International Journal of Fatigue 74 (2015) 38–45

Contents lists available at ScienceDirect

International Journal of Fatigue

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

Validation of a high-cycle fatigue model via calculation/test

comparisons at structural scale: Application to copper alloy
sand-cast ship propellers
Anthony Ezanno a,b, Cédric Doudard a,⇑, Sylvain Moyne a, Sylvain Calloch a, Thierry Millot c,
David Bellèvre d
a
Laboratoire Brestois de Mécanique et des Systèmes – LBMS EA4325, ENSTA Bretagne/UBO/ENIB, 2 rue François Verny, 29806 Brest, France
b
DGA Techniques navales, Avenue de la Tour Royale, BP 40915, F-83050 Toulon Cedex, France
c
DCNS Centre de Nantes – CESMAN, Indret, 44620 La Montagne, France
d
DGA Techniques hydrodynamiques, 27105 Val-de-Reuil, France

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

Article history: An in-house developed software is proposed to study structures made of cast alloy and subjected to high-
Received 24 July 2014 cycle fatigue. This in-house developed software is based on the implementation of a probabilistic two-
Received in revised form 10 December 2014 scale model for high cycle fatigue life prediction. To validate the proposed approach, calculation/test
Accepted 17 December 2014
comparisons at the scale of an engineering component (i.e., copper alloy sand cast ship propellers) have
Available online 31 December 2014
been performed.
So, the first part of the present paper concerns the description of the model for fatigue life prediction
Keywords:
and of the fatigue tests performed on two marine propellers. In a second part, the performance of the
Probabilistic two-scale model
Crack initiation
proposed model is estimated by comparing experimental and theoretical results in the case of full-scale
Casting defects fatigue tests. The results are then discussed and analyzed.
Sand cast aluminum bronze Ó 2014 Elsevier Ltd. All rights reserved.
Full-scale fatigue tests

1. Introduction example, their capability to describe or even to predict the S–N

curves (i.e., amplitude Stress–Number of cycles to failure) of
Fatigue, and more precisely High-Cycle Fatigue (HCF), is still materials.
one of the main causes of the fracture in service of mechanical Engineers and designers often ask themselves the relevant
engineering components and structures [1,2]. Guaranteeing and question concerning the capability of the models for fatigue life
ensuring the reliability of mechanical systems in relation to fatigue prediction to predict the fatigue strength of components (i.e., of
phenomena is always a difficult question for engineers. Indeed, structures) e.g., [3,4]. It is true that this question is not so easy.
high-cycle fatigue is an aspect of the mechanics of materials and There are various reasons for this:
structures which is particularly insidious because of its progressive
and masked character. Moreover, high-cycle fatigue is a complex – The first is due to the volume effect e.g., [4–6]. Indeed, the fati-
phenomenon in so far as it depends on a great number of parame- gue properties depend on the volume of the specimens that
ters (e.g., chemical composition of the material, microstructure, have been used to determine them. The fatigue limit is often
mechanical loading, environment, temperature, size of the struc- lower when it has been determined by using voluminous spec-
tures. . .). Over the last few years, significant advances have been imens than when it has been determined by using less volumi-
made in the comprehension of both microstructural mechanisms nous specimens. The passage from the scale of a specimen to
of fatigue damage and modeling [2]. Today, a great number of the scale of a structure is not so trivial.
models for fatigue life prediction of materials have been proposed. – The second is due to the stress heterogeneity effect [4–7]. This
In the majority of the cases, the models for fatigue life prediction effect is observed, for example, when one compares the fatigue
are validated at the scale of Representative Elementary Volume limit of a material determined by using tension–compression
(REV) (i.e., at the scale of fatigue specimen) by showing, for fatigue tests to the fatigue limit determined by bending fatigue
tests. It is generally found that the bending fatigue limit of a
⇑ Corresponding author. Tel.: +33 2 98 34 89 27; fax: +33 2 98 34 87 30. material is higher than the tension–compression fatigue limit
E-mail address: [email protected] (C. Doudard). of the same material [7–10]. One can be faced with such a

http://dx.doi.org/10.1016/j.ijfatigue.2014.12.008
0142-1123/Ó 2014 Elsevier Ltd. All rights reserved.
 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45 39

problem when we want to pass from the scale of a specimen to

the scale of a structure, because the stress heterogeneity could
be very different.
– ...

