Execution and Evaluation of Cyclic Tests at Constant Load Amplitudes - DIN 50100:2016
Execution and Evaluation of Cyclic Tests at Constant Load Amplitudes - DIN 50100:2016
Rainer Masendorf, Clausthal, Tests with constant load amplitudes are used to characterize the fatigue
and Christian Müller, Ingolstadt, strength behavior of material specimens and components. The S-N
Germany curve derived from these test results describes the relationship between
the load amplitude and the corresponding cycles to failure. Different
concepts for carrying out and evaluating fatigue tests make it difficult to
compare the results from different research institutes. The aim of the new
version of the German standard DIN 50100:2016 is to define a proce-
dure for determining an S-N curve for metallic alloys that does not allow
any scope for interpretation. It is assumed that the test results are sub-
ject to logarithmic normal distributions in both load and cycle direction.
It is further assumed that the S-N curve in the high-cycle fatigue regime
and the long-life fatigue regime can be approximated by a bilinear func-
tion. For the determination of the straight line of finite life, the pearl
Article Information string method and the load level method are available for determining
Correspondence Address the position parameter and the slope of the power function according to
Dr.-Ing. Rainer Masendorf Basquin. Long-life fatigue strength is determined using the staircase
Institut für Maschinelle Anlagentechnik
und Betriebsfestigkeit method and forms the knee point of the S-N curve on average with the
TU Clausthal straight line of finite life. For the long-life fatigue regime, a horizontal
Leibnizstraße 32
38678 Clausthal-Zellerfeld
course or a decrease with low inclination, depending on the material
Germany group examined, is assumed. In addition, DIN 50100:2016 contains infor-
E-mail: [email protected] mation on the accuracy of the estimation of the mean values and the scat-
Keywords ter of the characteristic values according to the sample size. The goal of
S-N curve, constant load amplitude, design of
experiment, sample size, modified staircase achieving comparability of S-N curves is supported by extensive exam-
method ples. An English translation of DIN 50100:2016 is available also.
Since the 1930s, intense efforts have been describing mean stress were added and of fatigue tests causes the same test re-
made to standardize the vocabulary of fa- published, with a supplement in German, sults to be interpreted differently by indi-
tigue tests; these efforts have led to the de- English, French and Russian provided in vidual experts (see Figure 1) [5]. For the
velopment of test guideline DVM 4001 – fa- 1953 [3]. In 1978, the changeover to SI computational lifetime estimation, the re-
tigue testing by the German Association for units occurred [4]. DIN 50100 remained lationship between the cycles to failure
Materials Testing in Engineering (DVM) [1]. unchanged and essentially contained the and the numerous influencing variables is
This test guideline was extended and pub- definition of terms until 2016. usually empirically derived from the test
lished in 1942 as pre-standard DIN 50100, Notes on the execution and evaluation results of many research institutes; how-
[2]. In the further course of development of of fatigue tests were not included. The ever, this procedure is unreliable without
DIN 50100 standards, various methods for lack of a method for evaluating the results knowledge of the test conditions and the
procedure for the test evaluation, and it there is the possibility that no failure of the mated. For this reason, DIN 50100:2016
contributes to the scatter of the estimated specimen will occur until a predefined uses the Basquin [10] approach to approxi-
S-N curve. A working group founded by number of cycles NG is reached. In this mate the curve in the HCF regime by a
the fatigue standards committee in 2012 case, the test result is evaluated as a runout straight line of finite life; see Equation (1).
has revised DIN 50100 for metallic alloys, [6]. The number of cycles NG is also called This approach includes two parameters,
with the goal of comparing the test results the ultimate number of cycles. position C and slope k, to describe the rela-
of various research institutes easily in the Graphically, the results of several tests tionship between the load amplitude La and
future [6]. An English translation of DIN are plotted in a diagram of load amplitude the number of cycles N to failure
50100 will be available soon. In conclusion, La against the number of cycles to failure N.
the evaluation of fatigue tests with constant Both axes of the diagram are logarithmi- N = C · L-ka (1)
load amplitude can also be standardized cally scaled, as shown in Figure 2 with data
internationally [7]. from Maennig [9]. The curve can be divided By taking the logarithm of both sides of
into three regimes: Equation (1), a straight-line equation is ob-
S-N curve – Wöhler curve • Low-cycle fatigue regime (LCF) is char- tained (Equation (2)). It can be concluded
acterized by up to approximately 104 cy- that the function can only be displayed as a
August Wöhler (1819-1914) derived from cles. In the LCF regime, the slope of the straight line if both axes of the diagram are
his test results the phenomenon of fatigue. S-N curve is significantly smaller than scaled logarithmically.
