Available online at www.sciencedirect.

com
Procedia
Engineering
Procedia
Procedia Engineering
Engineering 00 (2009)
2 (2010) 000–000
967–976
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia

Fatigue 2010

Fast characterization of high-cycle fatigue properties of a cast

copper–aluminum alloy by self-heating measurements under
cyclic loadings

A. Ezannoa, C. Doudarda, *, S. Callocha, T. Millotb, J.-L. Heuzéc

a
LBMS, ENSIETA/Université de Brest/ENIB (EA 4325), F-29806 Brest, France
b
CESMAN, DCNS BU Propulsion, F-44620 La Montagne, France
c
DGA/DET/CEP/MC/TCM, F-94114 Arcueil, France

Received 8 March 2010; revised 12 March 2010; accepted 15 March 2010

Abstract

Even if casting offers the possibility of producing complex shaped parts in a few operations, this process introduces inevitably
casting defects such as shrinkage cavities which are harmful regarding fatigue resistance. In this work, this initial defects
population is taken into account by indirect approach, i.e. without describing size and geometry of pores, to characterize rapidly
High-Cycle Fatigue (HCF) properties of cast materials by the aid of self-heating measurements under cyclic loadings. A
probabilistic multiaxial two-scale model is proposed to analyze the self-heating measurements in order to predict S/N curves. It is
assumed that fatigue damage is due to microplasticity that induces a dissipative energy which is related to the temperature
increase of the studied specimen under cyclic loadings for low stress amplitudes. To describe microplastic activity, a Poisson
distribution of elasto-plastic sites within an elastic matrix is considered. In this paper, the identification procedure has been
performed on the particular case of a sand cast copper-aluminum alloy used for casting marine propellers. The validation of the
proposed approach concerns the prediction of classical fatigue results. The results of this study show a good matching between
model predictions and experimental fatigue data, including scattering.
c 2010 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.

Keywords: cast copper-aluminum alloy; high-cycle fatigue; casting defects; microplasticity; Poisson Point Process; self-heating tests

1. Introduction

Among manufacturing processes, casting allows producing even huge structures with complex shapes in a few
operations. The liquid metal is poured into a mold which gives to the part its quasi-definitive form, as the metal
solidifies. However, this process has the disadvantage of introducing inevitably casting defects such as shrinkage
cavities and blowholes which appear during the solidification of the melt alloy. As a result, cast components contain
initial porosity which affect considerably their mechanical behavior and in particular their fatigue resistance. Indeed,

* Corresponding author. Tel.: +33 2 98 34 89 27; fax: +33 2 98 34 87 30.

E-mail address: [email protected]

1877-7058 c 2010 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
doi:10.1016/j.proeng.2010.03.105
968 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976

2 A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000

industrial feedback reports that these casting defects are often at the origin of the failure of castings. Therefore, some
authors have demonstrated the influence of such defects on fatigue strength of cast aluminum alloys [1–3], cast
magnesium alloys [4–5] or cast iron [6–8]. They conclude that shrinkage cavities act as stress concentrators within
the material. So, they are preferential sites for appearance of microplasticity phenomena which lead to the initiation
of a fatal fatigue crack.
Two different approaches can be applied for this family of metallic materials: direct approaches which take into
account size and shape of defects and indirect approaches where the initial population of porosities does not have to
be characterized. In this work, an indirect approach is adopted by the aid of self-heating measurements. The main
idea of this technique is to measure the change of the temperature specimen under cyclic loadings [9–14]. The
steady-state temperature is then plotted as a function of the stress amplitude applied to the specimen. For some
materials such as steels or cast iron, the first part of the curve shows nearly no change in temperature, whereas, in
the second part, the temperature increases significantly with the stress amplitude. A correlation between the mean
fatigue limit and the stress level leading to the temperature increase has been empirically proposed. The assumption
of a dissipation induced by microplasticity can explain this correlation. The present study is concerned with the
development and the identification of a probabilistic multiaxial two-scale model for High-Cycle Fatigue based on
the assumption that microplasticity occurs in sites according to a Poisson Point Process. Here, it is applied to the
particular case of a sand cast copper aluminum alloy.
This paper is organized as follows: in the first section, the material used for the application of this work is
introduced. Then, his mechanical behavior under cyclic loadings is studied. Next, a probabilistic two-scale model is
proposed. Finally, the identification and validation procedures are detailed.

