Professional Documents
Culture Documents
The Primer
by
Kenneth Runesson
There seems to be an ever increasing demand in engineering practice for more realistic
models as applied to metals as well as composites, ceramics, polymers and geological
materials (such as soil and rock). Consequently, a vast amount of literature is available
on the subject of nonlinear constitutive modeling, with strong emphasis on plastic-
ity and damage. Such modeling efforts are parallelled by the development of numerical
algorithms for use in Finite Element environment. For example, implicit (rather than ex-
plicit) integration techniques for plasticity problems are now predominant in commercial
FE-codes.
I am indebted to a great number of people who have contributed to the present volume:
Mr. M. Enelund, Mr. L. Jacobsson, Mr. M. Johansson, Mr. L. Mahler and Mr. T. Svedberg,
who are all graduate students at Chalmers Solid Mechanics, have read (parts of) the
manuscript and struggled with the numerical examples. Mr. T. Ernby prepared some of
the difficult figures. Ms. C. Johnsson, who is a graduate student in ancient Greek history
at Goteborg University, quickly became an expert in handling equations in LATEX. The
contribution of each one is gratefully acknowledged.
Goteborg in March 1996.
Kenneth Runesson
2nd revised edition:
I am grateful to Mr. Lars Jacobson and Mr. Magnus Johansson (in particular) for their
help in revising parts of the manuscript.
Goteborg in March 1997.
Kenneth Runesson
3rd revised edition:
Ms. EvaMari Runesson, who is a student in English at the University of Gothenburg (and
also happens to be my daughter) did an excellent job in mastering LATEXfor this edition.
iv
3 VISCOELASTICITY 29
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Prototype model: The Maxwell rheological model . . . . . . . . . . . . . . 30
3.2.1 Thermodynamic basis Constitutive relation . . . . . . . . . . . . 30
3.2.2 Prescribed constant stress (pure creep) . . . . . . . . . . . . . . . . 32
3.2.3 Prescribed constant strain (pure relaxation) . . . . . . . . . . . . . 32
3.3 Linear viscoelasticity Constitutive modeling . . . . . . . . . . . . . . . . 33
3.3.1 General characteristics . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.2 Laplace-Carson transform . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.3 Linear Standard Model (Generalized Maxwell Model) . . . . . . . . 37
3.3.4 Backward Euler method for linear standard model . . . . . . . . . . 40
3.4 Linear viscoelasticity Structural analysis . . . . . . . . . . . . . . . . . . 42
3.4.1 Structural behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4.2 Solution strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.3 Analysis of truss Elastic analogy . . . . . . . . . . . . . . . . . . 44
4 PLASTICITY 77
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2 Prototype rheological model for perfectly plastic behavior . . . . . . . . . . 79
4.2.1 Thermodynamic basis Yield criterion . . . . . . . . . . . . . . . 79
4.2.2 Plastic flow rule and elastic-plastic tangent relation . . . . . . . . . 80
4.4 Model for cyclic loading Mixed isotropic and kinematic hardening . . . 86
5 VISCOPLASTICITY 109
CHARACTERISTICS OF
ENGINEERING MATERIALS
AND CONSTITUTIVE
MODELING
In this chapter we give a brief introduction to the particular field within applied solid me-
chanics that deals with the establishment of constitutive models for engineering materials.
Some generally accepted constraints that must be imposed on constitutive models are dis-
cussed. Commonly occurring test conditions for obtaining results towards calibration and
validation are discussed briefly. Finally, the typical material (stress-strain) behavior of
the most important engineering materials (metals and alloys, cementitious composites,
granular materials) under various loading conditions is reviewed.
Common to all mechanical analysis of engineering materials and their behavior in struc-
tural components is the need for constitutive models that link the states of stress and
strain. From a mathematical viewpoint, the constitutive equations (that define the con-
stitutive model) are complementary equations to the balance and kinematic equations.
Figure 1.1: Typical microstructure of (a) Steel (perlitic grain structure, eutectoid compo-
sition), (b) Concrete, (c) Wood.
1
These are the necessary and sufficient conditions for any hyperstatic (statically indeterminate) struc-
ture, whereas it is not necessary to know the constitutive response to calculate the stresses in an isostatic
(statically determinate) structure.
Prediction of strain localization in shear bands and incipient material failure: Soft-
ening plasticity or damage coupled to plastic deformation
Most of the listed phenomena will be considered in some detail in this text. Clearly, the
task of the engineer is to choose a model that is sufficiently accurate, yet not unnecessarily
complex and computationally expensive. The questions that should be asked in regard
to the choice of a certain model are as follows:
Does the model produce sufficiently accurate predictions for the given purpose?
A list of constraints that must be placed on constitutive relations, that represent the
mechanical behavior of a continuous medium, is given below. Virtually all of these re-
quirements are intuitively obvious, although it is not trivial to express them properly in
mathematical language. Moreover, some constraints are important only in conjunction
with large deformations, say, at the modeling of material forming.
Constitutive relations, as well as other relations between physical entities, should not be
affected by arbitrary coordinate transformations.
This requirement is satisfied if proper tensorial relations are established.
The stress in a given body is determined entirely by the history of the motion of the body,
i.e. it is not affected by the future events.
This requirement is always satisfied if intrinsically time-dependent relations are estab-
lished with time as (one of) the independent coordinate(s). It may be violated if relations
between Laplace or Fourier transformed variables are set up directly. For example, care
must be taken when internal damping relations (expressing energy dissipation) are pro-
posed in the frequency domain, as discussed by Crandall (1970).
Constitutive relations must not be affected by arbitrary Rigid Body Motion (RBM) that
is superposed on the actual motion.
This requirement is most easily satisfied by employing objective tensor fields as the con-
stitutive variables. In particular, it is important to note that the ordinary time derivative
of common variables (stress, strain) is not objective. For example, the time rate of the
(Cauchy) stress tensor is not zero at RBM, even if the material does not feel any change
of stress, i.e. the stress components with respect to a corotating coordinate system do
not change. However, the non-zero time rate is merely a consequence of the rotation.
As a consequence, this time rate is not permissible in constitutive relations, at least not
for large material rotation. In small strain theory, which employs linear kinematics, the
requirement of objectivity can be ignored.
Response functions are unaffected by certain rotations of the chosen reference configu-
ration due to material symmetry. The most important special case is complete material
isotropy, which means that the response is equal in all directions or, more precisely, for
all possible spatial rotations of the chosen reference configuration. The precise definition
of symmetry is expressed mathematically in terms of the appropriate symmetry group (of
orthogonal transformations).
The 2nd law of thermodynamics states that the production of internal entropy, or rate of
material disorder, must be non-negative. This statement is equivalent to the statement
that dissipation of energy is never negative.
This law, whose mathematical formulation is the Clausius-Duhem Inequality, is discussed
in Chapter 3. It is a cornerstone for the further developments in the present text. In
particular, its automatic satisfaction is a key feature of standard dissipative materials,
which are considered in various contexts subsequently.
Phenomenological approach
The macroscopic model is established directly based on the observed characteristics from
elementary tests. The calibration is carried out mainly by comparison with experimental
results and/or with micromechanical predictions for well-defined boundary conditions on
the pertinent RVE, cf. the discussion above. Traditional models are sometimes simple
enough to admit the identification of the material parameter values one by one from
well-defined elementary experiments. The obvious example is the observation of the yield
stress of mild steel from a tensile test. However, the general approach is to optimize the
predictive capability of the model in the calibration procedure. The objective function
to be minimized is a suitable measure (norm) of the difference between the predicted
response and the experimentally obtained data.