Under cyclic loading, material strength is that in the high cycle fatigue regime. The
significantly lower than tensile strength high plastic strain amplitude in this re- log N = C – k · log La (2)
[8]. He developed testing machines for gime usually requires a strain-controlled
components for various load types and test procedure. LCF tests are not part of The position parameter C describes the in-
methods that are used to observe the defor- DIN 50100:2016. tercept of the cycle axis for a load ampli-
mations under normal operation. Due to • High-cycle fatigue regime (HCF) is char- tude of La = 1. In contrast to the usual
the pioneering achievements accomplished acterized by approximately 104 cycles mathematical representation, the inde-
using this test in the field of fatigue up to the cycle number at the knee point pendent variable La is plotted on the verti-
strength, a test with constant load ampli- NK. Depending on the given load ampli- cal diagram axis and the dependent varia-
tudes is also called a Wöhler test, and the tude La, the specimen will fail after N ble N on the horizontal diagram axis in the
resulting characteristic S-N curve is also cycles. The S-N curve can be approxi- S-N curve. This must be taken into account
called a Wöhler curve. mated by a straight line of finite life plot- when evaluating test results, for example,
ted within a double logarithmic scale. with spreadsheet programs.
Characteristics of S-N curves • Long-life fatigue strength (LLF) is char- The term “LLF strength” is introduced in
acterized by the number of cycles N > DIN 50100:2016 to replace terms such as
In the fatigue test, a specimen or a compo- the cycle number at the knee point NK. “fatigue limit” and “endurance limit”.
nent is stressed with a periodically chang- Depending on the given load amplitude These terms always assume a horizontal
ing load. The load amplitude La and the La, both failures and runouts will occur. course of the S-N curve beyond the knee
mean load Lm are constant during a Wöhler Figure 2 presents the results of a very point NK. Several tests have shown that
test. Depending on the magnitude of the large number of tests. The test results can cracks can also occur in axially stressed,
load amplitude La, it can be applied vari- be used to create an s-shaped curve fit. unnotched steel specimens at cycles
ously, often until a failure criterion, e. g., Usually, significantly smaller sample sizes N > 107. These crack initiations are usually
crack initiation or rupture, has been are available. In this case, the parameters caused by nonmetallic inclusions within
achieved. At small load amplitudes La, of an s-shaped curve fit cannot be esti- the volume of the material [11].
Figure 1: S-N curves after evaluation of the same dataset by different Figure 2: Fit to the s-shape of test results using a bilinear equation;
research institutes; data from [5] data from [9]
60 (2018) 10
FATIGUE TESTING 963
In the regime of LLF, the course of the Using small sample sizes, only quantiles • Nominal bending stress
S-N curve is divided into two types: close to the mean are observed in the ex- • Nominal torsional stress
• Type I: The horizontal course of the S-N periment, for example an interval of 10 % to • Nominal shear force stress
curve, called “LLF strength” (formerly 90 %. Within this small interval, it is impos- If the relationship between external loads
referred to as “fatigue limit” or “endur- sible to identify strong differences between and local stress or strains is known for
ance limit”), which is frequently ob- various probability distribution functions. specimens with stress concentrations
served in cubic-space-centered materi- Only when extrapolating to a small or high caused by notches, according to elasticity
als, e. g., ferritic steels. probability of survival P are there signifi- theory, local stress or strains, such as those
• Type II: Further decrease of the S-N curve cant differences; (see Figure 3). An investi- kinds listed below, can also be used as load
beyond the knee point following an as- gation of a large database with test series of variables L.
sumed straight line with a smaller inclina- sample sizes n >> 10 using modern statisti- • Normal stress
tion k2 compared to the HCF regime is fre- cal test methods has shown that log-normal • Shear stress
quently observed in face-centered or hex- distribution usually better approximates the • Normal strains
agonally packed materials, e. g., austenitic test results than does two-parametric • Shear strains
steel, aluminum, magnesium, or titanium. Weibull distribution [12]. Hence, statistical Note that strain-controlled tests with sig-
Performing tests in the LLF regime, an ulti- evaluation according to DIN 50100:2016 is nificant plastic strains cannot be evaluated
mate number of cycles NG must be prede- based on logarithmic normal distribution, using DIN 50100:2016.
