Accepted Manuscript

Fatigue Life Prediction Method for AISI 316 Stainless Steel under Variable-
Amplitude Loading Considering Low-Amplitude Loading below the Endurance
Limit in the Ultrahigh Cycle Regime

Lei He, Hiroyuki Akebono, Masahiko Kato, Atsushi Sugeta

PII: S0142-1123(17)30177-9
DOI: http://dx.doi.org/10.1016/j.ijfatigue.2017.04.006
Reference: JIJF 4316

To appear in: International Journal of Fatigue

Received Date: 2 March 2017

Revised Date: 13 April 2017
Accepted Date: 13 April 2017

Please cite this article as: He, L., Akebono, H., Kato, M., Sugeta, A., Fatigue Life Prediction Method for AISI 316
Stainless Steel under Variable-Amplitude Loading Considering Low-Amplitude Loading below the Endurance
Limit in the Ultrahigh Cycle Regime, International Journal of Fatigue (2017), doi: http://dx.doi.org/10.1016/
j.ijfatigue.2017.04.006

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Fatigue Life Prediction Method for AISI 316 Stainless Steel under Variable-Amplitude Loading
Considering Low-Amplitude Loading below the Endurance Limit in the Ultrahigh Cycle Regime
Lei He1, Hiroyuki Akebono1*, Masahiko Kato1, Atsushi Sugeta1
Department of the Mechanical Science and Engineering, Graduate School of Engineering, Hiroshima
University, 1-4-1 Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8527, Japan
*Corresponding author
Tel / Fax : +81-82-424-7538
E-mail address: akebono@hiroshima-u.ac.jp
Abstract
To investigate the effect of cycle ratio of combined high- and low-amplitude loading corresponding to
loading above and below the fatigue limit on fatigue life (cumulative fatigue damage), fatigue tests were
conducted under repeated two-step amplitude loading condition with varying cycle ratios (the number of
high amplitude cycles was kept at 10 cycles, the number of low amplitude loading cycle was set at four
different values 103, 104, 5 × 104, and 10/105 cycles in one block) using AISI 316, a type of commercial
austenitic stainless steel. The test results indicate that low-amplitude loading following high-amplitude
loading contributed to the fatigue damage and affected fatigue life. Furthermore, both the EN (European)
and KHK (The high pressure gas safety institute of Japan) standards, which describe predictive
approaches for variable-amplitude loading conditions that consider stress levels below the fatigue limit in
an S–N curve based on the oblique hyperbola model, gave inaccurate evaluation of fatigue life because of
the effect of cycles of low-amplitude loading below the fatigue limit in one block. Fatigue damage at
saturation in one block contributed by low-amplitude was proved. A reasonable predictive approach that
considering the number of cycles of low-amplitude loading in one block using linear cumulative fatigue
damage rule has been established based on repeated two-step amplitude loading test results, and its
precision was confirmed through random-amplitude loading tests.

Key words: AISI 316, Ultrahigh cycle, Variable amplitude loading, Below fatigue limit, Cycle ratios.

1 Introduction
Over the past several decades, a great deal of research has focused on the ultrahigh cycle fatigue
properties of structural materials because materials that fracture after more than 107 cycles have an
entirely different failure mechanism which called optical dark area (ODA) fracture mode than those that
fracture after fewer cycles [1–6]. In general, the stress levels that cause the fatigue fracture of materials in
the ultrahigh cycle regime are close to or below the conventional fatigue limit. As transportation systems
become more advanced, longer service periods are becoming increasingly necessary because an extended
design fatigue life can effectively reduce energy and resource consumption and waste production [7].
Numerous articles on the ultrahigh cycle fatigue properties of various materials, such as steels,
light-weight metals, and copper, under a constant loading amplitude have been published [8–10]. In some
condition, low-amplitude loading below the fatigue limit can improve fatigue resistance, this phenomenon
is called coaxing effect [11–12]. However, in practical engineering applications, most metallic structures
are subjected to variable-amplitude loading conditions; for very long life designs, metallic components
are generally designed to bear low-amplitude loading below the conventional fatigue limit with the
intermittent application of a small number of cycles of high-amplitude loading. Such low-amplitude
loading appears to be harmless to components, but unexpected failure has often occurred.
Ngiau and Mayer [13–14] have reported that low-amplitude loading below the fatigue limit contributes
to the fatigue damage of 2024-T351 aluminum alloy and cast 319-T7 aluminum alloy under block loading
and found that low-amplitude loading can accelerate not only crack initiation but also crack propagation,
Miller and Tokaji [15–16] found similar phenomenon that low-amplitude loading can promote fatigue
crack growth. Jono [17–18] have conducted more systematic experimental studies on this issue using
carbon steel and martensitic stainless steel. They have reported that low-amplitude loading below the
fatigue limit is detrimental and the number of cycles of low-amplitude loading applied in one block
affects the fatigue life. They have also derived a modified S–N curve based on measurements of plastic
deformation range pairs induced by high- and low-amplitude loading. However, the difficulty of
measuring the plastic deformation of low-amplitude loading because of its localization and extremely low
values (less than 0.04%) limits the practical applicability of this approach. Thus, it has been rarely applied
in engineering designs.
Austenitic stainless steel AISI 316 is commonly used in cooling pipelines of nuclear power generators
because of its high temperature and corrosion resistance and excellent mechanical properties. Under
operating conditions, cooling pipelines experience simultaneous thermal fluctuation due to the cooling
water and mechanical vibration, leading to the final fatigue fracture of pipeline in the high or ultrahigh
cycle regime [21]. Here, the thermal fluctuation and mechanical vibration occur during low- and
high-amplitude loading, respectively. Because of the abovementioned limitation of the approach by Jono
[17–18], this approach could not be employed in the present study. Furthermore, a modified version of the
rule developed by Miner [19] and Haibach [20] can only be implemented when the regression equation of
the S–N diagram is bilinear on a log–log scale. Vincent and Kamaya [21–22] demonstrated that the S–N
curve of austenitic stainless steel is not linear for fatigue tests conducted under both strain or stress
control. However, there are two predictive approaches to obtain a nonlinear S–N curve under
low-amplitude loading conditions: a High Pressure Gas Safety Institute of Japan (KHK) standard (0220)
[23] that can be used to evaluate fatigue damage below the fatigue limit by expanding the S–N diagram
from 107 to 108 cycles and passing a vertical line from the load corresponding to 108 cycles to half of that
load, and a European (EN) standard (13445–3) [24] that can be used to predict the fatigue life by drawing
an oblique line with a slope of –0.1 from 2 × 106 to 108 cycles. Figure 1 illustrates these two standards.
The horizontal lines in Fig.1 indicate the number of cycles below which the fatigue life is infinite; that is,
loads less than that at 108 cycles manifest non-damaging behavior.
The cumulative fatigue damage can be calculated using the Palmgren–Miner model [19,25], the only
parameters of which are the numbers of applied and total cycles at applied constant loading level. The
shortcoming of this model is that it omits the loading sequence effect under variable-amplitude loading
conditions. In the high–low (H–L) sequence, the cumulative fatigue damage is less than unity, whereas in
the low–high (L–H) sequence, it is greater than unity. To describe this condition, other cumulative
damage models, such as the damage curve model, the bilinear model including the crack initiation and
propagation stages, and the energy-based model, have been established since 1920 [26]. However, these
models either contain too many parameters or involve complex calculations, complicating their
implementation in actual engineering designs. Although the Palmgren–Miner model has apparent
shortcomings, it is still widely used in scientific research and practical engineering designs because of its
simplicity.