The arguments of the constitutive functions are observable variables (like stress, strain and
temperature) in addition to a sufficient number of nonobservable, or internal, variables
that represent the microstructural changes.
Statistical approach
Statistical models for describing material behavior are the least fundamental, in the
sense that they are normally established as response functions for specific loading and
environmental conditions. A variety of distributions can be used for describing the scatter
in strength data, whereby Weibulls statistical theory is quite often used.
Two principally different views can be distinguished with respect to the analysis of ma-
terial failure. From a classical standpoint, these approaches are related to the fact that
a material may behave in a ductile or brittle fashion, depending on material composition,
aging, temperature, etc.
The view of continuum damage mechanics is that the failure process starts with a gradual
deterioration of a continuously deforming material. After considerable inelastic deforma-
tion, due to the material ductility, the stress drops quite dramatically (in a displacement
controlled test) and deformations localize in a narrow zone (or band). This stage is defined
as the onset of fracture; cf. Figure 1.2. In many cases the localization is quite extreme
in the sense that a single macroscopic (discrete) crack starts to develop. Stresses can be
transferred across the crack until it is fully opened.
microcracks
=damage neck
develops
localization
zone=neck
brittle ductile
(a) (b)
Figure 1.2: Damage process (a) Localization (necking) in a bar of ductile material, (b)
Stress vs. strain characteristics.
The view of fracture mechanics is that a macroscopic crack (or flaw) has already occured,
and the main task is to determine whether the crack will propagate or not. A crack
that propagates only when the externally applied load is increased is termed stable. No
consideration is then given to the process leading to the (preexisting) fully open crack.
The analysis of crack stability is usually based on the assumption that the behavior close
to the crack tip is linear elastic (Linear Fracture Mechanics), such that the stress field
singularity at the crack tip is determined from linear elasticity, cf. Figure 1.3. The
simplest crack stability criterion is the (empirical) Griffith criterion, by which the crack is
deemed stable if the pertinent stress intensity factor, that depends on the applied loading,
does not exceed a critical value. This concept can be extended to cyclic loading, e.g. in
the shape of a threshold level of stress in Paris law.
Phenomenological constitutive laws are calibrated with the aid of experimental data that
are obtained from well-defined laboratory tests. The idea is to design the test in such a way
that the specimen is subjected to homogeneous states of stress, strain and temperature.
A few common test conditions in practice are listed below:
u
macroscopic
crack area under u curve =
L singular released fracture energy
stress field
at crack tip brittle
= (locally) u
(a) (b)
Figure 1.3: Fracture process (a) Preexisting edge cracks, (b) Far-field stress vs. extension
characteristics. Note: = u/L is not well-defined as local measure of strain!
Uniaxial stress
A cylindrical bar is subjected to a state of uniform (axial) stress, which may be tensile or
compressive, as shown in Figure 1.4(a). The strain state is cylindrical, i.e. nonzero radial
and tangential normal strains normally appear. Either the axial stress or the axial strain
is controlled. This elementary test condition is common for most materials, at least those
possessing cohesion. By definition, cohesion materials have shear strength that prevails
when the mean (normal) stress is zero. Frictional materials, whose shear strength vanishes
when the mean stress is zero, can not be tested under the uniaxial stress condition without
precompaction.
The conventional way of applying normal stresses, combined with shear stress, is to subject
a circular thin-walled tube to axial load, internal or external pressure, together with a
torsional moment, as shown in Figure 1.4(b). Since the wall thickness is small, the radial
stress varies approximately linearly through the thickness, and at the midplane of the tube
wall a well-defined triaxial stress state is obtained. Moreover, when torsion is applied,
the principal axes rotate due to additional shear stress. In the case there is no applied
pressure, a state of plane stress is obtained. This type of test is common for metals, but
has also been used for concrete and highly cohesive soil (such as clay).
A cylindrical specimen is subjected to external radial pressure and axial compressive load,
as shown in Figure 1.4(c). This is a commonly used test condition for granular materials,
such as powder and soil, as well as for rock and concrete. Two usual test procedures are
denoted Conventional Triaxial Compression (CTC) and Conventional Triaxial Extension
(CTE). In the CTC-test, an isotropic state of stress is first applied during the socalled
consolidation phase. Then the radial (=circumferential) stress is held constant, while the
axial compressive loading is further increased. This compression may be either stress-
or strain-controlled. In the CTE-test, isotropic stress is first applied to consolidate the
sample in the same fashion as for the CTC-test. However, the axial stress is then kept
constant while the radial pressure is further increased.
In order to assess the principal difference between these two test conditions, we consider
the corresponding principal stresses i < 0 (compression negative), where 1 2 3 .
Since i = 0, the CTC-test is defined by 1 = 2 > 3 and the axial stress is 3 , which is
the numerically largest principal stress. The CTE-test, on the other hand, is defined by
1 > 2 = 3 and the axial stress is now 1 . It is common to use these test results towards
the evaluation of a failure (or yield) criterion of the Mohr-Coulomb type, cf. Chapter 10.
Principal stresses can be applied independently in the cubical cell apparatus, as illustrated
in Figure 1.4(d). In practice, this is a quite complex device that has gained widespread
use for soil, rock and concrete.
r r
(a) (b)
z
2
r
r = = p
(c) (d)
Figure 1.4: Stress and strain states in (a) Tensile test, (b) Normal load-torsion test of
thin-walled tube, (c) CTC- and CTE-tests, and (d) Cubical cell test.
The basic behavior of a ductile metal is obtained under monotonic loading. Plastic yielding
will occur approximately at the same magnitude of stress in tension as in compression since
plastic slip is determined by the critical resolved shear stress along potential slip planes
(Schmids law). Yielding is independent of the magnitude of the mean stress, which
defines an ideal cohesive material. The further increase of stress beyond yielding is known
as hardening. Figure 1.5(a) shows the typical stress-strain relation in uniaxial tension at
monotonic loading of a hot-worked steel. The characteristic strength parameters are the
yield stress y and the ultimate strength (peak stress) u . Figure 1.5(b) shows the typical
yield surface in biaxial stress (approximately elliptical in reality for a polycrystalline
metal).
2
u
y
y
y
1
y
y u y
(a) (b)
Figure 1.5: (a) Stress-strain relation in uniaxial tension showing yielding, hardening and
ductile fracture. (b) Yield surface in biaxial stress.
The picture is complemented by unloading, followed by reversed loading which gives rise
to hysteresis loops, as shown in Figure 1.6. After significant straining the average un-
loading modulus will decrease significantly, which may be interpreted as a sign of internal
degradation (damage) and that fracture is approaching.
A A A A
y
B
E E E < E
t
(a) (b)
After the saturation level has been reached, the creep rate is rather constant. As will be
discussed later, the Norton creep law is traditionally adopted in this stage.
A A
0 0
B B
t t
rupture
creep A
0 A
B recovery
0
t t
tR
I II III
B
(a) (b)
Another aspect of viscous properties, besides creep, is the rate-dependence that is exhibited
in the stress-strain curve for certain materials. This is manifested by higher stiffness and
strength for larger loading rate, especially due to impact loading. In accordance with the
situation at creep, the rate-effect is more pronounced at elevated temperature. The typical
result at monotonic loading under prescribed strain rate for a rate-sensitive material is
shown in Figure 1.8. In a real structure the strain rate may vary considerably from the
loaded region to other parts. Hence, it is important to model rate-effects in such a fashion
that the rate-independent situation is obtained merely as a special case.