Nomenclature

NR number of cycles to failure

Ȉ0 stress amplitude [MPa]
Ȉ0* equivalent stress amplitude [MPa]
fr load frequency [Hz]
R load ratio, R = ımin/ımax
ș temperature variation [K]
θ steady-state mean temperature variation [K]
PF probability of failure
m Weibull modulus

2. Material

In this work, the studied material is a sand cast copper-aluminum alloy (CuproAlu). This alloy is often used to
manufacture marine propellers because of its excellent marine corrosion resistance. Its chemical composition is
given in Table 1. Its microstructure is revealed in Fig. 1 and consists of a coarse Į-phase within an eutectoid Į+țIII
phase where țIII-precipitate is a Ni-rich lamellar-shaped phase. During the slow cooling of the alloy, two other
precipitates can be discerned: a Fe-rich rosette-shaped phase (țII-precipitate) and a Fe-rich țIV-phase which
precipitates finely in the Į-grains. Casting defects are also visible on the microstructure as darker quirky shaped
areas. They all constitute a micro-porosity which appears to be homogeneously distributed in the material. Figure 2
is very helpful to understand how shrinkage porosity influences fatigue life of CuproAlu. It is easy to distinguish the
site of the initiation of a fatigue crack which led to failure of the specimen (Fig. 2a). A magnification of the
initiation site shows clearly the casting defect responsible of the fracture of the specimen.
 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976 969

A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000 3

Table 1. Chemical composition of the studied CuproAlu (wt%).

Al Ni Fe Mn Cu

8.5 - 10.2 4.0 – 5.5 3.0 – 5.5 0.0 – 1.5 Balance

Table 2. Mechanical properties of the CuproAlu (E = Young’s modulus, Ys = Yield stress, UTS = Ultimate Tensile Strength, El = Elongation at
failure).

E Ys UTS El

125 GPa 200 MPa > 650 MPa 18%

800
material: Cu Al9 Ni5 Fe4
. −3 -1
strain rate ε = 10 s
600
Stress, σ (MPa)

400

200

0
0 0.05 0.1 0.15 0.2
Strain, ε
(a) (b)

Fig. 1. (a) Microstructure of the studied CuproAlu; (b) monotonic tensile strain/stress curve of the studied CuproAlu.

(a) (b)

Fig. 2. Post-mortem analysis of a CuproAlu specimen under cyclic loadings: (a) Fatigue fracture surface; (b) Magnification of the fatigue fracture
origin highlighting a casting defect.
970 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976

4 A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000

3. Behavior of a cast copper–aluminum alloy under cyclic loadings

To evaluate the HCF properties of the studied material, self-heating measurements are used. They consist in applying
successive series of 1000 cycles for different increasing stress amplitudes, Ȉ0. For each stress amplitude, the change of the
temperature variation, ș = T – T0, (where T is the current temperature of the sample during the test measured by a thermocouple
and T0 the initial value) is recorded. The mean temperature seems to become stable after about 600 cycles and equals ș(Ȉ 0 ) (Fig.
3a). Figure 3b shows the change of the steady-state mean temperature with the stress amplitude for different loading ratios, R.
ș(Ȉ 0 ) suddenly increases as the stress level is close to the fatigue limit and is influenced by the mean stress, Ȉ m (Fig. 3b). The
greater the mean stress is, the earlier the temperature increases. However Figure 4 shows that the thermal equilibrium state
reached after 1000 cycles (Fig. 3a) changes during the fatigue life of the specimen (i.e., the thermal source changes with cycles).