Therefore, in this paper we propose to validate a high-cycle fati-

gue model (already validated at the scale of REV [11]) through cal-
culation/test comparisons at the scale of an engineering
component. In this study, particular attention is being paid to
large-dimension copper alloy sand cast ship propellers. This case
is particularly interesting for several reasons:

– The material, chosen for this study, contains a population of ini-

tial micro-flaws due to the casting process.
– The volume difference between the specimens and the compo-
nent (i.e., ship propeller) is very large.
– The stress heterogeneity in the fatigue specimens and in the Fig. 1. One of the two fixed pitch supra-divergent propellers under study.
component is also very different.
– ... Table 1
Chemical composition of Copper Aluminum (wt%).
The present paper is divided into five parts. The first part is ded- Cu Al Ni Fe Mn
icated to the two marine propellers and to the material under Balance 8.5–10.2 4.0–5.5 3.0–5.5 0–1.5
study. In the second part of the paper, the constitutive equations
of the high-cycle fatigue model are briefly examined [11]. The con-
sidered model belongs to the two-scale model family. It has been
developed in a probabilistic framework and takes into account
the presence of the initial micro-flaws by an indirect approach.
The third part of the paper concerns description of the in-house
developed fatigue software named ‘‘4Cast’’. This in-house devel-
oped fatigue software is dedicated to the fatigue life prediction of
cast structures. The architecture of the proposed software and
the different outputs are presented. In the fourth part of the paper,
a full-scale fatigue test campaign on two ship propellers is
described. The principle of the fatigue tests on marine propellers
is first given. Then, the fatigue tests are performed and validated.
After that, the fatigue results are given. In the last part of the paper,
the performance of the proposed model is estimated by comparing
experimental and theoretical results in the case of full-scale fatigue
tests. The results are then discussed and analyzed.

Fig. 2. Optical micrograph of the microstructure of Copper Aluminum alloy.

2. The two sand cast ship propellers under study
The mean size of this population of defects is close to 16 lm.
The two fixed pitch supra-divergent propellers under study Table 2 compiles some characteristics of the population of defects.
come from a patrol boat of the French Navy. Note that the two pro- Obviously, the presence of initial defects in the alloy plays a cru-
pellers are prototypes, which have been used for only a few days. It cial role in its fatigue life e.g., [14,20]. Nevertheless, in the light of
can be considered that they are without mechanical history. Each the characteristics of the population of defects present in our mate-
propeller has four blades. Their diameters measure 1.71 m and rial (Table 2), the casting flaws are ‘‘invisible’’ when using classical
weigh approximately 465 kg (Fig. 1). nondestructive testing methods (i.e., nondestructive testing meth-
They are made of a sand cast aluminum bronze CuAl9Ni5Fe4 ods used in a classical industrial context). That is the main reason
(i.e., Cupro Aluminum alloy). This alloy is commonly used in mar- that an indirect approach was adopted during this work. That
ine applications because of its excellent marine corrosion resis- means that the characteristics of the population of defects (i.e., size
tance (e.g., [12,13]). Table 1 gives the chemical composition of distribution, number of defects, . . .) will not be considered as ingre-
the studied material. dients of the proposed high-cycle fatigue model [11].
Fig. 2 shows a typical microstructure of the material. The micro-
structure consists of a coarse a-phase within a eutectoid a + jIII 3. A high-cycle fatigue model for cast materials
phase where jIII-precipitates are a Ni-rich lamellar-shaped phase.
Two other precipitates appear during the slow cooling of the alloy: In this part, the high-cycle fatigue model considered is briefly
an Fe-rich rosette-shaped phase (jII-precipitate) and an Fe-rich depicted. Only the necessary and sufficient ingredients for fatigue
jIV-phase which precipitates finely into the a-grains. More details life determination of a structure are presented. For more details,
on the characteristic microstructures of cast copper–aluminum an interested reader can refer to the reference [11].
alloys can be found for example in [12,13]. As previously stated, the casting process induces the presence
Moreover, as can be seen in Fig. 2, the casting process induces of initial micro-flaws in the material. The possible presence of
initial flaws in the material such as micro-shrinkage pores. Fig. 3 macro-flaws (i.e., detectable by classic nondestructive methods
shows the distribution of defects of identified size on a surface of used in an industrial context) in the structure would imply its
100 mm2. About 800 defects are present on the observed surface. deposition. In the considered high-cycle model, the presence of
40 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45