> 0
= 0
Damping, in the sense that free vibrations of a structure will decay with time and even-
tually die out, can be explained as the result of energy dissipation in the material. If the
amount of damping is dependent on the frequency of the vibrations, then the damping is
of viscous character and can be modelled within the framework of viscoelasticity or vis-
coplasticity. If, on the other hand, the damping is independent on the frequency, then the
damping is commonly denoted as hysteretic and can be modelled within the framework
of rate-independent plasticity.
An alternative way of assessing damping and rate effects is to consider forced vibrations
due to a sustained harmonic load with given frequency. The structural response, in terms
of strain and stress, is then normally observed to be dependent on the frequency of the
exciting load, which points towards a rate-effect.
The usual way of testing cyclic and, eventually, fatigue behavior is to subject the speci-
men to a (slow) cyclic variation of stress or strain with constant amplitude. If the applied
load level is below the macroscopic yield stress, but above a certain threshold, the cyclic
response is in the elastic range and no macroscopic plastic deformation is observed. How-
ever, after many load cycles a reduction of the apparent elasticity modulus is noted. This
(MPa)
200
damage threshold
100
N
-0.2 0.2 102 2 4 6 8 105
If the applied load level is high enough, the macroscopic yield stress will be exceeded, and
plastic strains will develop in each cycle. Not unlike the characteristics of a creep test,
three different stages of the deformation process may be distinguished:
Saturation Stage
Consider the early stage of cyclic loading with constant amplitude. The response is then
characterized as either cyclic hardening or cyclic softening. The typical behavior of cyclic
hardening is shown in Figure 1.10(a) for given strain amplitude (strain control) and in
Figure 1.10(b) for given stress amplitude (stress control). Cyclic hardening means that
the stress amplitude will initially increase in a few cycles to an asymptotic level in a strain
controlled test, whereas the strain amplitude will decrease in a stress controlled test. The
complementary behavior in the case of (initial) cyclic softening is shown in Figure 1.11(a)
and Figure 1.11(b). Hence, cyclic softening means that the stress amplitude will initially
decrease to an asymptotic level in a strain controlled test, whereas the strain amplitude
will increase in a stress controlled test.
In both stress- and strain-controlled cyclic loading, the ideal situation is that the respec-
tive strain or stress amplitude will shake-down quite rapidly to a stabilized stress-strain
= constant 3
max 2
1
min 1
Stabilized 2
max = min m = 0
(a)
3 2 1
= constant
max
min
t
Stabilized
m 6= 0 1 2
(b)
Figure 1.10: Initial cyclic hardening as shown in (a) Strain control, (b) Stress control.
= constant 1
2
max
t
min
2
1
m 6= 0 Stabilized
(a)
= constant
max
1 2 3
2 1
min
Stabilized (b)
max = min m = 0
Figure 1.11: Initial cyclic softening as shown in (a) Strain control, and (b) Stress control.
hysteresis loop, as shown in Figure 1.12(a). This loop is symmetrical in tension and com-
pression in the ideal situation. In reality, the stabilized (shake-down) amplitudes on the
tension and compression sides may not be symmetrical, even if the applied cyclic action is
symmetrical, as shown in Figure 1.10 and Figure 1.11. Denoting the stabilized maximum
values by max and max , and the minimum values by min and min , we define the cyclic
mean stress and cyclic mean strain, respectively, as
1 1
m = [max + min ], m = [max + min ] (1.1)
2 2
where it is noted that algebraic values are used. We thus conclude that, in general at
the saturation level, m 6= 0 in the strain-controlled test, whereas m 6= 0 in the stress-
controlled test.
However, it is also possible that stabilization does not occur at all (or is very slow). For
prescribed constant stress amplitude, this lack of stabilization is evident as ever increas-
ing plastic strain, or ratchetting, which is shown in Figure 1.12(b). Such ratchetting can
be expected when m 6= 0, in particular.
1 2
1 2
r
Shakedown
Ratchetting strain
(a) (b)
After certain amount of plastic deformation has accumulated in the hysteretic loops after
saturation, damage starts to develop. This damage development will eventually, say
after 1,000-10,000 cycles, result in cyclic softening until failure occurs. Hence, LCF is
characterized by relatively small values of NR , which is the number of cycles to failure
(NR < 10, 000).
The characteristics of LCF are observed in the strain-controlled as well as in the stress-
controlled environment. The elastic unloading modulus is continually decreasing in such
a way that the hysteresis-loops become more and more flattened. As a result, the
stress amplitude gradually decreases in the strain-controlled test, as shown in Figure 1.13,
whereas the strain-amplitude grows in an uncontrolled fashion in the stress controlled test.
Moreover, further ratchetting may be obtained due to the fact that the rate of damage
development is smaller in compression than in tension (and it is assumed to be zero in
the figure).
stabilized
(saturation) (MPa)
300
200 damage threshold
100
N
0.2 0.2 102 500 1000 1500 2000
Creep-fatigue is obtained when the stress varies in a cyclic fashion with a predefined hold-
time within each cycle. The failure is caused by the combined action of creep deformation
and deterioration of the stiffness due to LCF. This phenomenon is of particular importance
at the design of jet engines and other gas turbines, which operate under high temperature.
Relaxation-fatigue is the counterpart of creep-fatigue when the strain is allowed to vary
in a cyclic fashion with predefined hold-time.
Sometimes, the notion thermal fatigue refers to the situation where the stress/strain vari-
ation is due to cyclic temperature change. The effect becomes more pronounced for high
degree of static indeterminacy (when stresses are larger). Clearly, it is not possible to
control the stress or strain amplitude when the temperature is varied. The most gen-
eral loading situation is denoted thermomechanical fatigue, in which case a component is
subjected to cyclic variation of the mechanical load as well as the temperature.
At monotonic loading, cementitious materials (such as concrete) show nearly linear elastic
response at small load levels. However, the type of failure is entirely different in tension
and compression. Tensile failure will occur in a quite brittle (quasi-brittle) manner at
the tensile strength, = tu , whereby a macroscopic crack starts to develop and is
fully open when the stress has dropped to zero. The corresponding post-peak stress-
strain relationship is not well-defined, cf. the discussion in Section 2.2. This response is
depicted in Figure 1.14(a). Compressive failure, on the other hand, will occur in a ductile
manner after the compressive strength, = cu , has been reached. The stress drop in
the post-peak regime represents gradual crushing of the microstructure. Typically, the
ratio tu /cu is of the order 0.1. Quite often the response in compression close to failure is
modelled as elastic-plastic, whereby the yield criterion is strongly mean-stress dependant.
In order to compensate for the low tensile strength cementitious materials must in practice
be reinforced by steel bars, glass-fiber bars or distributed ductile fibers.
A typical failure criterion is that of Mohr-Coulomb (which is discussed in further detail
in Chapter 10). This criterion is shown in Figure 1.14(b) for biaxial stress states.
Structural failure in massive concrete structures can be very dramatic due to the large
amount of elastic energy that is stored in a large volume at the point of cracking. An
example of a major disaster was the failure of the Sleipner oil platform outside Stavanger,
Norway in 19XX.
Remark: Mean stress dependent yielding and failure is typical for different granular and
particulate materials, e.g. soil and powders. Another example is (graphitic) grey-cast
cu cu
(a) (a)
Figure 1.14: (a) Stress-strain relation in uniaxial tension and compression. (b) Failure
surface according to Mohr-Coulomb for biaxial stress states (plane stress).
iron, for which the ratio of yield stress in tension and compression, ty /cy , is of the order
3. 2
Granular materials, such as soil and (ceramic and metal) powders, show frictional charac-
teristics. A purely frictional material, such as sand, gravel or fragmented rock (ballast),
can sustain shear only in the presence of compressive normal stress between the particles.
