Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

Engineering Fracture Mechanics 73 (2006) 112–133

www.elsevier.com/locate/engfracmech

Multiaxial fatigue limit criterion for defective materials


a,* b
Y. Nadot , T. Billaudeau
a
Laboratoire de Mécanique et de Physique des Matériaux, UMR CNRS no. 6617, ENSMA, Téléport 2,
BP 40 109, 86961 Futuroscope Cedex, France
b
Transiciel, 1 chemin de la Crabe, BP 3059, 31 025 Toulouse Cedex 3, France

Received 30 August 2004; received in revised form 20 May 2005; accepted 24 June 2005
Available online 12 September 2005

Abstract

The objective of this paper is to quantify the influence of defect on the fatigue limit. Elastic–plastic simulations are
conducted to determine the stress distribution around defects for different geometries and different loading. It is shown
that a relevant mechanical parameter governing the fatigue limit for defect material could be the gradient of the hydro-
static stress. A multiaxial fatigue criterion is identified with three parameters and validated for different metallic mate-
rials under multiaxial conditions. Results are good and show that the gradient of the hydrostatic stress is a good
parameter to characterise the influence of a defect on the fatigue behaviour.
 2005 Elsevier Ltd. All rights reserved.

Keywords: Casting defect; Fatigue limit; Multiaxial criterion; Gradient effect; Artificial defect

1. Introduction

The fatigue behaviour of casting materials is strongly dependent on the defect population. Casting de-
fects are inherent to the foundry process and can be found in different materials: Aluminium alloys [1–4],
various Steels [5–7] or Cast Irons [8,9] for instance. Depending on the material and the process, various
populations of casting defects can be encountered: inclusion, shrinkage, gas pore, oxide or dross defect
[10]. The designer wants to know the maximum defect size allowable in the component. Consequently,
the influence of the defect on the fatigue behaviour must be characterised. Different ways have been pro-
posed to model the defect: let us have a look at these propositions before explaining the contains of the
present paper. It is generally admitted in the literature that the size of the biggest surface defect is the

*
Corresponding author. Tel.: +33 5 49 49 82 39; fax: +33 5 49 49 82 38.
E-mail addresses: yves.nadot@lmpm.ensma.fr (Y. Nadot), thomas.billaudeau@tso.transiciel.com (T. Billaudeau).

0013-7944/$ - see front matter  2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2005.06.005
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 113

Nomenclature

a material parameter describing defect influence (lm1)


d grain size (lm)
k coefficient in Ramberg Osgood relation
p exponent in Ramberg Osgood relation
n exponent in Papadopoulos criterion
E Young modulus of the material
G spatial gradient of the local hydrostatic stress (MPa/lm)
J1 max maximum value, during a load cycle, of the hydrostatic stress J 1 max ¼ 13 trðrÞ (MPa)
J 1 max calculated value of the maximum hydrostatic stress including gradient effect (MPa)
J2,a amplitude during the load cycle of the second invariant of the deviatoric stress tensor (MPa2)
Kt elastic stress concentration factor
R load ratio (between minimum and maximum stresses of the load cycle)
TD fatigue limit under torsion loading (MPa)
MPS maximum principal stress (MPa)
a material parameter in CrosslandÕs criterion
b material parameter in CrosslandÕs criterion (MPa)
epl plastic deformation along the loading axis for tension test
u ratio between TD and RD
rcr equivalent stress in the sense of Crossland criterion (MPa)
rcr equivalent stress including gradient effect in the sense of Crossland criterion (MPa)
rVM equivalent Von Mises stress (MPa)
r(N) local stress tensor at point N around the defect (MPa)
R(M) nominal stress tensor at point M corresponding to the uniform applied stress (MPa)
RD fatigue limit under tension loading (MPa)
DR
2 amplitude of the load cycle under tension (MPa)
DT
2 amplitude of the load cycle under torsion (MPa)
Det amplitude of the deformation in Ramberg Osgood relation
pffiffiffiffiffiffiffiffiffi amplitude of the stress in Ramberg Osgood relation
Dr
area equivalent defect size measured perpendicularly to the direction of the maximum principal
stress in MurakamiÕs equation

controlling parameter of material fatigue limit [5,11–14], therefore, many studies tend to propose a way to
model a single small surface defect. Nevertheless, we must not forget that a competition between surface
and internal defect can report initiation to the bulk material.
The simplest representation of a casting defect is a crack. This approach gives conservative results for
uniaxial tension loading and becomes really poor in the case of multiaxial loading. Many problems arise
with the dimension of the crack compared to the microstructure: this is the famous Ôshort crackÕ problem
for which Linear Elastic Fracture Mechanics is not applicable. Numerous papers have addressed this prob-
lem and one major conclusion is that a model based on short crack propagation needs careful identification
of initiation and micropropagation in relation with microstructure barriers in the fatigue p
stage
ffiffiffiffiffiffiffiffiffiI. The con-
cept of defect size differs from the authors but an interesting parameter is the so-called area parameter
proposed by Murakami [14]. He proposes to correlate the fatigue limit of defect material with the defect
pffiffiffiffiffiffiffiffiffi
size ( area) whereas no consideration is made about the shape. This proposition highly simplifies such
114 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