where N(X) is the average number of active sites defined by

Z
NðXÞ ¼ kðMÞdV; ð2Þ
X

with k(M), the density of active sites for each point M in the consid-
ered domain, X.
In this paper, it is assumed that k(M) follows a power law of the
equivalent stress amplitude, R0 ðMÞ,
m
1 hR0 ðMÞ  Q i

kðMÞ ¼ ; ð3Þ
V 0 S0  a pm ðMÞ
where V 0 Sm
0 , a , Q and m are four material parameters and hi is
⁄

Macauley’s brackets (i.e., positive part of ‘’). The equivalent stress

amplitude, R0 ðMÞ, and the average hydrostatic pressure, pm(M),
are defined by the following equations
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
3
Fig. 3. Histogram of the defect size distribution. R 
0 ðMÞ ¼ Maxt @ ðRðtÞ  Rm Þ : B : ðRðtÞ  Rm ÞA; ð4Þ
2

Table 2 1
Characteristics of the population of defects on an analyzed surface of 98 mm2. pm ðMÞ ¼ TraceðRm Þ; ð5Þ
3
Area of analyzed Total number Mean size of Cumulated area of
surface (mm2) of defects defects (lm) defects (mm2)
where Rm is the average stress tensor depending on the cyclic load-
ing path, RðtÞ, and B is a fourth order operator given by
98 789 15.9 0.22

B ¼ J þ 2b K; ð6Þ

initial micro-flaws is taken into account by an indirect approach with b⁄ a material parameter. K and J are the projector operators in
(i.e., the shape, the size and the distribution are not ingredients
of the model). When a cyclic loading is applied to the material, hydrostatic and deviatoric spaces, respectively, defined by the
micro-plasticity appears around different cast defects. It is following equations
assumed that this micro-plasticity governs the fatigue damage ini- 1
tiation [20]. As shown by [20], none of the defects are identical so K:R¼ TraceðRÞI; ð7Þ
3
that the micro-plasticity appears for different stress levels around
the defects. Fig. 4 shows a schematic representation of the evolu- J : R ¼ R  K : R; ð8Þ
tion of the mean number of active sites (i.e., defects all around
which micro-plasticity occurs) as a function of the loading where I is the unit tensor.
amplitude. In the present study, the weakest link theory is considered [25]
To describe the micro-plasticity activity, a set of elasto-plastic which is usually done in high-cycle fatigue domain. Thus, the prob-
inclusions (in the mechanical sense of the term), randomly distrib- ability of failure initiation, PF, is equal to the probability of finding
uted within an elastic matrix, is considered [21,22]. It is assumed at least one active site in the considered domain, given by
that a Poisson Point Process describes the activation of the sites.
Then, the probability, Pk(X), of finding k active sites in a domain
PF ¼ PkP1 ðXÞ ¼ 1  Pk¼0 ðXÞ: ð9Þ
X of volume VX is given by [23,24] By using Eq. (3), the probability of failure initiation reads
k m 
V eff hRF  Q i
 
½NðXÞ
Pk ðXÞ ¼ exp½NðXÞ; ð1Þ PF ¼ 1  exp  ; ð10Þ
k! V 0 S0  a pmF
Number of active sites

Loading amplitude, Σ0

Fig. 4. Schematic representation of the activation of the sites versus the loading amplitude [11].
 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45 41

where V eff ¼ V X Hm is the effective volume, pmF ¼ MaxX ðpm ðMÞÞ and
RF ¼ MaxX R0 ðMÞ , the maximum values, in the domain X, of the
hydrostatic pressure and the equivalent stress, respectively [26].
The stress heterogeneity factor, Hm, is given by
m
1 hR0 ðMÞ  Q i S0  a pmF
Z 
Hm ¼ dV: ð11Þ
VX X hRF  Q i S0  a pm ðMÞ
The mean fatigue limit, R1 , and the standard deviation, R1 , are
then given by, respectively
1 
1
dPF V0 m 1
Z  
R1 ¼ RF  
 dRF ¼ ðS0  a pmF Þ C 1þ þ Q; ð12Þ
0 dRF V eff m
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Z 1  dP
F
R1 ¼ R2F  R1
2 dRF
0 dRF
 1 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
V0 m 2 1
   
