Sie sind auf Seite 1von 28

Article

International Journal of Damage


Mechanics
0(0) 1–28
Modelling low-cycle fatigue tests ! The Author(s) 2016
Reprints and permissions:
using a gradient-enhanced sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1056789516653244
continuum damage model ijd.sagepub.com

Heraldo S da Costa Mattos

Abstract
This study is concerned with the modelling of uniaxial low-cycle fatigue tests using a gradient-enhanced
continuum damage model. The aim of the paper is to show that a simple but adequate description can be
performed using such kind of phenomenological approach. The model accounts for the coupling between
plasticity and damage, including a mathematically consistent description of the softening behaviour and size
dependency. It successfully predicts low-cycle fatigue and can deal with regular cycles. A consistent
procedure to identify experimentally the constitutive parameters that appear in the theory is proposed.
Lifetime predictions for different loading amplitudes are compared with experimental data for two alu-
minium alloys showing a good agreement. The one-dimensional model can be interpreted as a particular
version of a more general gradient-enhanced damage model, which is summarized at the end of the paper.

Keywords
Continuum damage mechanics, low-cycle fatigue, elasto-plasticity, non-local damage model, damage
accumulation, lifetime prediction

Introduction
The accurate modelling of low-cycle fatigue tests under random load paths is a basic step to under-
stand important physical phenomena such as crack initiation and propagation. Different approaches
have been attempted in this area, with a variety of cumulative damage theories developed for
metallic materials. Since the pioneer works of Coffin and Manson (the so-called Manson-Coffin
law, Coffin Jr, 1954; Manson, 1954) and of Miner (the linear damage accumulation rule, also known
as Miner’s rule, Miner, 1945) developed decades ago, many other alternative studies concerned with
the fatigue damage accumulation in metallic materials under variable strain amplitude have been

Laboratory of Theoretical and Applied Mechanics, Graduate Program in Mechanical Engineering, Department of Mechanical
Engineering, Universidade Federal Fluminense, Niterói, RJ, Brazil
Corresponding author:
Heraldo S da Costa Mattos, Laboratory of Theoretical and Applied Mechanics, Graduate Program in Mechanical Engineering,
Department of Mechanical Engineering, Universidade Federal Fluminense, Rua Passo da Pátria 156, 24210-240, Niterói, RJ, Brazil.
Email: heraldo@mec.uff.br

Downloaded from ijd.sagepub.com by guest on September 19, 2016


2 International Journal of Damage Mechanics 0(0)

performed up to now (e.g. Fatemi and Yang, 1998; Jinescu, 2012; Kwofie and Rahbar, 2013;
Lv et al., 2015; Manson and Halford, 1986; Seweryn et al., 2008; Sun et al., 2014; Zuo et al., 2015).
The present work adopts a phenomenological approach presenting a special type of non-local
continuum damage model. Although this kind approach to structural integrity problems is not new
(ductile damage, creep damage, fatigue damage), continuum damage mechanics is still a very active
area of research, in particular the adequate modelling of low-cycle fatigue in metals and alloys (see
Martinez et al., 2015). A large number of papers were published in the literature concerning the
modelling, in the framework of continuum damage theory, of the fatigue damage under a large
variety of one-dimensional and multiaxial loading paths including complex non-proportional load-
ing paths (Lemaitre, 1992; Lemaitre and Chaboche, 1992; Lemaitre and Desmorat, 2005;
Murakami, 2012). However, most of these damage models are local, different from the damage
model presented in this paper.
One of the main advantages of continuum damage models is that they do not require cycle
counting strategies (Anes et al., 2014; Bisping et al., 2014). However, it is not a simple task to do
a mathematically correct and physically realistic description of the damage accumulation and of the
strain-softening behaviour due to the degradation of the microstructure. In the case of metallic bars,
the cyclic plastic strain induces a strain-hardening and also a degradation of the structure (fatigue
damage). On the other hand, the degradation of the structure induces a softening behaviour in the
engineering stress–strain curves. Hence, it is very important to model the coupling between plasticity
and damage in order to perform an adequate lifetime prediction.
The main feature of the present study is that the evolution of the damage variable is supposed to
be affected by its second-order spatial derivative. One of the main differences between the present
model and others found in literature is how the plastic multiplier affects this additional term. In the
present case, there is no direct product between the Laplacian of the damage variable and the plastic
multiplier. The model accounts for the coupling between plasticity and damage, including a math-
ematically consistent description of the softening behaviour and size dependency. The use of con-
tinuum mechanics models into design and structural integrity assessment of inelastic mechanical
components is often restricted by the difficulties to find references about material parameters
obtained experimentally. In the present study, a simple but consistent procedure to identify experi-
mentally the constitutive parameters that appear in the theory is presented. These parameters are
obtained through strain-controlled low-cycle fatigue push–pull tests. The determination of the stable
hysteresis loop in elasto-plastic behaviour is an important step concerning low-cycle fatigue of
metallic structure.
A numerical procedure to approximate the resulting governing equations using a Fractional Steps
Method is summarized in Appendix 1. The coupled mathematical problem is split into a sequence of
three simpler problems: elastic predictor, diffusion problem and evolution problem. The first step,
elastic predictor, is a non-linear two-point boundary value problem solved by a shooting technique.
In the second step, the ‘‘diffusion’’ of the damage into the body, an implicit method such as Crank-
Nicolson method and Crout reduction is used to solve tridiagonal linear systems. In the third and
last step, an Euler explicit method was used, together with a projection technique to solve the
evolution problem.
The effectiveness and usefulness of the procedure is checked by comparing numerical simulation
of non-monotonic uniaxial tests in AA 6351 and AA 2011 specimens at room temperature with
experimental results. The plastic strain, damage, isotropic and kinematic hardening are strongly
coupled in these materials. The determination of the stable hysteresis loop in elasto-plastic behav-
iour is an important step concerning low-cycle fatigue of metallic structures. The theoretical and
experimental stable loops are compared for different load amplitudes; besides, lifetime predictions

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 3
for different strain amplitudes are compared with experimental data, showing a good agreement. It
is important to remark that low-cycle fatigue uniaxial testing is so far the most important technique
used to rank the susceptibility of different materials to non-monotonic load histories. Most models
require a large number of fatigue tests in order to provide basic parameters to be directly used in
engineering design, mainly to estimate the influence of the loading history on the damage accumu-
lation. Hence, the study of plastic oligocyclic fatigue in metallic materials is still up to now strongly
dependent on these tests. However, one advantage of the proposed model is that only one fully
reversed fatigue test is required to identify all material parameters that appear in the theory.
The one-dimensional model can be interpreted as a particular version of a more general gradient-
enhanced damage model, which is summarized in Appendix 1. One of the main shortcomings of
these non-local continuum damage theories is that the general theory may seem too complex for a
neophyte with many cumbersome formulas. This is especially true for the so-called gradient-
enhanced continuum damage theories, which are being used for a long time but still a very active
area of research (see Alessi et al., 2015). The goal of Appendix 1 is not to provide a ‘‘validation’’ of
this theory, but to show that the present study can be used as a first step towards a more general
three-dimensional theory.

One-dimensional model
In this section, one-dimensional constitutive equations are proposed to model plastic oligocyclic
(low-cycle) fatigue in metallic specimens subjected to cyclic loading. Typically, in a fatigue test,
rupture occurs in less than 1000 cycles. The model is developed within the framework of continuum
damage mechanics considering an auxiliary damage variable D 2 ½0, 1 associated with the dissipa-
tive mechanism of rupture. If D ¼ 0, the section is not damaged and if D ¼ 1 it no longer can resist
to the mechanical loading.

Preliminary definitions
The mechanical system in analysis is the useful portion (gauge length) of a round elasto-plastic
specimen with gauge length Lo and cross section Ao submitted to a prescribed elongation ðtÞ
(Figure 1). FðtÞ is the axial force applied at the extremities in order to impose this elongation history.

