenomena. What is necessary for reliable
on specimen shapes, specimen sizes and materials, though it is well known fatigue design is to elucidate the essential structure of S-N curve and the
that the qualities of materials having even the same name are different related influential factors based on the fatigue process from the beginning of
depending on suppliers of every material, for example commercial carbon stress cycle to the final fracture.
steels, tool steels, bearing steels (such as SAE52100), spring steels and
aluminum alloys. 2. S-N data of materials containing defects and the essential
Therefore, an equation of S-N curve fitted to a material supplied by one meaning of S-N curve
manufacturer does not necessarily represent the fatigue properties of other
materials with the same composition and heat treatment produced by other 2.1. Correlation among S-N curves of series of specimens containing
suppliers and accordingly the data fitting procedure has to be repeated on defects with different sizes
every material and supplier. Such a method cannot be used especially for
the prediction of S-N curve of materials containing defects. Moreover, as If we test specimens containing defects, the S-N data will be an S-N
explained later, it is difficult to use such types of models for the prediction of curve lower than defect free specimens. This feature is schematically
fatigue life under variable amplitude loading. It must be noted that Miner’s expressed as Fig. 3. Curves A, B and C show respectively the S-N data of
rule, Palmgren-Miner rule [8,12] or modified Miner’s rule cannot be used in the specimens containing a defect with size dA, dB and dC (dA < dB < dC).
combination with this kind of curve fitting approach, because it is difficult Naturally the fatigue lives Nf are short in the order of Series C, Series B
to correctly assess the fatigue damage by stress cycles with an amplitude and Series A (NfC < NfB < NfA) and the fatigue limits decrease in the
below the fatigue limit. It must be also noted that Miner’s rule and Modified order σwA, σwB and σwC (σwC < σwB < σwA). The slope of the S-N curve, C2
Miner’s rule (or its variations [13–18]) do not consider the sequence effect in Eq. (1), is nearly constant although a slight decrease is observed for
of load application on fatigue damage counting. Thus, the models made larger defect sizes. Historically the determination of S-N curve as Fig. 3
without the basis of mechanics fail in the fatigue life prediction problems of has been an important objective of fatigue testing. However, S-N curves
defective materials under variable amplitude loading. This paper analyzes such as Curves A, B and C have been treated only as separate experi­
the essential structure of S-N curve and presents the mechanistic view of the mental results and the correlations among these curves have not been
fatigue phenomena which is useful for the prediction of fatigue life and studied in depth.
fatigue limit and can contribute to reliable fatigue design in presence of S-N data for materials manufactured by the additive manufacturing
defects. (AM) process usually show a large scatter in fatigue lives and fatigue
S-N curve of high cycle fatigue is composed of the following three limit. The scatter is caused by the different size of defects at fatigue
parts. fracture origins, i.e. mostly the largest defect contained in the surface
layer of an individual AM specimen. As an assumption, if several AM

2
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

√̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅
area: μm. area is the representative size of a defect and is defined by
the square root of the area obtained by projecting the defect onto a plane
perpendicular to applied stress. The constant C is C = 1.43 for a surface
√̅̅̅̅̅̅̅̅̅
defect and C = 1.56 for an internal defect [20,29,30]. The area
parameter model can be used as the powerful tool for the prediction of
fatigue limit, the horizontal line of S-N curve. If a specimen does not
contain a definite defect, the fatigue limit can be approximately esti­
mated by σw = 1.6HV [31,32]. The details of the model are explained in
Ref. [20,29,30]. Using this model, the fatigue limits (for example σwA,
σwB and σwC in Fig. 3), namely the horizontal lines of Fig. 2 can be
predicted.
Therefore, if we have two fatigue data points in the finite life regime
(Fig. 2) as the minimum condition, we can approximately determine the
knee point Nknee as the intersection point of the finite life line and the
horizontal line of the fatigue limit σw. The existence of the knee point is
mechanically related to the fact that fatigue limit is determined when an
initiated crack after some initial growth [20,29,30] becomes non­
propagating. It follows that the fatigue limit is not a critical stress for
Fig. 3. S-N data for specimen Series A, B and C containing defects with crack initiation but the threshold stress at which the initiated crack
different sizes. The sizes of defects are dA, dB, and dC (dA < dB < dC) for spec­ ceases growth. Cracks can become nonpropagating when they reach a
imen Series A, B and C.
microstructural barrier such as a grain boundary. In many materials the
specimen series having a defect with the same size in each series are crack closure phenomenon [33–35] is the dominant mechanism. Three
prepared, separate S-N curves for each defect size will be obtained as in types of crack closure mechanisms were reported, plasticity induced
Fig. 3. However, since it is impossible to prepare such specimen series crack closure [33–35], oxide induced crack closure [36,37] and surface
intentionally, apparently one set of S-N data which are composed with a roughness induced crack closure [38,39]. Among these three mecha­
mixture of separate S-N data such as Curves A, B and C are obtained in nisms plasticity induced crack closure [33–35] is the most important
actual fatigue test and a large scatter will be observed. So far, this kind of mechanism. Without understanding the phenomenon of nonpropagating
S-N data or S-N curves has been reported without further pursuit as the cracks, it is impossible to understand the essential structure of S-N curve.
typical fatigue properties of various materials. Although some papers In some exceptional problems such as very high cycle fatigue (VHCF),
treat the scatter of fatigue lives in S-N data as a normal distribution, fatigue cracks initiate at nonmetallic inclusions which trap hydrogen
there is no theoretical justification that the scatter of S-N data obeys the and the hydrogen enhances the elimination of distinct fatigue limit
normal distribution as indicated by Murakami et al. [19,20]. If we as­ (knee point) by the mechanism of hydrogen induced fatigue crack
sume that the gauge part of one cylindrical fatigue specimen is growth [20,40]. Another case of the elimination of distinct fatigue limit
composed of 10 sub-smaller circular plate specimens, the fatigue failure is VHCF of Ti alloys in which crack initiates at a large internal α phase or
of the specimen occurs at the weakest sub-specimen. Although the fa­ at a group of α phases adjacent to each other having preferential
tigue strengths of the 10 sub-specimens in one specimen are thought to orientation to applied stress [41,42]. In these cases, the clear knee point
obey the normal distribution, the statistical distribution of the fatigue due to the mechanism of nonpropagating crack does not necessarily
strength of the failed specimens should obey the statistics of extremes appear and the phenomena must be treated as exceptional cases.
distribution [19,21–28].
This paper elucidates the essential structure of S-N curves by 2.3. Structure of S-N curve: Unification of the data by the normalized S-N
decomposing the mutually important correlations of S-N curves such as curve
Curves A, B and C in Fig. 3 from the viewpoint of the effects of small
defects and small cracks. As explained later, this viewpoint is a basic Fig. 3 schematically illustrates a group of S-N curves of specimens
condition for the fatigue life prediction under variable amplitude containing defects with different sizes. If the stress σ higher than fatigue
loading. limit σw for each defect, dA, dB and dC (dA < dB < dC), is applied, fatigue
failure occurs in a finite life. Therefore, the driving force is not the ab­
solute value of the applied stress σ but should be measured as the value
2.2. Fatigue limit of specimens containing defects and the knee point of S-
of the stress σ relative to fatigue limit σw.
N curve
Namely, σ/σw represents the driving force and accordingly de­
termines S-N curve. In other words, the relative value of σ/σw for σ/σw >
In order to elucidate the essential structure of S-N curve, the fatigue
1 effectively controls the fatigue life Nf. Thus, plotting S-N data of Fig. 3
limit σw, the knee point, Nknee, and the slope of the S-N curve will be
in terms of σ/σw in the ordinate, we should have another type of S-N
investigated on specimens containing defects for various materials. First
curves as in Fig. 4. Although the knee points, Nknee, for Curves A, B and C
of all, the horizontal line of S-N curve, the fatigue limit σw and the size of
√̅̅̅̅̅̅̅̅̅ may not be exactly the same value, the values of Nknee for the cases of
defect, area, at fatigue fracture origin will be investigated. The fatigue
√̅̅̅̅̅̅̅̅̅ small defects are within a narrow range for many materials.
limit σw of specimens containing a small defect with the size area has
Test data for a medium carbon steel is shown in Fig. 5 (a). Fig. 5 (b)
been studied in detail and can be predicted for many metallic materials
√̅̅̅̅̅̅̅̅̅ shows the same data normalized by the fatigue limit σ/σw. This S-N
by Eq. (4) of the area parameter model [20,29,30]. curve will be called the normalized S-N curve in the following discussion
√̅̅̅̅̅̅̅̅̅
σ w = C(HV + 120)/( area)1/6 (4) and is expressed as