6 6

Steady-state temperature variation, θ (K)

material: Cu Al9 Ni5 Fe4 material : Cu Al9 Ni5 Fe4
R=-1, f =10Hz, Σ =200 MPa f =10Hz
r
Temperature variation, θ (K)

r 0

4.5 4.5 R = -1
θ R = -0.4

3 3

1.5 1.5

0 0
0 300 600 900 1200 0 50 100 150 200
Number of cycles, N Stress amplitude, Σ (MPa)
0
(a) (b)
Fig. 3. Self-heating effects on a cast CuproAlu alloy: (a) change of the temperature during 1000 cycles at constant stress amplitude; (b) change of
the steady-state temperature with the load amplitude for different loading ratio.

6
material: Cu Al9 Ni5 Fe4
R=-1, f =10Hz, Σ =200 MPa
Temperature variation, θ (K)

r 0

4.5

1.5

0 5 5 5 5
0 1.5 10 3 10 4.5 10 6 10
Number of cycles, N

Fig. 4. Change of the temperature variation with the number of cycles N.

 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976 971

A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000 5

Afterwards, a probabilistic multiaxial two-scale model capable of describing both self-heating of metallic
materials under cyclic loadings at low amplitude and HCF fatigue behavior is proposed. Its constitutive equations
are described in the following of the paper.

4. A probabilistic multi-axial two-scale model for HCF

4.1. Activation process of microplastic sites

In this paper, it is assumed that damage fatigue is caused by the microplasticity appearance. To describe this
phenomenon, it is considered a set of elasto-plastic inclusions, called sites, randomly distributed within an elastic
matrix [15]. The volume of each site is denoted by V0. The distribution of active sites (i.e., sites where
microplasticity occurs) is assumed to be described by a Poisson point process. The probability of finding k active
sites in a domain ȍ of volume VΩ is given by [16][17]

[ − N(Ω)]k
Pk (Ω) = exp[ − N (Ω)] (1)
k!

where N(ȍ) is the average number of activate sites in a domain ȍ of volume VΩ defined by

N(Ω) = Ȝ Vȍ (2)

with Ȝ, the density of active sites, called intensity of the Poisson point process. Furthermore, some authors have
shown the gradual appearance of the microplasticity with the stress amplitude [18]. Consequently, it is considered
that the average density of active sites follows a power law of the equivalent stress amplitude, Ȉ0*,
m

1 § Ȉ0* ·
m
§ 3
¨
1 © 2
( ) ( )
·
Σ0 : B : Σ0 ¸
¹
Ȝ ( Ȉ0 ) = ¨ ¸ = (3)
V0 © S0 ¹ V0 (S0 ) m

1
where B = J + 2b* K with K : Ȉ = Ȉ kk I and J : Ȉ = Ȉ − K : Ȉ , according to [19], and where V0S0m and m are two
3
material parameters. It is well-known that porous materials mechanical behavior is hydrostatic-pressure dependent
[20]. In the case of cast alloys, it justifies the addition of a third material parameter, b* [21] relating to first invariant
of associated tensor. This apparition scenario of microplasticity is illustrated in Fig. 5.

Fig. 5. Schematic representation of the activation scenario of the sites versus the equivalent stress amplitude.
972 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976

6 A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000

4.2. Elasto-plastic behavior of a site

To describe the dissipation induced by the microplasticity activation, the stresses in sites must be evaluated by a
law of localisation and their thermo-mechanical behavior must be defined.
The stress tensor, ı , in a site can be related to the macroscopic stress tensor, Ȉ , and the corresponding plastic
strain tensor, İ p , by

ı = Ȉ − 2ȝ (1 − ȕ ) J : İ p − 3ț (1 − Į ) K : İ p (4)

where α = (1+Ȟ)/3(1 − Ȟ) , ȕ = 2(4 − 5Ȟ) /15(1 − Ȟ) and μ, ν, κ are the shear modulus, Poisson’s ratio and the bulk
modulus, respectively. The plastic strain tensor is not a deviatoric tensor due to the hydrostatic stress dependence of
the mechanical behavior.
The concepts of continuum thermodynamics are used to model the thermo-mechanical behavior of sites [22]. An
additive decomposition of strain tensor İ in an elastic part İ e and a plastic part İ p is assumed for each site as