Experimental results
Stress amplitude, σa (MPa)

σ10 8
0.1

0.5σ10 8
EN Standard
KHK Standard

103 104 105 106 107 108 109

Number of cycles to failure, Nf

Fig.1 Illustration of EN and KHK standards (log-log)

In this study, to elucidate the influence of the frequency ratio of combined high- and low-amplitude
loading corresponding to loading above and below the fatigue limit on fatigue life (cumulative fatigue
damage), a systematic experimental investigation was conducted. Constant-amplitude loading fatigue
tests were first performed to obtain the basic data and calculate the regression equation using the oblique
hyperbola model in log–log coordinates. Second, to investigate the precision of the EN and KHK
standards and the effect of the frequency ratio of low-amplitude loading on the fatigue damage, repeated
two-step amplitude loading tests were conducted. Furthermore, a new evaluation method was developed
using the data obtained from the repeated two-step tests; this method considers the number of cycles of
low-amplitude loading in one block. Finally, random-amplitude loading tests were conducted to examine
the accuracy of the proposed predictive approach of fatigue life.
2 Experimental procedure
2.1 Materials and specimens
AISI 316, a type of commercial austenitic stainless steel, was used in this investigation. The chemical
composition and mechanical properties of this steel are given in Tables 1 and 2, respectively.
Hourglass-shaped specimens were manufactured from several round bars, with minimum diameters of 4.5
mm, and a round notch radius of 20 mm, respectively. Figure 2 shows the detailed dimensions of the
specimen. Each specimen was polished to a mirror-like surface using emery paper ranging from 360 to
2000 grit and alumina powder (1.0 μm) prior to fatigue testing.
 Table 1 Chemical composition of AISI 316 (mass%)
C Si Mn P S Ni Cr Mo Fe
0.04 0.30 1.25 0.038 0.025 10.13 16.01 2.00 Bal.

Table 2 Mechanical properties of AISI 316

Ultimate tensile
Yield strength Elongation Reduction of area
strength
263 MPa 570 MPa 62% 75%

100 4.5

Φ10.16 R=20 10.08

Φ15.24

Fig.2 Shape and dimensions of specimen

2.2 Fatigue test

Loading-controlled fatigue tests were conducted at room temperature in atmospheric conditions with a
stress ratio of R = –1 using a MTS 810 fatigue testing machine. During the constant-amplitude loading
fatigue tests (frequencies of 10–20 Hz), the stress level at which rupture did not occur at loadings over 107
cycles was defined as the fatigue limit σw.
In repeated two-step amplitude loading fatigue tests (frequencies of 1 and 200 Hz), the high and low
stress levels σH and σL were set to values above and below the fatigue limit, respectively. Figure 3 shows a
plot of this loading spectrum condition. At a constant value of σH/σL, the number of cycles of σH in one
block was kept at 10 cycles, and four different values of the number of cycles of σL in one block were
applied to investigate the effect of variable cycle ratios (nH/nL) on the fatigue life or fatigue damage.

σH σW
σL
R=-1

Time

nH nL

Fig.3 Loading spectrum of one block in the repeated two-step amplitude loading test
 Random-amplitude loading tests were conducted using a Servopulser fatigue testing machine
(Shimadzu Corporation) and controlled with a program created by Tanagashima [27,28]. The obtained
random-waveform voltage signals were input into the controller of the Servopulser as analog signals
using a 50-pin miniature ribbon connector. In the present study, the following stationary Gaussian
stochastic process χ(t) with a mean value of zero was applied to generate a random waveform:
(1)

where
(2)
N is the divided number of the power spectrum; ωk is the angular frequency; t is the time; and φk is the
uniform stochastic variable number, which can be obtained using the multiplicative congruential method;
S(ωk) and ω express power spectrum density function and the divided width of the angular frequency,
respectively. Pseudo-wideband waveforms with loading amplitudes both above and below the fatigue
limit were employed in a series of random-load fatigue tests. The extreme value of each random
waveform was set to 1000, and a testing frequency of 10 Hz was selected. The maximum and minimum
voltage signals corresponding to the maximum and minimum loading amplitudes were limited to ±8 V for
the purpose of protecting the loading cell of the Servopulser fatigue testing machine. The power spectrum
distribution is shown in Fig. 4(a) with fl of 1 Hz and fu of 10 Hz, respectively. Figure 4(b) shows an
example of the pseudo-wideband random waveform used in this study. Effective cycles were obtained
using the rainflow cycle counting method, and the cumulative fatigue damage values were subsequently
calculated using the Palmgren–Miner linear damage rule, which is given by
(3)