N = C(σ/σ w )m for σ > σ w (5)

where, the units of the quantities in Eq. (4) are σw: MPa, HV: kgf/mm2,

3
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

pore. Fig. 6 (a) shows the original S-N data of Ti-6Al-4V, and Fig. 6 (b) is
the normalized S-N data made by the method of Fig. 4 [44,45]. The
values of σw of the specimens which ran out 107 cycles at the previous
stress σ were tested again at a higher stress until the specimen fractured
√̅̅̅̅̅̅̅̅̅
and the size of the defect at fracture origin, area, was identified to
calculate the value of σ/σw.
Fig. 7 (a) and (b) show the S-N data and the normalized S-N data for
AM Ni-based superalloy 718 manufactured by two different processes,
Batches A and B [46]. It is interesting to see that although the original S-
N data for Batch A are plotted separately from Batch B, the normalized S-
N data are almost overlapped. This is because the size of the defect at
fracture origin is taken into consideration for calculation of the fatigue
limit σw of individual specimens.
Fig. 8 shows the S-N data and the normalized S-N data made by
replotting the original data reported by Meneghetti et al. [47]. It must be
noted that, in Fig. 8(b), there are many specimens which failed below
the predicted fatigue limit σw, i.e. σ/σw < 1. The reason will be discussed
in detail in Section 5 (Causes of scatter of S-N data).
Fig. 4. σ/σw - N curves (Normalized S-N curves) made from Fig. 3. Fatigue limit The sizes of defects in AM materials are in general not constant and
√̅̅̅̅̅̅̅̅̅
σw, σw = C(HV + 120)/( area)1/6. have a large scatter. The size and number of defects are larger in low
quality AM materials. Since the individual AM specimens contain
The S-N curves normalized by the fatigue limit σw calculated by Eq. different defects in size, location and number, the fatigue limit of one
(4) play an extremely important role to understand the structure of S-N series of specimens is not constant and then the fatigue life also varies
curves. For example, as explained later, the analysis of S-N data based on depending on individual specimens. This fact has not been recognized in
the method of Fig. 4 is useful for the interpretation of S-N data of AM some existing papers and the scatter of fatigue limits and fatigue lives
materials in which the size of defects has a large scatter. In the following, have been interpreted in the manner of Weibull model. Now, if we as­
the normalized S-N data such as Fig. 4 for various materials will be sume that we can prepare a series of specimens which contain a defect
investigated. with the same size as shown Fig. 9, we can apply the same method
explained in Section 2 for the artificial defect.
3. Interpretation of S-N data of additively manufactured (AM) However, since in reality the size of defects necessarily has some
materials scatter, S-N curves cannot be expressed with one representative S-N
curve and the individual data points are interpreted as one point on the
It is well known that S-N data of AM materials have a large scatter. several groups of S-N curves explained in Section 2. The several groups
AM materials are manufactured by special processing techniques of S-N curves such as Curves A, B and C in Fig. 3 and Fig. 4, which are not
different from those for wrought materials and it is therefore difficult in recognized explicitly in AM materials, have been simply interpreted as
principle to avoid various types of defects such as lack of fusion and gas the scatter of fatigue life. In such a case, the width of the scatter band of

Fig. 5. Normalized S-N data. Material: Annealed 0.46% carbon steel.

4
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

Fig. 6. S-N data of AM Ti-6Al-4V and the normalized S-N data of Ti-6Al-4V [44,45]. EBM: Electron Beam Melting, DMLS: Direct Metal Laser Sintering. Rotating
bending fatigue.

Fig. 7. S-N data of AM Ni-based superalloy 718 and the normalized S-N data. Tension-compression fatigue.

Fig. 8. S-N data of AM maraging steel [47] and the normalized S-N data. Tension-compression fatigue.