problem since natural defects have a complex morphology. Murakami proposes to determine the area as
the projection of the defect on the plane perpendicular to the direction of the maximal principal stress
(MPS). This way to deduce the size of the defect from the loading gives a powerful tool for defect material
and it has been largely validated for different metallic materials under uniaxial conditions. Problems come
for multiaxial conditions: even if the defect size includes the loading directions, the use of the maximum
principal stress is not sufficient because of the lack of consideration of the stress triaxiality. A biaxial exten-
sion for defect material is proposed by Endo [15,16]. The author considers the influence of the second prin-
cipal stress. The criterion is then written with the ratio between the applied tension load and torsion load
over their respective fatigue limit.
Another way is to consider the defect equivalent to a notch. The first level of description consists in using
the elastic stress concentration factor (Kt), or the fatigue stress concentration factor (Kf), but the main lim-
itations are the extension to multiaxial loading and the fact that the stress gradient is not directly taken into
account. The second level in notch approach consists in considering the stress distribution around the
notch. This allows a better description of the local response of the material and non-local effects are roughly
described by the mean of a stress gradient. Some approaches give interesting results for macroscopic
notches [17–20]. Taylor [21] proposes a mix between crack and notch approaches by the use of mode I
opening mechanisms like in cracks but using stress values at a given distance of the notch. Finally, the third
level of description, probably the most realistic one, consists in performing elastic–plastic simulations to
determine the stress distribution around the defect.
Some of the previous approaches have been compared in [22] and authors show that each can be really
appropriated for a given problem when the initial geometry of the flaw is controlled and when the loading path
is limited to well defined cases. In the general case of various defect geometries and multiaxial loading, it seems
necessary to perform a detailed experimental analysis of fatigue mechanisms in order to point out the key fac-
tors to use the relevant mechanical approach. A recent study performed by the authors [23] has been conducted
on a low carbon steel with artificial defects of different geometries and sizes; the main results are presented
below in Section 2. This study gives the bases to propose the fatigue criterion described in this paper.
After a brief summary of the experimental study, we present the elastic–plastic finite element simula-
tions. The role of normal and shear stress as well as stress distribution on initiation at the fatigue limit
is studied and compared to fatigue mechanisms. A classical initiation multiaxial fatigue criterion is used
as a base to propose a multiaxial fatigue limit criterion for defect material. The fourth section explains
the motivations for the use of a stress gradient to characterise the defect and the criterion is modified by
the introduction of the gradient of the hydrostatic stress. In the last section, a detailed experimental vali-
dation is proposed for three different materials and various load cases.

2. Fatigue mechanisms at the tip of the defect

The experimental results at the origin of the present paper are detailed in Ref. [23] by the authors. A brief
summary is presented here to highlight the construction of the present criterion. The material tested is a low
carbon steel C36 (chemical composition in Table 1). Artificial defects with controlled geometry and size
(from 50 lm to 1000 lm) have been introduced at the surface of fatigue samples. Fatigue limit tests have
been conducted under tension, torsion and combined tension–torsion loading. A careful experimental

Table 1
Chemical composition of the C36 steel
C Mn Si S P Cr Ni Co Mo W
0.36 0.60 0.27 0.009 0.034 0.14 0.07 0.07 0.01 0.01
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 115

procedure has been employed in order to observe and understand the fatigue mechanisms at the fatigue
limit in order to reveal the mechanisms governing the fatigue limit. The analysis of the experimental
SEM observations is completed by the knowledge of stress distribution around defects. From this study,
we can draw some major physical conclusions:

• The first stages of crack nucleation at the tip of the defect occurs in the maximum shear plane and in the
maximum loaded part of the defect under both tension and torsion loading (see Fig. 1). A comparison
with defect free material shows that initiation mechanisms are similar. It seems therefore appropriate to
use a multiaxial criterion based on initiation on maximum shear planes influenced by the normal stress.
This is a non-classical conclusion for fatigue analysis of defective materials where initiation is often
neglected.
• The stress distribution around defects gives rise to a high stressed volume generally located in the plane
perpendicularly to the direction of the maximum principal stress. Consequently, the macroscopic
crack that leads to failure of the sample propagates in that plane in mode I (governed by the opening
stress).

Fig. 1. Comparison between fatigue mechanisms and stress distribution around a spherical defect under tension and torsion. Von
Mises equivalent stress, linear elastic simulation.
116 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

• The influence of the defect on the fatigue limit cannot be analysed by the consideration of the maximum
loaded point at the tip of the defect: a gradient effect has to be considered.
• The previous conclusions are bounded to a class of size and geometry of defects. A discussion of theses
limits is presented at the end of the present paper.

These results have been used to build the architecture of the present criterion where the physical observa-
tions are the main motivations.

3. Stress distribution around defects

3.1. Finite element analysis

In order to characterise the 3D stress distribution around the defect, finite element simulations have
been conducted for each type of defect and each type of loading. Symmetry and boundary conditions
are directly representative of the fatigue samples used: the applied load corresponding to the experimental
fatigue limit. Thus, results presented in this paper are always representative of the fatigue limit of the tested
sample. Linear elastic simulations are performed to optimise the mesh size and to determine the Kt factor of
each defect. Secondly, the stress level at the maximum loaded point (using equivalent Von Mises stress) of
the defect is compared to the yield stress of the material. For most of our defects, plasticity occurs
around the defect so that we need to perform elastic–plastic simulations. Consequently, all the simulations
are conducted with an elastic–plastic Prandtl–Reuss type model based on Von Mises energetic criterion and
with a linear isotropic hardening law. Identification of the hardening law is done using cyclic stabilised
experimental stress–strain curve (see Fig. 2) and all simulations are done for one load cycle. The plasticity
model used here is not adequate to describe cyclic loading but this is a first step to evaluate elastic–plastic
stress distribution and tests are mainly conducted under proportional loading with alternated stress.
The cyclic loading is included in the cyclic stabilised experimental stress–strain curve used to identify the
model.