Where Ni is number of cycles to failure under constant amplitude loading and ni is applied number of
cycles at the same amplitude loading.
S(f)

0
fl f(Hz) fu

Fig.4(a) Power spectrum distribution

 8 260 MPa

Voltage(V)
0
50
-8 -260 MPa Time(s)

Fig.4(b) Instance of wide-band random waveform

Each fracture surface was examined with a scanning electron microscope (SEM; Hitachi, S-3000N). To
understand the fatigue crack propagation behavior under variable amplitude loading, the fracture surface
was quantitatively analyzed for certain specimens because beach marks and striations were observed on
the fracture surfaces.

3 Experimental results
3.1 Constant amplitude loading
Figure 5 shows the S–N curves obtained during the constant-amplitude loading tests. The black curve
in the figure is the regression equation for the data points tested under constant-amplitude loading, the
model of which is a double-logarithmic oblique hyperbolic model given by
(4)
Where σ and N indicate stress amplitude and number of cycles to failure, respectively.
This model was computed on the basis of a standard by the Society of Materials Science, Japan, (JSMS)
entitled “Standard Evaluation Method of Fatigue Reliability for Metallic Materials-Standard Regression
Method of S–N curve” [29]. The objective model showed better agreement with the experimental results
than did the double-logarithmic bilinear model which is often used in fatigue research.
In the high stress region (above 280–350 MPa), fatigue life increased slowly with decreasing applied
stress amplitude. However, fatigue life increased rapidly at stresses 40 MPa below the high stress region
(280–240 MPa). A fatigue limit of 236 MPa was determined using the abovementioned JSMS standard
[29]. To verify the validity of the determined endurance limit, a constant-amplitude loading fatigue test
was conducted at the abovementioned stress level; the survival of the specimen in this test indicates that
the calculated value represents the fatigue limit.
 400

Stress amplitude,σa (MPa)

200

EN Standard
KHK Standard
100
103 104 105 106 107 108 109
Number of cycles to failure, Nf

Fig.5 S–N curve of AISI 316 under constant-amplitude loading

3.2 Repeated two-step amplitude loading

3.2.1 Test results
Repeated two-step amplitude loading tests were conducted with the stress σH of the high-amplitude
loading phase maintained above the fatigue limit at a constant value of 260 MPa and the stress σL of the
low-amplitude loading phase maintained below the fatigue limit at a value varying from 80 to 225 MPa.
In these tests, the loading frequency ratio nH/nL (cycle ratio) was set to four different values (10/103,
10/104, 10/5 × 104, and 10/105) to determine the influence of the number of cycles of σL below the
endurance limit in one block on the cumulative fatigue damage.
The evaluation method of the EN and KHK standards are also plotted in Fig.5. The fatigue test results
are given in Table 3. All of the samples except one failed under these two-step amplitude loading
conditions with loading phases above and below the fatigue limit; furthermore, under these conditions,
the fatigue limit disappeared [13–18, 30]. At constant stress levels with varying cycle ratios, the number
of blocks to failure of the 10/103 series was the longest among the four considered cycle ratios. Because
of the large scatter of the 10/103 series at low-amplitude loading of 200 and 100 MPa, fatigue tests were
conducted twice to ensure that precise data could be acquired. For the test results of the remaining cycle
ratios, which exhibited roughly similar number of blocks to failure for stresses above 160 MPa. This
phenomenon will be discussed in the next section. For the 10/105 series under σH/σL = (260
MPa)/(225MPa), failure did not occur; in this case, σL was 0.95σw, and the number nL of low-amplitude
cycles in one block was 105 cycles, Lu [31] investigated the strengthening effect of low amplitude loading
below the fatigue limit under repeated two-step amplitude loading for 20MnCr5, and found that at certain
loading level and number of cycles of low amplitude loading, fatigue strength can be improved even
though number of cycles of low amplitude loading in one block loading was less than 10 7 cycles. Based
on Lu’s conclusion, it can be considered that the fatigue strength of abovementioned survived specimen
was improved because of coaxing effect occurred [32].
The relationship between the fatigue life and stress amplitude loading of σL is shown in Fig.6. The
horizontal lines in Fig. 6 represent the fatigue life calculated under the assumption that σL did not damage
the specimens; that is, the number of blocks to failure was computed based on damage resulting from
only σH. Each block of this two-step loading sequence contained 10 cycles of high-amplitude loading, the
number of cycles to failure at constant-amplitude loading of σH (260 MPa) was 4.5×104, so the number
of blocks to failure was B = 4.5 × 103 blocks; under this assumption, the fatigue life in terms of the
number of cycles could be computed as B(nH + nL) with a varying cycle ratio [33]. However, for each
series, the obtained data were below the corresponding horizontal line described above, indicating that
low-amplitude loading below the endurance limit contributes to the fatigue damage of the specimen or
materials under variable-amplitude loading conditions.