Figure 1. Mechanical system considered in the analysis.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


4 International Journal of Damage Mechanics 0(0)

uðz, tÞ is the axial displacement of the centroid of the cross section z at instant t. The model is
conceived to study the fatigue behaviour in the presence of small plastic strains. Besides the lifetime
prediction, it only allows to access macroscopic quantities (forces and displacements) and all other
variables must be interpreted as global average parameters. ðtÞ ¼ ðF=Ao Þ is the axial stress compo-
nent. This kind of global approach is often used in a thermodynamic framework to describe damage
evolution in complex tensile tests of inelastic structures and materials (da Costa Mattos and
Chimisso, 2011; da Costa Mattos et al., 2008, 2012). The main feature of the present study is that
the evolution of the damage variable is supposed to be affected by its second-order spatial derivative.
This hypothesis, as discussed in the next sections, allows to model the strain localization (although 
is constant along the length, the deformation is non-homogeneous) and of the dependency of the
fatigue behaviour on the length of the specimen. This one-dimensional model can be interpreted as a
particular case of a more general three-dimensional damage model, which is summarized in the last
section of the paper.
The mechanical behaviour of the system is supposed to be governed by the following set of
equations

@
¼0 ð1Þ
@z
@u
¼" ð2Þ
@z

 ¼ ð1  DÞEð"  "p Þ ð3Þ

X ¼ ð1  DÞac ð4Þ

Y ¼ ð1  DÞ½v1 ð1  expðv2 pÞ þ y  ð5Þ



@"p 1, if ð  XÞ  0
¼ _ Sg; Sg ¼ ð6Þ
@t 1, otherwise

@p
¼ _ ð7Þ
@t

@c @"p b _
¼  X ð8Þ
@t @t a

_  0,  ¼ j  Xj  Y  0, _ ¼0
 ð9Þ
     
@D _ p 2 2 ev2 p @2 D
¼ Eð"  " Þ þ ac þ 2 v1 p þ þ py þ k 2 ð10Þ
@t 2So v2 @z
Together with the following set of boundary and initial conditions

Dðz ¼ 0, tÞ ¼ 0; Dðz ¼ Lo , tÞ ¼ 0 8t
ð11Þ
uðz ¼ 0, tÞ ¼ 0; uðz ¼ Lo , tÞ ¼ ðtÞ 8t

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 5

uðz, t ¼ 0Þ ¼ 0; ðz, t ¼ 0Þ ¼ Xðz, t ¼ 0Þ ¼ Yðz, t ¼ 0Þ ¼ 0;
ð12Þ
"p ðz, t ¼ 0Þ ¼ pðz, t ¼ 0Þ ¼ cðz, t ¼ 0Þ ¼ Dðz, t ¼ 0Þ ¼ 0; for 0  z  Lo
Rt

" is the axial strain and "p the plastic strain. pðtÞ ¼ 0
"_p ðÞ
d is the accumulated plastic strain.  is
the yield function and _ is the plastic multiplier. c is an auxiliary variable related to the kinematic
hardening induced by the plastic strain. In the framework of thermodynamic of irreversible pro-
cesses, ("p ,p,c) are called the internal variables, while , Y and X are the thermodynamic forces
associated to them. E, K, N, v1 , v2 , p , , and k are material constants. a, b, So are material
parameters given by the following expressions
      2
100L a2 100L b2 100L So
a ¼ a1 ; b ¼ b1 ; So ¼ S1o ð13Þ
Lo Lo Lo
with a1 , a2 , b1 , b2 , S1o and S2o being the material constants. Generally, a, b and So are constants for
materials with negligible cyclic hardening or softening.
Equation (1) is the equilibrium equation and equation (2) the definition of the axial strain.
Equations (3) to (5) will be called the state laws and equations (6) to (8) the evolution laws. The
loading–unloading conditions in equation (9) can be shown to be equivalent to the Karush-Kuhn–
Tucker complementary conditions. Using these complementary conditions, it is simple to verify
that _ ¼ 0 if  5 0. As a consequence, using the evolution laws, it is possible to conclude that
@"p @c @p
@t ¼ @t ¼ @t ¼ 0 if  5 0 (elastic behaviour). Equation (10) is the balance equation for the damage
variable, with the angle bracket hxi (usually called the McCauley bracket) defined as
hxi ¼ maxf0, xg. From this definition, it is simple to verify that @D @t  0. The Laplacian of the
damage variable in equation (10) (that will be called the ‘‘diffusive term’’ in the damage equation)
allows the modelling of strain localization due to the damage and to introduce some size depend-
ency (which is indeed observed in experiments). Since the increase of damage generally leads to a
local softening behaviour, the rate-independent damage models based on a local approach may
lead to a physically unrealistic description of strain localization phenomenon. This additional term
regularizes the classic local damage models and avoids all these shortcomings. The Laplacian of
the damage variable appears in many nonlocal damage models (a balance equation for the damage
variable instead of an evolution law), but with different interpretations. Many studies within the
context of continuum damage mechanics were performed using such kind of approach in the last
decades, but very few are concerned with low-cycle or ultra-low-cycle fatigue analysis (see Costa
Mattos and Sampaio, 1995; Costa Mattos et al., 1992; da Costa Mattos et al., 2009, in the case of
brittle-elastic behaviour, and Chimisso and Costa Mattos, 1994; Forest, 2009; Frémond and
Nedjar, 1996; Nedjar, 2001; Saanouni and Hamed, 2013; Saanouni, 2012, in the case of inelastic
behaviour).
The main difference between this equation and other gradient-enhanced damage models is that
they usually have the following abstract form (f is a general function of the state variables which
depend on the model)
 
@D @2 D
¼ _ f ð", "p , p, c, DÞ þ k 2 ð14Þ
@t @z

In the present model, there is no direct product between the Laplacian of the damage variable and
the plastic multiplier _ (see equation (10)), because it seems to better describe the strain softening
phenomenon in the case of low-cycle and ultra-low-cycle fatigue. However, both approaches give

Downloaded from ijd.sagepub.com by guest on September 19, 2016


6 International Journal of Damage Mechanics 0(0)

Figure 2. Boundary conditions – The fatigue damage at the extremities of the specimen is negligible.

promising results and we do not claim that the one proposed in this paper is the only good one. The
goal is to present an alternative that cannot be neglected in future studies.
Using equation (10) together with the boundary conditions for the damage variable in equation
(12), it is also possible to prove that @D _ @D
@t 6¼ 0 only if  6¼ 0 (or, what is equivalent, that @t ¼ 0 when
_ ¼ 0). Therefore, the ‘‘diffusive term’’ in equation (10) is only activated when the system is undergo-
ing an inelastic deformation. Equation (10) that governs the evolution of D is similar to the heat
equation with a non-linear heat source term. k and So play a role, respectively, analogous to the
thermal conductivity and to the specific heat. The smaller the k is, the more localized the damage is
(and, consequently, the strains " and "p ). The variable k is related to the spatial distribution of D and
So is related to the evolution of D at a given point: the bigger the So is, the bigger must be the
mechanical energy in order to obtain a given damage rate.
The boundary conditions for the damage variable in equation (12) are obviously an approxima-
tion in which it is assumed that the damage at the extremities of the system is negligible compared
with the damage at its useful portion (Figure 2). The initial conditions (11) describe a ‘‘virgin’’ initial
state (undamaged and no accumulation of plastic strain).
Figure 3 presents a schematic representation of the internal variables ", "p and D in a section of
the specimen when it is loaded monotonically and then unloaded until F ¼ 0. In traditional
‘‘homogeneous’’ damage models, the damage variable can be measured experimentally from the
variation of the system stiffness. In the present approach, the damage variable in each section of
the system can be measured by unloading the specimen and measuring the variation of the slope
of the curve AFo  @u
@z obtained experimentally. In a phenomenological approach, this damage vari-
able is associated with the variation of the stiffness of the useful portion of the specimen. Such a
variation is due to the degradation induced by the plastic deformation (with a big influence of the
accumulated plastic strain). Experimentally, in most metals and alloys, the following behaviour is
observed: initially the damage evolution is almost negligible (strain localization in slip bands,
micro-cracks, etc.) and then the evolution is very fast (a macro-crack appears). This makes the
idea of introducing a ‘‘critical’’ damage interesting (see Figure 3, in which N is the number of
cycles).

Summary of the techniques to identify the material parameters


The identification of the parameters fE, , y g (Young’s modulus, Poisson’s ratio and yielding stress)
is classical and will not be discussed in this paper. Cyclic tests with controlled elongation of the

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 7

Figure 3. Schematic representation of the variable Dat a section z. N is the number of cycles.