5
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

defects. Although the difference in fatigue lives at a constant stress level

looks like scatter, this scatter almost disappears in the normalized S-N
data of Fig. 11 (a-2). This trend is the same in Fig. 11 (b-2), (c-2) and (d-
2). Note that the data points for 300M steel [52] in Fig. 11 (d-1) shows
the largest scatter in fatigue life at the same stress level, the data are
uniquely rearranged in the normalized S-N data in Fig. 11 (d-2). The
data points for 300 M steel fall above σ/σw of one in Fig. 11 (d-2). This is
because the specimens have been shot peened [52] and have compres­
sive residual stress at the fatigue fracture origins, but the fatigue limits
were calculated by Eq. (4) for zero mean stress.
The knee point Nknee of steels having Vickers hardness lower than
HV = 400 is distinct and ranges around Nknee = ~105–106. On the other
hand, Nknee for heat-treated steels such as bearing steels and low alloy
steels with HV > 400 is not distinct as Fig. 11 (b-1) and (b-2). The fatigue
fracture origins of these hard steels are mostly at nonmetallic inclusions.
The reason for nondistinct knee point is presumed to be the influence of
hydrogen trapped at nonmetallic inclusions [20,40]. The effect of
Fig. 9. Defect at fatigue fracture origin observed at fatigue fracture origin of hydrogen on VHCF is attributed to hydrogen induced fatigue crack
√̅̅̅̅̅̅̅̅̅
AM Ti-6Al-4V by EBM without HIP [44,45]. The size of the defect, area = 454 growth which weakens the plasticity induced crack closure mechanism
μm. This defect is plotted with the mark ○* in the statistics of extremes data and eventually the distinct knee point disappears [20,40].
of Fig. 10.

fatigue lives can be predicted by using the statistics of extremes data 5. Causes of scatter of S-N data
(Fig. 10) [44,45] of the size of defects observed at the fatigue fracture
origin (e.g. Fig. 9) in combination with the method of the normalized S- There are three main factors as the cause of scatter of S-N data.
N curves (Fig. 4). If such predictions are necessary before fatigue test,
similar defect data of statistics of extremes on the as-processed AM • Factor 1: Scatter of defect sizes contained in materials.
material can be used for this purpose [44–46]. • Factor 2: Variability of the microstructure.
• Factor 3: Poor alignment of testing machine and specimens.
4. S-N data of other materials and the normalized S-N data
Factor 1: Scatter depends on presence of initial defects. If the scatter
Fig. 11 shows S-N data and the normalized S-N data for various steels of defects is small, the scatter of fatigue life and fatigue limit is small.
[43,48–52] including several data points for specimens containing Typical examples have been shown in Fig. 11.
Factor 2: In smooth specimens containing no definite defect, a crack
which initiates in the microstructure corresponds to an initial defect. In
this case, fatigue crack initiates in a grain preferentially oriented to the
direction of applied stress or along grain boundary. Such a fatigue crack
is mechanically equivalent to an initial defect but the size is dependent
on the grain size and the orientation. For such reasons, the scatter of
fatigue data of smooth specimens is larger than that of specimens con­
taining an artificial defect with constant size larger than grain size.
Factor 2 influences Factor 1 to some extent if there is variation of
microstructure around the initial defect. It must be noted that many
grains yield and plastic deformation occurs at high applied stresses, so
that mechanical behavior of microstructure is averaged over many
grains resulting in less scatter of fatigue lives Nf. However, by decreasing
the applied stress close to fatigue limit, the scatter of crystallographic
slip behavior (plastic deformation) among grains increases inside one
specimen and the growth of initiated cracks is influenced by the un-
yielded grains resulting in large scatter in crack growth rate and fa­
tigue lives Nf. Presence of many nonpropagating cracks at the fatigue
limit in low and medium carbon steels is the evidence of this phenom­
enon [20,29,30]. It seems that this phenomenon occurs in combination
with the crack closure mechanism which is sensitive to crack tip
deformation.
Factor 3: It must be noted that poor alignment of the testing machine
and specimens can frequently be a cause for the scatter in fatigue life and
strength even if specimens contain no apparent defect. For example, it is
well known that machine vibration in rotating bending fatigue test
significantly influences the test results. The typical causes of machine
vibration in rotating bending test are as follows, (1) the poor alignment
of specimen attachment to machine, (2) the resonant vibration of testing
Fig. 10. Statistics of extremes of defects at fracture origin observed at fatigue machine due to the loss of damping capability by the temperature rise of
fracture origin of AM Ti-6Al-4V alloy [44,45]. V0 = ~65 mm3: Control volume bearings sometime after start of operation when the number of machine
of surface annual zone of one specimen under stress higher than 90% of rotation is close to the resonance vibration, and (3) building vibrations
nominal stress. The second largest defect of EBM marked by ○* is shown during fatigue testing [53].
in Fig. 9. As an additional factor of Factor 3, we need to pay attention to the

6
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

Fig. 11. S-N data of various materials and the normalized S-N data. Source data (a-1), (b-1), (c-1) and (d-1). Normalized S-N data (a-2), (b-2), (c-2) and (d-2).

7
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

effect of residual stress. Regularly distributed residual stress introduced approximate S-N curve? Knowing the minimum number of data for
on purpose by several industrial techniques to increase fatigue strength reliable prediction of an S-N curve is useful for efficient fatigue tests and
does not become the main cause of scatter of fatigue strength and life. minimizing number of fatigue specimens. However, needless to say this
However, we have to note that irregularly distributed residual stress plan can be achieved based on the condition of correct alignment of
created by such as the additive manufacturing process [54,55] causes testing machines and specimens.
distortion of specimens and eventual scatter of fatigue strength and life. According to the method explained in this paper, the necessary and
If fatigue tests are carried out using specimens with distortion or a minimum data for determining an S-N curve are the Vickers hardness HV
√̅̅̅̅̅̅̅̅̅
fatigue testing machine with poor alignment, the resultant S-N data have of the material in question, the size of the defect, area, and two data
no consistency and will have a large scatter. Discussion on fatigue points. As shown in Fig. 2, connecting two data points, S-N curve above
mechanisms and models based on such inconsistent data does not fatigue limit σw can be approximately determined. This line is denoted
guarantee the correct solution. Usually it has not been explicitly by Line (a). The horizontal line of fatigue limit σw is denoted by Line (c)
mentioned in published papers that the alignment of testing machine which can be determined by Eq. (4). The intersection of Line (a) and Line
and specimens is precisely checked and this problem has not been dis­ (c) is the knee point Nknee.
cussed in public. The detailed explanation by Meneghetti et al. [47] on If additional data are prepared, the value of slope of S-N curve, m,
this problem gives a useful and constructive information to be consid­ and the value of C in Eq. (5) can be revised for determination of more
ered. Fig. 8 is the replot of the data reported in Ref. [47]. Fig. 8 (b) (AM reliable S-N curve. If we can establish this method, it is not necessary to
maraging steel) has clearly revealed many data points of fatigue failure carry out so many fatigue tests.
under σ/σw < 1 in the normalized S-N curve. The main cause of the The advantage of the formulation of S-N curve of this method is not
unusually low fatigue strength is presumed to be the distortion of limited to one series of specimens containing a defect with the same size.
specimens as pointed out by the authors of Ref. [47]. According to the Extending the method in combination with the crack growth rate anal­
discussion by the authors of Ref. [47], there is a possibility that the ysis as explained in Section 7, S-N curves for other series of specimens
transition zone of the threshold stress intensity factor ΔKth from small containing defects with different size can be predicted.
crack to long crack is influenced by the specimen distortion and residual Another advantage of the present method is the application to the
stress. The authors of the present paper have recognized the problem of prediction of the scatter band in S-N curve for the materials such as AM
the transition zone from small crack to long crack [20,56–59]. However, materials containing defects with various sizes. Although the size of the
it is very difficult to discuss the scatter problem of Fig. 8 (Ref. [47]) from defect at fatigue fracture origin for AM materials, cast irons, powder
the viewpoint of the transition zone, because the data are presumed to metals and hard steels is unknown before fatigue testing, the scatter
include the effects of mixture of several factors such as residual stress, band of S-N data can be predicted with minimum number of fatigue tests
specimen distortion and the crack size transition zone. in combination with the statistics of extremes data of defects and
According to the survey by the authors of the present paper, the nonmetallic inclusions.
problem of poor alignment of testing machines and specimens is Fig. 12 shows an example of the method applied to the prediction of
becoming more serious. If fatigue fracture origins of hard steels having the lower band (the shortest fatigue life) of the scatter of S-N curve for
HV > 400 are mostly at specimen surface, bent specimens should be AM material of Ti-6Al-4V. In the prediction, the statistics of extremes
suspected. The brief conclusion on this problem is that the quality of data of Fig. 10 was used to predict the largest defect by assuming the
fatigue test data is different depending on testing laboratories in uni­ number of specimens 200. Numbers in the figure represent the values of
√̅̅̅̅̅̅̅̅̅
versities and industries. If unusually low fatigue strength data are re­ area for each specimen. Thus, the collections of the statistics of ex­
ported [60–62], this factor should be suspected. tremes data of defects for AM materials [44–46,63,64] will be useful for
the determination of the lower bound of fatigue limit and fatigue life.
6. The minimum data necessary for the prediction of S-N curve