2

¼ ðS0  a pmF Þ C 1þ C 1þ ; ð13Þ
V eff m m
R1
where CðpÞ ¼ 0 t p1 expðtÞdt is the gamma function [27]. Eqs.
(11) and (12) account for volume and stress heterogeneity effects
associated to the Weibull model [28,29].
Finally, the S–N–P curve (i.e., amplitude Stress–Number of
cycles to failure–Probability of failure) associated with a failure
probability, PF, is given by [11]
A
NðPF Þ ¼ ; ð14Þ
hR  R1 ðPF Þi2

F

where A is a material parameter and R1 ðP F Þ the fatigue limit for a

failure probability, PF.
In the next section, the implementation of the considered
model in the high-cycle fatigue in-house developed software is
presented. Fig. 5. The flowchart of the 4Cast postprocessor.

4. The high-cycle fatigue in-house developed postprocessor: 0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

4Cast 3
R ¼ Maxi¼f1;...;ng @

0 ðRi  Rm Þ : B : ðRi  Rm ÞA: ð16Þ
2
4Cast is high-cycle fatigue in-house developed software dedi-
cated to the fatigue life prediction of cast structures. This program At this stage, an Indicator of Critical Zones, ICZ, can be intro-
is used as a postprocessor of an elastic finite element analysis. Thus, duced. This indicator defined by
for a given structure, with its boundary conditions, and submitted to
a given cyclic loading, the role of 4Cast postprocessor is to determine:
hR0  Q i
ICZ ¼ ; ð17Þ
S0  a pm
– the global fatigue failure probability of the analyzed structure, permits the determination of critical zones of the structure where
– the number of cycles to the initiation of fatigue failure for a the fatigue initiation problems could happen. The greater the indi-
given probability, cator ICZ, the more critical the zone. Nevertheless, this indicator
– the critical areas (i.e., the different zones where the fatigue ini- does not give the risk of fatigue initiation. This information is given
tiation could occur). by the global fatigue failure probability of the structure, PF, which is
given by Eq. (10) in which the effective volume, Veff, is
Fig. 5 shows the flowchart of the 4Cast postprocessor. One can
Z   m
see the four main stages. First, a loading cycle is partitioned in n hR0 ðMÞ  Q i S0  a pmF
V eff ¼ dV
different steps. For each step, an elastic finite element analysis is X

hRF  Q i S0  a pm ðMÞ


performed to determine the stress field in the analyzed structure. X  Z 

As soon as the stress fields have been determined for all the steps, ¼ Ne dV e  ½He  ; ð18Þ
Ve
the values of average stress, Rm ðMÞ, can be calculated at each inte- elts

grating point of each finite element. The average stress is obtained where Ne is the finite element basis function and He the heterogene-
via a minimization procedure defined by ity factor determined at each node, j, of the considered finite ele-
2 0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi13 ment and given by
3 
Rm ¼ MinX 4Maxi¼f1;...;ng @ Ri  X : B : ðRi  XÞA5: ð15Þ !m
2 hR0 jj  Q i S0  a pmF
H e jj ¼ : ð19Þ
hRF  Q i S0  a pm jj
From the practical point of view, the algorithm proposed by
Yildirim [30] has been used to solve this minimization/maximiza- Finally, in the case where the probability, PF, is not null, the
tion problem. Then, the equivalent stress amplitude, R0 , can be number of cycles to fatigue initiation, Pu, is obtained by combining
calculated at each integrating point by Eqs. (10) and (14)
42 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45