Table 3 Fatigue test results under repeated two-step amplitude loading

Stress (MPa) Number of blocks to failure
σH/σL 10/103 10/104 10/5×104 10/105
260/225 ------ 549 ------ 4036(Run out)
2447
260/200 483 548 1023
757
260/180 ------ 1081 ------ ------
260/160 2176 1107 1180 1153
260/140 3253 1820 ------ 1958
260/120 3676 3337 2473 2281
4266
260/100 4329 2010 3944
2544
260/80 ------ 5382 ------ 3631
 10/103 10/5×104
10/104 10/105
109
10/105

Number of cycles to failure, Nf

10/5×104
108
10/104

107

10/103

106
0 50 100 150 200 250
Stress amplitude of σL (MPa)

Fig.6 Comparison of experimentally obtained fatigue life and that calculated by assuming the specimen
experiences no damage when the applied stress is below the fatigue limit

Cumulative fatigue damage values calculated using the linear damage model (Eq.(3)) according to the
EN and KHK standards are summarized in Table 4. In this study, the fatigue limits calculated using the
EN and KHK standards are 165 and 120 MPa, respectively. The values marked by asterisks indicate that
the low-stress amplitude is below the corresponding fatigue limit. The cumulative fatigue damage will be
described in more detail in the next subsections, which present analysis based on the EN and KHK
standards.

Table 4 Cumulative fatigue damage calculated based on the EN and KHK standards
Stress(MPa) Cumulative fatigue damage value(EN) Cumulative fatigue damage value(KHK)
σL 10/103 10/104 10/5×104 10/105 10/103 10/104 10/5×104 10/105
90.29 4.67
260/225 ------ 1.31 ------ ------ 0.14 ------
(Run out) (Run out)
0.72 0.57
260/200 0.40 4.95 7.16 0.12 0.36 1.18
0.22 0.18
260/180 ------ 0.61 ------ ------ ------ 0.28 ------ ------
260/160 0.49* 0.17* 0.18* 0.18* 0.51 0.28 0.77 1.33
260/140 0.73* 0.29* ------ 0.31* 0.76 0.47 ------ 2.26
260/120 0.82* 0.52* 0.39* 0.36* 0.82* 0.52* 0.39* 0.36*
0.95* 0.95*
260/100 0.68* 0.31* 0.62* 0.68* 0.31* 0.62*
0.57* 0.57*
260/80 ------ 0.84* ------ 0.57* ------ 0.84* ------ 0.57*

3.2 Cumulative fatigue damage based on EN and KHK standards

The fatigue damage calculated using the EN standard for stresses above 165 MPa (see Fig.5 and Table
4) demonstrate that the EN standard generally gave a nonconservative estimate of the fatigue life for the
10/103 and 10/104 series, even though one calculated fatigue damage value for the latter series was greater
than unity (fatigue damage of 1.31 at L = 225 MPa). On the other hand, for the 10/5 × 104 and 10/105
series with low-amplitude loadings above 165 MPa, the EN standard gave estimated fatigue damage
values much higher than unity. When the low-amplitude loadings were below 165 MPa, the cumulative
fatigue damage increased gradually as the low-amplitude loading stress decreased. In this region, all of
the damage values were less than unity. However, the cumulative fatigue damage still contained a
contribution from σL, even at extremely low loading amplitudes (below 160 MPa).
For the KHK standard, the cumulative fatigue damage values for low-amplitude loading stresses above
120 MPa were analyzed. Table 4 shows that the KHK standard gave nonconservative fatigue life
estimates for the 10/103, 10/104, and 10/5 × 104 series. However, the cumulative fatigue damage values
for the 10/105 series were greater than unity, too large to apply in engineering designs. For σL < 120 MPa,
every fatigue damage value was less than unity; this result is very similar to that obtained for the EN
standard.
These results suggest that both the EN and KHK standards gave nonconservative estimates for the
fatigue life under repeated two-step amplitude loading tests when a small number of low-amplitude
loading cycles are included in the loading spectrum. Furthermore, conservative estimates were given by
both evaluation methods when a large number of low-amplitude loading cycles were included in the
loading spectrum. It appears that a critical number of cycles, above which the estimated fatigue life
becomes conservative, may exist. Therefore, it is essential to establish a more accurate evaluation method
that includes the influence of low-amplitude loadings below the fatigue limit for practical application to
engineering designs.
3.3 Fracture morphology
Figures 7 and 8 illustrate representative fracture micrographs taken by a digital microscope (KH-1300,
Hirox Co., Ltd.) and an SEM (S-3000N, Hitachi, Ltd.). In Fig.7, the sample ruptured under
constant-amplitude loading, and its fracture surface appears to contain one initiation site. Figure 8 shows
fracture micrographs of specimens fractured under repeated two-step amplitude loading. In Fig.8, a single
initiation site was generated at the surface of the specimen, as with the specimen in Fig.7. Beach marks
and striations formed on fracture surface as a result of the variable-amplitude loading and the sufficiently
low stress σL of the low-amplitude loading phase (Fig.8). In addition, the entirety of the initiation sites
originated at the surfaces of the samples; thus, it is very likely that the stress σH of the high-amplitude
loading phase dominated the nucleating period under repeated two-step amplitude loading. Therefore, the
fracture models for constant-amplitude and repeated two-step amplitude loading scenarios are almost the
same.
 a b

b Initiation site

1mm 50μm

Fig.7 (a) Low- and (b) high-magnification micrographs of a fracture surface under constant-amplitude
loading with a = 240 MPa and Nf = 7.04  105 cycles

a b b c

c Striation
Initiation site

1mm 50μm 2μm

Fig.8 (a) Low- and (b), (c) high-magnification micrographs of a fracture surface under repeated two-step
amplitude loading with H/L = (260 MPa)/(120 MPa), nH/nL = 10/105, and N = 2281 blocks