Large numbers of fatigue tests have been carried out all over the
world. How many fatigue tests are necessary for quick determination of

Fig. 12. Lower bound (the shortest fatigue

life) for the fatigue life of the additively
manufactured Ti-6Al-4V calculated based the
statistics of extreme values of defects for 200
specimens (Fig. 10) [44,45]. Details of the
computations are given in the Appendix as
well as determining the constants C* and m*
(C* = 10− 4.2, m* = 2.0) in Eq. (7). The pre­
√̅̅̅̅̅̅̅̅̅
dicted largest values of defects, areamax, for
√̅̅̅̅̅̅̅̅̅
200 speciemns are areamax = 183 μm for
√̅̅̅̅̅̅̅̅̅
DMLS and areamax = 831 μm for EBM from
Fig. 10. Numbers in the figure associated
with the data points indicate the value of
√̅̅̅̅̅̅̅̅̅
area at fracture origin in each specimen.
DMLS: Direct Metal Laser Sintering, EBM:
Electron Beam Melting.

8
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

7. Perspective for fatigue life prediction under variable limit for a critical crack size.
amplitude loadings: Conditions for applicability of Miner’s rule Historically, modified Miner’s rule was proposed as a convenient
solution. In modified Miner’s rule, the straight line of finite life regime of
Service loadings in machine components are usually not constant S-N curve is extended downward below the initial fatigue limit (Fig. 13)
amplitude. Therefore, fatigue life prediction of actual components and and a fictitious fatigue life (the dashed line in Fig. 13) is assumed for
structures under variable amplitude loadings is of practical importance. convenience. However, the mechanistic basis of Miner’s rule has not
Miner’s rule [12] is often used for fatigue life prediction under variable been fully discussed in case of high cycle fatigue [20,65]. If we simply
amplitude loading because of simplicity. In Miner’s rule, fatigue damage apply this rule to actual components, an unusually large safety factor
Di is linearly accumulated such that one cycle of the stress σi produces a may be necessary in some service loading conditions.
fatigue damage of 1/Ni where Ni is the constant amplitude fatigue life at However, if we understand correctly the essential structure of the S-
a stress of σi. In a variable amplitude loading history the total accumu­ N curve of specimens containing defects, we can consider the continuous
lation of fatigue damage D is calculated as follows. decrease in fatigue limit σw with number of cycles for fatigue design. If
∑ subsequent applied stress σi+1 is lower than the fatigue limit σw at the
D= (ni /Ni ) (6) end of stress σi, the damage by σi+1 can be ignored. Thus, for the correct
estimation of the fatigue damage, D, it is rational to consider the cor­
where, the individual cycles, ni: each with their own mean and ampli­ relation between the applied stress σi and the fatigue limit σw which
tude is identified with a suitable cycle counting scheme. continuously decreases with number of cycles, i.e. growth of crack. In the
Here, how should we count the contribution of a stress σi below σw to stage of the present discussion, to avoid the complicated phenomenon in
D? In Miner’s rule, a cycle of stress below the initial fatigue limit σw,initial small crack problems, we ignore the effect of retardation or acceleration
will not cause any fatigue damage. In this discussion, we must note that of crack growth after overloading.
the value of σw continuously decreases with crack growth as a function of One example of this method will be introduced in the following. First
crack size throughout the lifetime, and the fatigue life Nf at the stress of all, it is necessary to elucidate the correlation between the S-N curve
below σw becomes ∞. This situation is schematically explained by with the form Nf = C(σ/σw)m and the growth process of small crack. The
Fig. 13. Therefore, as a result, stress cycles of σi < σw,initial that were basic approach adopted here is similar to the method explained in the
initially nondamaging become damaging once σi exceeds the fatigue reference [20,65]. Then,
da
= C* (σ/σ w − 1)m* an* (7)
dN
√̅̅̅̅̅̅̅̅̅
where, a = area is defect size, and C*, m* and n* are material
constants.
It must be noted here that σw is the fatigue limit which decreases with
√̅̅̅̅̅̅̅̅̅
every cycle continuously as a function of a (= area) and that σw can be
expressed by Eq. (4). The term (σ/σw – 1) in Eq. (7) is a function of a
√̅̅̅̅̅̅̅̅̅
(= area) and expresses the effective driving force, which increases
cycle-by-cycle since σw decreases with an increase in a. By integrating
this equation, Nf and growth process of small crack can be connected.
Since Eq. (7) cannot be integrated in a closed form, the numerical
integral is needed to calculate predicted Nf. Presuming n* = 1 is
reasonable by considering the data reported in existing papers [66–69],
at least two data points are necessary to determine C* and m*. This
tentative assumption of n* = 1 is to be reasonable if we consider the
analogy of the present model to the Dugdale model [70], in which the
Fig. 13. Concept of Modified Miner’s rule. plastic zone size ahead of crack tip, rp, is linearly proportional to crack

Fig. 14. Comparison between the original data and predictions by using two selected data points. (a) the original S-N data [43] and (b) normalized S-N data.