Moreover, in a purely frictional material the tensile strength is zero (tu = 0). Many fine-
grained materials, such as clay and powders that have been subjected to precompaction,
show combined frictional and cohesive characteristics, which means that the material can
sustain some shear stress even without any normal stress. In particular, this is the case for
clayey soils and for rock and concrete. Hence, frictional/cohesive features can be trans-
lated into mean-stress dependent failure criteria, cf. the discussion in Section 2.5, and
it can be concluded that soils, powders and concrete do, in fact, have much in common
when it comes to the modeling of failure characteristics.
The inelastic deformations of granular materials contain a volumetric component (con-
trary to the case for most metals). The deformation is dilatant at dense initial packing,
whereas it is contractant at loose initial packing. Dilatant behavior is associated with
softening response (negative hardening), whereas contractant response is accompanied
by hardening. In reality there may be a significant elastic-plastic coupling in the sense
that the inelastic volume change affects the elastic moduli. The mechanical response is
normally tested in a triaxial stress apparatus under cylindrical stress conditions, cf. the
CTC-and CTE-conditions discussed in Section 2.3.
Remark: In metals it necessary to account for evolving porosity close to failure, whereby
the yielding characteristics resemble those of a powder compact. 2
Natural fine-grained soils (in particular clay) show a more complex mechanical response
due to the presence of fluid (water and air) in the open pores. The resulting hydro-
mechanical interaction of such poro-mechanical materials introduces time-dependent de-
formation at constant applied load. Such a time-delayed deformation process is denoted
consolidation in soil mechanics (which must not be confused with creep due to viscous
character of the solid particles). Basically, consolidation is the process of squeezing a
sponge filled with water. Oil reservoirs constitute a complex geological system of solid,
liquid (oil/water) and gas in a mixture state.
In the extreme case the permeability is so small that virtually no seepeage of fluid can
take place in the pore system, which is termed undrained condition. In the special case of
water-saturated pores, such an undrained state corresponds to overall incompressibility
of the granular material.
In addition to consolidation, fine-grained soils show creep under constant loading. Such
creep, which is sometimes denoted secondary consolidation, can be observed experi-
mentally in the triaxial apparatus or in the oedometer (which imposes a state of uniaxial
THERMODYNAMICS A BRIEF
SUMMARY
2.1.1 General
The free energy per unit volume of a dissipative material is defined as (, k ), where is
the (macroscopic) strain, whereas k constitute a finite set of, say N, internal variables
that represent irreversible microstructural processes in the material. A typical example
(that we shall consider later in more detail) is the plastic deformation that is caused by
dislocations of crystal planes in a metal.
From (, k ) we may calculate the stress and the socalled dissipative stresses (that
are energy-conjugated to k ) from the constitutive equations:
def
= , = , = 1, 2, . . . , N (2.1)
k
Remark: The relations (2.1) are consequences of the 2. law of thermodynamics (which
is not proven here) and are sometimes known as Colemans relations. 2
Hence, for given values of the state variables and k , we may always calculate the
dependent state variables and from the relation (2.1). In order to link values of
k to the observable variable , further constitutive relations must be established. Such
relations are expressed as rate equations of the form
k = f (, k ; ), = 1, 2, . . . , N (2.2)
The functions f (, k ; ) must be chosen in such a fashion that the dissipation inequality
N
X
D= k 0 (2.3)
=1
is satisfied for all possibles values of the strain rate .
Remark: The inequality (2.3), which is known as the Clausius-Duhem inequality, is also
a consequence of the 2. law of thermodynamics. 2
In conclusion, any material model that satisfies the relations (2.1) to (2.3) is consistent
with fundamental thermodynamic requirements.
Strain control
A strain-driven solution strategy is the natural approach in finite element codes based on
the displacement method. By integrating (2.2) for given strain history, i.e. (t) is known,
and given initial values k (0) = 0, we may calculate k (t). It is then straightforward to
calculate (t) from (2.1)1, at any time when the arguments (t) and k (t) are known.
Stress control
When the stress history, (t), is known, we assume that it is possible to invert (2.1)1 , to
obtain = (, k ). This expression can then be inserted into (2.2), which can now be
integrated for k (t).
Remark: In practice it is customary to use the strain-driven algorithm even in this case;
however, it is necessary to carry out iterations in order to compute (t) such that the
prescribed value (t) is obtained from (2.1)1 . 2
k = f (, k ) (2.4)
where f are bounded state functions. Examples of model classes are viscoelasticity and
viscoplasticity, which are discussed in greater detail in Chapter 3 and 5.
Since f is bounded, there will be insignificant (zero) change of k during a step loading
in time of , i.e. for very rapid change of . This means that we obtain elastic response
(that is defined by constant k ) during such a loading.
k = f() (, k ) (2.5)
()
where f are bounded state functions. Since the rate equations (2.5) are linear in ,
the corresponding material response is often termed incrementally linear. The most
important example is plasticity, which is treated in Chapter 4.
That the material is rate-independent may be illustrated by its indifference to a change
of clock. To show this, we assume that a different clock, i.e. another time-scale, is
introduced as the strictly monotonic function s(t), i.e. s(t) > 0. It is then possible to
def def
invert s(t) to give t = t(s), from which we can obtain k (s) = k (t(s)) and (s) = (t(s)).
Upon using the chain rule, e.g. (t) = (d/ds)s(t), it follows from (2.5) that
dk () d
f ((s), k (s)) s(t) = 0 t (2.6)
ds ds
which gives
dk d
= f() ((s), k (s)) (2.7)
ds ds
Upon integrating (2.7), we note that the same value of k is obtained for given value of s
independent of the function s(t). Hence, the response is not dependent on the clock or
real time but only on the history (s). Hence, the speed of the straining process (in real
time) is of no relevance for the solution of (2.5), as shown schematically in Figure 2.1.
s1(t)
s s2(t) k
s k
t1 t2 t s s
VISCOELASTICITY
In this chapter, we outline the elements of linear, as well as nonlinear, viscoelasticity. The
Maxwell model is taken as the prototype model. Both (Laplace) transform technique and
numerical integration are described for handling the time-dependence of the constitutive
relations. Structural analysis of a truss and a beam cross-section is outlined. Finally, the
modeling of damping (in dynamic analysis) using viscoelasticity is discussed.
3.1 Introduction
One of the simplest rheological models featuring combined viscous and elastic response
is the Maxwell model, as depicted in Figure 3.1. It is characterized by a spring (with
elasticity modulus E) serially connected to a dashpot (with viscosity coefficient ). Since
the dashpot represents a dissipative element, its strain will conveniently be treated as an
internal variable, that is subsequently denoted v .
1
Relaxation functions do not exist for all rheological models.
e = v v
The simplest choice of the free energy that gives the desired constitutive behavior is
1
= E( v )2 (3.1)
2
from which we obtain
= = E( v ) (3.2)
This constitutive equation can readily be derived from the rheological model in Figure 3.1.