4 Discussion
4.1 Modified S–N curve below the endurance limit
The cumulative fatigue damage values predicted by the EN and KHK standards were not in agreement
with the theoretical value of unity. Consequently, a rational approach should be established to evaluate the
cumulative fatigue damage below the fatigue limit. The fictitious fatigue life Nf* of σL at each of the tested
loading levels and cycle ratios were calculated using the linear cumulative damage model, given by
(5)

based on the assumption that the cumulative fatigue damage equals unity. Where block and NH indicate
number of blocks to failure under repeated two-step amplitude loading test and number of cycles to
failure under stress amplitude of 260 MPa.
The results are plotted in Fig.9. At a constant loading level, the smaller cycle ratios corresponded to
longer fictitious fatigue lives Nf*, and at a constant cycle ratio, the magnitude of the fictitious fatigue life
increased with decreasing low-amplitude loading level σL. The regression equations of the fictitious
fatigue lives Nf* of the four series with different cycle ratios are plotted in different colors in Fig.9. The
four equations showed very similar slopes, with values of –0.228, –0.231, –0.331, and –0.296 on a
log–log scale for the 10/103, 10/104, 10/5 × 104, and 10/105 series, respectively.
Figure 10 plots the relationship between percentage of cumulative fatigue damage contributed by σL
and corresponding stress amplitude σL. Figure 10 shows that at high low-amplitude loading levels σL (such
as 225 MPa), the vast majority of the cumulative fatigue damage (approximately 90%) was attributable to
σL; even at extremely low loading amplitudes, such as 80 MPa, σL still inflicted a small amount of fatigue
damage to specimen, the black line displays the curve of best fit for the data in the 10/10 4, 10/5 × 104, and
10/105 series, which indicates a damage saturation statement, this will be further discussed in the next
subsection. The regression equation of the 10/103 series and most of that for the 10/104 series lie along the
predictive line of the EN standard, and for loading levels above 165 MPa, the regression equation of the
10/5 × 104 and 10/105 series lie to the right of the predictive line. Thus, the EN standard yields an under-
or overestimation of the contribution of the low-amplitude loading phase to the fatigue damage at
different cycle ratios depending on the position of the variable regression equations. The KHK standard
gave conservative or nonconservative cumulative fatigue damage for the same reason.

400
Stress amplitude of σL(MPa)

200

10/103
10/104
10/5×104
100
10/105

EN Standard
KHK Standard
50
103 104 105 106 107 108 109
Number of cycles of Nf *

Fig.9 S–N curves of AISI 316 below the fatigue limit together with the fictitious fatigue life calculated
under the assumption that the cumulative damage value equals unity (Eq.(5))
 250
10/103
10/104

Stress amplitude of σL(MPa)

10/5×104
200
10/105

150

100

50
0 20 40 60 80 100
Percentage of cumulative fatigue damage
contributed by σL(%)

Fig.10 Contribution of low-amplitude loading to the cumulative fatigue damage

4.2 Damage contributed by low-amplitude loading under repeated two-step amplitude loading
The cumulative fatigue damage of one block for the 10/103 series were found to be quite different from
that of the other series under equal loading amplitudes; additionally, the magnitudes of the cumulative
fatigue damage were nearly equivalent among the remaining series at equal loading amplitudes, although
there was some scatter, as shown in Fig.11. This indicates that beyond a certain number of cycles of
low-amplitude loading, further loading does not contribute to the cumulative fatigue damage of the
specimen. In Fig. 11, this limiting number of cycles was less than 104 cycles based on the above
discussion.
Figure 12(a) illustrates the damage process below the fatigue limit under repeated two-step amplitude
loading. After being subjected to high-amplitude loading (10 cycles in one block) the specimen underwent
low-amplitude loading, which contributed to the fatigue damage of the specimen, until the low-amplitude
ceased to inflict further damage. At this point, the cumulative fatigue damage is said to be saturated by
low-amplitude loading, and the number of cycles to saturation is denoted by nSa. The number of cycles to
saturation was calculated as
(6)

where DSa and D0 are the cumulative fatigue damage at saturation and the cumulative fatigue damage
after one cycle of low-amplitude loading for the 10/103 series. Before the saturation limit was computed,
to reduce the effect of the amount of scatter on the computational results, the average cumulative damage
at the saturation limit was calculated for the 10/104, 10/5 × 104, and 10/105 series at equal stress levels
σH/σL, the results of saturation limit is shown in Fig.12(b). Because of the amount of scatter in the test
results for the 10/103 series at σL = 100 and 200 MPa, the scatter of the results shown in Fig.12(b) is fairly
large. The mean number of cycles to saturation was 5680 cycles. Analogously, in Fig.10, the curve of best
fit shows the contribution of σL to the cumulative fatigue damage at saturation.
In Fig.11, the dashed line shows the regression equation of the damage at saturation in one block, the
equation is calculated by least squares procedure and is given as Eq.(7). The fictitious fatigue life Nf*
below the fatigue limit can be calculated if the number nL of cycles of low-amplitude loading reaches the
number of cycles to saturation (nL ≥ 5680) using the following equations:
(7)
(8)

which can be combined to obtain the relationship between the fictitious fatigue life and the stress level of
the low-amplitude loading as
(9)