Fig. 2. (a) Smooth cylindrical fatigue sample containing artificial surface defect. (b) Artificial defect at the free surface, SEM image (see
Appendix B). (c) Detail of the mesh around the defect. (d) Plastic stress–strain cyclic stabilised experimental curve, smooth cylindrical
sample without defect under uniaxial tension loading.
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 117

3.2. Analysis of stress distribution

Fig. 3 presents an illustration of the stress distribution around a spherical defect. For tension loading
(Fig. 3a), stress distribution on the highest loaded plane is nearly constant (the highest variation along
the equator line of the MPS is less than 6%). In other words, every point on this equator line is loaded
as the Hot Spot (maximum loaded point: point A). Moreover, the stress distribution from the defect surface
to the bulk is homogeneous: on the highest loaded plane, the evolution of stresses on a direction normal to
the surface of the defect is the same from every point of the equator line. It means that stresses in the highest
loaded plane can be represented by the line (plotting line) ahead of the Hot Spot (point A). In the case of
torsion loading (Fig. 3b) it can be noted that the maximum loaded plane (in the sense of shear stress) cor-
responds to the macroscopic plane for crack propagation during fatigue stage II (mode I) as well as for
tension loading. In this plane, there is only a smooth variation of shear stress from 330 MPa at point A
to 300 MPa at the surface (point B). It is therefore possible to consider, as a first approximation, that
the plotting line from point A can be representative of the stress distribution around defects under torsion
loading.
In the following, we will analyse the stress distribution for all defects and load cases with a particular
attention to the role of hydrostatic and deviatoric stresses spatial evolutions.

Fig. 3. (a) Details of the mesh for tension simulations. (b) Details of the mesh for torsion simulations.
118 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

3.2.1. Hydrostatic stress


Fig. 4 shows the evolution of the maximum principal stress along the plotting line from point A for sev-
eral defects and both tension and torsion loading (all details in Table 2). Remember that the nominal stress
applied to the meshed sample is the experimental fatigue limit of the sample. It can be concluded from
Fig. 4 that local MPS at point A is not the governing mechanical parameter: local value of the MPS at point
A varies from 160 to 380 MPa. This confirms the experimental observation showing that cracks initiate in
shear planes at the tip of the defect even if the macroscopic crack propagates, after nucleation, perpendic-
ularly to the direction of the MPS in opening mode I. One could argue the fact that we could consider the
effect of gradient of MPS. We can compare 1000 lm spherical defect under tension (case no. 8) and torsion
(case no. 12): the same stress gradient gives rise to different local values (230 MPa under tension and
160 MPa under torsion). This gives an argument to show that the stress triaxiality at the tip of the defect
is an important parameter to consider.

3.2.2. Shear stress


Fig. 5 presents the evolution of the equivalent Von Mises stress calculated from point A along the Plot-
ting line for the same defects and loading than in Fig. 4. It appears that the comparison of the same 170 lm
spherical defect under tension (case no. 6) and torsion (case no. 10) leads to opposite results than for MPS.
Similar local stress (respectively 285 and 290 MPa) whereas stress evolutions are different. It seems there-
fore impossible to analyse the results with the only consideration of the Von Mises stress even when ana-
lysing non-local effects.
Experimentally, we observe that the fatigue crack nucleation appears in the maximum shear plane. From
the simulations we have made, it appears that the governing parameters are necessarily based on the shear

MPS (MPa)
400
Case 1
Case 2

350 Case 3
Case 4
Case 5
300 Case 6
Case 7
Case 8
250 Case 9
Case 10
Case 11
200 Case 12

150

100
0 50 100 150 200 250 300 350 400
Distance from the defect (µm)

Fig. 4. Evolution of the maximum principal stress (MPS) versus distance from the defect for different cases (C36 steel).
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 119

Table 2
Description of the different experimental cases for the C36 steel
Case number Loading type Defect type Defect Rmax Rmoy Tmax Tmoy
size (lm) (MPa) (MPa) (MPa) (MPa)
Case 1 Tension No defect 240 0
Case 2 Tension Elliptical H 170 200 0
Case 3 Tension Elliptical H 400 157 0
Case 4 Tension Elliptical V 170 235 0
Case 5 Tension Spherical 95 230 0
Case 6 Tension Spherical 170 195 0
Case 7 Tension Spherical 400 150 0
Case 8 Tension Spherical 880 135 0
Case 9 Torsion No defect 169 0
Case 10 Torsion Spherical 170 160 0
Case 11 Torsion Spherical 380 145 0
Case 12 Torsion Spherical 950 125 0
Case 13 Tension Spherical 370 280 154
Case 14 Tension Spherical 850 246 135
Case 15 Torsion Spherical 310 173 43
Case 16 Tension–torsion Spherical 360 128 0 72 0
Case 17 Compression–torsion Spherical 360 0 64 145 0

Von Mises stress (MPa)


350
Case 1
330 Case 2
Case 3
310 Case 4
Case 5
290
Case 6
Case 7
270
Case 8

250 Case 9
Case 10
230 Case 11
Case 12
210

190

170

150
0 50 100 150 200 250 300 350 400

Distance from the defect (µm)

Fig. 5. Evolution of the Von Mises equivalent stress versus distance from the defect for different load cases (C36 steel).

stress associated with the hydrostatic stress. This is a well known conclusion for fatigue behaviour of defect
free metallic materials. The conclusion is not classical for defect materials because many authors consider
120 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

them with a quasi-brittle behaviour with mode I governing cracking mechanisms. From both mechanical
nominal and local viewpoints, this result reinforces the experimental conclusion of the previous paper
[23] showing that defect materials can be treated with the same basic mechanisms than the defect free
materials.