Figure 4. Cyclic test with prescribed elongation.

gauge length can be used to identify the other material parameters that appear in the theory
(Figure 4). In strain-controlled fully reversed cyclic tests, the mechanical response of most metals
and alloys tends to be stabilized (periodic) after a certain number of cycles. When fatigue damage
became important, the material no longer has a periodic response. The main hypothesis to be
adopted in the parameter identification is that damage can be considered negligible until the stabi-
lized cycle. The parameters fE, y , a, b, v1 , v2 g can be identified from one cyclic test until stabilization.
The procedure is exactly the same as proposed in Costa Mattos and Soares Filho (2010) for a
particular model of elasto-plasticity without damage.
The parameters k and So , can be identified experimentally from m fatigue tests with constant
amplitude of elongation ðLÞi ði ¼ 1, mÞ by adjusting the life ðNÞ ^ i predicted numerically with the

Downloaded from ijd.sagepub.com by guest on September 19, 2016


8 International Journal of Damage Mechanics 0(0)

experimental results ðNÞi . If the linear least-square method is considered, the error function ðSo , kÞ
to be minimized is the sum of the squares of the discrepancies

m h
X i2
ðSo , kÞ ¼ ^ o , k; ðLÞi Þ
ðNÞi  NðS ð15Þ
i¼1

where NðS^ o , k; ðLÞi Þ is the life (number of cycles until D ¼ 1) obtained solving numerically
problems (1) to (13) considering a cyclic loading history similar to the one presented in
Figure 4. A reasonably simple numerical technique to solve this problem is presented in the
next section.
In general, for most metals and alloys at room temperature, if a specimen following the ASTM
E606/E606M-12 standard method is considered, it has been observed that k0 ¼ ðk=L2o Þ  0:01 s1 . If
such an approximate value is considered, only one experimental fatigue test is necessary in order to
identify So . In this case, So is the root of the function ðSo Þ.

^ o , LÞ
ðSo Þ ¼ N  NðS ð16Þ

^ o ; LÞ is the life obtained solving numerically problems (1) to (13) with a cyclic loading
where NðS
history with amplitude L and with Lo being the gauge length of the specimen and k ¼ ð0:01L2o Þ s1 .
Of course, if more tests are performed, a better approximation can be obtained considering the
arithmetic mean of different values obtained from different experimental tests.

Numerical approximation
A reasonably simple numerical technique can be adopted for approximating the resulting non-linear
problem formed by equations (1) to (13). The solution of this non-linear problem is necessary in
order to identify the parameters k and So from fatigue tests. This technique was used for obtaining
the results presented in this paper. The numerical procedure for the analysis of this low-cycle fatigue
problem is not new, since it is very similar to the one presented in Costa Mattos and Chimisso (1997)
in the context of the study of the necking phenomenon of elasto-plastic bars. Although the present
paper is not concerned with the analysis (consistency, stability, convergence, etc. that have already
been performed) of the numerical technique used to obtain the results, it is interesting to show the
main features of the numerical technique.
Equations (1) to (13) form a non-linear system of partial differential equations together with a
unilateral constraint. The way used to obtain a numerical approximation for the problem is based
on the operator split method (Chorin et al., 1978; Ciarlet and Lions, 1990; Marchuck, 1975). The
objective of this technique is to take advantage of some additive decomposition properties of the
mathematical operator to solve a sequence of simpler problems instead of a unique complex
problem. The approximation techniques based on the operator split method and the associated
product formula algorithms have been applied in various areas of continuum mechanics and, in
particular, in elasto-plasticity and elasto-viscoplasticity (e.g. Ortiz et al., 1983; Simo and Miehe,
1992).
The following five-step procedure is proposed to find an approximation for f, u, ", "p ,
p, c, D, X, Yg from time tn (in which they are supposed to be known) to time tnþ1 ¼ tn þ t:

 u,
(i) A first approximation f,  for f, u, "g at instant tnþ1 is obtained by solving numerically the
 "g
following two-point boundary value ‘‘equilibrium’’ problem (the following notation is adopted

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 9
for all functions of z and t: f ðz, tn Þ ¼ fn ðzÞ

@ nþ1
¼0 ð17:1Þ
@z
@u nþ1
¼ "nþ1 ð17:2Þ
@z

nþ1 ¼ ð1  Dn ÞEð"nþ1  "pn Þ ð17:3Þ

u nþ1 ðz ¼ 0Þ ¼ 0; u nþ1 ðz ¼ Lo Þ ¼ nþ1 ð17:4Þ

(ii) A first approximation D for D at instant tnþ1 is obtained by solving numerically the following
linear partial differential equation

@D @2 D
¼k 2 ð18:1Þ
@t @z
 ¼ 0, tÞ ¼ 0;
Dðz  ¼ Lo , tÞ ¼ 0
Dðz ð18:2Þ

 tn Þ ¼ 0 for 0  z  Lo
Dðz, ð18:3Þ
 for f, Dg and a final approximation for f"p , p, X, Yg at instant
 Dg
(iii) A second approximation f,
tnþ1 and at each point z 2 ½0, Lo  is obtained by solving numerically the following initial
value problem (using the simplifying notation y_ ¼ dy
dt for the time derivative of any arbitrary
variable y)
    
_ _ ev2 p  t Þ ¼ D ðzÞ
D ¼ Eð"  "p Þ2 þ ac2 þ 2 v1 p þ þ py ; Dðz, n nþ1 ð19:1Þ
2So v2
(
p _ 1, if ð  XÞ  0
"_ ¼ Sg; Sg ¼ ; "p ðz, tn Þ ¼ "pn ðzÞ ð19:2Þ
1 otherwise

_
p_ ¼ ; pðz, tn Þ ¼ pn ðzÞ ð19:3Þ

b_
c_ ¼ "_p  X; cðz, tn Þ ¼ cn ðzÞ ð19:4Þ
a

_ ¼ ð1  DÞE"_p  DEð


_ "  "p Þ;  tn Þ ¼ ðzÞ
ðz,  ð19:5Þ

X_ ¼ ð1  DÞac_  D_ ac;
^ Xðz, tn Þ ¼ Xn ðzÞ ð19:6Þ

_ 1 ð1  ev2 p Þ þ y ; Yðz, tn Þ ¼ Yn ðzÞ


Y_ ¼ ð1  DÞv1 v2 ev2 p _  D½v ð19:7Þ

_  0, F ¼
  X
 Y  0, F ¼ 0 ð19:8Þ

Downloaded from ijd.sagepub.com by guest on September 19, 2016


10 International Journal of Damage Mechanics 0(0)

(iv) A final approximation for D at instant tnþ1 is obtained by the following projection procedure


If D  
nþ1 5 Dn then Dnþ1 ¼ Dn , otherwise Dnþ1 ¼ D nþ1 ð20Þ

(v) Since Dnþ1 and "pnþ1 were obtained in the previous steps, a final approximation for f, u, "g at
instant tnþ1 can be obtained by solving numerically the following two-point boundary value
‘‘equilibrium’’ problem

@nþ1
¼0 ð21:1Þ
@z
@unþ1
¼ "nþ1 ð21:2Þ
@z

nþ1 ¼ ð1  Dnþ1 ÞEð"nþ1  "pnþ1 Þ ð21:3Þ

unþ1 ðz ¼ 0Þ ¼ 0; unþ1 ðz ¼ Lo Þ ¼ nþ1 ð21:4Þ

The numerical solution of problems in steps (i), (ii) and (iv) is straightforward. Solution tech-
niques for these simple problems can be found in standard books of numerical analysis. The numer-
ical solution of problem (iii) is also simple, but requires a projection technique which is described
below.