The fictitious fatigue life at stress levels corresponding to the experimental conditions was computed with
the corresponding cycle ratios. The calculation results were in good agreement with the experimental data,
as shown in Fig.13. Therefore, Eq.(9) can be employed to assess the fictitious fatigue life below the
fatigue limit under cycle ratios whose loading conditions satisfy the previously mentioned saturation
criterion.
However, when the cycle ratio is not in the saturation range (i.e., nL < 5680 cycles), the regression line
of the 10/103 series plotted in Fig.13 can be applied to predict the fatigue life under variable-amplitude
loading. As previously mentioned, the regression equations of the 10/103, 10/104, 10/5 × 104, and 10/105
series had very similar slopes; the average slope of the four lines was computed and is included in Fig.13
as a dashed line replacing the regression line of the 10/103 series to improve the evaluation accuracy, the
intersection abscissa between S–N curve under constant amplitude loading and above-mentioned line is at
106 cycles. Thus, the following modified S–N curve is proposed. When nL < 5680 in one block, the fatigue
life can be evaluated using the abovementioned regression equation with the average slope of the four
investigated series (dashed line in Fig.13), and when nL ≥ 5680 in one block, Eqs.(9) and (5) should be
used to predict the fatigue life.
 10-2
10/103 10/5×104

Cumulative fatigue damage value of

10/104 10/105

1 block contributed by σL
10-3

10-4

10-5
50 100 150 200 250
Stress amplitude of σL(MPa)

Fig.11 Relationship between the cumulative fatigue damage during one block of low-amplitude loading
and the corresponding loading stress amplitude L

n Sa
Saturation cycles
1000
Damage

Damage Non-damage

nH nL

Fig.12(a) Schematic of the damage process under low-amplitude loading below the fatigue limit

105
Number of cycles to saturation

104

103

102
0 50 100 150 200 250
Stress amplitude of σL(MPa)

Fig.12(b) Number of cycles to saturation

 Solid mark : Experimental results
Open mark : Estimated fictitious fatigue life
400

Stress amplitude of σL(MPa)

200

10/103
10/104
10/5×104
100
10/105

EN Standard
KHK Standard
50
103 104 105 106 107 108 109
Number of cycles of Nf *

Fig.13 Comparison of fictitious fatigue life obtained using the linear cumulative fatigue damage rule and
Eqs.(7)–(9)

It is well known that the fatigue damage is induced by the motion of a dislocation that forms when
metallic materials are subjected to cyclic loading. Numerous pioneering works have been conducted to
establish a rational hypothesis or model [34,35] to reveal the fatigue damage mechanism.
In this study, on the basis of experimental results, the influence of the cycle ratio on the fatigue life was
qualitatively interpreted. A simplified damage process considering the effect of the cycle ratio was
developed. In this model, dislocation glide types, such as climbing, jogging, and cross-slipping, were
ignored. The process was divided into the following four steps. (I) When the specimen is subjected to
cyclic loading above the fatigue limit, dislocations are generated and glide along the favored grain
orientation slip plane, and a persistent slip bands (PSB) forms. (II) After the transition from
high-amplitude loading to low-amplitude loading below the fatigue limit, dislocations continued to glide
until they arrive at the grain boundary, which arrests the movement of the dislocations. (III) After that,
dislocations pile up along the grain boundary because the loading is too low for them to pass though the
grain boundary or induce slip systems in other grains. As discussed before, after a certain number of
cycles, the damage reaches the saturation limit. After the number of cycles to saturation, the remaining
low-amplitude loading cycles in one block inflict no further damage to the specimen. (IV) During a
subsequent loading block, piled-up dislocations receive sufficient force from high-amplitude loading to
pass through the barrier; therefore, the PSB may extend into a neighboring grain. Steps (I)–(IV) continue
repeatedly until final fracture.
5 Validity of proposed approach
To verify the validity of the modified S–N curve described in section 4.1, proving tests were conducted
under wide-band random-amplitude loading with an extreme value of 1000. The maximum stress in the
random-amplitude loading tests was maintained at 260 MPa, which is equal to that in the repeated
two-step amplitude loading tests. The details of triggering a random-amplitude waveform were previously
introduced in section 2.2. Figure 14 shows the relationship between the random-amplitude waveform
output from the loading cell and the local strain behavior, which was measured by an attached strain
gauge. The dashed and solid lines indicate the output voltage signal and the local strain value, respectively.
The random-amplitude waveform generated by program was considered to be feasible. The rainflow
cycle counting method was applied to count the effective number of cycles in one block of the random
waveform.
It is known that the fatigue life of various materials is affected by the mean stress, and thus the
well-known Goodman and Smith–Watson–Topper (SWT) equations were used to modify the influence of
the mean stress on the fatigue life. These two equations are respectively given as
σ (10)
σ σ

σσ (11)
where, σa, σm, σmax and σb indicate stress amplitude, mean stress, maximum stress and tensile strength of
material, respectively. The SWT equation includes not only the mean stress σm and the stress amplitude σa
but also the maximum stress σmax.
The test results are summarized in Table 5 for the first and second random-amplitude loading tests W1
and W2. In both W1 and W2, the cumulative fatigue damage values using the SWT equation are greater
than those modified using the Goodman equation. Therefore, the following analysis of the
random-amplitude loading test data is based on the SWT modification.
The percentage of all cycles that were high-amplitude loading cycles above the fatigue limit in one
random-amplitude waveform block were 0.54% and 0.53% for W1 and W2, respectively, as counted
using the rainflow cycle counting method. Moreover, the cumulative fatigue damage values of one block
were 9.11 × 10–5 and 6.08 × 10–5 for W1 and W2, respectively. High-amplitude loading above the fatigue
limit contributed 71.59% and 68.72% of the cumulative fatigue damage in W1 and W2 in one block. Thus,
high-amplitude loading produced the majority of the damage to specimen during random-amplitude
loading tests. However, nearly 61% of the cycles were below 50 MPa. Combining with the best fit curve
in damage saturation statement shows in Fig.10, the contribution by low-amplitude loading below 50
MPa to the cumulative fatigue damage was extremely small. Under this condition, it can be considered
that fatigue damage below 50 MPa can be omitted.
When the cycles with stress levels below 50 MPa were removed, it was found that the cumulative
fatigue damage of one block in the random-amplitude tests rarely changed. This is in good agreement
with the results of the repeated two-step amplitude loading tests. That is, stress levels between 50 and 236
MPa (the fatigue limit) contributed the remaining approximately 30% fatigue damage in random
waveform of one block. The total cumulative fatigue damage values were 0.64 and 0.61 for W1 and W2,
respectively. A possible reason for the overestimation of the fatigue life is that the specimens were
subjected to extremely complicated loading conditions, leading to a shorter fatigue life than predicted.
Although the predicted damage values were lower than unity, because random-amplitude loading is an
extremely complex loading condition, the obtained outcomes appear reasonable. Moreover, the proposed
modified S–N curve below the fatigue limit can be used to predict fatigue life under variable-amplitude
loading when combined with the design fatigue curve.