A Concerning the connection between the propeller and the

NðPu Þ ¼  1=m 2 : ð20Þ

lnð1Pu Þ 2
bracket stiffener, the propeller is embedded by wedging on a
1 lnð1P F Þ

hR  Q i
F cone-shaped shaft, fixed by bolts on a bracket stiffener (Fig. 8).
The orientation of the propeller is obtained thanks to a system of
In this section, the probabilistic high-cycle fatigue model dedi- two pieces with an inclined plane wedge placed between the
cated to cast materials and presented in Section 3 has been imple- cone-shaped shaft and the bracket stiffener (Fig. 8).
mented in a fatigue in-house postprocessor, 4Cast. To access the Concerning the connection between the propeller and the two
predictive capacity of the proposed approach, a comparison hydraulic actuators, a clevis assembling system (one on the blade,
between fatigue tests carried out on two ship propellers and one on the actuator) linked with a textile sling has been chosen
results predicted by the proposed high-cycle fatigue model will (Fig. 6). The use of this system on the propeller blade imposes
be performed in the following sections. First, in the next section, the drilling of the blade for the setting up of a bolt supporting
the fatigue tests developed on marine propellers are presented. the two clevises on each side of the blade (Fig. 9). This bolt has
Finally, the experimental results will then be compared with the been manufactured according to the concept of cold rolling to
prediction of the numerical approach. ensure fatigue resistance [32]. The contact between the clevises
and the curved surface of the blade is ensured by an epoxy resin
5. Full-scale fatigue tests on ship propellers CHOCKFASTÒ Orange, (Fig. 9). The contact area is 80 mm2.

This section deals with the description of the fatigue tests per- 5.2. Fatigue tests
formed on the two ship propellers. First, the experimental equip-
ment and the main technical choices are presented. It should be The high-cycle fatigue model proposed in this paper is based on
noted that the objective of these tests is not to reproduce a loading an initiation criterion. Consequently, the fatigue tests on the pro-
representative of the hydrodynamic loading seen by a ship propel- peller blades are carried out up to crack appearance. It implies
ler in service [13]. But our objective is to perform fatigue tests with the choice of a crack detection technique as well as the definition
well-controlled boundary conditions facilitating the comparisons of a critical crack length for the quantification of the number of
between experimental results and the numerical predictions. The cycles at initiation.
experimental set-up is then validated by comparison of some In this work, one considers initiation to be as soon as at least
experimental results under static or cyclic loading and elastic finite one crack of minimum length equal to 3 mm is detected. This cri-
element simulations. Finally, the results of fatigue tests obtained terion allows full-scale experimental fatigue results to be linked to
on the two ship propellers are given. model predictions obtained with small characterization specimens.
Among the existing different crack detection techniques, the
5.1. Experimental equipment and experimental setup only one able to respect this criterion, while keeping in mind that
the studied area is quite large, is penetrant testing. Fig. 10 illus-
The fatigue test consists in applying a cyclic bending loading on trates this technique applied in the case of a blade propeller.
each propeller blade separately thanks to two antagonist servo-
hydraulic actuators (Fig. 6). Each actuator is linked to the propeller
with a sling. Moreover, the direction of the loading (i.e., direction of
the axes of the hydraulic actuators) has been chosen to be normal
to the surface of the blade at force application point.
Fig. 7 shows the location of the force application point on the
blade. This point has been chosen as the intersection between
the blade reference line, which passes through the center of all
the sections of the blade, and the section at 0.7 times the maxi-
mum radius, R0, of the propeller (Fig. 7). For an average propeller,
it is in this section that the hydrodynamic thrust is maximum [31].
So particular attention has been paid to the design of the con-
nection between, on the one hand, the propeller and the bracket
stiffener and, on the other hand, the blade and the two slings
(Fig. 6). Fig. 7. Definition of the location of the force application point.

Fig. 6. Experimental setup for fatigue tests on a patrol boat propeller blade.
 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45 43

ratio (Table 3). In the next part, these results are compared to the
predictions of the aforementioned high-cycle fatigue model.

6. Comparison between experimental results and HCF model

predictions

6.1. Material parameter identification

4Cast is a high-cycle fatigue in-house postprocessor based on the

use of a two-scale probabilistic model [11]. This probabilistic HCF
model depends on six material parameters (namely, m, a⁄, Q, b⁄,
V 0 Sm
0 and A). m describes the scatter of fatigue results, a accounts
⁄

for the effect of the mean stress on fatigue properties, Q is a yield

microplasticity appearance, b⁄ accounts for the effect of hydrostatic
pressure on microplasticity evolution, V 0 Sm0 is a scale parameter for
the fatigue response and A the only parameter introduced in the
Fig. 8. Propeller hub axis orientation. description of S–N–P curves [11]. All the material parameters can
be calibrated from results of classic fatigue tests. Table 4 gives the
identified values of the 6 material parameters and the different fati-
gue tests used to calibrate this set of material parameters.