fined, from which point a test is evaluated as which is completely described by its mean A static mean load Lm can be superim-
a runout. Only up to this ultimate number of and its standard deviation. posed on load amplitude La. Load ratio RL
cycles NG is the course of the S-N curve ex- describes the ratio between a minimum load
perimentally proven. Extrapolation of the Different kinds of loads Lmin and a maximum load Lmax; Equation (3).
S-N curve beyond the ultimate number of
cycles NG is prohibited within the scope of Fatigue tests can be carried out on the ba- L min
RL = (3)
DIN 50100:2016. DIN 50100:2016 recom- sis of different kinds of loads L. L max
mends including the number of cycles NG DIN 50100:2016 uses the general term
in the index of the characteristic value for load L for loading a specimen or compo- For the determination of an S-N curve, all
LLF strength, e. g., or LaL,107 or LaL,1E7. nent. Loads can include the following: tests are carried out using the same load
If no better values from experience exist, • Forces ratio RL or the same mean load Lm.
then the following values for the ultimate • Bending moments
number of cycles NG are recommended: • Torsional moments Specimen design and
NG = 5 · 106: for cubic-space-centered ma- • Displacements specimen manufacturing
terials, e. g., ferritic steels or • Angle of torsions
cast iron If a cross section or a section modulus can A prerequisite for the determination of a valid
NG = 107: for cubic face-centered or hex- be defined for the specimen, then the nom- S-N curve is the use of specimens of the same
agonally packed materials, e. g., inal stress, such as the kinds listed below, quality within a test series with regard to ma-
austenitic steel, aluminum, can also be used as load variables L terial, heat treatment, manufacturing technol-
magnesium, or titanium • Nominal tensile/compressive stress ogy, geometry and surface roughness.
Distribution functions
for S-N curves
60 (2018) 10
964 FATIGUE TESTING
No generally valid specimen geometry the LLF regimes. In the HCF regime, a fail- are evaluated by a linear regression using
can be specified in DIN 50100:2016 be- ure criterion is always achieved, e. g., crack the least number of squares for all n value
cause the geometry of specimens depends initiation or rupture. The result of a test in pairs of load amplitude La,i and the number
on various conditions, e. g., the geometry of the HCF regime is the number of cycles N of cycles Ni of a test series. It is important
the semi-finished product, the testing ma- at load amplitude La. DIN 50100:2016 rec- to choose the number of cycles as the de-
chine, or the type of stress. When choosing ommends the pearl string method and the pendent variable in the regression algo-
specimen geometry, it must be ensured load level method for the execution and rithm (Equations (4) and (5)).
that no cracks occur in the area of the spec- evaluation of HCF tests. In the LLF regime,
imen clamping and in the transition zone an ultimate number of cycles NG is speci- see Equation 4 (4)
between the clamping head and the test fied beyond which a test is evaluated as a
cross section. Specimens with cracks out- runout. The results of tests in the LLF re- 1⎛ n
( )
n
(⎞
∑ log Ni +k⋅∑ i=1 log La,i ⎟⎠
n ⎜⎝ i=1
)
side of the test cross-section are rejected as gime are samples of “failures” and C = 10 (5)
invalid. If an S-N curve has to be deter- “runouts”. In DIN 50100:2016 tests in the
mined for a component, e. g., a crankshaft, LLF regime are carried out and evaluated Using the pearl string method, it is as-
a conrod, or a rotor shaft, then the load using the staircase method. sumed that the standard deviation does not
must be applied according to the require- High cycle fatigue – Pearl string method. vary between the load levels. Therefore,
ments of operation. The conditions men- The tests are carried out on different load test results can be shifted parallel to the fi-
tioned above concerning the position of levels in the HCF regime; (see Figure 4). nite life line to a fictitiously load level La,fict
cracks, etc. can become invalid for compo- The advantage of the pearl string method is with the number of cycles Ni,fict to calculate
nents. For example, when testing a its ability to conduct a test according to in- standard deviation (Equation (6)).