Numerical solution of problems formed by equations (17.1) to (17.4) and (21.1) to (21.4). The
following equation is obtained by combining equations (17.2) and (17.3)

@u nþ1  nþ1


¼ þ "pn ð22Þ
@z ð1  Dn ÞE

The value of u nþ1 is known at the boundaries (equation (17.4)). Hence, we have
Z Lo  
nþ1
nþ1 ¼ þ "pn dz ð23Þ
0 ð1  Dn ÞE

Since  nþ1 is constant along the length (equation (17.1)), the value  nþ1 can be approximated by
integrating numerically equation (23). Different classical techniques can be used (the Trapezium
Rule and Simpson’s Rule, for instance). Once  nþ1 is known, "nþ1 can be obtained using equation
(17.3)

nþ1
"nþ1 ¼ þ"p ð24Þ
ð1  Dn ÞE n

u nþ1 can also be approximated through a numerical integration technique using equation (17.2). The
numerical solution of the problem formed by equation (21) can be performed in a similar way.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 11
Numerical solution of problem (18). Different classical techniques can be used for this simple
linear parabolic partial differential equation (finite difference of finite element techniques). It is
suggested to adopt the unconditionally stable Crank-Nicholson finite difference method together
with a Crout Reduction scheme to solve the resulting tridiagonal algebraic problem at each time
step. For further details see Ames (1969).

Numerical solution of problem (19). In order to solve numerically such a non-linear system of
ordinary differential equations, an Euler-type scheme together with a projection technique may
be used.

 
D nþ1 ¼ Dn þ D ð25:1Þ

"pnþ1 ¼ "pn þ "p ð25:2Þ

pnþ1 ¼ pn þ p ð25:3Þ

cnþ1 ¼ cn þ c ð25:4Þ

 nþ1 ¼  n þ  ð25:5Þ

Xnþ1 ¼ Xn þ X ð25:6Þ

Ynþ1 ¼ Yn þ Y ð25:7Þ

with
    
p ev2 pn
D ¼ Eð"nþ1  "pn Þ2 þ aðcn Þ2 þ2 v1 pn þ þ pn  y ð26:1Þ
2S0 v2

1, if ð nþ1  Xn Þ  0
"p ¼ pSg; Sg ¼ ð26:2Þ
1, otherwise

b
c ¼ "p  pXn ð26:3Þ
a

 ¼ ð1  D nþ1 ÞE"p  DEð"nþ1  "pn Þ ð26:4Þ

X ¼ ð1  D nþ1 Þac  Dacn ð26:5Þ

Y ¼ ð1  D nþ1 Þv1 v2 ev2 pn p  D½v1 ð1  ev2 pn Þ þ y  ð26:6Þ

It is simple to verify that all increments can be written as functions of p. p is obtained as
follows

If
 nþ1  Xn
 Yn  0 then p ¼ 0; otherwise, p is the root of the function #ðpÞ given by

#ðpÞ ¼
nþ1 þ   Xn  X
 Yn  Y ð27Þ

Downloaded from ijd.sagepub.com by guest on September 19, 2016


12 International Journal of Damage Mechanics 0(0)

The root of # can be found analytically and is given by the following expressions

In traction (Sg ¼ 1)

 nþ1 þ Eð1  D nþ1 Þð"nþ1  "pn Þ  Xn  Yn


p ¼ ð28:1Þ
½1  D nþ1 ½E þ a  bXn þ v1 v2 ev2 pn  þ An ½Eð"nþ1  "pn Þ  acn  v1 ð1  ev2 pn Þ  y 

In compression (Sg ¼ 1)

 nþ1  Eð1  D nþ1 Þð"nþ1  "pn Þ þ Xn  Yn


p ¼ ð28:2Þ
½1  D nþ1 ½E þ a þ bXn þ v1 v2 ev2 pn  þ An ½Eð"nþ1  "pn Þ þ acn  v1 ð1  ev2 pn Þ  y 

with
    
1 ev2 pn
An ¼ Eð"nþ1  "pn Þ2 þ aðcn Þ2 þ 2 v1 pn þ þ pn y ð28:3Þ
2So v2

Materials and experimental procedures


Two different aluminium alloys have been considered in this study: an AA 6351 alloy and an AA
2011 alloy. Fatigue test specimens were precision machined from the bar using a numerical-control
lathe. The specimens were machined with the stress axis parallel to the extrusion direction. They
were smooth and cylindrical in the gauge section, which measured 12 mm in diameter and 25 mm in
length. The length to diameter ratio of the fatigue specimens was chosen to ensure that it would not
buckle during the fully reversed cyclic training.
The machined surface did not need any further polishing since no circumferential scratches or
surface machining marks were present. The experimental tests have been carried out by means of a
servo hydraulic machine equipped with a 100 kN dynamical load cell and dynamical clip-on extens-
ometer with a gage length of 25 mm. Monotonic tests were performed to determine elastic modulus,
and yield strength was performed in displacement control using as input a ramp with a rate of
0.01 mm/s. Besides, fully reversed low-cycle fatigue tests were carried out according to ASTM E606/
E606M-12 standard method. The frequency utilized in the fatigue tests was 0.1 Hz, to assure an
isothermal process (the gradient of temperature during the testing program must not exceed a range
of  2 K as required by the ASTM standard) and the imposed load wave shape was triangular (see
Figure 4). In order to analyse cumulative effects, besides tests with constant amplitude loading,
fatigue tests with different amplitudes have also been performed.
ASTM standard test method provides parameters for the definition of fatigue failure of the tested
specimen: the percent reduction of the applied load or the presence of a macroscopic crack on the
specimen surface. However, both alloys do not show any softening before failure, so the monitoring
of the percent reduction of the applied load is not a suitable parameter for evaluating damage.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 13
To identify surface cracks is an even more difficult task since the defect appears in different ways,
depending on the strain level and on the material; furthermore, the cracks do not always start on the
surface, but also in the inner part. A sharp variation of the dynamic stiffness, calculated for every
cycle from the load cell and extensometer data, was therefore taken as the failure condition. In fact,
this variation is a symptom of the presence of a macroscopic crack. The variation is not always
represented by a reduction, but sometimes it may appear as an increase. This, as verified a posteriori,
depends upon the point from which the crack starts. In fact, the crack usually originates on the
surface, but, even if it is certainly between the extensometer blades, its circumferential position
determines either an apparent increasing or decreasing of the specimen stiffness. The section is no
longer symmetrical with respect to the load and a flexural stress appears. The neutral axis does not
coincide with the specimen axis, but rather it moves toward the extensometer which, as a conse-
quence, undergoes less strain with respect to the undamaged specimen case. The specimen appears to
be more rigid and consequently the dynamic elastic modulus increases. Vice versa, in the case of a
crack originating from a point between the extensometer blades, the measured strain is higher, the
specimen appears less stiff and a decreasing dynamic elastic modulus is found. The loss of loop
stability (as seen in Figure 5) is another criterion for stopping the test.
To assure load alignment on the specimen, before each cyclic test, the strain measured by means
of four strain gages glued to the median plane of the specimen were verified to be compatible to pure
membrane strain condition. Four strain gages were glued to an unloaded specimen for thermal
compensation and the final electrical configuration was composed by two full bridges, each one
containing two active strain gages located on the same diameter of the specimen and connected in
order to eliminate membrane strain an sum flexural strains.

Figure 5. AA 2011 – Loop instability due to a macroscopic crack.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


14 International Journal of Damage Mechanics 0(0)

Results and discussion


Fatigue test in AA 2011 specimens
An AA 2011 aluminum alloy heat-treated T6 was used in the experiments. Copper, alone or asso-
ciated with magnesium, gives good mechanical properties and makes the alloy suitable for heat
treatment. An important characteristic of this class of alloys is the high value of the fracture tough-
ness. Another important characteristic is the cyclic hardening observed in fatigue tests (see in
Figure 6, the comparison between the cyclic curve and the first part of the quasi-static stress–
strain curve).
The following fatigue tests with constant amplitude have been performed: (i) a fatigue test with
elongation amplitude ðL=Lo Þ ¼ 0:007. The experimental lifetime Nf1 was 800 cycles; (ii) a fatigue
test with elongation amplitude ðL=Lo Þ ¼ 0:01. The experimental lifetime Nf2 was 280 cycles.
Besides, in order to have a preliminary idea about damage accumulation in fatigue, the following
tests have been performed (Figure 7): (iii) a fatigue test in which initially it was applied on the
specimen 132 cycles of elongation amplitude ðL=Lo Þ ¼ 0:007 and then it was cycled with a pre-
scribed elongation amplitude ðL=Lo Þ ¼ 0:01 until rupture. The experimental lifetime Nf was 317
cycles; (iv) a fatigue test in which initially it was applied on the specimen 132 cycles of elongation
amplitude ðL=Lo Þ ¼ 0:01 and then it was cycled with prescribed elongation amplitude
ðL=Lo Þ ¼ 0:007 until rupture. The experimental lifetime Nf was 537 cycles. The experimental
results of all tests are summarized in Table 1.
Cumulative effects, whether linear or nonlinear, are very important in fatigue analysis.
Preliminary experimental results show that linear damage accumulation (such as Palmgreem-
Miner rule) is not adequate for the ASTM AA 2011 aluminium alloy. Palmgreen-Miner rule is
based on the assumption that damage is accumulated additively when it is defined by the associated
life ratio ðni =Nfi Þ where ni is the number of cycles applied under a given load for which the numbers