6.2. Results for the loadings at R = 1

The three loadings with a loading ratio of 1 have been simu-

lated and analyzed with our post-processor 4Cast. For each loading
amplitude (i.e., 40, 45 and 50 kN), the numerical analysis gave a
probability of failure near 1, suggesting that propeller blades have
a high risk of crack initiation. These numerical results have been
validated by the experimental observations reported in the previ-
ous part of this paper. Indeed, for the three cyclic loadings with a
loading ratio of 1, cracks were observed after a certain number
of cycles (Table 3).
Fig. 13 shows the distribution of the critical zone indicator, ICZ,
and the experimental initiation locus on the two faces of the pro-
peller blade. One can see that the center of the suction side is the
more critical area for crack initiation (Fig. 11). This prediction is
in good agreement with experimental observations. Indeed, all
the crack initiation sites observed during the tests were located
in the central part of the suction side (Fig. 11).
The comparisons between experimental numbers of cycles to
Fig. 9. Textile sling anchorage at the blade.
crack initiation with those determined by the theoretical approach,
for different probabilities of crack initiation, are reported in Fig. 12.
During this campaign, several loading amplitudes at different The blue points illustrate the experimental number of cycles to
loading ratios and at a load frequency of 5 Hz have been studied. crack initiation (on the right Y-axis) versus the number of cycles
The different loading conditions are given in Table 3. This table to crack initiation predicted by the model for a probability of 50%
also indicates the initiation fatigue life of each tested propeller (on the X-axis). On the other hand, the numbers of cycles to crack
blade and the number of detected cracks at the end of the fatigue initiation predicted by the model for probabilities of 10%, 50% and
test. 90% (on the left Y-axis) are represented as a function of the number
This fatigue test campaign on propellers has allowed us to gather of cycles to crack initiation predicted by the model for a probability
several pieces of information concerning the fatigue properties of of 50% (on the X-axis).
our propellers, i.e. the number of cycles at initiation and their local- All these results show that the theoretical predictions are in
ization on the blade for different loading amplitudes and loading good agreement with experimental results.

Fig. 10. Penetrant testing on a blade for fatigue crack detection: suction side of a blade after 140 000 cycles under ±50 kN.
44 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45

Table 3
Experimental results of fatigue tests on propeller blades.

Propeller Loading ratio Amplitude loading Number of cycles to Total number of Number of initiated
number R (kN) initiation cycles cracks
Port propeller 1 1 ±50 – 135 000 2
2 1 ±50 105 000 140 000 4
3 1 ±45 165 000 220 000 5
4 1 ±40 600 000 700 000 10
Starboard 1 1 ±40 300 000 500 000 8
propeller 2 0.1 40 ± 33 600 000 700 000 9
3 0.1 41.5 ± 34 350 000 550 000 8
4 – – – – –

Table 4
The tests used for the calibration of the six material parameters.

Material parameters The tests used for the identification Identified value

V 0 Sm
0
Mean fatigue limit under torsion at Rr ¼ 1 6.5  1017 mm3 MPa7
Q S–N–P curve under tension–compression at Rr ¼ 1 26 MPa
m S–N–P curve under tension-compression at Rr ¼ 1 7
b⁄ S–N–P curve under tension-compression at Rr ¼ 1 0.1
A S–N–P curve under tension-compression at Rr ¼ 1 2.67  109 MPa2
a⁄ Mean fatigue limit under tension-compression at Rr ¼ 0:5 1.1

Fig. 13. Comparison between experimental and theoretical results concerning

Fig. 11. Cracking risk map at loading ratio R = 1 and locus of crack initiation crack initiation for the loadings with a loading ratio of 0.1.
observed after fatigue tests.