The appropriate rate equation that determines the development of the internal variable
(viscous strain) is given as
1
v = v (3.3)
If we introduce the dissipative stress v associated with v from the constitutive equation
v = = E( v ) (3.4)
v
we may use this identity to express the rate equation (3.3) as
1 E
v = = ( v ) (3.5)
where t = /E is the natural relaxation time. It appears that (t) may be solved for
prescribed (t). Alternatively, (t) may be solved for prescribed (t).
where H(t) is Heavisides function defined as H(t) = 0 when t < 0, and H(t) = 1 when
t 0. Since v (0) = 0, we obtain from (3.5)
0 t
v (t) = (3.9)
E t
Combining this expression with the constitutive expression for in (3.2), we obtain
1 t
(t) = C(t)0 , with C(t) = 1+ (3.10)
E t
where C(t) is the creep function for the Maxwell model, which is depicted in Figure 3.2(a).
We may, instead of prescribed stress, assume that the strain 0 is applied suddenly at
t = t0 , whereafter it is held constant in time, i.e. (t) is written as
where it was used, again, that v (0) = 0. Combining this expression with the constitutive
equation for in (3.2), we obtain
t
(t) = R(t)0 , with R(t) = Ee t (3.13)
2
where R(t) is the relaxation function for the Maxwell model which is depicted in Fig-
ure 3.2(b).
Remark: It is noted that the strain 0 /E and the stress E0 represent the instantaneous
elastic response preceding the time-dependent creep and relaxation processes, respectively.
2
2 1
It is noted that R(t) 6= C(t) . However, such a simple inversion is possible for the corresponding
Laplace transforms, which is discussed below.
C R
E
E
1
E
1
t t
t t
(a) (b)
Figure 3.2: (a) Creep function, (b) Relaxation function for Maxwell material.
That creep and relaxation functions exist can be taken as the definition of a linear vis-
coelastic material response. Hence,
C(t)0
0
t t
(a) (b)
Figure 3.3: (a) Constant stress loading, (b) Creep response due to linear viscoelastic
behavior.
This behavior may also be described in terms of straight isochrone curves, as shown in
Figure 3.4(a,b) for a (viscoelastic) solid and a (viscoelastic) fluid, respectively. These are
defined by the following properties of the creep function:
For example, according to this definition, the Maxwell model represents a viscoelastic
fluid, whereas the Kelvin model (which is discussed below) represents a viscoelastic solid.
t=0 t=0
0 0
t1 > 0 t1 > 0
t2 > t1
t2 > t1
t t
(a) (b)
Figure 3.4: Straight isochrone curves for, (a) Solid and, (b) Fluid behavior, which are
characteristic for linear viscoelastic models.
It is possible to obtain the total strain response for a prescribed stress history (t) by
using the expression (3.10) and the principle of superposition. By applying the small
stress amplitude d(t ) at time t t, we obtain the strain response
d
d(t, t ) = C(t t )d(t ) = C(t t ) (t )dt , t t (3.16)
dt
This expression may be integrated in the form of a hereditary (or convolution) integral
Z t Z t
d d
(t) = d(t, t ) dt = C(t t ) (t )dt or = C (3.17)
0 0 dt dt
where the star () denotes convolution product.
Remark: The expression in (3.17) is valid also for non-differentiable functions (t) if the
derivative d/dt is taken in the sense of a distribution. For example,
d
(t) = 0 H(t t0 ) (t) = 0 (t t0 ) (3.18)
dt
where (t) is the Dirac delta distribution. Formal use of (3.17) then gives
Z t
(t) = C(t t )0 (t t0 )dt = C(t t0 )0 (3.19)
0
which is precisely the definition of the creep function. If, in particular, 0 is applied at
t = 0 due to rapid initial loading, then the expression in (3.10) is retrieved. Henceforth,
we shall therefore always assume that (0) = 0 and (0) = 0 with possible step values
applied at t = 0+ . 2
The dual formulation of (3.17) for the total stress response due to a prescribed strain
history (t) is obtained in an analogous fashion by using the relaxation function. We
obtain Z t
d d
(t) = R(t t ) (t )dt or = R (3.20)
0 dt dt
The analysis of linear viscoelastic materials can be reduced to the analysis of linear elastic
materials by means of symbolic calculus in the form of the Laplace-Carson transform. For
the present purpose, we consider functions f (t), which are piecewise differentiable and
which vanish identically for t 0. The transform is defined as (f ) (s):
Z
(f ) (s) = s f (t)est dt (3.21)
0
A table of transforms for frequently occuring functions are given in the Appendix of this
chapter.
A very useful property is given in the following theorem:
def d
() (s) = R (s) = (R) (s)() (s) (3.23)
dt
Proof: Homework!
We may check explicitly that the results in (3.24) and (3.25) are valid for the Maxwell
model. For this model we derived the creep and relaxation functions
1 t t
C(t) = 1+ , R(t) = Ee t (3.26)
E t
With the requirement that C(t) = R(t) = 0 for t < 0, both functions have a finite jump
at t = 0.
Using the tabulated transforms in the Appendix, we obtain
1 1 t s
(C) (s) = 1+ , (R) (s) = E (3.27)
E t s t s + 1
First, we conclude that
(C) (s)(R) (s) = 1 (3.28)
Secondly, we obtain
Z t
dR dR
C = C(t)R(0) + C(t t ) (t )dt =
dt dt
Z t 0
t tt 1 tt
1+ 1+ 1+ e dt = 1 (3.29)
t 0 t t
and
Z t
dC dC
R = R(t)C(0) + R(t t ) (t )dt =
dt 0 dt
Z t
t tt 1
e t + e t dt = 1 (3.30)
0 t
As a prototype of a general linear viscoelastic material, one may take the Linear Standard
Viscoelastic Model, whose rheological design is shown in Figure 3.5.
E1
1
E2
2
EN
N
v v
The model consists of N Maxwell chains coupled in parallel. The expression for , that
was given for the Maxwell model, is extended in the following obvious fashion:
N
1X
= E ( v )2 (3.31)
2 =1
Remark: It appears that E() , as defined in (3.33)2, is the elastic stiffness of the model at
infinite loading rate, in which case no viscous strain will develop in the dashpots (v = 0).
Moreover, the considered Linear Standard Model can represent solid behavior only if it
has non-zero elastic stiffness E(0) at zero loading rate. Clearly, this is the case only if one
of the dashpots disappear, e.g. N = which gives E(0) = EN . These situations will be
elaborated further in Section 3.7 in conjunction with dynamic behavior. 2
The rate equations for the N internal variables v are chosen as
1
v = , = 1, 2, . . . , N (3.34)
and it appears that
N N
X X 1
D v = ( )2 (3.35)
=1
=1
Since all > 0, it is clear that D > 0, i.e. the CDI is satisfied also for this general
model.
Upon substituting from the constitutive equations (3.32)2 into (3.34), we obtain the
uncoupled rate equations
1
v = ( v ), = 1, 2, . . . , N with t = (3.36)
t E
We have now established the complete set of constitutive equations from which the creep
and relaxation functions can be derived.
It turns out to be convenient to start with the relaxation function R(t). By applying the
Laplace-Carson transform to (3.36), with the initial values v (0) = 0, we obtain
1
(v ) = (3.37)
t s + 1
which may be combined with (3.32)2 to give
E t s
( ) = (3.38)
t s + 1
Finally, upon combining this result with (3.32)1 gives
N
X
E t s
() = (R) () with (R) = (3.39)
t s+1
=1
It appears readily that R(0) = E() , whereas R() = 0 if t < for all .
Remark: If, in the :th Maxwell chain, its spring stiffness is infinite (E = ), then
R(t) does not exist. 2
We may now invert (R) in (3.39) to obtain (C) . Hence,
N
!1
1 X t s
() = (C) () , (C) = = E (3.41)
(R) =1
t s + 1
It appears that no simple expression of C(t) can be found, except in special cases. Below
we (re)consider the Maxwell and Kelvin models.