İ = İe + İp (5)

and the cumulated plasticity strain, p, is defined by

2 p -1 p
p = İ : B : İ (6)
3

A normality rule is considered for the plasticity model

∂f
İ p = Ȝ (7)
∂σ

with Ȝ , the plastic multiplier and f, the yield surface defined by

f=
3
2
( ) ( )
σ − Χ : B : σ − Χ − ıy − R ≤ 0 (8)

where ıy is the yield stress of a site. This constitutive equation includes a mixed hardening. The back stress Χ and R
are expressed by

2 2§ C ·
X= C : İp = ¨C j J + k K ¸ : İp (9)
3 3© 2b * ¹

R = Q [1 - exp(-bp) ] . (10)

In Eq. (9) and (10), Cj and Ck are two parameters of the linear kinematic hardening law whereas Q is the asymptotic
value of the isotropic hardening variation and b, a material parameter.
With the chosen framework, the intrinsic dissipated energy D site (Ȉ 0 , ı y , N) in one site for the loading cycle, N,
and a given yield stress ıy and equivalent stress amplitude Ȉ0* can be expressed as
 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976 973

A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000 7

D site (Ȉ 0* , ı y , N) = ³ (ı − X) : İ p − R p VSdt (11)

cycle

where ¢ . ² are the Macauley brackets (i.e. positive part of ‘.’) and one assumes that the isotropic hardening
variation is small during a cycle.

4.3. Description of mean temperature evolution under cyclic loading

dȜ
With a Poisson point process, VΩ dȈ is the mean number of inclusions activated for an equivalent stress
dΣ
amplitude ranging from Ȉ and Ȉ+dȈ in a domain of volume Vȍ, i.e. the number of sites whose mean fatigue limit
lies between Ȉ and Ȉ+dȈ ,and whose dissipated energy during a load cycle is D site (Ȉ 0 , ı y , N) . The global dissipated
energy density ǻ(N) per cycle is then given by

Ȉ 0*
d Ȝ(Ȉ)
ǻ(N) = ³ D site (Ȉ 0* , Ȉ, N) dȈ . (12)
0 dȈ

The thermal response of the model can be determined by integrating the heat conduction equation in which the mean
microplastic dissipation (i.e., the cyclic dissipation multiplied by the load frequency, fr) is introduced

ș ǻ(N) f r
ș + = (13)
IJ eq ȡc

where the characteristic time IJ eq is a parameter depending on the heat transfer boundary condition.

4.4. Description of S/N curves

By using the same framework and the weakest link theory that is usually considered in HCF, this approach can be
used to describe HCF results. In the present case, it is assumed that initiation occurs within the volume of the
materials, so that the probability of failure of a domain ȍ of volume Vȍ is given by

ª mº
V § ȈF ·
PF = 1 - exp «- eff ¨ ¸ » (14)
« V0 ¨ S0 ¸ »
¬ © ¹ ¼

m
1 § Ȉ 0* (M ) ·
where Veff = Vȍ H m is the effective volume, H m = ³ ¨¨ ¸ dV is the stress heterogeneity factor and
Vȍ ȈF ¸
© ¹
Ȉ F = max Ω (Ȉ 0* (M) ) . The HCF life prediction is often characterized by the mean fatigue limit Σ ∞ and the standard
deviation given by, respectively
1
§ V ·m § 1
Σ ∞ = S0 ¨¨ 0 ¸¸ ī¨1 + ·¸ (15)
© Veff ¹ © m¹
974 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976

8 A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000

1
§ V ·m ª 2º ª 1º
Σ ∞ = S0 ¨¨ 0 ¸¸ ī «1 + » − ī 2 «1 + » (16)
© Veff ¹ ¬ m¼ ¬ m¼

∞
where ī(p ) = ³ t p−1 exp(− t ) dt is the gamma function. Equations (15 and 16) account for volume and stress
0
heterogeneity effects associated to the Weibull model [23][24].
To model the S-N curve associated with a failure probability PF for a tensile fatigue test, an energetic approach
can be used as proposed by [25]. This approach is based on the hypothesis that crack initiation is associated to a
critical value of the cumulative dissipated energy, Ecd, for a site whose yield stress is equal to the fatigue limit. In
this case, the number of cycles to failure (i.e., the propagation stage is neglected with respect to the initiation stage)
is related to Ecd by

( )
NR
³ D site Ȉ 0 , Σ ∞ (PF ), N dN = E cd . (17)
0

In the next part, it is proposed to identify and validate the entire approach in the case of a cast CuproAlu alloy.