threaded connection, the crack usually ap- formation on previous specimens. This al- −k
⎛L ⎞
pears next to the bold head. This crack is lows the option to approach the transition Ni,fict = Ni ⋅ ⎜ a,fict ⎟ (6)
often next to the clamping head; however, areas to the LCF and LLF regimes by shift- ⎜⎝ L ⎟⎠
a,i
it is also a desired behavior. ing load levels stepwise. As a disadvan-
tage, the test results for the middle of the The mean of the number of cycles N50 %,fict
Execution and HCF regime only offer a small contribution on the fictitious load level La,fict and on
evaluation of tests to the estimation of the slope k of the standard deviation ∼slogN are calculated by
straight line of finite life, at least from the Equations (7) and (8).
When carrying out fatigue tests, a distinc- point of view of reliability. Slope k and po-
tion must be made between the HCF and sition C of the finite life line, Equation (2), see Equation 7 (7)
k=−
( ) ( )
n ⋅ ∑ i=1 log L a,i ⋅log Ni − ∑ i=1 log L a,i ⋅ ∑ i=1 log Ni
n n n
( ) Note the number of degrees of freedom of
Equation 4
( ) ( )
2
⎛ 2⎞
n ⋅ ∑ i=1 ⎜ log L a,i ⎟ − ⎛ ∑ i=1 log L a,i ⎞
n n
n – 2 in the denominator of Equation (8).
⎝ ⎠ ⎝ ⎠
Position C and slope k (two parameters)
have to be estimated by the test results.
Therefore, the number of degrees of free-
1 n log N
Equation 7 log N50%,fict = ∑ log Ni,fict → N50%,fict = 10 50%,fict
n i=1
dom has to be set to n – 2. The empirical
standard deviation is underestimated for
small sample sizes and is corrected by
Equation (9) [13].
1
( )
2
⋅ ∑ i=1 log Ni,fict ⋅log N50%,fict
n
Equation 8 s! log N =
n−2 n −1.74
s! log N,korr = s! log N ⋅ (9)
n−2
60 (2018) 10
FATIGUE TESTING 965
tions (10) and (11)). In this manner, a load- equally spaced load levels Li. Since nor- In the staircase method, the test procedure
dependent standard deviation can be evalu- mal distribution with logarithmic char- (load level of the next specimen) depends
ated. The disadvantage of the load level acteristic is also assumed for LLF on the result of the preceding one:
method is that it is not possible to detect if a strength, a constant factor dlog can be • After a failure, the next specimen is
chosen load level is in the transition area to defined between the neighboring load tested on the next lower load level with
the LCF or HCF regime after the test series levels. Considering a logarithmic scale, equidistant increment dlog
has been completed. this factor leads to visually equally • After a runout, the next specimen is
spaced load levels Li. To obtain one to tested on the next higher load level with
see Equation 10 (10) three load levels with runouts and fail- an equidistant step dlog.
ures, the stair case factor dlog is calcu- Due to the test procedure used, the test re-
1
( )
2
⋅ ∑ i=1 log Ni − log N50%
n
slog N = (11) lated as a function of an estimated stand- sults are automatically concentrated around
n −1 ard deviation of the population slogL,GG; the mean of LLF strength (see Figure 6).
Equation (15). Good results are achieved The evaluation of an interrupted staircase
In analogy to the pearl string method, if the stair case factor dlog is approxi- sequence is prohibited using the ad-
standard deviation slogN is underestimated mately the same size as the standard de- vanced IABG-method [15] as proposed in
for small samples sizes n and is corrected viation slogL,GG of the population. DIN 50100:2016. Test results at the begin-
by Equation (12) [13]. ning of the staircase sequence are only
dlog = 10SlogL,GG (15) taken into account in the evaluation if the
n − 0.74 load level is validated during the staircase
slog N,korr = slog,N ⋅ (12)
n −1 For some materials and components, sequence. An ordinal digit i is assigned to
DIN 50100:2016 contains data for typical each of the occupied load levels. The lowest
Using the means N50 % of the two load levels standard deviations of the population valuable load level is assigned to the ordi-
La1 and La2, slope k of the finite life line is slogL,GG to help the test engineer choose the nal digit 0, and the higher load levels are
calculated (Equation (13)). staircase factor dlog. Load levels Li are cal- assigned to 1, 2, etc. (see Figure 6). The
culated on the basis of an estimated LLF number of events fi (sum of runouts and
⎡ ⎛N ⎞⎤
⎢ log ⎜ 50%,La1 ⎟ ⎥ strength LaL,NG (Equation (16)). failures) is calculated for each load level.