3.2.3. Equivalent stress


In the present study, the loading path is proportional with a constant amplitude. The purpose of the
study is to introduce a defect parameter in a multiaxial fatigue criterion. On the basis of the previous con-
clusions, we propose to use Crossland criterion [24] in order to illustrate the role of the defect. It is also
possible to do this with other types of criteria as critical plane criteria for instance [24–26]. Crossland cri-
terion uses a combination of the two invariants of the stress tensor to calculate the threshold line for ini-
tiation of a fatigue crack. He supposes that the amplitude of the shear stress (J2,a) and the maximum value
of the hydrostatic stress (J1 max) are the governing parameters of the fatigue limit. This criterion has largely
been validated for many metallic materials in the case of constant amplitude proportional loading. The cri-
terion can be written as follows:
pffiffiffiffiffiffiffi
rcr ¼ J 2;a þ a J 1 max 6 c ð1Þ

This criterion is identified with two alternated fatigue limits: tension and torsion. For our C36
steel: a = 0.3804 and c = 169 MPa. We can make the first basic test of such criterion by applying the
pffiffiffiffiffiffiffi
criterion at point A for each test. Results are presented in Fig. 6 were J 2;a is plotted versus J1 max. Remem-
ber that the nominal stress applied is always the experimental fatigue limit. The plain line represents the
identified threshold: a perfect prediction would lead to results plotted exactly on this line. The two
dotted lines are representative of 10% error. Fig. 6b presents the results as a relative error for each
experiment:
pffiffiffiffiffiffiffi 
J 2;a þ aJ 1 max  c
errorð%Þ ¼ 100 ð2Þ
c
It can be concluded that application of Crossland criterion at the tip of the defect (point A) gives bad results
because the error can be up to 50%. Independently of the loading, the results are bad for high acuity defects
and small ones while they are better for smooth defects and big defects. This first conclusion can be related
to the stress evolution observed in Figs. 4 and 5: the bad results obtained for the local application of the
criterion correspond to the highest stress gradients. This conclusion has to be compared to the gradient ef-
fect for fatigue limit of smooth specimens under different loading (tension, rotary bending and pure
bending).

4. Multiaxial criterion

The purpose of this section is to present the gradient based multiaxial criterion proposed to describe the
influence of defects. The first part motivates the use of stress gradient as the defect parameter, the second
one discusses the measurement of the gradient and the last one presents the criterion.

4.1. Stress gradient approach

The conclusions of the previous section show that the local approach considering elastic–plastic stresses
at the hot spot of the defect gives poor results. The analysis of stress distribution from the tip of the defect
to the bulk shows that significant different gradient of stresses are noted depending on the loading and the
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 121

J 2a (MPa)
250
Case 1
230
Case 2

Case 3
210
Case 4
190
Case 5

Case 6
170
Case 7
150
Case 8

Case 9
130
Case 10
110
Case 11

Case 12
90

70

50
0 50 100 150 200 250
(a) J1max (MPa)

50

40

30

error 20
%

10

0
Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8 Case 9 Case 10 Case 11 Case 12

(b) -10

Fig. 6. (a) Application of Crossland criterion at point A using local approach. (b) Same result presented as an error percentage.

type of defect. The effect of the macroscopic stress gradient is well known in fatigue, Papadopoulos [27]
proposed a criterion including stress gradient effect after an experimental analysis where he concludes that
the gradient effect is one order of magnitude higher than pure size effect. Moreover, he analyses the role of
122 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

hydrostatic and deviatoric gradient and concludes that the gradient of the deviatoric part of the stress dis-
tribution is not influent. The criterion proposed includes the gradient on the hydrostatic part of the stress
tensor.
In the case of the local stress distribution around defects at the fatigue limit (see Figs. 4 and 5), a similar
conclusion can be drawn. The gradient of the deviatoric part clearly appears less influent than the gradient
of the hydrostatic part. Stress level at point A obtained for torsion tests differs less than 11% (Fig. 5) while
the stress gradient is different. It appears therefore the same conclusion than Papadopoulos: the stress gra-
dient of the deviatoric part of the stress tensor does not seem to influence the fatigue limit. Nevertheless,
this conclusion based on the relative comparison of stress distributions does not present a physical
motivation. In the following, we will only consider the gradient of the hydrostatic part of the stress
tensor.

4.2. Stress gradient measurement

The purpose is here to calculate the gradient of the hydrostatic stress at point A. The definition of the
gradient given by Papadopoulos is
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s 2  2  2
oJ 1 max oJ 1 max oJ 1 max
G¼ þ þ ð3Þ
ox oy oz

As explained in Fig. 3a, due to the regular geometry of artificial defects, we propose to reduce the study
of the stress distribution around defects to the Ôplotted lineÕ from point A. It comes therefore at
point A

oJ 1 max
G¼ ð4Þ
oy

As seen in Fig. 4, due to the defect geometry, the stress gradient is not constant along the plotting line: sig-
nificative evolution from point A to the bulk material is observed. It is necessary to find the definition of the
calculation of the stress gradient which is appropriate to the problem. Different calculations are possible
(see Fig. 7 for illustration):

• At point A
oJ 1 max
G¼ ð5Þ
oy
• Over a physical distance (grain size)
DJ 1 max
G¼ ð6Þ
d
• Up to the stress level of the defect free fatigue limit
DJ 1 max
G¼ ð7Þ
y ðr ¼ rD Þ
• Up to the size of the defect
DJ 1 max
G ¼ pffiffiffiffiffiffiffiffiffi ð8Þ
area
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 123

190
J1max (MPa)

170

150

130
(5)

110

(6) (7)
(8)
90

80
MPa
70

50
0 50 100 150 200 250 300 350 400
22 µm 170 µm Distance from the defect (µm)

Fig. 7. Different ways to measure the stress gradient at the tip of the defect (case 6).