5. Identification and validation stages

5.1. Identification stage

The proposed model depends on five material parameters, namely, m, Q, b, V01/m S0 and Ecd. To identify these
parameters, it is proposed to perform a self-heating test until failure for the last stress amplitude. Fig. 6a shows the
identification of the parameters m and V01/m S0 by using the self-heating measurements. The change of the
temperature variation during the fatigue life for the last level allows determining two parameters Q, b (Fig. 6b). The
number of cycles to failure NR (550 000 cycles) has been obtained for the same level of stress amplitude. It permits
to identify Ecd.

6 6
Steady-state temperature variation, θ (K)

material : Cu Al9 Ni5 Fe4 material: Cu Al9 Ni5 Fe4

R=-1, f =10Hz
r
R=-1, f =10Hz, Σ =200 MPa
Temperature variation, θ (K)

r 0

4.5 4.5
Q = 26 MPa & b = 0.0095

3 3

m=7
1.5 1.5

N = 550 000
R

0 0
0 50 100 150 200 0
5
1.5 10 3 10
5
4.5 10
5 5
6 10
Loading amplitude, Σ (MPa) Number of cycles, N
0
(a) (b)

Fig. 6. Identification of the model: (a) on a self-heating curve; (b) on the change of the temperature with N.
 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976 975

A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000 9

5.2. Validation stage

On the one hand, the validation of the identified model is based on the prediction of the temperature variation
with N for two stress amplitude levels. Fig. 7a shows the comparison between the predictions and the experimental
results. One can see that the temperature variations during fatigue life specimens are predicted in a reasonable way.
On the other hand, the validation is based on the comparison between the predictions of the identified model and
experimental fatigue results (Fig 7b). The S/N curves and the scatter are in agreement with the model.

10 240
material: Cu Al9 Ni5 Fe4 material : Cu Al9 Ni5 Fe4
R=-1, f =10Hz R=-1
r 220
Temperature variation, θ (K)

Stress amplitude, Σ (MPa)

7.5
200

0
Σ =220 MPa 180 P =90%
0 F
5
160
P =50%
F
140
2.5 model
Σ =180 MPa
0 no-failure
120
specimen P =10%
F
0
5 5 5 5
100 5 6 7 8
0 1.5 10 3 10 4.5 10 6 10 10 10 10 10
Number of cycles, N Number of cycles to failure, N
R

(a) (b)

Fig. 7. Comparison between the predictions of the identified model and the experimental results: (a) change of the temperature with N for two
stress amplitude levels (Ȉ0=180 MPa and Ȉ0=220 MPa); (b) tensile fatigue test.

6. Conclusion

This paper deals with the characterization of HCF properties of a cast copper-aluminum alloy under cyclic
loadings by self-heating measurements. Self-heating tests are related to HCF results by a probabilistic multiaxial
two-scale model. The proposed approach is based on a Poisson distribution of sites where microplasticity takes
place.
The identification procedure needs the use of a self-heating curve. The validation of this entire framework and
the identification procedure is first obtained by comparison between experimental thermal effects for the last stress
amplitude level of a self-heating test with the model predictions. The second validation of the model concerns the
prediction of classical fatigue results (S/N curve). The thermal effects during fatigue life of specimens as well as the
S/N curves are accurately reproduced by the model for a cast CuproAlu alloy.