Figure 6. AA 2011 – Cyclic and quasi-static curves for AA2011.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 15
of cycles to rupture (under periodic conditions) would be Nfi . Considering such linear theory, the
fatigue criterion for different loading amplitudes is
X ni
¼1 ð29Þ
Nfi

If criterion (29) is valid, sequence effects are negligible (no matter the sequence of the cycles ni , the
final life would be the same). In these simple tests, it can be verified that
X ni
¼ C, with C ¼ 0:826 in test 1 andC ¼ 0:978 in test 2 ð30Þ
Nfi

In order to access the main features of the model (mainly the adequacy of the parameters experi-
mental identification), simulations of fatigue tests with amplitude different from the ones used to
identify the parameters are compared with the experimental results. A reasonably adequate estimate
of constants fa1 , a2 g, fb1 , b2 g and fS1o , S2o g in a given interval   ðL=Lo Þ  can be performed
considering three tests with the following elongation amplitudes

L1 =Lo ¼ ; L2 =Lo ¼ ð þ Þ=2; L3 =Lo ¼ ð31Þ

Figure 7. Cumulative damage in fatigue. Tests with two different strain amplitudes: ðL1 =Lo Þ ¼ 0:01 and
ðL2 =Lo Þ ¼ 0:007.

Table 1. AA 2011 – Cumulative fatigue tests.

n1 n2 N ¼ n1 þ n2

Test 01 ½ðL=Lo Þ ¼ 0:007 ½ðL=Lo Þ ¼ 0:01 317


132 185
Test 02 ½ðL=Lo Þ ¼ 0:01 ½ðL=Lo Þ ¼ 0:007 537
132 404

Downloaded from ijd.sagepub.com by guest on September 19, 2016


16 International Journal of Damage Mechanics 0(0)

Figure 8. AA 2011 – Evolution of the damage distribution along the length of the specimen for two different values
of k.

Table 2. AA 2011 – Material parameters.

E (GPA) y (MPa) a1 (GPA) a2 b1 b2 v1 (MPa) v2 k0 (s1) So (MPa)

74.00 280.00 15.119 1.326 203.00 1.359 112.00 10.80 0.01 3.278

In the present paper, the following values for  and were adopted

 ¼ 0:006 and ¼ 0:014 ð32Þ

As mentioned before, a reasonably adequate estimate of parameter k is ðk=Lo Þ ¼ 0:01s1 . It is


important to observe that the deformation tends to be more localized at the middle of the specimen
for smaller values of the parameter k. Figure 8 shows as an example the simulation of the damage
distribution along the specimen for two different values of k. For high values of k, damage is nearly
constant over the specimen length. Table 2 presents the mechanical parameters identified for
this alloy.
Figure 9 presents the theoretical and experimental stabilized cycle obtained for an elongation
amplitude ðL=Lo Þ ¼ 0:012, not used for the parameters identification, showing a very good agree-
ment. The adequacy of model predictions for different elongation amplitudes can be verified in
Figure 10 where the lifetime predictions (in cycles) for different elongation amplitudes are compared
with the experimental results.
The predicted lifetimes for the tests presented in Table 1 are presented in Table 3. The model
predictions are in very good agreement with the experiments. Only one test was performed per each
prescribed elongation amplitude. Of course, the scatter of results is important in fatigue analysis.
However, in this paper, the main point is that all the material parameters of the model were
identified experimentally using only one cyclic test with prescribed strain amplitude and that all
other results (stabilized cycles, predicted lifetimes, considering different loading levels) were obtained
as a consequence of the simulation using the proposed model.

Fatigue tests in AA 6351 specimens


Different from the aluminium alloy studied in the previous section, the cyclic hardening until sta-
bilization is negligible for this alloy. The isotropic hardening is also negligible @Y
@t  0 in this case.
The material parameters are depicted in Table 4.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 17

Figure 9. AA 2011 – Stabilized loop in a cyclic test withðL=Lo Þ ¼ 0:012.

Figure 10. AA 2011 – Lifetime prediction (in cycles) for different elongation amplitudes.

The prediction of the life N for different elongation amplitudes L and different parameters
ðSo , kÞ is presented in Figure 11. Curve (a) is the experimental curve, curve (b) is the model predic-
tion obtained taking k0 ¼ 0:1 s1 and So ¼ 15, 000 MPa, curve (c) is the model prediction obtained
taking k0 ¼ 0:01 s1 and So ¼ 15, 000 MPa, curve (d) is the model prediction obtained taking

Downloaded from ijd.sagepub.com by guest on September 19, 2016


18 International Journal of Damage Mechanics 0(0)

Table 3. AA 2011 – Cumulative fatigue tests (lifetime predictions).

n1 n2 N ¼ n1 þ n2

Test 01 ½ðL=Lo Þ ¼ 0:007 ½ðL=Lo Þ ¼ 0:01 318


132 186
Test 02 ½ðL=Lo Þ ¼ 0:01 ½ðL=Lo Þ ¼ 0:007 549
132 417

Table 4. AA 6351 – Material parameters.

E (GPA) y (MPa) a1 (GPA) a2 b1 b2 v1 (MPa) v2 k0 (s1) So (MPa)

70.00 264.00 29.28 0.00 328.00 0.00 0.00 0.00 0.01 15,000

Figure 11. AA 6351 – Lifetime prediction (in cycles) for different elongation amplitudes and different parameters
ðSo ,kÞ.

k0 ¼ 0:01 s1 and So ¼ 10, 000 MPa. Figure 12 shows the ðF=Ao Þ  ð=Lo Þ curves (first cycle) for two
different elongation amplitudes. Some cycles after stabilization, fatigue damage becomes important
and the material has no longer a periodic response. The model allows describing the cyclic softening
induced by the fatigue damage as it can be verified in Figure 13. Closure effects are important, but
can be reasonably neglected in this one-dimensional model, since they are more important only at
the end of the fatigue life. It is necessary to observe that most of the fatigue models try to avoid or to
circumvent this issue. The closure effects could be included in the model as in Costa Mattos et al.
(1992). However, it would be necessary to include additional definitions (and equations) in the paper
and the predicted lifetimes would not be necessarily so different from the engineering point of view.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 19

Figure 12. AA 6351 – Stabilised loop for two different elongation amplitudes.

Figure 13. AA 6351 – Curve ðF=Ao Þ  ð=Lo Þ in a fatigue test with ðL=Lo Þ ¼ 0:014.