The Maxwell model (which represents a fluid) is obtained from the Linear Standard Model
by the special choice of one single chain with
E1 E and t1 t (3.42)
The Kelvin model (which represents a solid), as shown in Figure 3.6(a), is obtained from
the Linear Standard Model by the choice of two chains with
E1 = , E2 = E and 1 = , 2 = (3.45)
and is shown in Figure 3.6. This is a more tricky model, since (R) can not be obtained
directly from (3.39). Rather, we consider Figure 3.6 to obtain
def
= 1 + 2 = s + E = E(1 + t s) with t = (3.46)
E
Vol 0 March 7, 2006
40 3 VISCOELASTICITY
which gives
1 1
(R) = E(1 + t s), (C) = (3.47)
E 1 + t s
The Three-parameter model (which is also called the Standard Solid in the literature),
shown in Figure 3.6(b), is the simplest special case of the Linear Standard Model.3 It is
obtained by the choice of two chains with
We note that E(0) is the value of R(t) at very slow loading, i.e. for t = . Likewise, E()
is the value of R(t) at very rapid loading corresponding to t = 0.
Homework: Find the explicit expression for C(t). 2
When it is difficult to find the inverse from Laplace transform space back to the time-
domain, the constitutive relations may instead be integrated in a step-by-step fashion.
Here we shall apply the Backward Euler (BE) method to the Linear Standard Model in
the case of prescribed strain history.
3
It is too simple to represent realistic solid behavior.
(a)
E2 = E ( 2 = )
( E1 = ) 1 =
(b)
E2 ( 2 = )
E1
1 =
where
1
n v n t
= a (t) , Ev = a (t)E with a (t) = 1 + (3.54)
t
n+1
The (total) stress is then obtained from (3.32)1 as
N
X
n+1 n+1 n v
= = + E v (3.55)
=1
where
N
X N
X
n v n v v
= , E = Ev (3.56)
=1 =1
Remark: Pure elastic response is obtained when t , in which case we obtain from
(3.54), (3.55) and (3.33):
n+1
= n + E() 2 (3.57)
Maxwell model
In the special case of the Maxwell model, which is obtained if we set N = 1 with E1 E
and t1 t , we obtain from (3.53) and (3.55)
1
n+1 n+1 tr t
= a(t) with a(t) = 1 + (3.58)
t
n+1 tr
where is the elastic trial stress defined as
n+1 tr
= n + E (3.59)
The behavior of a linear viscoelastic structure will depend on its statical (in)det-
erminacy. Here, we shall outline some general features, which will be illustrated later in
conjunction with the discussion of truss structures.
Isostatic structures
For an isostatic structure, the stresses are (by definition) uniquely determined by the
applied load. For the important special case when the load is constant in time (after initial
step loading), the stresses are also constant. Hence, the entire structure is subjected to a
state of pure creep. If, in addition, the creep functions of every material point are affine,
i.e. a common relative creep function C(t) can be identified, then C(t) applies also to the
entire structural response.
Hyperstatic structures
For a hyperstatic structure, the stresses are not uniquely determined by the applied load
but depend on the material properties. Hence, the stresses will generally redistribute
with time even for a load that is constant in time. This phenomenon is denoted structural
relaxation. The exception is, again, the situation when a common relative creep function
C(t) for each material point can be identified. Then the structural relaxation is zero (for
initial step loading), and C(t) applies to the entire structural response.
It follows from the relations (3.22) and (3.23) that linear viscoelasticity becomes quite
analogous to linear elasticity upon Laplace-Carson transformation. As a consequence,
it is always possible (in theory) to take advantage of this fact when solving structural
problems. The equivalent linear elastic problem in transform space is then solved (an-
alytically or numerically), which is followed by inversion back to the time-domain. For
real structures, such inverse transformation must generally be carried out numerically. A
powerful algorithm was developed by Talbot (1979). The entire procedure is illustrated
in Figure 3.7.
Solution
Figure 3.7: Solution procedure for viscoelastic problem based on elastic analogy.
The elastic analogy strategy outlined above may be efficient when the response is sought
only at one point (or a few points) in time. In most cases, however, it is desirable to know
the entire time response, which makes it more advantageous to employ a step-by-step
procedure in time. This means that the constitutive relations are solved by numerical
integration.
Here we shall briefly repeat the crucial relations and steps in the (finite element) analysis
of a truss structure built of uniform bars of linearly elastic material. The truss structure
under consideration, which is depicted in Figure 3.8, is subjected to time dependent
loads collected in the load vector P (t). The energy-conjugated displacement components,
collected in p(t), are sought.
P1
node
P2 p2
}
normal force N i
p1 elongation n i bar element
length L i
Each uniform bar is assumed to have the length Li , cross-section area Ai and relaxation
function Ri (t). Its elongation and normal force are denoted ni and Ni respectively. After
applying the Laplace-Carson transform, we obtain the following constitutive relations for
the i:th bar (in complete analogy with the corresponding linear elastic bar):
Li (Ri ) Ai
(ni ) = (Ni )
or (Ni )
= (ni ) (3.60)
(Ri ) Ai Li
We may collect these relations for all bars in the matrix relations
where (F e ) and (S e ) are the diagonal compliance and stiffness matrices, respectively,
in transform space of the element assembly (denoted by subindex e). More explicitly,
Next, we shall outline the structural analysis pertinent to a statically determinate (iso-
static) and statically indeterminate (hyperstatic) truss, respectively.
(P ) = AT (N ) (3.63)
It is noted that the matrix AT represents only geometric relations. (In fact, A is the
natural transformation, that is defined below for the hyperstatic truss.)
From virtual work considerations (Clebschs theorem), we may establish the kinematically
dual relation of (3.63) as:
(p) = B T (n) (3.65)
which may be combined with (3.61) and (3.64) to give the structural flexibility relation
in transform space:
as shown in Figure 3.9. Considering (3.64), we immediately see that this relation can be
backtransformed in trivial fashion to the time-domain as
where N 0 is the instantaneous normal force that is uniquely determined (from equilib-
rium) by the suddenly applied load P 0 at t = 0.
As to the structural displacements p(t), it appears from (3.66) that no simple inversion
can be obtained in general. However, in the special case that the relaxation functions are
affine, i.e.
(Ri ) (s) = Ei (R) (s), i = 1, 2, . . . (3.69)
(s) is the transform of a common relative relaxation function R(t), then we
where (R)
may conclude from (3.62) and (3.66) that
1 1
(p) = F (P )
= F ( C)
(P )
with ( C)
= (3.70)
(R) (R)
where F represents the elastic structural flexibility activated for a suddenly applied load.
For example,
where p0 is the initial elastic response due to the suddenly applied load P 0 at t = 0.
Figure 3.9: Transformation diagram for structural analysis based on elastic analogy (Inner
loop is well-defined only for isostatic structures).