7 7 loadings (i.e., 40 ± 33 and 41.5 ± 34 kN), the numerical analysis

10 10
gave a probability of failure near 1, suggesting that propeller blades
have a high risk of crack initiation. These numerical results were
at initiation for a determined P F
Experimental number of cyles

P =90%
Predicted number of cycles

validated by the experimental observations reported in the previ-

F

PF=50% ous part of this paper. Indeed, for the two cyclic loadings with a
at initiation, N a exp

loading ratio of 0.1, cracks were observed after a certain number

6 6 of cycles (Table 3).
10 10
P =10% Fig. 13 shows the distribution of the critical zone indicator, ICZ,
F
and the experimental initiation locus on the two faces of the pro-
peller blade. One can see that the center of the pressure side is
the more critical area for crack initiation (Fig. 13). This prediction
is in good agreement with experimental observations. Indeed, all
5 5 the crack initiation sites observed during the tests were located
10 10
5 6 7 in the central part of the pressure side (Fig. 13).
10 10 10
The comparisons between experimental numbers of cycles to
Predicted number of cycles at initiation for
crack initiation with those determined by the theoretical approach,
P =50%, N (P =50%)
F a num F for different probabilities of crack initiation, are reported in Fig. 14.
The same type of representations as that already presented in
Fig. 12. Comparison of experimental and theoretical results concerning crack
initiation for the loadings with a loading ratio of 1. Fig. 12 is used in Fig. 14. The results show that the theoretical pre-
dictions are in good agreement with experimental results.
6.3. Results for the loadings at R = 0.1 Consequently, the proposed probabilistic fatigue model the
parameters of which are identified at the scale of Representative
The two loadings with a loading ratio of 0.1 were simulated and Elementary Volume, is relevant to predict the fatigue life of compo-
analyzed with our post-processor 4Cast. For the two considered nents. The volume effect on fatigue properties associated with the
 A. Ezanno et al. / International Journal of Fatigue 74 (2015) 38–45 45

10
7 7 [3] Kuguel R. The highly stressed volume of materials as a fundamental parameter
10
in the fatigue strength of metal members. Department of theoretical and
applied mechanics report 169. Urbana (IL): University of Illinois; 1960.
Experimental number of cyles

at initiation for a determined P

PF=90% [4] Chantier I, Bobet V, Billardon R, Hild F. A probabilistic approach to predict the

Predicted number of cycles

PF=50% very high-cycle fatigue behavior of spheroidal graphite cast iron structures.
at initiation, N a exp

Fatigue Fract Eng Mater Struct 2000;23:173–80.

[5] Avalle M, Belingardi G, Cavatorta MP, Doglione R. Casting defects and fatigue
strength of a die cast aluminum alloy: a comparison between standard
106 10
6 specimens and production components. Int J Fatigue 2002;24:1–9.
PF=10% [6] Shirani M, Härkegård G. Large scale axial fatigue testing of ductile cast iron for
heavy section wind turbine components. Eng Fail Anal 2011;18:1496–510.
[7] Cazaud R. La fatigue des métaux. Paris, France: Dunod; 1948.
[8] Bathias C, Baïlon J-P. La Fatigue des matériaux et des structures. 2nd
ed. Hermes Sciences Publicat.; 1997. revue et augmentée édn.
[9] Banvillet A, Palin-Luc T, Lasserre S. A volumetric energy based high cycle