One of the main goals of this paper is to show that most of the parameters necessary to the use of
this kind of model can be identified experimentally in simple tests, that the results are quite reason-
able from the physical point of view, and that the simulation (in the none-dimensional case) can be
performed using an ordinary personal computer.
In Figure 14, the axial displacement distribution along the useful length of the specimen for
different time instants can be observed. While damage is negligible, the displacement distribution
is linear. Close to the rupture, it can be verified that each half of the specimen tends to a rigid body
motion – uðz, tÞ ! 0 if 0  z 5 Lo =2 and uðz, tÞ ! ðtÞ ¼ uðz ¼ Lo , tÞ if ðLo =20Þ 5 z  Lo .
The damage distribution along the length at different instants is shown in Figure 15, and in
Figure 16 the evolution of the damage variable D at the middle of the specimen in a fatigue test
with ðL=Lo Þ ¼ 0:014 can be observed. Figure 15, together with Figure 14, shows that the one-
dimensional model allows a reasonably adequate description of the strain localization phenomenon
(a non-homogeneous deformation considering a constant axial force). Within the range of loading
presented in Figure 11, the material presents a cyclic softening due to the damage evolution after
stabilization but not a strictly softening response. In order to show the model’s ability to describe
softening behaviour, a monotonic tensile test under prescribed elongation is presented in Figure 17.
The simulation of a typical tensile test is in good agreement with experimental data. The proposed
theory allows a non-monotone (‘‘softening’’) behaviour of the external force necessary to apply a
given elongation , but the parameter So must be changed, since it is no longer related to fatigue
damage (but ductile damage). For the simulation of tensile tests, it was considered So ¼ 65 MPa and
the bar was divided in 20 sections ½z ¼ ði  1ÞLo ; i ¼ 1, 21. The distribution of the damage variable

Downloaded from ijd.sagepub.com by guest on September 19, 2016


20 International Journal of Damage Mechanics 0(0)

Figure 14. AA 6351 – Evolution of the axial displacement along the length in a fatigue test with ðL=Lo Þ ¼ 0:014.

Figure 15. AA 6351 – Evolution of the damage distribution along the length in a fatigue test with ðL=Lo Þ ¼ 0:014.

Figure 16. AA 6351 – Evolution of the damage variable D at section z ¼ Lo =2 in a fatigue test with ðL=Lo Þ ¼ 0:014.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 21

Figure 17. AA 6351 – Curve    in a tensile test together with the distribution of the damage variable along the
length.

along the length is also presented and it can be verified that it tends to localize at the middle of the
specimen.

Conclusions
The present paper discusses the possibility of modelling uniaxial low-cycle fatigue tests using a
special kind of gradient-enhanced continuum damage theory. In general, such kind of non-local
damage model is considered too complex to be used by engineers and designers and the aim of the
paper is to show that a simple but adequate description can be performed using such kind of
phenomenological approach.
The proposed model deals with regular cycles and with any other arbitrary load-path without the
need of cycle counting strategies. It enables a convenient macroscopic description of the inelastic
cyclic behavior, accounting for the strain-softening and strain localization induced by the damage
process as well as for the nonlinear damage accumulation and size dependency. A simple procedure,
requiring only one fatigue test with prescribed strain amplitude, is also proposed to perform the
experimental identification of all material constants that appear in the theory. The adequacy of the
model predictions is checked through simulations of low-cycle fatigue tests in AA 2011 and AA 6351
specimens. The results are in good agreement with experimental data; they show that the model is
physically coherent and and that it is interesting to pursue the study of plastic oligocyclic fatigue in
metallic materials using this approach. Of course, an intensive experimental program is still neces-
sary in order to fully validate the proposed model, since the scatter of results is important in fatigue
analysis. The present paper is a first step towards a simple but consistent modeling of fatigue damage
accumulation in elasto-plastic structures.

Declaration of Conflicting Interests


The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or pub-
lication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publica-
tion of this article: The author thanks the Research Foundation of the State of Rio de Janeiro (FAPERJ) and
the Brazilian National Council for Scientific and Technological Development (CNPq) for supporting part of
the work presented here.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


22 International Journal of Damage Mechanics 0(0)

References
Alessi R, Marigo JJ and Vidoli S (2015) Gradient damage models coupled with plasticity: Variational formu-
lation and main properties. Mechanics of Materials 80: 351–367.
Ames WF (1969) Numerical Methods for Partial Differential Equations. New York, USA: Academic Press
Incorporation.
Anes V, Reis L, Li B, et al. (2014) New cycle counting method for multiaxial fatigue. International Journal of
Fatigue 67: 78–94.
ASTM E606/E606M-12 (2014) Standard test method for strain-controlled fatigue testing. ASTM Book of
Standards, Volume: 03.01.
Bisping JR, Peterwerth B, Bleicher C, et al. (2014) Fatigue life assessment for large components based on
rainFow counted local strains using the damage domain. International Journal of Fatigue 68: 150–158.
Bonora N and Newaz GM (1998) Low cycle fatigue life estimation for ductile metals using a non linear
continuum damage mechanics model. International Journal of Solids and Structures 35(6): 1881–1894.
Chimisso FE and Costa Mattos H (1994) Modeling the softening behavior of damageable elasto-plastic bars.
In: Aliabadi MH, Carpinteri A, Kaliszky S, et al. (eds) Localized Damage III. Computer Aided Assessment
and Control. Southampton, UK: Computational Mechanics Publications, pp. 297–304.
Chorin AJ, Hughes TJR, McCracken MF, et al. (1978) Product formulas and numerical algorithms.
Communications on Pure and Applied Mathematics 31(2): 205–256.
Ciarlet PG and Lions JL (1990) Handbook of Numerical Analysis. Vol. I, Amsterdam, Netherlands: North-
Holland Elsevier Science.
Coffin Jr LF (1954) A study of the effects of cyclic thermal stresses on a ductile metal. Transactions of the
American Society of Mechanical Engineers 76: 931–950.
Costa Mattos HS and Chimisso FEG (1997) Necking of elasto-plastic rods under tension. International Journal
of Non-linear Mechanics 32(6): 1077–1086.
Costa Mattos HS and Sampaio R (1995) Analysis of the fracture of brittle elastic materials using a continuum
damage model. Structural Engineering and Mechanics 3(5): 411–427.
Costa Mattos HS, Fremond M and Mamiya EM (1992) A simple model of the mechanical behavior of ceramic-
like materials. International Journal of Solids and Structures 24: 3185–3200.
Costa Mattos HS and Soares Filho PF (2010) Approximate local elasto-plastic solution for notched plates
undergoing cyclic tensile loading. Materials & Design 31(9): 4336–4347.
Cowin SC and Nunziato JW (1983) Linear materials with voids. Journal of Elasticity 13: 125–147.
da Costa Mattos HS, Bastos IN and Gomes JACP (2008) A simple model for slow strain rate and constant load
corrosion tests of austenitic stainless steel in acid aqueous solution containing sodium chloride. Corrosion
Science 50: 2858–2866.
da Costa Mattos HS, Bastos IN and Gomes JACP (2014) A thermodynamically consistent modelling of stress
corrosion tests in elasto-viscoplastic materials. Corrosion Science 80(2014): 143–153.
da Costa Mattos HS and Chimisso FEG (2011) Modelling creep tests in HMPE fibres used in ultra-deep-sea
mooring ropes. International Journal of Solids and Structures 48(1): 144–152.
da Costa Mattos HS, Monteiro AH and Palazzetti R (2012) Failure analysis of adhesively bonded joints in
composite materials. Materials & Design 33: 242–247.
da Costa Mattos H, Pires-Domingues SM and Rochinha FA (2009) Structural failure prediction of quasi-brittle
structures: Modeling and simulation. Computational Material Science 46: 407–417.
Fatemi A and Yang L (1998) Cumulative fatigue damage and life prediction theories: A survey of the state of
the art for homogeneous materials. International Journal of Fatigue 20(1): 9–34.
Forest S (2009) Micromorphic approach for gradient elasticity, viscoplasticity and damage. ASCE Journal of
Engineering Mechanics 135: 177–131.
Frémond M and Nedjar B (1996) Damage, gradient of damage and principle of virtual power. International
Journal of Solids Structures 33(8): 1083–1103.
Jinescu VV (2012) Cumulation of effects in calculating the deterioration of fatigue loaded structures.
International Journal of Damage Mechanics 21: 671–695.