References

[1] Linder J, Axelsson M, Nilsson H. The influence of porosity on the fatigue life for sand and permanent mould cast aluminium. Int J
Fatigue 2006;28:1752–58.
[2] Avalle M, Belingardi G, Cavatorta MP, Doglione R. Casting defects and fatigue strength of a die cast aluminium alloy: a comparison
between standard specimens and production components. Int J Fatigue 2002;24:1–9.
976 A. Ezanno et al. / Procedia Engineering 2 (2010) 967–976

10 A. Ezanno et al./ Procedia Engineering 00 (2010) 000–000

[3] Buffière J, Savelli S, Jouneau PH, Maire E, Fougères R. Experimental study of porosity and its relation to fatigue mechanisms of model
Al-Si7-Mg0.3 cast Al alloys. Mater Sci Eng A 2001;316:115–26.
[4] Lu Y, Taheri F, Gharghouri M, Han H. Experimental and numerical study of the effects of porosity on fatigue crack initiation of HPDC
magnesium AM60B alloy. J Alloy Compd 2009;470:202–13.
[5] Lee S, Gokhale A, Patel G, Evans M. Effect of process parameters on porosity distributions in high-pressure die-cast AM50 Mg-alloy.
Mater Sci Eng A 2006;427:99–111.
[6] Agha HY, Béranger A, Billardon R, Hild F. High cycle fatigue behaviour of spheroidal graphite cast iron. Fatigue Fract Engng Mater
Struct 1998;21:287–96.
[7] Chantier I. Tolérance aux défauts initiaux et effets de surface: dimensionnement à la fatigue de pièces de fonderie. ENS Cachan, 2000.
[8] Nadot Y. Tolérance aux défauts dans les métaux soumis à des sollicitations cycliques multiaxiales. HDR, 2007.
[9] Welter G. Essais d'endurance par traction et compression. Wlad Inst Met 1937;4:30–9.
[10] Luong MP. Infrared thermography of fatigue in metals. SPIE 1992;1682:222–33.
[11] La Rosa G, Risitano A. Thermographic methodology for rapid determination of the fatigue limit of materials and mechanical
components. Int J Fatigue 2000;22:65–73.
[12] Galtier A, Bouaziz O, Lambert A. Influence de la microstructure des aciers sur leurs propriétés mécaniques. Mécanique et Industrie
2002;3:457–462.
[13] Doudard C, Calloch S, Hild F, Cugy P, Galtier A. Identification of the scatter in high cycle fatigue from temperature measurements. C.R.
Mecanique 2004;332:795–801.
[14] Doudard C, Calloch S. Influence of hardening type on self-heating of metallic materials under cyclic loadings at low amplitude. Eur J
Mech A-Solids 2009;28:233–40.
[15] Van KD. Sur la résistance à la fatigue des métaux. Sciences et techniques de l'armement, Mémorial de l'artillerie française, 1973.
[16] Gulino R, Phoenix SL. Weibull strength statistics for graphite fibres measured from the break progression in a model graphite/glass/
epoxy microcomposite. J Mater Sci 1991;26:3107–18.
[17] Jeulin D. Modèles morphologiques de structures aléatoires et changement d'échelle. 1991.
[18] Cugy P, Galtier A. Microplasticity and temperature increase in low carbon steel. In: Blom AF, editor. Proceedings of the Eight
International Fatigue Congress (FATIGUE 2002), West Midlands, UK: EMAS; 2002.
[19] Besson J, Guillemer-Neel C. An extension of the Green and Gurson models to kinematic hardening. Mech Mater 2003;35:1–18.
[20] Aubertin M, Li L. A porosity-dependent inelastic criterion for engineering materials. Int J Plasticity 2004;20:2179–208.
[21] Berg CA. Plastic dilation and void interaction. In: Melvin F, Kanninen MF, editors. Inelastic Behavior of Solids, New York: McGraw-
Hill; 1970, p. 171–210.
[22] Lemaitre J, Chaboche JL. Mechanics of Solid Materials, Cambridge (UK): Cambridge University Press, 1990.
[23] Weibull W. A statistical theory of the strength of materials. Roy Swed Inst Eng Res 1939; Report 151.
[24] Weibull W. A statistical distribution function of wide applicability. ASME J Appl Mech 1951;18:293–7.
[25] Doudard C, Calloch S, Cugy P, Galtier A, Hild F. A probabilistic two-scale model for high-cycle fatigue life predictions. Fatigue Fract
Engng Mater Struct 2005;28:279–88.