In the case of a hyperstatic truss the equilibrium equations have no unique solution
(N ) . The dual kinematic relation is obtained from virtual work considerations (Clebschs
theorem) as
(n) = A(p) (3.72)
We may combine this relation with (3.63) and (3.61)2 to give the structural stiffness
relation in transform space:
whereby
(N ) = (S e ) A((S) )1 P 0 N 6= 0 (3.78)
and it follows that there will be a time-dependent redistribution of stresses even for the
constant load P 0 . This phenomenon is termed structural relaxation. 2
We shall consider the Linear Standard Model. From (3.55) we obtain the normal force
n+1 def
Ni = Ni in the i:th bar:
E v Ai
Ni = n Niv + i ni (3.79)
Li
N= n
N v + S ve n (3.80)
where
E1v A1
L1
0
E2v A2
S ve =
0 L2
(3.81)
The pertinent equilibrium and kinematic relations for the truss structure are given as
P = AT N and n = A p (3.82)
By combining these relations with the constitutive relation (3.79), we obtain the structural
stiffness relation
P = AT n N v + S v p with S v = AT S ve A (3.83)
where S v is the algorithmic structural stiffness matrix. We may now solve for p from
def
(3.83) to obtain the updated displacement vector n+1 p = p as
p = n p + p with p = (S v )1 ( P AT n N v ) (3.84)
Step loading
p = (S e )1 P with P = P n P (3.85)
This is purely elastic response due to a step loading P , that is applied instantaneously
at the time t = tn .
0
P (t) = P 0 H(t), P = P (0) = P 0 (3.86)
p = n p + p with p = (S v )1 (P 0 AT n N v ), n 0 (3.87)
h1 h
z , t = 0 , t =
b
Figure 3.10: (a) Typical cross-section of sandwich beam. Stress distribution for Maxwell
material in the core for (b) t = 0 and (c) t = .
where c,0 is the initial elastic curvature in response to the suddenly applied moment at
t = 0, and 0 (z) is the initial (elastic) stress distribution:
M0 E(z)M0
c,0 = e
, 0 (z) = z (3.95)
S Se
Now, returning to the sandwich beam, we conclude that
bh31 b(h3 h31 )
(S) = (R1 ) I1 + (R2 ) I2 , I1 = , I2 = (3.96)
12 12
where Ri (t), i = 1, 2, are the relevant relaxation functions for the inner core and surface
layers, respectively. The moments of inertia Ii , i = 1, 2, are defined w.r.t. the neutral
axis, and we note that I = I1 + I2 .
Consider the case when the core of the sandwich beam responds as a Maxwell-model,
(2)
whereas the surface layers are elastic (t = ). For simplicity, the (unrealistic) assump-
tion is made that the E-modulus is the same, i.e. E1 = E2 = E. Since the relaxation
function for the Maxwell model is given by its transform
t s
(R) (s) = E (3.97)
t s + 1
we obtain
" #
t s 1 I2 t s
(S) = EI1 + EI2 ((S) )1 = 1 1 (3.98)
t s + 1 EI2 I t s + II2
We have phrased ((S) )1 in a form that provides directly for the inverse transform. With
M0 applied at t = t0 , we obtain
M0 I2 II2 tt
c (t) = 1 1 e (3.99)
EI2 I
with the limiting values
M0 M0
c (0) = , c (t) = when t = (3.100)
EI EI2
To work out the corresponding stress-distribution with time is left as homework, (cf.
Figure 3.10(b,c)). Consider the cases I2 0 (homogenous Maxwell model) and I2 I
(homogeneous elasticity).
Numerical integration in time and space (over the beam cross-section) is a versatile and
effective technique, which is general w.r.t. the choice of the material properties. For
this purpose, we consider the double-symmetric cross-section as shown in Figure 3.11.
The cross-section is subdivided into 2nint lamellas of width bi and thickness zi , where
subindex int stands for integration.
n+1 def
The moment M = M at time t = tn+1 is given as
Z h/2 nint
X
M= (z)zb(z)dz 2 i zi Ai with Ai = bi zi (3.101)
h/2 i=1
symm
h
2
zi h
2
zi
z
bi
def
where n+1 i = i is the updated stress value in the i:th lamella.5 Again, we shall consider
the Linear Standard Model, for which was given in (3.55). We thus use the relation
M = n M v + S v c (3.103)
where
Z h/2 nint
X
n v n v n v
M = (z)zb(z)dz = 2 i zi Ai (3.104)
h/2 i=1
and
Z h/2 nint
X
v
S = E v (z)z 2 b(z)dz = 2 Eiv Ii with Ii = zi2 Ai (3.105)
h/2 i=1
n+1 def
from which c is solved to give the updated curvature c = c
c = n c + c with c = (S v )1 (M n M v ) (3.106)
Step loading
c = (S e )1 M with M = M n M (3.107)
which gives
For a nonlinear viscoelastic material, the creep curves are nonlinear functions of the
initially applied stress 0 , as indicated in Figure 3.12. In practice, the major effort has been
spent on modeling only the stationary creep stage, which is also indicated in Figure 3.12.
(In order to be able to model both the transient and stationary stages with a unified
model, one has to resort to hardening viscoplasticity, as discussed in Chapter 5.)
It is clear that = (0 , t), where the dependence on t may be of quite general character,
i.e.
(t) 6= C(t)0 (3.111)
Such behavior is characterized by nonlinear isochrone curves for the creep behavior, as
shown in Figure 3.13.
t t
(a) (b)
Figure 3.12: (a) Constant stress loading, (b) Creep response due to nonlinear viscoelastic
behavior.
0 t=0
t1 > 0
t2 > t1
The most commonly used creep law for the modeling of stationary creep in metals, which
is due to Norton (1929), is obtained by generalizing the Maxwell model. This may be
done in such a fashion that the rate law for the viscous strain is a simple power law in
the stress:
n
v 1 || c
= (3.112)
c ||
where c is the creep modulus, nc is the creep exponent and is a relaxation time (whose
value is normally chosen constant). The value of c will depend on the choice of nc .
In addition, c is strongly temperature-dependent and c 0 when the temperature
approaches the melting point (T Ts ). The creep exponent nc is less sensitive to tem-
perature increase, but nc 1 when T Ts . Corresponding values of c and nc can be
found in, for example, Hult (1984), p 106.
Remark: From (3.112) follows that the same creep rate is modelled in tension and
compression (for the same magnitude of stress). 2
Upon combining (3.112) with Hookes law in (3.2), we obtain the governing constitutive
equation nc
E ||
+ = E (3.113)
c ||
Remark: Equations (3.112) and (3.113) represent the conventional choice of parameters
based on the a priori chosen value of . Another possibility is to conform directly to the
Maxwell model by postulating the creep law
n
v 1 || c
= (3.114)
t E ||
where E is the (static) value of the elasticity modulus. We then have the relation
nc
c
t = (3.115)
E
in which case t 0 when c 0 for high temperature. We may then rewrite (3.113) as
n
E || c
+ = E (3.116)
t E ||
which format is used subsequently. The Maxwell model s readlily retrieved when nc = 1
2
Creep
In the creep situation, when = 0 > 0 is held constant (after rapid loading), we obtain
the solution of (3.113), with the notation in (3.115), as
1 0 nc 0 0 nc t
(t) = , (t) = + , t>0 (3.117)
t E E E t
which is characteristic for stationary (or stage II) creep, as shown in Figure 3.14(a).