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 23
Kwofie S and Rahbar N (2013) A fatigue driving stress approach to damage and life prediction under variable
amplitude loading. International Journal of Damage Mechanics 22: 393–404.
Lemaitre J (1992) A Course on Damage Mechanics. Berlin, Germany: Springer-Verlag.
Lemaitre J and Chaboche JL (1994) Mechanics of Solid Materials. Cambridge: UK: Cambridge University
Press.
Lemaitre J and Desmorat R (2005) Engineering Damage Mechanics: Ductile, Creep, Fatigue and Brittle Failures.
Amsterdam: Springer.
Lv Z, Huang H, Zhu S, et al. (2015) A modified nonlinear fatigue damage accumulation model. International
Journal of Damage Mechanics 24: 168–181.
Manson SS (1954) Behavior of materials under conditions of thermal stress. Technical Report National Advisory
Committee for Aeronautics-TR-1170. Cleveland, OH: Lewis Flight Propulsion Lab.
Manson SS and Halford GR (1986) Re-examination of cumulative fatigue damage analysis – An engineering
perspective. Engineering Fracture Mechanics 25(5/6): 539–571.
Marchuck (1975) GI Methods of Numerical Analysis. Berlin, Germany: Springer-Verlag.
Martinez X, Oller S, Barbu LG, et al. (2015) Analysis of ultra low cycle fatigue problems with the Barcelona
plastic damage model and a new isotropic hardening law. International Journal of Fatigue 73: 132–142.
Miner MA (1945) Cumulative damage in fatigue. Journal of Applied Mechanics 67A: 159–164.
Murakami S (2012) Continuum Damage Mechanics: A Continuum Mechanics Approach to the Analysis of
Damage and Fracture. Dordrecht: Springer-Verlag.
Nedjar B (2001) Elastoplastic-damage modeling including the gradient of damage: Formulation and compu-
tational aspects. International Journal of Solids and Structures 38: 5421–5451.
Ortiz M, Pinski PM and Taylor RL (1983) Operator split methods for the numerical solution of the elasto-
plastic dynamic problem. Computational Methods in Applied Mechanics and Engineering 39: 137–157.
Saanouni K (2012) Damage Mechanics in Metal Forming: Advanced Modelling and Numerical Simulation.
Hoboken, NJ: John Wiley & Sons, Inc.
Saanouni K and Hamed M (2013) Micromorphic approach for finite gradient-elastoplasticity fully coupled
with ductile damage: Formulation and computational aspects. International Journal of Solids Structures
50(14–15): 2289–2309.
Seweryn A, Buczynski A and Szusta J (2008) Damage accumulation model for low cycle fatigue. International
Journal of Fatigue 30: 756–765.
Simo JC and Miehe C (1992) Associative coupled thermoplasticity at finite strains: Formulation, numerical
analysis and implementation. Computational Methods in Applied Mechanics and Engineering 98: 41–104.
Steglich D, Pirondi A, Bonora N, et al. (2005) Micromechanical modeling of cyclic plasticity incorporating
damage. International Journal of Solids and Structures 42: 337–351.
Sun Q, Dui H-N and Fan X-L (2014) A statistically consistent fatigue damage model based on Miner’s rule.
International Journal of Fatigue 69: 16–21.
Zuo F, Huang H, Zhu S, et al. (2015) Fatigue life prediction under variable amplitude loading using a non-
linear damage accumulation model. International Journal of Damage Mechanics 24(5): 767–784.

Appendix 1. A general thermodynamically consistent modeling of the


oligocyclic fatigue damage
The one-dimensional model proposed in the previous sections is a preliminary step towards a non-
local three-dimensional continuum damage model developed within a thermodynamic context. In
the case of materials with softening behaviour, the classical local damage models are usually ill-
posed and size dependent. The main feature of the gradient-enhanced damage models is that a
balance equation is adopted for the damage variable instead of an evolution law. The presence of
the gradient of the damage variable in the governing equation for the damage variable regularizes
the problem, preventing the occurrence of discontinuous damage gradients and, consequently, the

Downloaded from ijd.sagepub.com by guest on September 19, 2016


24 International Journal of Damage Mechanics 0(0)

occurrence of discontinuous displacement gradients. This fact allows an adequate simulation of


severe local deformations without the numerical difficulties of mesh dependence that arise in
others continuum damage models. This kind of approach has been adopted in the last decades by
different groups and very promising results were obtained up to now (Bonora and Newaz, 1998;
Chimisso and Costa Mattos, 1994; Costa Mattos and Sampaio, 1995; Costa Mattos et al., 1992; da
Costa Mattos et al., 2009; Nedjar, 2001; Steglich et al., 2005).

Preliminary definitions and summary of the basic balance equations


In this section, the basic local equations that govern the evolution of such kind of continuum are
summarized. It is not the goal of the present paper to perform a deep analysis of the basic principles
that govern the damageable continuum. The proposed principles may be regarded as a special case
of the theories of continuum with microstructure and can be found in Costa Mattos and Sampaio
(1995). In particular, these governing principles are very close to those proposed in the theory of
elastic materials with voids (Cowin and Nunziato, 1983). Nevertheless, the definition and the phy-
sical interpretation of the additional kinematic variable and also the proposed constitutive equations
make both theories very different. In the theory of elastic materials with voids the additional variable
is related with the change in solid volume fraction. The present theory assumes that the damage is
related with micro-cracks and not with micro-voids, and hence the damaged material is not con-
sidered a porous medium and the damage variable is not directly related with a volume change. The
differences between these geometric discontinuities at the microscopic level are more evident when
large deformations, heat transfer or wave propagation are considered.
A continuous damageable body is defined as a set of material points B which occupies a region 
with boundary  of the Euclidean space at the reference configuration. For the sake of simplicity,
the hypothesis of small deformation and isothermal processes will be assumed throughout this
section. In this theory, besides the classical variables that characterize the kinematics of a continuum
medium (displacements, velocities and accelerations of material points), an additional variable
D 2 ½0, 1 is introduced. A point, in such continuum theory, is representative of a given ‘‘volume
element’’ of the real material and it is endowed with a microstructure that accounts for the kinetic
energy and internal power of the microscopic motions associated to the microscopic geometric
discontinuities. This variable is related with the microscopic motions and can be interpreted as a
measure of the damage state of the ‘‘volume element’’. If D ¼ 0, all the links between material points
are preserved and the initial material properties are also preserved. If D ¼ 1, a local rupture is
considered. Since the degradation is an irreversible phenomenon, the variation rate @D @t must be
negative or equal to zero.
The necessary mechanic field variables are introduced as primitive quantities. Specifically, there
exists a stress tensor , a body force b and an external force f . In addition, microscopic forces
associated to the damage variable are introduced in the theory: a microscopic stress vector H , a
microscopic internal force M and a microscopic external force q. The microscopic external force q
must be introduced in the theory in order to take into account the non-mechanical actions (chemical
or electromagnetic) that affect the damage state of the material even if there is no mechanical
deformation. Generally, this additional term is not necessary in low-cycle fatigue analysis and the
microscopic external force can be neglected (q ¼ 0). However, it can be important, in some practical
engineering situations, such as the case of stress-corrosion damage. The fatigue of austenitic stainless
steels in acid media is a good example (see da Costa Mattos et al., 2014).

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 25
The evolution of the damageable body is supposed to be governed by the following balance
relations (and boundary conditions) in the local form (see Costa Mattos and Sampaio, 1995;
Cowin and Nunziato, 1983)
( 
divðÞ þ b ¼ 0 in  with  ¼  T in  M þ divðHÞ ¼ 0 in 
ð33Þ
 n ¼ f on  H n ¼ q on 

The first set of equations in (33) are local versions of the balance of linear momentum of the
balance of angular momentum. The second set of equations in (33) is a local version of the balance
of the momentum of the microscopic forces. The above equations can also be alternatively obtained
using the Virtual Power Principle, similarly as in Frémond and Nedjar (1996) or da Costa Mattos
et al. (2009). ‘‘Damage’’ in continuum mechanics usually refers to the loss of mechanical stiffness,
deEned at the scale of a representative elementary volume. In general, damage is related to the
nucleation and propagation of microscopic cracks in the bulk of this volume element. This addi-
tional balance equation does not appear in classical continuum mechanics and accounts for the
microscopic stresses (and microscopic movements) within a volume element. n is the unit outward
normal to the surface . From now on, it will be assumed that there are no chemical or electro-
magnetic external actions on the body and hence, q ¼ 0.

Constitutive equations
The basic governing equations presented in the previous section are valid for any kind of damage-
able body. In this section, we restrict ourselves to elasto-plastic behavior and isothermal processes.
The theory can be summarized in the four steps below.

State variables. Under the hypothesis of small deformations and isothermal processes, the local
state of a elasto-plastic material is supposed to be a function of the total strain ", of the plastic strain
"p , of the damage variable D, of its gradient gradðDÞ and also of a scalar variable p associated to the
isotropic hardening and of a second order tensor variable c associated with the kinematic hardening.