F
multiaxial fatigue criterion. Int J Fatigue 2003;25:755–69.
10
5
10
5 [10] Doudard C, Hild F, Calloch S. A probabilistic model for multiaxial high cycle
5 6 7 fatigue. Fatigue Fract Eng Mater Struct 2007;30:107–14.
10 10 10
[11] Ezanno A, Doudard C, Calloch S, Heuzé J-L. A new approach to characterizing
Predicted number of cycles at initiation for and modeling the high cycle fatigue properties of cast materials based on self-
P =50%, N (P =50%) heating measurements under cyclic loadings. Int J Fatigue 2013;47:232–43.
F a num F
[12] Centre technique des industries de la fonderie. Atlas métallographique des
alliages cuivreux. Paris; 1989.
Fig. 14. Comparison of experimental and theoretical results concerning crack [13] Helle H. An investigation of ship propeller fatigue [PhD thesis]. Delft University
initiation for the loadings with a loading ratio of 0.1. of Technology; 1981.
[14] Nadot Y, Mendez J, Ranganathan N. Influence of casting defects on the fatigue
passage from the scale of a specimen to the scale of ship propellers limit of nodular cast iron. Int J Fatigue 2004;26:311–9.
is correctly taken into account. [15] Ammar HR, Samuel AM, Samuel FH. Porosity and the fatigue behavior of
hypoeutectic and hypereutectic aluminum–silicon casting alloys. Int J Fatigue
2008;30:1024–35.
7. Conclusion [16] Murakami Y. Material defects as the basis of fatigue design. Int J Fatigue
2012;41:2–10.
[17] Koutiri I, Bellett D, Morel F, Pessard E. A probabilistic model for the high cycle
The present paper was devoted to the validation of a probabilis- fatigue behavior of cast aluminum alloys subject to complex loads. Int J
tic high cycle model through calculation/test comparisons at the Fatigue 2013;47:137–47.
[18] Mu P, Nadot Y, Nadot-Martin C, Chabod A, Serrano-Munoz I, Verdu C. Influence
scale of an engineering component. The considered components of casting defects on the fatigue behavior of cast aluminum AS7G06-T6. Int J
in this study were copper alloy sand cast ship propellers. Fatigue 2014;63:97–109.
The proposed probabilistic model for cast materials was devel- [19] Charkaluk E, Constantinescu A, Szmytka F, Tabibian S. Probability density
functions: from porosities to fatigue lifetime. Int J Fatigue 2014;63:127–36.
oped in a probabilistic two-scale framework taking into account [20] Nadot Y, Billaudeau T. Multiaxial fatigue limit criterion for defective materials.
the presence of a population of initial micro-flaws by an indirect Eng Fract Mech 2006;73(1):112–33.
approach. It accounts for scale effects without extra parameters. [21] Doudard C, Calloch S, Hild F, Cugy P, Galtier A. Identification of the scatter in
high cycle fatigue from temperature measurements. Comptes Rendus
The parameters of the model are identified by using classic fatigue Mécanique 2004;332(10):795–801.
tests performed on specimens. [22] Doudard C, Calloch S, Cugy P, Galtier A, Hild F. A probabilistic two-scale model
The proposed model was implemented in in-house software. for high-cycle fatigue life predictions. Fatigue Fract Eng Mater Struct
2005;28(3):279–88.
This software can be used to determine the fatigue life of cast
[23] Gulino R, Phoenix SL. Weibull strength statistics for graphite fibres measured
structures. The comparison between fatigue tests on marine pro- from the break progression in a model graphite/glass/epoxy microcomposite. J
pellers and the calculations given by the proposed software show Mater Sci 1991;26(11):3107–18.
that the scale effects are correctly taken into account by the pro- [24] Jeulin D. Modèles morphologiques de structures aléatoires et changement
d’échelle, thèse d’état. Université de Caen; 1991.
posed approach. This good result is obtained thanks to the proba- [25] Freudenthal AM. Statistical approach to brittle fracture. Academic Press; 1968.
bilistic aspect of our approach. p. 591–619.
[26] Hild F, Billardon R, Marquis D. Hétérogénéité des contraintes et rupture des
matériaux fragiles. CR Acad Sci 1992;315(Série II):1293–8.
Acknowledgments [27] Abramowitz M, Stegun IA. Handbook of mathematical functions. New York
(USA): Dover Publications, Inc.; 1965.
[28] Weibull W. A statistical theory of the strength of materials. Roy Swed Inst Eng
The authors are highly grateful to Direction Générale de l’Arm- Res, Report 151; 1939.
ement (DGA) from the French ministry of Defense and DCNS for [29] Weibull W. A statistical distribution function of wide applicability. ASME J
their support and the French Navy for the loan of the propellers. Appl Mech 1951;18(3):293–7.
[30] Yildirim EA. Two algorithms for the minimum enclosing ball problem. SIAM J
Optim 2008;19(3):1368–91.
References [31] Aucher M. Hélices marines. In: Techniques de l’Ingénieur, traité Génie
Mécanique; 1996.
[32] Fares Y, Chaussumier M, Daidié A. Étude comparative des caractéristiques
[1] Suresh S. Fatigue of materials. Cambridge University Press; 2003.
mécaniques de boulons en acier 42CrMo4 de classe 8.8 selon le processus de
[2] Bathias C, Pineau A. Fatigue of materials and structures. John Wiley & Sons;
fabrication. Méc Ind 2008;9(4):315–22.
2013.