8 300
6
200
4

Local strain(×10-6 )
Output voltage(V)

100
2
0 0
-2
-100
-4
-200
-6 Local strain
Output voltage
-8 -300
0 1 2 3
Time(s)

Fig.14 Relationship between output voltage and local strain

Table 5 Random-amplitude loading test results

Mean stress Damage in 1 Loading above Total
No. Total blocks
modified block fatigue limit damage
Goodman 8.52×10-5 0.54% 0.60
W1 7068
SWT 9.11×10-5 0.54% 0.64
Goodman 5.94×10-5 0.53% 0.59
W2 9989
SWT 6.08×10-5 0.53% 0.61

6 Conclusions
In order to investigate the effect of cycle ratio of combined high- and low-amplitude loading
corresponding to loading above and below the fatigue limit on fatigue life, fatigue tests were conducted
under variable-amplitude loading condition using AISI 316, a type of commercial austenitic stainless steel.
On the basis of the obtained outcomes, the following conclusions were obtained.
 Under repeated two-step amplitude loading tests with a varying cycle ratio, both the EN and KHK
standards overestimated the fatigue life when the number of cycles of low-amplitude loading in one
 block was small. With the stress level held constant, the fatigue life increased as the number of
cycles of low-amplitude loading increased. Thus, it was demonstrated that low-amplitude loading
following high-amplitude loading contributed to the fatigue damage. Under this loading condition,
the damage reached saturation after a critical number of low-amplitude loading cycles. Damage
saturation statement contributed by low-amplitude loading below the fatigue limit in one block was
proved.
 Micrographs of a fracture surface revealed that the fatigue crack always originated at the specimen
surface regardless of whether the specimen was subjected to constant-amplitude, repeated two-step,
or random-amplitude loading. Even if the fatigue life exceeded 107 cycles. Under variable-amplitude
loading conditions, high stress levels dominated at the fatigue crack initiation site.
 A predictive approach using linear cumulative fatigue damage rule based on repeated two-step
amplitude loading tests was established, and its validity was verified through random-amplitude
loading tests. The proposed approach yields accurate estimations of the cumulative fatigue damage
for materials with S–N curves based on the oblique hyperbola model that undergo combined high-
and low-amplitude loading above and below the fatigue limit.

Acknowledgment

The author is grateful to China Scholarship Council for supporting scholarship in the author's doctoral
course.