with
pffiffiffiffiffiffiffiffiffi
DJ 1 max ¼ J 1 max ð0Þ  J 1 max area ð9Þ

The first proposition is the mathematical definition of the spatial gradient at point A. Such gradient gives
very bad results because as mentioned before, the gradient is not constant. With the stress distribution
due to plastic deformation at the tip of the defect, the calculated gradient can be equal to 0 even if
50 lm fare from the tip of the defect, the stress is really lower than at the tip. It seems that the local
gradient at point A (5) is not the relevant gradient parameter. The second proposition is physically moti-
vated by considering that the first grain plays a major role in the crack initiation. Some authors [28] have
successfully applied such type of approach for the fatigue of metallic materials with parameter d which is
often higher than the grain size. The third proposition is motivated from a mechanical point of view: the
fatigue limit would play the role of a threshold stress under which stresses do not participate to fatigue
damage. This point of view can give good results and be modified by considering a stress lower than the
fatigue limit (the stress level for the initiation of non-propagating cracks). Morel [29] have already built a
fatigue criterion for notch specimens based on the stress gradient as proposed by Papadopoulos. This
criterion gives interesting results. Fig. 7 illustrates the different ways to calculate the gradient of the
hydrostatic stress. The last proposition is not physically based and proposes to calculate the gradient over
the size of the defect. This is based on the observation of the stress distribution around all tests: the stress
distribution around defects is influenced over a distance comparable to the defect size for nearly all tests
performed.
124 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

Tests have been conducted with the four measurements proposed. The first one (5) gives 50% mean error,
the second one (6) 10%, the third one (7) 7% and the fourth one (8) 6% on cases 2–8 on C36. It seems more
appropriate to consider the gradient as a mean value of the evolution of the stress from the defect to the
bulk than consider the local gradient at the tip of the defect. For reasons of simplicity, we propose to use
the fourth method to calculate the gradient in the criterion, reminding that the measurement over a given
stress is also a good method.

4.3. Multiaxial gradient criterion for defect material

The fundamental governing parameters to describe the influence of casting defects on the fatigue limit
are now determined. We can introduce the gradient of the hydrostatic stress (8) into (1), it comes
 
pffiffiffiffiffiffiffi G
rcr ¼ J 2;a þ aJ 1 max 6 c with J 1 max ¼ J 1 max 1  a ð10Þ
J 1 max

The gradient G is normalised by the local hydrostatic stress and the criterion (10) introduces the mate-
rial parameter a which has the dimension of the inverse of a length. We could imagine to enhance the
description of the gradient effect by another function of the gradient, for example Papadopoulos [27]
proposed
  n 
 G
J 1 max ¼ J 1 max 1  a ð11Þ
J 1 max

The influence of the gradient needs in this case one more material coefficients to be identified. We propose
to test in the next section the expression (10) on three different materials in order to evaluate the capability
of the criterion, considering n = 1.

4.4. Criterion identification

The criterion is limited to surface defects, the problem of internal defects will be discussed in the last
section. The criterion proposed can be identified by different ways depending on the type of material
concerned.

4.4.1. The defect free material exists


It is the case for impacted materials, for in service defect creation (corrosion, damage, etc.) and generally
for materials where it is possible to test samples with or without defects (machining surfaces of cast mate-
rials). In this case the identification of a and b needs alternating fatigue limit under tension and torsion for
the defect free material. The identification of a needs one alternating fatigue limit under tension for a given
defect size.

4.4.2. The material always contains surface defects


It is the case of most of industrial casting materials. The identification of a and b needs alternating fa-
tigue limit under tension and torsion for one given size. The identification of a needs one alternating fatigue
limit under tension for another given defect size.
The other material data needed is the elastic–plastic stabilised cyclic stress–strain curve in order to cal-
culate stress level at point A and stress gradient.
The next section proposes a detailed validation procedure of the criterion for different materials, a large
population of defects and different loading tests including combined tension–torsion.
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 125

5. Experimental validation

The criterion proposed (10) is validated in this section for three different materials containing artifi-
cial defects. The first one is the C36 from the present study but results are enhanced by different loading
paths. The second one is a nodular cast iron tested in our laboratory under tension and torsion. The third
one is a S 45 C carbon steel tested by Murakami under tension and torsion. All the results are commented
in order to define the limits of the present criterion and finally, the application to natural defects is
discussed.

5.1. Validation for C36 carbon steel

The identification for this material is based on three experiments:

Fatigue limit under tension, defect free material: RD = 240 MPa.


Fatigue limit under torsion, defect free material: TD = 169 MPa.
Fatigue limit under tension, 170 lm spherical defect: RD = 195 MPa.

The parameters of the criterion are therefore: a = 0.3804, c = 169 MPa and a = 377 lm1. Fig. 8 pre-
sents the experimental results for all the set of cases studied. We can conclude that the criterion gives good
results because the mean error is 7.8% and 2 cases only lead to more than 12% error. Under pure tension
and load ratio R = 1 (cases 2–8), results are independent of the geometry: the parameter ÔaÕ identified on a
spherical defect gives good results to describe the evolution for elliptical defects. The results for spherical
defects of different sizes are also good. Under tension for load ratio R = 0.1 (cases 13 and 14), results are

20
error
%

10

0
Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8 Case Case Case Case Case Case Case Case
10 11 12 13 14 15 16 17

-10

-20

-30

Fig. 8. Prediction error with the proposed criterion for C36 steel.
126 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

not good: the error is up to 26%. As mentioned in the Section 3.1, the plasticity model used is poor because
it does not consider kinematics hardening. This is the reason why the local stress state at the tip of the
defect has to be described in a better way. Under torsion loading and R = 1 (cases 10–12), results are
fairly good. This gives an experimental argument to use in the criterion the local value of the shear stress
without considering the effect of its gradient. Under torsion and R = 0.5 (case 15), the result is also good.
Under combined loading (cases 16 and 17), both compression–torsion and tension–torsion tests give good
results.
This set of experiments on C36 steel leads to the conclusion that the criterion is able to describe the
geometry and the size of the defect for different multiaxial loading cases except for tension with mean stress.
This is probably due to the plasticity model used which is not able to give a realistic description of the stress
state around the defect in this case.
A general discussion of the results is proposed in Section 5.4.