Free energy – State laws. Following the classical assumption of thermodynamic of irreversible
processes, the free energy is supposed to be a function of the state variables. Thus, the following
expression is proposed for the free energy

ð", "p , c, p, D, gradðDÞÞ ¼¼ ð1  DÞ½ e ð"  "p Þ þ p ð pÞ þ c ðcÞ þ k2 ðgradðDÞ gradðDÞÞ ð34Þ

with

E h  i
e ¼ ½trð"  "p Þ2 þ ð"  "p Þ : ð"  "p Þ ð35Þ
2ð1 þ Þ 1  2

p ¼ v1 ½ p þ expðv2 pÞ þ py ð36Þ


1 _
c ¼ a ðc : cÞ ð37Þ
2

Downloaded from ijd.sagepub.com by guest on September 19, 2016


26 International Journal of Damage Mechanics 0(0)
_
where E is the Young modulus,  is the Poisson’s ratio, y is the yielding stress. k, v1 , v2 and a are
positive constants and "e ¼ ð"  "p Þ is the elastic strain tensor. The term k2 ðgradðDÞ gradðDÞÞ is
considered so as to give to D a diffusive behavior, thus smoothing the field D in .
The so-called thermodynamic forces ð, X, Y, G, HÞ, related to the state variables
ð", "p , c, p, D, gradðDÞÞ, are defined from the free energy by the state laws

@ ð1  DÞE h  i
¼ ¼ trð"  "p Þ1 þ ð"  "p Þ ð38Þ
@" 1 þ  1  2

@ _
X¼ ¼ ð1  DÞð a cÞ ð39Þ
@c

@
Y¼ ¼ ð1  DÞ½v1 ð1  expðv2 pÞ þ y  ð40Þ
@p

@
G¼ ¼ ð e þ p þ cÞ ð41Þ
@D
@
H¼ ¼ kgradðDÞ ð42Þ
@gradðDÞ

To complete the constitutive equations, additional information about the dissipative behavior
must be given. This information can be obtained from a plastic potential  and are called the
evolution laws.

Plastic potential – Evolution laws. The potential  is presumed to have the form
 ¼ Jð  XÞ  Y þ g ðX, G; ", "p , p, c, DÞ  0 ð43Þ

with
 1=2
3
Jð  XÞ ¼ ð  XÞdev : ð  XÞdev ð44Þ
2
_
b ab G2 1
g¼ _
ðX : XÞ  ðð1  DÞ2 c : cÞ þ  ð e þ p þ cÞ
2 ð45Þ
2a 2 2So 2So

ð  XÞdev is the deviatoric part of ð  XÞ and Jð  XÞ is the Von Mises equivalent stress. b is a
positive parameter. The following evolution laws are postulated

@"p @ 3
ð  XÞdev
¼ _ ¼ _ 2 ð46Þ
@t @ Jð  XÞ

@c @ @"p b _
¼ _ ¼  X ð47Þ
@t @X @t a

Downloaded from ijd.sagepub.com by guest on September 19, 2016


da Costa Mattos 27
@p @
¼ _ ¼ _ ð48Þ
@t @Y
8 
_
@D < M þ _ @ ¼ M þ ð e þ p þ cÞ
, if 0  D 5 1
¼ @G So ð49Þ
@t :
0, if D ¼ 1

_  0,   0, _ ¼0
 ð50Þ

_ is the Lagrange multiplier associated with the condition   0. From equation (50) it is easy
@"p
to@cverify
that, if  5 0 then _ ¼ 0 and, from equations (46) to (50), it follows that @p @t ¼ 0, @t ¼ @t ¼ 0.
Consequently, the material will behave elastically. M is the microscopic internal force associated
with D (equation (33)). hxi ¼ maxf0, xg. Equation (49) assures that @D @t  0 in all processes.
It can be proved that the state laws (38) to (42) and the evolution laws (46) to (50) define a
complete set of thermodynamically admissible constitutive equations.
Introducing equations (42) and (49) in the second balance equation in (33), the following balance
equation is obtained for 0  D  1

@D ð e þ p þ cÞ
¼ kr2 D þ ð51Þ
@t So

where the symbol r2 denotes the Laplacian operator with respect to the present position:
r2 D ¼ divðgradðDÞÞ. Equation (51) that governs the evolution of D is similar to the heat equation
with a non-linear heat source term. k plays a role analogous to the thermal conductivity. The smaller
is k the more localized is the damage (and, consequently, the deformations " and "p ). The variable k
is related to the spatial distribution of D and So is related to the evolution of D at a given point: the
bigger is S0 , the bigger must be the energy ð e þ p þ c Þ, in order to obtain a given rate D. _
Closure effects are important, but can be reasonably neglected in the one-dimensional case, since
they are more important only at the end of the fatigue life. It is necessary to observe that most of the
fatigue models try to avoid or to circumvent this issue. The unilateral nature of damage can be
included in the model through the free energy as in Costa Mattos et al. (1992). However, it would be
necessary to include additional definitions (and equations) in this Appendix 1.
Equations (1) to (10) can be obtained as a particular case of this general theory assuming a one-
_
dimensional state of stress and taking a ¼ 23 a. All other material constants are similar in both
theories. The material constants identified in the one-dimensional model can be used as a starting
guess in iterative techniques to identify the constants that appear in the general theory (inverse
problems using the finite element technique and considering structures with more complex geome-
tries). Due to the large dispersion of experimental results, in some cases, it may be not interesting to
expend a very long time trying to improve the identifications of So , k and its more practical to use the
proposed one-dimensional procedure as an approximate method to identify these constants. In the
one-dimensional context, the tensors  and " are supposed to be given by the following expressions
0 1 0 1
ðtÞ ¼ ðFðtÞ=Ao Þ "ðx, tÞ ¼ ððtÞ=Lo Þ
B C B C
¼@ 0 A; "¼@ 
"ðx, tÞ A ð52Þ
0 
"ðx, tÞ

Downloaded from ijd.sagepub.com by guest on September 19, 2016


28 International Journal of Damage Mechanics 0(0)

where FðtÞ is the external force and ðtÞ the prescribed elongation. 
is a variable such that
0 5 
5 0:5. Considering the material isotropy and that trð"p Þ ¼ trðXÞ ¼ trðcÞ ¼ 0, the variables
ðÞdev , "p , X and c can be expressed in the following form
0 1 0 1
2 2
3 ðtÞ 3 Xðx, tÞ
B C B C
ðÞdev ¼ B
@  ðtÞ
3
C;
A X¼B
@  Xðx,
3
tÞ C
A ð53Þ
 ðtÞ
3  Xðx,
3

0 1 0 1
"p ðx, tÞ cðx, tÞ
B "
p
ðx, tÞ C B  cðx,2 tÞ C
"p ¼ @ 2 A; c¼@ A ð54Þ
p
 " ðx,
2

 cðx,2 tÞ

It is simple to verify that the components of the tensor variables  , "p , X and c are not
dev
independent, in this case. From the definition of the von Mises equivalent stress, it is possible to
conclude that Jð  XÞ ¼ jXj. Finally, neglecting the microscopic external forces and the external
body force and using the above results with the local balance equation 33 and constitutive equations
(38) to (42), (46) to (50), it is possible to obtain the following set of equations

@
¼0 ð55Þ
@z
@u
¼" ð56Þ
@z

 ¼ ð1  DÞEð"  "p Þ ð57Þ

X ¼ ð1  DÞac ð58Þ

Y ¼ ð1  DÞ½v1 ð1  expðv2 pÞ þ y  ð59Þ


     
@D _ ev2 p @2 D
¼ Eð"  "p Þ2 þ ac2 þ 2 v1 p þ þ py þ k 2 ð60Þ
@t 2So v2 @z

@"p 1, if ð  XÞ  0
¼  Sg; Sg ¼ ð61Þ
@t 1, otherwise

@p
¼ _ ð62Þ
@t

@c @"p b _
¼  X ð63Þ
@t @t a

_  0,  ¼ j  Xj  Y  0, _ ¼0
 ð64Þ

Downloaded from ijd.sagepub.com by guest on September 19, 2016

Das könnte Ihnen auch gefallen