9
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

length. If the prediction results contradict with the experimental results, test. The completely reversed tension-compression fatigue test with R =
the value of n* must be modified. Once C*, m* and n* are determined − 1 was performed by MTS servo-hydraulic fatigue testing machine
based on two or three fatigue data points, Eq. (7) can be used extendedly under frequency of 10 Hz for short fatigue life test and 20 Hz for rela­
to other arbitrary loading history. Details of the numerical integral tively long fatigue life test close to fatigue limit.
applied in the present study are explained in the Appendix. Specimens were machined after annealing the original 0.47% C (JIS
First, the method was applied to an annealed medium carbon (0.46 S45C) steel bars. The distortion of all the specimens after machining was
%C, JIS S45C) steel under constant amplitude loading. Fig. 14 shows (a) confirmed to be < 10 μm at the mid point of the specimens. Afterward,
the original S-N data [43] and (b) the normalized S-N data, which were the specimens were polished by emery papers followed by electro­
obtained at a stress ratio, R, of − 1 using specimens having various sizes polishing for the observation of crack growth behavior. In order to check
of drilled hole. Selecting the two experimental data points obtained for the alignment of fatigue testing machine, four strain gauges were
√̅̅̅̅̅̅̅̅̅ √̅̅̅̅̅̅̅̅̅
area = 185 μm (cf., red diamond in Fig. 14) and area = 74 μm (cf., attached to every specimen before test at the equal quarter points of the
red triangle in Fig. 14), we determined the constants as C*= 10− 4.4, m* circumferential direction at the part of 13 mm diameter (Fig. 15 (a)).
= 2.4 and n* = 1.0 based on Eq. (7). The predictions based on Eq. (7) for The bending strain for setting specimens to fatigue testing machine was
other defect sizes are in good agreement with the experiments. strictly controlled within ± 2.5 με.
The second application of the method was planned and performed as A drilled hole with 100 μm diameter (d) and depth (h), d = h = 100
√̅̅̅̅̅̅̅̅̅
Fig. 15. The fatigue behaviors were predicted for medium carbon steel μm ( area = 92 μm), (Fig. 15 (a)) was introduced onto specimen
specimens containing a small artificial defect under a two-level stress surface.

Fig. 15. Basic experiments for investigating step loading.

10
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

The initial fatigue limit σw,initial of the specimen containing the 100 stresses, σ1 = 270 MPa (Nf 1 = 30,407 cycles) and σ2 = 235 MPa (Nf2 =
μm hole can be predicted by Eq. (4) to be σw,initial = ~199 MPa for HV = 120,264 cycles), C*, m* and n* in Eq. (7) were determined as C* =
176 (Average of 12 measurements with standard deviation 4.7). 10− 3.4, m* = 2.8 (and n* = 1.0) by the method explained in Appendix.
Four fatigue tests were conducted as follows to evaluate the validity The resulting S-N curve from the model is shown as the solid and dashed
of the method. line in Fig. 15 (b). These values, C*, m* and n*, were used to predict the
total fatigue life for the step loading test by integrating Eq. (7).
(1) Fatigue tests:
(3) Fatigue life prediction for step loading tests:
Two constant-amplitude tests were conducted, at σ1 = 270 MPa with
a resulting fatigue life of 30,407 cycles and at σ2 = 235 MPa with a The third test: The calculated number of cycles Nstep 1 for crack growth
√̅̅̅̅̅̅̅̅̅
fatigue life of 120,264 cycles. to area = 251 μm under σ1 = 270 MPa was Nstep 1 = 23,000 cycles. At
√̅̅̅̅̅̅̅̅̅
The third test of σ1 = 270 MPa for 22,300 cycles followed by σ3 = 185 the second step, the number of cycles Nstep 2 for crack growth to area =
MPa (10% higher than the predicted fatigue limit σw2) until failure of an 1000 μm under σ3 = 185 MPa was calculated as Nstep 2 = 434,000 cycles.
additional 562,430 cycles. In this calculation, the cycles after crack size 1000 μm was neglected,
The fourth test of σ1 = 270 MPa for 23,400 cycles followed by σ4 = because it is well known that remaining life after crack size 1000 μm is
135 MPa (20% lower than the predicted fatigue limit σw2) was stopped very short. Then, the calculated total number of cycles to failure Nf, step 1
after an additional 107 cycles. + step 2 = Nstep 1 + Nstep 2 = 23,000 + 434,000 = 457,000 cycles (the red
The crack length measured on specimen surface at the moment of square mark □ in Fig. 15 (b)). Since the experimental value is Nf, step 1 +
√̅̅̅̅̅̅̅̅̅
stress-switching in the third and fourth tests was ~ 400 μm ( area = step 2 = 585,830, the prediction is considered acceptable.
~250 μm). The fourth test: At the first step, σ1 = 270 MPa was applied for 23,400
cycles. Fig. 16 (a) shows the crack before switching to the second step.
(2) Determination of C*, m* and n*: The total length of crack emanating from the hole of 100 μm diameter
was 392 μm.
Based on two fatigue data obtained by tests under constant aplitude The second step test of σ4 = 135 MPa (20% lower than the predicted

Fig. 16. Crack growth and nonpropagation at the step loading test (the fourth test) by (a) the first step, σ1 = 270 MPa and N = 23,400 cycles and (b) the second step
test, σ4 = 135 MPa (20% lower than the predicted fatigue limit σw2) after an additional N = 107 cycles.

11
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

Fig. 17. Step loading having the alternation of low and high stress cycle blocks.