5.2. Validation for nodular cast iron

This material contains very small natural defects such as microshrinkages. Nevertheless, the criterion is
validated here for artificial spherical defects in order to test different defects from 95 to 1000 lm. The dif-
ferent cases are presented in Table 3 and the validation of the criterion in Fig. 9. In the case of materials
containing natural casting defects, it is not possible to identify the parameters on the defect free material so
that in this case the following experiments are used to identify the criterion:

Fatigue limit under tension, 300 lm spherical defect: RD = 180 MPa.


Fatigue limit under tension, 700 lm spherical defect: RD = 135 MPa.
Fatigue limit under torsion, 400 lm spherical defect: TD = 185 MPa.

The parameters are therefore: a = 1.306, c = 232 MPa and a = 460 lm1. The criterion gives good results
in all the range of sizes and loading studied including tension with mean stress. It is to be noted that in this
case the test case number 11 is done on a big defect of 1000 lm. No experiment is conducted directly on
small spherical defect (100–400 lm) under tension with mean stress but an extrapolation of the results
shows that results become bad in this case for smaller defects. The general feature of the results is the same
as for the C36.

Table 3
Description of the different experimental cases for the FGS 52
Case number Loading type Defect type Defect size Rmax Rmoy Tmax Tmoy
(lm) (MPa) (MPa) (MPa) (MPa)
Case 1 Tension Spherical 200 210 0
Case 2 Tension Spherical 300 180 0
Case 3 Tension Spherical 400 165 0
Case 4 Tension Spherical 700 135 0
Case 5 Tension Spherical 1000 120 0
Case 6 Torsion Spherical 200 205 0
Case 7 Torsion Spherical 300 195 0
Case 8 Torsion Spherical 400 185 0
Case 9 Torsion Spherical 700 165 0
Case 10 Torsion Spherical 1000 155 0
Case 11 Tension Spherical 1000 250 137
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 127

20
error
%

10

0
Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8 Case 9 Case 10 Case 11

-10

-20

Fig. 9. Prediction error with the proposed criterion for FGS 52.

Table 4
Description of the different experimental cases for the S 45 C
Case number Loading type Defect type Defect Rmax Rmoy Tmax Tmoy
size (lm) (MPa) (MPa) (MPa) (MPa)
Case 1 Flexion Drill hole 45 225 0
Case 2 Flexion Drill hole 90 205 0
Case 3 Flexion Drill hole 180 182 0
Case 4 Flexion Drill hole 0
Case 5 Flexion Drill hole 450 157 0
Case 6 Torsion Drill hole 180 133 0
Case 7 Torsion Drill hole 270 126 0
Case 8 Torsion Drill hole 450 120 0
Case 9 Torsion No defect 142 0
Case 10 Flexion No defect 240 0

5.3. Validation from S 45 C

In this section, we propose to validate the criterion on a carbon steel type S 45 C and from experiments
conducted by Murakami. All the experimental details can be founded in [14] and Table 4 presents the re-
sults used for our validation. The material cyclic behaviour is reported in Appendix A, it has been taken
from the literature on the bases of the type of steel used by Murakami and the hardness of the material.
For this material, defects are artificial type (drill hole) so that the experiments used to identify the criterion
are the following:

Fatigue limit under rotative bending, defect free material: RD = 240 MPa.
Fatigue limit under torsion, defect free material: TD = 142 MPa.
Fatigue limit under tension, 180 lm cylindrical defect: RD = 205 MPa.
128 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

40
error
30
%
20

10

0
Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8
-10

-20

-30

-40

-50

-60

-70

-80

-90

-100

Fig. 10. Prediction error with the proposed criterion for S 45 C.

The parameters are therefore: a = 0.043, c = 142 MPa and a = 2208 lm1. Under tension and R = 1,
results in Fig. 10 are good except for the smaller defect (45 lm). This shows that the criterion can describe
a large population of defect morphology because a drill hole is different than our defects. This is the first
limit of this criterion: a minimum size has to be given depending on the material, this is discussed in the next
section. Results under torsion are not very good: the error varies from 19% to 32%. Remember that for this
material, the elastic plastic stabilised cyclic stress–strain curve is taken from the literature and is not
the same as the S 45 C test by Murakami. Nevertheless, this set of experiments gives encouraging
results for the application of the proposed criterion to a various class of defect size, morphology and load-
ing path.

5.4. Discussion of the results

In regard to the simplicity of the proposed criterion, results are fairly good. Thirty-four experimental
results have been obtained on different materials, defects and loading: 8 simulations give more than 12%
error. The general conclusion that can be drawn for this global analysis of the result is that the gradient
of the hydrostatic part of the stress distribution at the tip of the defect seems to be a good parameter to
represent the defect influence on the fatigue resistance in the range of the fatigue limit under multiaxial
loading. This way to characterise a defect includes the following parameters:

• material by the mean of the cyclic stress–strain curve,


• defect morphology (size and shape) in relation with the macroscopic loading.

As proposed by Papadopoulos, this criterion does not introduce the gradient of the shear part of the stress
tensor. This has no physical motivation but results are nevertheless quite good. A tentative to consider the
gradient of the shear stress has been tested but it needs a new parameter. This new parameter needs to be
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 129

identified with another fatigue limit for a given defect size. The comparison of the present criterion
with a version including the effect of the shear stress give the following mean result (cases 2–8 for the
C36):

• gradient of the hydrostatic stress only: error of 6.5%,


• gradient of the equivalent stress (hydrostatic + shear stress): mean error 6%.