⎢ ⎜⎝ N50%,La2 ⎟⎠ ⎥ Since no distinction is made between fail-
k = −⎢ ⎥ (13) Li = LaL,NG · (dlog)i i = ..., -2, -1, 0, 1, 2 (16) ures and runouts, a fictitious test result can
⎢ ⎛L ⎞ ⎥
a1
⎢ log ⎜ ⎟ ⎥
⎢⎣ ⎝ L a2 ⎠ ⎥⎦
1 n
∑ log Ni → N50% = 10 50%
log N
log N50% = Equation 10
Parameter C, which describes the position n i=1
of the line of finite life, Equation (1), corre-
sponds to the intercept of the number of
cycles for La = 1, Equation (14).
60 (2018) 10
966 FATIGUE TESTING
60 (2018) 10
FATIGUE TESTING 967
For a steel specimen that is forged and thus whether the slope of the finite life line research results have become state of the
machined, a standard deviation of the pop- will be underestimated. Thus, test results art and have also been incorporated into
ulation slogL,GG = 0.030 is estimated for LLF far from the center are favorable for esti- DIN 50100:2016. With such accomplish-
strength LaL,NG.GG. Depending on the de- mating the parameters with a high accu- ments, Prof. Zenner has become a quasi-co-
sired permissible error, the user can racy. Previous knowledge of the approxi- author with us, and we would like to thank
choose the required sample size from mate position of the fatigue strength line is him emphatically.
DIN 50100:2016 (see Table 3). required for the successful application of The new version of DIN 50100:2016, in-
Considering the examples above with the load level method. cluding an English translation, was pre-
the given assumptions, the parameters of The term “LLF strength” is introduced pared by an ad hoc working group of the
the population, such as the mean and the and replaces the terms “fatigue limit” and DIN Working Committee Fatigue Testing in
standard deviation, will appear within a “endurance limit”. The goal is to indicate the course of 19 meetings from 2012 to
confidence of 80 %. The boundaries of the that the S-N curve can continue to decrease 2018. The members of the working group,
confidence band are equal to the permissi- even after the ultimate number of cycles at Matthias Ell, BAM Berlin, Christoph Hen-
ble errors given in the Tables 1 to 3. the knee point. LLF strength is determined kel, AMAG Ranshofen, Hellmuth Klingel-
Tables 1 to 3 state that a mean value can using a modified staircase method. In this höffer, BAM Berlin, Franz Klubberg, RWTH
be well estimated even with relatively method, the test results concentrate around Aachen, Rainer Masendorf, TU Clausthal
small sample sizes n. The accuracy of the the mean of LLF strength automatically. No and Rainer Wagener, Fraunhofer LBF, have
estimation of the standard deviation, on the precise prior knowledge of the position of created an easy-to-use standard through
other hand, is always poor, even for large LLF strength is required. Although the their engagement and constructive discus-
sample sizes n. Standard deviation is nec- mean value of LLF strength can be esti- sions that will simplify the comparison of
essary for calculating values of small prob- mated well by means of just a few speci- the test results in the future.
abilities of failure. In these cases, mens (n ≥ 15), the estimation of the stand-
DIN 50100:2016 recommends the use of ard deviation is inaccurate even for sam- References
standard deviations from the literature that ples sizes of n = 100. If standard deviation
1 DVM 4001: Fatigue testing; definitions and sym-
result from averaging the standard devia- has to be used for extrapolation of small bols (in German), 2nd draft DVM, Berlin, Zeitschrift
tions of a large number of tests. The esti- probabilities of failure, the use of standard für Metallkunde 28 (1936), No. 4, pp. 102-104
mated standard deviations in single tests deviations obtained from averaging many 2 DIN 50100 pre-standard: Test for fatigue
will lead to inaccurate results. Typical test series of comparable materials or com- strength; definitions, symbols, execution
standard deviations are given by Adenstedt ponents is recommended. The use of stand- (in German) (1942)
[17]. For some material groups and compo- ard deviation from a single test series can 3 DIN 50100: Fatigue test (Woehler test);
definitions, symbols, execution, evaluation
nent types, DIN 50100:2016 also contains lead to inaccurate results.