fatigue limit σw2) was stopped after an additional 107 cycles. Fig. 16 (b) (1) Existing models for S-N curves, though having the name of a
shows the nonpropagating cracks emanated from the crack of Fig. 16 (a) model, are basically made by curve fitting procedures and cannot
after switching the stress from σ1 = 270 MPa to σ4 = 135 MPa for N = be applied to the prediction of fatigue life and fatigue limit of
107 cycles (Runout). The total crack length was 432 μm including the materials containing various defects.
length l of nonpropagating cracks, lleft = 12 μm and lright = 28 μm. Thus, (2) Replotting the S-N data to the normalized S-N data, σ/σw vs N, is
the result of the fourth test verifies that the application of stress below useful for analyzing the essential structure of S-N curves and for
the fatigue limit calculated by Eq. (4) for the crack size at moments is checking the validity of the original S-N data. If many data are
nondamaging and it is not necessary to count the cycles as a damage. plotted below the predicted fatigue limit σw, based on the initial
√̅̅̅̅̅̅̅̅̅
The results of these fatigue tests reveal the simple mechanism of the size of defect area i.e. at σ/σw < 1, the bad alignment of testing
effect of step loading and sequence on fatigue damage. machines and specimens must be suspected.
By considering the decrease in fatigue limit with crack growth under (3) The effective driving force for fatigue crack growth is not the
variable amplitude or step loading, we can quantitatively count the absolute value of the applied stress σ. The net amount of the stress
damage caused by varying stress amplitudes based on the model of Eq. above the fatigue limit (σ/σw – 1) determines the fatigue crack
(7). For example, if a stress amplitude σ is lower than σw at the end of growth rate. In this analysis, it must be noted that the value of σw
previous stress cycles, the effect of the additional cycles of σ should not continuously varies at throughout the lifetime until the final
be counted as fatigue damage. fracture occurs.
If we assume a step loading like Fig. 17 in which the blocks of low σl (4) S-N curves of specimens having defects with different initial sizes
and high stress σh cycles are alternately repeated, we need to note the have mutual correlations. If S-N curve of one series of specimens
variation of fatigue damage given by these blocks from the early stage of is determined, S-N curves of other series can be predicted based
loading to the later stage of fatigue loading. Even if the damage given by on the crack growth analysis developed by the S-N curves of the
the low stress block at the early stage of loading can be ignored, the first series of specimens. The scatter of S-N data of AM materials
damage given by the same block of stress cycles at the later stage is can be predicted by this analysis in combination with the statis­
gradually induced, because the crack grows at the beginning by the tics of extremes of defects such as lack of fusion.
repetition of the block of high stress σh and eventually after the size of (5) The analytical method for S-N curves developed in this paper can
√̅̅̅̅̅̅̅̅̅
the crack exceeds the critical value areacrit for low stress σl, the block of be extensively applied to fatigue under step loadings and variable
√̅̅̅̅̅̅̅̅̅
low stress σl contributes to the damage. The critical value areacrit is amplitude loadings. The reason for unsuccessful results of appli­
calculated by equating the low stress σl to the fatigue limit σw in Eq. (4). cation of Miner’s rule is pointed out in terms of continuously
Thus, even if the block of low stress σl is non-damaging at the early stage decreasing fatigue limit σw under fatigue cycling.
of stress cycle, the effect must be taken into consideration after some
repetitions of the block of high stress σh. At the same time, the contri­ Funding
bution of one block of high stress σh on fatigue damage is not constant
from the beginning to end but continuously varies as well. This means This research did not receive any specific grant from funding
that the sequence of low and high stress applications definitely in­ agencies in the public, commercial, or not-for-profit sectors.
fluences the damage and eventually the fatigue life under variable
amplitude loading. Declaration of Competing Interest
The results of the more systematic experiments and detailed analyses
of fatigue under several step loadings and variable amplitude loadings The authors declare that they have no known competing financial
will be reported in a separate paper based on the method developed in interests or personal relationships that could have appeared to influence
this paper, and the validity of existing models [12–18,71,72] will be the work reported in this paper.
examined.
Acknowledgments
8. Conclusions
The authors would like to thank Prof. Darrell Socie of University Il­
In order to elucidate the essential structure of S-N curve such as σ = linois for his useful discussion and comments throughout completing
C1 NC2 , this paper discussed a prediction method for fatigue life and fa­ this paper. The authors are grateful to Dr. Takeshi Horikawa, Emeritus
tigue limit of materials containing various defects and the nature of professor of Ryukoku University, for useful discussion on the failure of
scatter of fatigue strength from the mechanistic viewpoint of small crack. modified Miner’s rule under the high-low stress and sequence combi­
The conclusions are summarized as follows. nation under step loading fatigue test. The authors are grateful also to

12
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

Dr. Yoichi Yamashita of IHI for useful discussion on the difference be­ problems. The discussion with Dr. Yamashita was successfully reflected
tween the very high cycle fatigue failure from nonmetallic inclusions, in the present paper. The authors would like to thank Dr. Yoshihiro
essentially small crack problems, and the fatigue life in long crack Fukushima for technical assistance with the experiments.

Appendix A. Determination of C*, m* and n*

A1. Definition of the objective function O

The error E of the prediction Nf pred compared to the experimental value Nf was evaluated in a logarithmic scale.
( ) ( )
E = log Nf pred − log Nf (A1)

where, Nfpred was determined by integrating Eq. (7) in the main text by assuming an appropriate incremental cycle ΔN, for example ΔN = 100.
The objective function O (C*, m*, n*) is defined as the average value of square of Ei for one experimental data point as Eq. (A2).
∑M 2
E
O = i=1 i (A2)
M

where, M is the number of data points.

A2. Minimization of the objective function O by variation of C* and m*

In the present paper, n* was assumed to be 1 based on the tendencies of existing data [66–69]. The optimum C* and m* values were determined to
minimize O. To seek an optimum set of C* and m* values, O was calculated for each C* and m*. For this analysis, at least two data-points of (σ, Nf) are
necessary to determine C* and m*.
Once the optimum values of C* and m* are determined, Eq. (7) can be extendedly applied not only to the prediction of fatigue lives of specimens
containing defects with various sizes under constant amplitude loading but also to the fatigue damage counting for step loadings.

A3. Results of analysis

The optimum values of C* and m* for Ti-6Al-4V were determined to be 10− 4.2 and 2.0, respectively. The comparison between the prediction and
the experimental data are given in Fig. A1. In Fig. A1, two data points marked by large red symbols were selected for the calculation. The predictions
including fatigue limits are in good agreement with the experimental data.

Fig. A1. Comparison between original data and predictions. Two large red symbols indicate the datasets selected for the calculation of C* and m*. HV = 369 for EBM,
HV = 378 for DMLS (see also Fig. 6).

References [4] Wöhler A. Versuche zur Ermittlung der auf die Eisenbahnwagenachsen
einwirkenden Kraft und die Widerstandsfahigkeit der Wagen Achsen. Z Fur Bauwes
1860;10:583–614.
[1] Ohnami M. Fracture and society. Ohmsha 1992.
[5] Basquin OH. The exponential law of endurance tests. Am Soc Test Mater Proc 1910;
[2] Poncelet JV. Introduction a la mecanique industrielle, physique ou experimentale.
10:625–30.
Thiel 1841.
[6] Stromeyer CE. The determination of fatigue limits under alternating stress
[3] Wöhler A. Bericht uber die Versuche, welche auf der Konigl. Niederschlesisch-
conditions. Proc R Soc London Ser A, Contain Pap a Math Phys Character 1914;90:
Markischen Eisenbahn mit Apparaten zum Messen der Biegung und Verdrehung
411–25.
von Eisenbahnwagen-Achsen wahrend der Fahrt angestellt wurden. Z Fur Bauwes
[7] Battelle Memorial Institute. Prevention of the Failure of Metals Under Repeated
1858;8:641–52.
Stress. New York: John Wiley & Sons, Inc.; 1941. p. 1–295.