It seems therefore that the most influencing parameter is the gradient of the hydrostatic stress.
Concerning the way to measure the stress gradient, Eq. (8) has been used here: it is based on the spatial
distribution of stresses at the tip of the defects studied. The disadvantage of this measurement is that it is
dependent on the defect size. It is also possible to use this criterion by calculating the stress gradient by
considering Eq. (7): results obtained are also good. It is probably a better method for application on real
components.
Two tests on the C36 give bad results: cases 13 and 14. When the mean stress is different than 0, results
become poor. This can be explained by the plasticity model which is not able to describe the influence of the
mean stress due to the isotropic hardening law. The application of this criterion to various industrial load-
ing needs to use a plasticity model including combined isotropic and kinematics hardening in order to de-
scribe the major effects of cyclic loading and to get a more accurate value of the spatial distribution of
stresses.
The torsion test with mean stress (C36, case 15) is very interesting: the experimental result (see Table 2)
shows that the fatigue limit under torsion is influenced by the mean torsion stress for defective materials.
This result is different than the defect free material behaviour where the fatigue limit is not influenced by the
mean shear stress. This is due to the fact that the defect induces triaxiality so that the stress state is no more
pure shear stress. This experimental result has been also mentioned by [30] or [31]; the last one shows that
the fatigue limit of springs under torsion is influenced by the mean stress and the steel used for spring con-
tains microdefects. The criterion gives good prediction in this case by considering only the shear stress at
the tip of the defect.
For very small defects (C36, case 5—FGS 52, case 1—S 45 C case 1), results are poor or very bad. We
have to remind that when a defect becomes smaller with a size close to the microstructural parameter (grain
size for example), its influence on the fatigue limit is lower and lower so that it can no longer influence the
fatigue limit. For the C36 tested, defects lower than 100 lm under tension and 200 lm under torsion have
no influence on the fatigue limit. This is a transition between the defect that can be considered as a stress
concentrator with a mechanical analysis and microstructural effect where the defect cannot be considered as
a notch and where the macroscopic hardening law is not available to calculate the microscopic stress dis-
tribution around the defect. With the present approach, the size that does not influence the fatigue limit of
the material is defined as the minimum size to apply the criterion. A correct application of the present cri-
terion needs also to verify that the macroscopic plasticity model is used properly at the scale of the smallest
defect.
For very big defects, the gradient effect becomes lower (G tend to 0) so that results tend to a local ap-
proach. Results are generally good for defects up to 1 mm so that it seems that the criterion has no limit of
application for big defects. It could be interesting to test this criterion for macroscopic notches.

5.5. Application to other materials with natural defects

The present paper is devoted to the validation of the criterion for various artificial defects in differ-
ent materials. This is a first step to validate the proposition in cases where the defect morphology is well
130 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

defined. In this section, some elements are given for the application of the present criterion to natural
defects.
The criterion has been validated for three different materials but the plastic behaviour is not so different
(see Appendix A). It is necessary to validate this proposition for a high strength steel. We could imagine
that the influence of a defect is not the same from a ductile to a fragile material.
The validations to natural defects such as shrinkage or oxide is also under progress but needs new data.
A shrinkage is a complex geometry defect with two scales (see Appendix B): the scale of the global geometry
and the scale of the local roughness. The question is now: do we have to consider the local roughness of a
defect or do we have to look at the defect with global geometry. The answer is not simple and depends on
the ductility of the material. Generally speaking, we can consider that ductile materials are less influenced
by the local roughness than fragile ones under fatigue loading. The application of the criterion as proposed
in this paper to natural defects seems therefore possible for ductile materials but it needs to be studied for
ÔfragileÕ materials. Furthermore, natural defects can be located in the bulk of the material where the envi-
ronment is different. It is therefore necessary to conduct a complete study to identify the three parameters of
the criterion in the case of internal defects.

6. Conclusion

From this study, the following concluding remarks can be done:

• The elastic–plastic analysis of the stress level around defects shows that the local values of stresses at
the maximum loaded point cannot be a good parameter to calculate the fatigue limit for defect
materials.
• The study of the stress distribution around defects shows that the stress gradient of the shear part of the
stress tensor does not influence the fatigue limit. At the contrary, the stress gradient of the hydrostatic
part of the stress tensor has a strong influence on the fatigue limit.
• A multiaxial fatigue criterion for defective materials is proposed: the defect is represented by the gradient
of the hydrostatic stress and this parameter is introduced in CrosslandÕs criterion. The criterion contains
three coefficients identified from fatigue limit experiments.
• Experimental validation of the proposed criterion is done for three different materials: C36 steel, nodu-
lar cast iron and S 45 C steel. Three different type of defect are tested under tension, torsion and
combined tension–torsion. Results are very encouraging because in most cases the error is lower than
15%.

Appendix A. Cyclic stabilised stress–strain curve for the different materials tested
 1p
Det Dr Dr
¼ þ
2 2E 2k

E (GPa) k (MPa) p
C36 210 1232 0.214
Fonte GS52 178 1149 0.15
S 45 C 210 1228 0.185
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 131