(in German), Beuth, Berlin, Germany (1953)
values for typical standard deviations. For test evaluation, it is assumed that 4 DIN 50100: Fatigue test (Woehler test);
test results are subject to a logarithmic nor- definitions, symbols, execution, evaluation
Conclusions mal distribution in the direction of the cy- (in German), Beuth, Berlin, Germany (1978)
cles and in the direction of the load. 5 J. M. Finney, J. Y. Mann: Fatigue S/N Data in
The aim of the new version of To limit the scope for interpretation, relation to variability in predicted life, Proc.
DIN 50100:2016 is to standardize the plan- DIN 50100:2016 contains numerical and of the Conf. on Aircraft Structural Fatigue,
Melbourne (1976), pp. 149-178
ning, execution and evaluation of fatigue graphical examples for the evaluation and
6 DIN 50100: Load controlled fatigue testing —
tests with constant load amplitudes to ob- representation of fatigue tests with con- Execution and evaluation of cyclic tests at con-
tain comparable results at different re- stant amplitudes. stant load amplitudes on metallic specimens
search laboratories. An English translation is available also and components (in German), Beuth, Berlin,
In DIN 50100:2016, the S-N curve is di- to allow the procedure for fatigue tests with Germany (2016)
vided into the areas of high-cycle fatigue constant load amplitude to also be stand- 7 DIN 50100: Load controlled fatigue testing —
(HCF) (each specimen fails) and long-life fa- ardized internationally. Execution and evaluation of cyclic tests
at constant load amplitudes on metallic
tigue (LLF) (at the same load amplitude fail-
specimens and components, Beuth, Berlin,
ures and runouts can occur), which differ in Acknowledgements Germany
the execution and evaluation of the tests. 8 H. Zenner, K. Hinkelmann: Fatigue of Compo-
Two methods are proposed in Prof. Zenner has influenced a generation of nents-August Wöhler (1819-1914) A Historical
DIN 50100:2016 for determining the fatigue strength researchers. Many of his Review, DVM, Berlin, Germany (2017)
straight line of finite life in the HCF re-
gime. Using the pearl string method, load
levels are chosen with respect to the re- Required sample size n
Permissible
sults of all previously tested specimens Error (%) 10 14 20 50 100
within this series. No precise previous
knowledge of the position of the finite life Fo 4.7 3.8 3.3 1.9 1.4
Estimated LLF strength LaL,NG
line is required. Fu -4.5 -3.6 -3.2 -1.9 -1.4
With the load level method, two load lev- Fo 216.4 173.5 126.8 49.3 30.5
els at which the tests will be carried out are Estimated standard deviation slogL
Fu -68.4 -63.4 -55.9 -33.0 -23.4
defined in advance. It is not possible to de-
termine whether the tests are in the transi- Table 3: Staircase method, required sample size n as a function of the required permissible error for a
tion area to the LCF or LLF regimes and confidence of 80 %, standard deviation of the population slogL,GG = 0.030
60 (2018) 10
968 FATIGUE TESTING
Bibliography
DOI 10.3139/120.111238 of Technology (TUC), Germany and has been a Dr.-Ing. Christian Müller, born in 1984, stud-
Materials Testing scientific employee at the Institute for Plant ied Mechanical Engineering at Clausthal Univer-
60 (2018) 10, pages 961-968 Engineering and Fatigue Analysis (IMAB) of TUC sity of Technology (TUC) and was a scientific em-
© Carl Hanser Verlag GmbH & Co. KG since 1994. His PhD thesis (2000) considered the ployee at the Institute for Plant Engineering and
ISSN 0025-5300 influence of the prestraining cyclic material prop- Fatigue Analysis (IMAB) of TUC between 2010
erties of thin sheets. He has been a leading engi- and 2015. He completed his PhD thesis on the
The authors of this contribution neer at IMAB (TUC) in Clausthal, Germany since statistical evaluation of S-N curves in 2015. Since
2000. The focus of his work is fatigue testing of then, he has worked in the field of the fatigue
Dr.-Ing. Rainer Masendorf, born in 1964, studied materials and components for determining strength of high-voltage batteries at Audi
Mechanical Engineering at Clausthal University fatigue properties. Ingolstadt, Germany.
60 (2018) 10