13
Y. Murakami et al. International Journal of Fatigue 146 (2021) 106138

[8] Palmgren AG. Die Lebensgauer von Kugellagern (Life length of Roller Bearings. In [43] Murakami Y, Fukuda S, Endo T. Effect of Micro-hole on Fatigue Strength, 1st
German), Z. Vereinesd Dtsch. Ingenieure VDI Z; 1924. ISSN, 0341-7258. Report, Effect of micro-hole (Dia.: 40/50/80/100 & 200 μm) on the Fatigue
[9] Weibull W. A statistical representation of fatigue failures in solids. Elander 1949: Strength of 0.13% and 0.46% Carbon Steels. Trans Japan Soc Mech Eng 1978;A 44:
1–50. 4003–13.
[10] Socie D. Private communication. [44] Masuo H, Tanaka Y, Morokoshi S, Yagura H, Uchida T, Yamamoto Y, et al.
[11] For example, D’Antuono P. An analytical relation between the Weibull and Basquin Influence of defects, surface roughness and HIP on the fatigue strength of Ti-6Al-
laws for smooth and notched specimens and application to constant amplitude 4V. Int J Fatigue 2018;117:163–79.
fatigue. Fatigue Frac Engng Mater Struc 2020;43(5):991–1004. [45] Murakami Y, Masuo H, Tanaka Y, Nakatani M. Defect analysis for additively
[12] Miner MA. Cumulative damage in fatigue. J Appl Mech 1945;12(3):A159–64. manufactured materials in fatigue from the viewpoint of quality control and
[13] Corten HT, Dolan TJ. Cumulative fatigue deamage. In: Proc. of the Intern. statistics of extremes. Procedia Structural Integrity, Fatigue Design 2019;19:
Conference on Fatigue of Metals. London; 1956. p. 235–46. 113–22.
[14] Haibach E. Modifizierte Lineare Schadensakkumulations-Hypothese zur [46] Yamashita Y, Murakami T, Mihara R, Okada M, Murakami Y. Defect analysis and
Berucksichtigung des Dauerfestigkeitsabfalls mit Fortschreitender Schadigung. fatigue design basis for Ni-based superalloy 718 manufactured by selective laser
LBF: Darmsadt, Tech Mitt 1970;50:50–70. melting. Int J Fatigue 2018;117:485–95.
[15] Kawada Y, Misawa H. The effect of stress amplitudes below endurance limit on the [47] Meneghetti G, Rigon D, Gennari C. An analysis of defects influence on axial fatigue
cumulative fatigue life [in Japanese] J Soc Materials Sci Japan 1968;17(173): strength of maraging steel specimens produced by additive manufacturing. Int J
123–7. Fatigue 2019;118:54–64.
[16] Koibuchi K. Study of fatigue damage rule under service loadings [in Japanese] [48] Konuma S, Furukawa T. Relationship between Fatigue Properties and Hardness for
J Soc Materials Sci Japan 1968;17(173):169–72. High Hardness Steels Tempered at Several Temperatures [in Japanese] J Soc Mater
[17] Nakamura H, Horikawa T. An experimental study of stress cycle pattern which Sci Japan 1986;38:1128–32.
gives the minimum value of Σ(ni/Ni) (Service loading and fatigue strength, Report [49] Nisitani H, Ogata T, Endo M. Differences of the fatigue processes of 0.45% C steel
16) (in Japanese). Trans JSME 1970;36–287:1059–66. quenched and tempered at various tempering temperature. Trans Japan Soc Mech
[18] Nakamura H. Historical View on Products, Matters and Lives [in Japanese]. Nagata Eng 1984;A 50: 453:1104–9.
Bunsho-Do; 1997. [50] Nishijima S, Suzuki E, Niikura Y, Uchibori K, Terashita M, Morii A. Inter-laboratory
[19] Murakami Y, Machida H, Miyakawa S, Takagi T. A new quality control method tests on microscopic evaluation of non-metallic inclusions in steels for spring use.
based on statistics of extremes for preventing recalls for mass production products. Trans Jpn Soc Spring Res 1987;32:52–74.
Trans Jpn Soc Mech Eng 2017;83(853):1–16. [51] Kawada Y, Nakazawa H, Kodama S. The effects of the shapes and the distributions
[20] Murakami Y. Metal Fatigue: Effects of Small Defects and Nonmetallic Inclusions. of inclusions on the fatigue strength of bearing steels in rotary bending. Trans Jpn
2nd ed. Elsevier & Academic Press; 2019. Soc Mech Eng Ser I 1963;29(206):1674–83.
[21] Murakami Y. Inclusion rating by statistics of extreme values and its application to [52] Bag A, Delbergue D, Bocher P, Lévesque M, Brochu M. Statistical analysis of high
fatigue strength prediction and quality control of materials. J Res Natl Inst Stand cycle fatigue life and inclusion size distribution in shot peened 300M steel. Int J
Technol 1994;99(4):345–51. Fatigue 2018;118:126–38.
[22] Freudenthal AF, Gumbel EJ. Physical and statistical aspect of fatigue. In: Dryden, [53] Endo M, Matsuo T. A practical method for fatigue limit prediction in ductile cast
Karman T, editors. Advances in Applied Mechanicsm, vol. IV, H-1. New York, NN: irons. Fatigue Fract Eng Mater Struct 2019;42:2106–19.
Academic Press; 1956. p. 117− 58. [54] Meneghetti G, Rigon D, Cozzi D, Waldhauser D, Dabala M. Influence of build
[23] Gumbel EJ. Statistics of Extremes. New York: Columbia University Press; 1958. orientation on static and axial fatigue properties of maraging steel specimens
[24] Murakami Y, Kodama S, Konuma S. Quantitative evaluation of effects of non- produced by additive manufacturing. Procedia Struct Integrity 2017;7:149–57.
metallic inclusions on fatigue strength of high strength steels. I: Basic fatigue [55] Yang Y, Wang D. A study on the residual stress during selective laser melting (SLM)
mechanism and evaluation of correlation between the fatigue fracture stress and of metallic powder. Int J Adv Manuf Technol 2016;87:647–56.
the size and location of non-metallic inclusions. Int J Fatigue 1989;11(5):291–8. [56] Chapetti MD. A simple model to predict the very high cycle fatigue resistance of
[25] Murakami Y, Usuki H. Quantitative evaluation of effects of non-metallic inclusions steels. Int J Fatigue 2011;33:833–41.
on fatigue strength of high strength steels. II: Fatigue limit evaluation based on [57] Schönbauer BM, Mayer H. Effect of small defects on the fatigue strength of