Appendix B. Different types of defect


132 Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133

References

[1] Ting JC, Lawrence FV. Modelling the long-life fatigue behaviour of a cast aluminium alloy. Fatigue Fract Engng Mater Struct
1993;16(6):631–47.
[2] Ehrström JC, Roy D, Garghouri M, Fougeres R. Modelling the effect of micro-pores on the fatigue life AA7050 aluminium alloy
thick plates. In: Brown MW, De Los Rios ER, Miller KJ, editors. Proceedings of ECF 12: fracture from defects, Sheffield, 1998.
p. 217–22.
[3] Lu ZJ, Ciepalowicz AJ, Devlukia J, Evans WJ. Fatigue life prediction for a suspension arm made of cast aluminium alloy. In:
Evans WJ, Evans RW, Bache MR, editors. Proceedings of component optimisation from material properties and simulation
software, 1999. p. 343–50.
[4] Grandt Jr AF, Scheumann TD, Todd RE. Modelling the influence of initial material inhomogeneities on the fatigue life of
notched components. Fatigue Fract Engng Mater Struct 1993;16(2):199–213.
[5] De Kazinczy F. Effect of small defects on the fatigue properties of medium-strength cast steel. J Iron Steel Inst
1970:851–5.
[6] Jayet-Gendrot S, Gilles P, Migne C. Behaviour of duplex stainless casting defects under mechanical loadings. Fatigue Fract
ASME 1997;1.
[7] Heuler P, Berger C, Motz J. Fatigue behaviour of steel castings containing near-surface defects. Fatigue Fract Engng Mater Struct
1992;16(1):115–36.
[8] Nadot Y, Mendez J, Ranganathan N, Beranger AS. Fatigue life assessment of nodular cast iron containing casting defects.
Fatigue Fract Engng Mater Struct 1999;22:289–300.
[9] Starkey MS, Irving PE. Prediction of fatigue life of smooth specimens of SG Iron by using a fracture mechanics approach. In:
Amzallag C, Leis BN, Rabbe P, editors. Low-cycle fatigue and life prediction. ASTM STP, vol. 770. American Society for Testing
and Materials; 1982. p. 382–98.
[10] Nadot Y, Denier V. Fatigue failure of suspension arm: experimental analysis and multiaxial criterion. Engng Failure Anal
2004;11(4):485–99.
[11] Tanaka K, Nakai Y, Yamashita. Fatigue growth threshold of small cracks. Int J Fract 1981;17(5):519–33.
[12] Lukáš P, Kuntz L, Weiss B, Stickler R. Nondamaging notches in fatigue. Fatigue Fract Engng Mater Struct 1986;9(3):195–204.
[13] Weiss B, Stickler R, Blom AF. A model for the description of small three dimensional defects on the high cycle fatigue limit. In:
Miller KJ, De Los Rios ER, editors. Proceedings of ‘‘short fatigue cracks’’, ESIS 13. London: Mechanical Engineering
Publications; 1992. p. 423–38.
[14] Murakami Y. Metal fatigue: effects of small defects and nonmetallic inclusions. Elsevier; 2002.
[15] Endo M, Murakami Y. Effects of an artificial small defect on torsional fatigue strength of steels. J Engng Mater Technol
1987;109:124–9.
[16] Endo M. Effects of small defects on the fatigue strength of steel and ductile iron under combined axial/torsional loading. In:
Ravichandran KS, Ritchie RO, Murakami Y, editors. Proceedings of third engineering foundation international conference; small
fatigue cracks: mechanics, mechanisms and applications, Hawaii, 1999. p. 375–87.
[17] Yao W. The prediction of fatigue behaviours by stress field intensity approach. Acta Mech Solida Sinica 1996;9(4):337–49.
[18] Qilafku G, Azari Z, Kadi N, Gjonaj M, Pluvinage G. Application of a new model proposal for fatigue life prediction on notches
and key-seats. Int J Fatigue 1999;21:753–60.
[19] Qilafku G. Loading mode and notch effect in high cycle fatigue. In: Pluvinage G, Gjonaj M, editors. Actes de the NATO advanced
research workshop on notch effects in fatigue and fracture, Durres, Albania, 2000. p. 221–37.
[20] Qilafku G, Kadi N, Dobranski J, Azari Z, Gjonaj M, Pluvinage G. Fatigue of specimens subjected to combined loading. Role of
hydrostatic pressure. Int J Fatigue 2001;23:689–701.
[21] Taylor D, Zhou W, Ciepalowicz AJ, Devlukia J. Mixed-mode fatigue from stress concentrations: an approach based on
equivalent stress intensity. Int J Fatigue 1999;21:173–8.
[22] Nadot Y, Mendez J, Ranganathan N. Influence of casting defects on the fatigue limit of nodular cast iron. Int J Fatigue
2004;26:311–9.
[23] Billaudeau T, Nadot Y, Bezine G. Multiaxial fatigue limit for defective materials: mechanisms and experiments. Acta Mater
2004;52:3911–20.
[24] Crossland B. Effect of large hydrostatic pressures on the torsional fatigue strength of an alloy steel. In: Proceedings of the
international conference on fatigue of metals, Institution of Mechanical Engineers, London, 1956. p. 138–49.
[25] Findley WN. A theory for the effect of mean stress on fatigue of metals under combined torsion and axial load or bending. Trans
ASME J Engng Indus Ser B 1959;81.
[26] Matake T. An explanation of fatigue limit under combined stress. Bull Jpn Soc Mech Engrs 1980;2(1).
[27] Papadopoulos IV, Panoskaltsis VP. Invariant formulation of a gradient dependent multiaxial high-cycle fatigue. Engng Fract
Mech 1996;55(4):513–28.
Y. Nadot, T. Billaudeau / Engineering Fracture Mechanics 73 (2006) 112–133 133

[28] Flavenot JF, Skalli N. LÕépaisseur de couche critique ou une nouvelle approche du calcul en fatigue des structures soumises à des
sollicitations multiaxiales. Méca Matér Electr 1983;397:15–25.
[29] Morel F, Palin Luc T. A nonlocal theory applied to high cycle multiaxial fatigue. Fatigue Fract Engng Mater Struct
2002;25:249–665.
[30] Marquis G, Socie D. Long-life torsion fatigue with normal mean stresses. Fatigue Fract Engng Mater Struct 2000;23:293–300.
[31] Beretta S. Application of multiaxial fatigue criteria to materials containing defects. Fatigue Fract Engng Mater Struct
2003;26:551–9.

You might also like