Reference
[1] Murakami Y., Takada M., Toriyama T. Fatigue. Super-Long Life Tension-Compression Fatigue
Properties of Quenched and Tempered 0.46% Carbon Steel. International Journal of Fatigue 1998; 16(9):
661-667.
[2] Sakai T., Lian B., Takeda M., Shiozawa K., Oguma N., Ochi Y., Nakamura T. Statistical duplex S-N
characteristics of high carbon chromium bearing steel in rotating bending in very high cycle regime.
International Journal of Fatigue 2010; 32(3): 497–504.
[3] Sakai T., Oguma N., Morikawa A. Microscopic and nanoscopic observations of metallurgical
structures around inclusions at interior crack initiation site for a bearing steel in very high-cycle fatigue.
Fatigue and Fracture of Engineering Materials and Structures 2015; 38(11): 1305–1314.
[4] Shiozawa K., Hasegawa T., Kashiwagi Y., Lu L. Very high cycle fatigue properties of bearing steel
under axial loading condition. International Journal of Fatigue 2009; 31(5): 880–888.
[5] Shiozawa K., Lu L., Ishihara S. S-N curve characteristics and subsurface crack initiation behaviour in
ultra-long life fatigue of a high carbon-chromium bearing steel. Fatigue and Fracture of Engineering
Materials and Structures 2001; 24(12): 781–790.
[6] Shiozawa K., Morii Y., Nishino S., Lu L. Subsurface crack initiation and propagation mechanism in
high-strength steel in a very high cycle fatigue regime. International Journal of Fatigue 2006; 28(11):
1521–1532.
[7] Sakai T., Nakagawa A., Oguma N., Nakamura Y., Ueno A., Kikuchi S., Sakaida A. A review on fatigue
fracture modes of structural metallic materials in very high cycle regime. International Journal of Fatigue
2016; 93:339-351.
[8] Grad P., Reuscher B., Brodyanski A., Kopnarski M., Kerscher E. Mechanism of fatigue crack
initiation and propagation in the very high cycle fatigue regime of high-strength steels. Scripta Materialia
2012; 67(10): 838–841.
[9] Zhang M., Wang W., Wang P., Liu Y., Li J. Fatigue behavior and mechanism of FV520B-I welding
seams in a very high cycle regime. International Journal of Fatigue 2016; 87: 22–37.
[10] Yang F., Yin S. M., Li S. X., Zhang Z. F. Crack initiation mechanism of extruded AZ31 magnesium
alloy in the very high cycle fatigue regime. Materials Science and Engineering A 2008; 491(1–2):
131–136.
[11] Lu X., Zheng S.L. Strengthening of transmission gear under low-amplitude loads. Materials Science
and Engineering A 2008; 488: 55–63.
[12] Lu X., Zheng S.L. Changes in mechanical properties of vehicle components after strengthening
under low-amplitude loads below the fatigue limit. Fatigue and Fracture of Engineering Materials and
Structures 2009; 32: 847-855.
[13] Ngiau C., Kujawski D. Sequence effects of small amplitude cycles on fatigue crack initiation and
propagation in 2024-T351 aluminum. International Journal of Fatigue 2001; 23(9): 807–815.
[14] Mayer H., Ede C., Allison J. E. Influence of cyclic loads below endurance limit or threshold stress
intensity on fatigue damage in cast aluminium alloy 319-T7. International Journal of Fatigue 2005; 27(2):
129–141.
[15] Miller K.J., O’Donnell W.J. The fatigue limit and its elimination. Fatigue and Fracture of
Engineering Materials and Structures 1999; 22: 545-557.
[16] Tokaji K., Ando Z., Yamada T. Study on the Effect of Specimen of Understress on Carbon Steel
under Fatigue Life of Smooth Rotating Torsion. The Japan Society of Mechanical Engineers 1981;
24(188): 273-281.
[17] Jono M., Hanai M., Kikukawa M. Effect of Frequency Ratio of Overstress to Under-stress on
Cumulative Damage below Fatigue Limit (When Low Level Stress has Tensile Mean Stress). The Society
of Materials Science, Japan 1982; 361(32): 1137-1143. (in Japanese)
[18] Jono M., Song J., Uchida H. Effects of Cycle Ratio of Over-stress to Under-stress on Cumulative
Fatigue Damage of AISI410 JIS Stainless Steel. The Japan Society of Mechanical Engineers 1984;
453(50): 973-979. (in Japanese)
[19] Minear MA. Cumulative damage in fatigue. J Appl Mech(ASME) 1945; 12(3): A159-64.
[20] Haibach E. Betriebsfestigkeit- Verfahren und Daten zur Berechnung(structural durability- Methods
and data for calculation). 2nd ed. Düsseldorf: VDI-Verlag; 2003.
[21] Vincent L., Le Roux J.-C., Taheri S. On the high cycle fatigue behavior of a type 304L stainless steel
at room temperature. International Journal of Fatigue 2012; 38: 84–91.
[22] Kamaya M., Kawakubo M. Loading sequence effect on fatigue life of Type 316 stainless steel.
International Journal of Fatigue 2015; 81: 10–20.
[23] Standard for Facilities of Ultra High Pressure Gas. KHKS-0220. The High Pressure Gas Safety
Institute of Japan 2010. (in Japanese)
[24] BS EN 13445-3:2009, Unfired Pressure Vessels. Design, BSI.
[25] Palmgren A. Die Lebensdauer von Kugellagern. Verfid-mwsrrchinik, Berlin, 1924; 68: 339-341.
[26] Fatemi a., Yang L. Cumulative fatigue damage and life prediction theories: a survey of the state of
the art for homogeneous materials. International Journal of Fatigue 1998; 20(1): 9–34.
[27] Tanegashima R., Akebono H., Kato M., Sugeta A. 3-Dimensional observation of the interior fracture
mechanism and establishment of cumulative fatigue damage evaluation on spot welded joints using
590MPa-class steel. International Journal of Fatigue 2013; 51: 121–131.
[28] Tanegashima R., Ohara I., Akebono H., Kato M., Sugeta A. Cumulative fatigue damage evaluations
on spot-welded joints using 590MPa-class automobile steel. Fatigue and Fracture of Engineering
Materials and Structures 2015; 38: 870–879.
[29] Sakai T., Sugeta A. Standard Evaluation Method of Fatigue Reliability for Metallic Materials:
Standard Regression Method of S-N Curves 2007. JSMS.
[30] Joy-A-Ka S., Ogawa Y., Akebono H., Kato M., Sugeta A., Sun Y., Fujii H. Fatigue Damage
Evaluation of Friction Stir Spot Welded Cross-Tension Joints Under Repeated Two-Step Force
Amplitudes. Journal of Materials Engineering and Performance 2015; 24(6): 2494-2502.
[31] Lu X., Zheng S.L. Strengthening and damaging under low-amplitude loads below the fatigue limit.
International journal of fatigue 2009; 31: 341-345
[32] Makajima M., Jung J. W., Uematsu Y., Tokaji K. Coaxing Effect in Stainless Steels and
High-Strength Steels. Key Engineering Materials 2007; 346: 235–238.
[33] Tanaka T., Fuji T. Studies on Fatigue Damage Caused by Stresses Below the Endurance Limit (The
Effect of Program Period and Fatigue Failure by Stresses Below the Endurance Limit). The Society of
Materials Science, Japan 1972; 232(22): 146-152. (in Japanese)
[34] Mughrabi H. Cyclic slip irreversibilities and the evolution of fatigue damage. Metallurgical and
Materials Transactions B: Process Metallurgy and Materials Processing Science 2009; 40(4): 431–453.
[35] Tanaka K., Mura T. A Dislocation Model for Fatigue Crack Initiation. Journal of Applied Mechanics
1980; 48(1): 97–103.
Highlights
 Number of cycles of low amplitude loading in one block remarkably affected fatigue life.
 Fatigue damage saturation limit has been confirmed under repeated two-step amplitude loading tests.
 Both EN and KHK standards gave inaccuracy evaluation of fatigue life under repeated two-step
amplitude loading condition.
 A rational prediction approach has been established considering low-amplitude loading below the
fatigue limit.
 The proposed predictive approach has been verified by random-amplitude loading tests.