statistics for extreme values of inclusion size. Int J Fatigue 1989;11(5):299–307. martensitic stainless steels. Int J Fatigue 2019;127:362–75.
[26] Beretta S, Murakami Y. Largest-extreme-value distribution analysis of multiple [58] Rigon D, Meneghetti G. An engineering estimation of fatigue thresholds from a
inclusion types in determining steel cleanliness. Met Trans B 2001;32B:517–23. microstructural sizeand Vickers hardness: application to wrought and additively
[27] Beretta S, Anderson CW, Murakami Y. Statistical analysis of defects for fatigue manufactured metals. Int J Fatigue 2020;139:105796.
strength prediction and quality control of materials. Fatigue Fract Eng Mater Struct [59] Romano S, Nezhadfar PD, Shamsaei N, Seifi M, Beretta S. High cycle fatigue
1998;21:1049–65. behavior and life prediction for additively manufactured 17–4 PH stainless steel:
[28] ASTM International. Designation: E2283-03. Standard Practice for Extreme Value Effect of sub-surface porosity and surface roughness. Theor Appl Fract Mech 2020;
Analysis of Nonmetallic Inclusions in Steel and Other Microstructural Features; 106:102477.
2003. [60] Melander A, Rolfson M, Nordgtren A, Jansson B, Hedberg H, Lund T. Influence of
[29] Murakami Y, Endo M. The area parameter model for small defects and nonmetallic inclusion contents on fatigue properties of SAE 52100 bearing steels. Rep Swedish
inclusions in fatigue strength: Experimental evidence and applications. In: Inst Met Res 1990;IM-2589:343–8.
Blom AF, Beevers CJ, editors. Theoretical Concepts and Numerical Analysis of [61] Muralidharan U, Stover JC, Kavadas P. Effect of residual titanium on fatigue
Fatigue. West Midlands: EMAS Ltd; 1992. p. 51–71. resistance of bearing steels. Proc Fatigue ’93 1993;3:1483–9.
[30] Murakami Y, Endo M. Effects of defects, inclusions and inhomogeneities on fatigue [62] Kubota M, Fukuda M, Komoda R. Effect of hydrogen on fatigue limit of SCM435
strength. Int J Fatigue 1994;16(3):163–82. low-alloy steel, Procedia Structural Integrity. Fatigue Des 2019;2019(19):520–7.
[31] Garwood MF, Zurburg HH. Erickson MA. Correlation of Laboratory Tests and [63] Romano S, Brandã A, Gumpinger J, Gschweitl M, Beretta S. Qualification of AM
Service Performance. Interpretation of Tests and Correlation with Service. parts: Extreme value statistics applied to tomographic measurements. Mater Des
Philadelphia, PA: ASM; 1951. p. 1− 77. 2017;131:32–48.
[32] Nishijima S. Statistical analysis of fatigue data. J Soc Mater Sci Japan 1980;29 [64] Romano S, Bruckner-Foit A, Brandao A, Gumpinger J, Ghidini T, Beretta S. Fatigue
(316):24–9. properties of AlSi10Mg obtained by additive manufacturing: Defect-based
[33] Elber W. Einfluss der plastischen zone auf die rissausbreitung unter modelling and prediction of fatigue strength. Eng Frac Mech 2018;187:165–89.
schwingbelastung. Materialprufung 1970;6:189–93. [65] Yamashita Y, Murakami Y. Small crack growth model from low to very high cycle
[34] Elber W. Fatigue crack closure under cyclic tension. Eng Fract Mech 1970;2:37–44. fatigue regime for internal fatigue failure of high strength steel. Int J Fatigue 2016;
[35] Elber W. The significance of fatigue crack closure. Damage tolerance in aircraft 93:406–14.
structures. ASTM STP 1971;486:230–42. [66] Murakami Y, Harada S, Endo T, Tani-ishi H, Fukushima Y. Correlation among
[36] Endo K, Komai K, Ohnishi K. Effects of stress history and corrosive environment on growth law of small cracks, low-cycle fatigue and applicability of Miner’s rule. Eng
fatigue crack propagation. Memo Fac Eng Kyoto Univ 1969;31(1):25–46. Frac Mech 1983;18(5):909–24.
[37] Ritchie RO, Suresh S, Moss CM. Near-threshold fatigue crack growth in a 2 1/4 Cr- [67] Murakami Y, Miller KJ. What is fatigue damage? A view point from the observation
Mo pressure vessel steel in air and hydrogen. J Eng Mater Technol Trans ASME of low cycle fatigue process. Int J Fatigue 2005;27:991–1005.
1980;102(3):293–9. [68] Nisitani H, Goto M. Effect of stress ratio on the propagation of small crack of plain
[38] Walker N, Beevers CJ. A fatigue crack closure mechanism in titanium. Fatigue Fract specimens under high and low stress amplitudes (Fatigue under axial loading of
Eng Mater Struct 1979;1:135–48. annealed 0.45% C steel). Trans Jpn Soc Mech Eng 1984;50(453):1090–6.
[39] Minakawa K, McEvily AJ. On crack closure in the near-threshold region. Scri Metall [69] Nisitani H, Oda Y, Yakushiji T. Consideration of the small-crack growth law based
1981;15:633–6. on COD (Low-cycle bending fatigue test of an annealed 0.45%C steel in the SEM).
[40] Murakami Y, Nomoto T, Ueda T. Factors influencing the mechanism of superlong JSME Int J, Ser I 1990;33(2):243–8.
fatigue failure in steels. Fatigue Fract Eng Mater Struct 1999;22(7):581–90. [70] Dugdale DS. Yielding of steel sheets containing slits. J Mech Phys Solids 1960;8:
[41] Yoshinaka F, Nakamura T, Takeuchi A, Uesugi M, Uesugi K. Initiation and growth 100–4.
behavior of small internal fatigue cracks in Ti-6Al-4V via synchrotron radiation [71] Murakami Y, Matsuda K. Cause of unsuccessful results of Miner’s rule: Behavior of
microcomputed tomography. Fatigue Fract Eng Mater Struct 2019;42(9): small fatigue crack growth under repeated two-step loadings. In: Ravichandran KS,
2093–105. Ritchie RO, Murakami Y, editors. Small Fatigue Cracks: Mechanics, Mechanisms
[42] Yoshinaka F, Nakamura T, Takaku K, Shiozawa D, Nakai Y, Uesugi K. Observation and Applications; Elsevier Sci. Ltd.; 1999. p. 119− 30.
of the initial process of internal fracture in very high cycle fatigue in Ti-6Al-4V by [72] Endo T, Matsuishi M, Mitsunaga K, Kobayashi K, Takahashi K. Rain Flow Method,
synchrotron radiation μCT imaging [in Japanese] Trans Japan, Soc Mech Eng the proposal and the applications. Bull Kyushu Inst Technol Sci Technol 1974;28:
2017;83:856. 33–62.

14