Relaxation
In the relaxation situation, when = 0 > 0 is held constant (after rapid loading), we
obtain from (3.113) the problem
E nc
+ = 0, (0) = 0 = E0 (3.118)
t E
0 nc = 1
0 nc
0 nc > 1
E
E 1
t t
t t
(a) (b)
Figure 3.14: (a) Creep and (b) Relaxation curves for a Norton-material
whose solution is
t
0 e t , nc = 1
(t) = h i n 11 (3.119)
(nc 1) + (nc 1) t c
, nc 6= 1
0 E nc 1 t
where n+1 tr is the elastic trial stress given in (3.59). We may rearrange in (3.120) to
obtain
E n+1
1 + n+1 = n+1 tr (3.121)
| |
| {z }
>0
which shows that sign(n+1 ) = sign(n+1 tr ). Now, taking the absolute value of both sides
of (3.121) gives
def def def
e + E = etr with e = ||, etr = | tr |, n+1
= (3.122)
we may combine (3.122), (3.123) to obtain a set of equations from which e and can
be solved. Two different iterative schemes will be discussed below. After convergence of
such a scheme it is possible to obtain n+1 as
(
n+1 tr
n+1 e if >0
= n+1 tr
(3.124)
e if <0
R (e , ) = e etr + E = 0
t
(3.128)
R (e , ) = (e ) t =0
with
def d nc e nc 1
(e ) = (e ) = (3.131)
de E E
The inverse of J can be obtained explicitly as
" #
t 1 1
1 E def t 1
J 1 = t
, ha = E + (3.132)
ha 1 1 t
Like in the case of linear viscoelasticity, the structural response will depend on the statical
(in)determinacy. Structural relaxation will always take place (even if all truss members
have the same material properties) for a hyperstatic structure. In practice, most engi-
neering analyses for design purposes deal only with the stationary state that is achieved
a long time after a time-independent structural load has been applied to the structure.
Because of the inherent material nonlinearity, the analysis of the transient structural
behavior must be based on numerical integration of the constitutive relations. Iterations
are needed to find the incremental displacements in each time step.
Ni = Ai i (3.135)
and the normal forces are collected in the column matrix N (n).
Remark: In the case of linear elasticity, we have a = 1. Moreover, the Maxwell model is
retrieved at the choice nc = 1, whereby we obtain
1
tr t || t
= 1 ; a= 1+ (3.136)
t | tr | t
P = AT N and n = A p (3.137)
Iterations are required in order to find the solution p from (3.137)1. For given load
n+1 def
P = P , we may devise the following Newton procedure (k being the iteration count):
p(k+1) = p(k) + p
(AT N )
S v(k) T
a p = (P A N
(k)
) with S va = (3.138)
(p)
and with p(0) chosen as the converged value of p in the previous timestep.
In (3.138), we introduced the algorithmic tangent stiffness (ATS) matrix S va , which is a
nonlinear function of the incremental solution p for a given timestep.
where, for each bar, the algorithmic tangent stiffness modulus Eav is defined as
E
Eav = nc 1 2 (3.140)
||
1+ nc t
t E
(N ) (N )
S va = AT A = AT A (3.141)
(n) (n)
where it was used that d() = d(n)/L. An explicit expression for Eav pertinent to the
Norton law is obtained upon differentiating the relation (3.120). We obtain
d() = E d() E d() (3.144)
||
with nc 1
t 1 ||
d() = nc d() (3.145)
t E E ||
Upon rearranging terms in this expression, we obtain
E
d() = nc 1 d() (3.146)
||
1 + nc t
t E
Figure 3.15: (a) Non-stationary creep of Norton material, (b) Stationary stress distribu-
tion in double-symmetric cross-section.
practice, the only one that can be analyzed analytically. Again, only double-symmetric
cross-sections are considered (for simplicity).
Since no elastic strain develops at the steady state, we have from (3.111)
n
1 | | c
= , (z) = c z (3.147)
t E | |
where we used the notation and for the steady state values at t = . From (3.147),
we conclude that
n
1 | | c 1 1
| | = | | = E|z| nc (t c ) nc (3.148)
t E
Finally, we may solve for c from (3.150) and insert into (3.149) to obtain the stress
distribution from the generalization of the Navier formula (for elastic response) as follows:
M0 n1 1
= |z| c z (3.151)
In
Remark: In is a generalized moment of inertia, and the usual moment of inertia is define
as I = I1 (for nc = 1). In this latter case it appears that the classical linear stress
distribution, which is pertinent to linear elastic response, is retrieved. 2
Remark: The stress distribution in (3.151) is the same as for the nonlinear elastic Bach
material defined by the stress-strain law
nc
|| M n1 1
= 0 ; = |z| c z (3.152)
0 || In
which is depicted in Figure 3.16. The reader should show this as homework! 2
C
C
In the case of a cross-section with height h, the maximum stress (for z = h/2) at t =
max
( ) and at t = 0 (0max ) are given as
1 max
1 1
max M0 h nc max M0 h I1 h nc
= and 0 = ; max = (3.153)
In 2 I1 2 0 In 2
C C C
Figure 3.17: Stress distributions in beam cross-section at stationary creep for different
values of the creep exponent nc .
n+1 def
The updated stress = was given in (3.121), i.e
tr
= tr E (3.157)
| tr |
where
tr = n + E with = zc (3.158)
As discussed previously, must usually be calculated in an iterative fashion for given tr .
Iterations are used to find the solution c from (3.156). For a given moment M0 in a
creep situation, we may devise the Newton procedure
(k+1)
c = (k)
c + c (3.159)
Figure 3.18 shows the stress relaxation with time for the rectangular cross-section sub-
jected to the moment M0 applied at t = 0. (The creep exponent is nc = 2.)
h
2
zcoordinate
t=
0
t=0
h
2 0 0max
stress
Figure 3.18: Stress relaxation with time in rectangular cross-section for the creep exponent
value nc = 2.
where nint is now the total number of integration points across the height of the beam
cross-section. The updated stress is still given by (3.157) and (3.158), where i is now
defined by
i = + zi c (3.166)
Hence, for given N = N0 and M = M0 , we solve for and c from the system
N(, c ) = N0
M(, c ) = M0 (3.167)
The algorithmic stiffness moduli in (3.169) are obtained as follows: First, we obtain
Z h+
d(M) = d()zb(z)dz =
h
Z h+ ! !
Z h+
Eav (z)zb(z)dz d() + Eav (z)z 2 b(z)dz d(c )(3.172)
h h
Z h+ nint
X
v
Sa,NM = Eav (z)zb(z)dz = (Eav )i zi Ai (3.174)
h i=1
Z h+ nint
X
v
Sa,MM = Eav (z)z 2 b(z)dz = (Eav )i Ii (3.175)
h i=1
Assume now that the y-axis is located at the center of gravity, i.e.
Z h+
zb(z)dz = 0 (3.176)
h
h c
t>0 M0
y
t=0 N0
+ z
h
z (z) = + c z
3.7.1 Preliminaries
The viscoelastic response can be assessed, and the used model calibrated, from obser-
vations made under quasistatic or dynamic conditions. Creep and relaxation are typical
quasistatic phenomena (where inertial forces can be ignored), whereas rapid loading gives
rise to inertial forces that must be accounted for. In the latter case, the viscoelastic dis-
sipation of energy results in damping of the response (as compared to the purely elastic
response). In the literature, the Kelvin model has traditionally been used to describe
the damped behavior of vibrating structures. This is usually denoted viscous6 damping,
which represents a pronounced dependence of the rate of loading (as we shall see later),
whereas the other extreme that the damping is not dependent on the rate of loading is
termed hysteretic damping. In conclusion, we may state that creep and damping are two
sides of the same coin.
The viscoelastic characteristics may be observed from free and forced vibrations. Free
vibrations occur after an initial disturbance of the static state of equilibrium and will
die out eventually. Forced vibrations, on the other hand, represent the stationary (har-
6
Since all realistic models are generically of viscous character, we shall use this term henceforth without
specific reference to the Kelvin model.
monic) motion due to (harmonic) loading. This is the only loading situation considered
henceforth. Moreover, we restrict our consideration to linear viscoelastic response.
Consider the uniform bar in Figure 3.20 with its mass L lumped to one end. The forced
vibrations are caused by the applied harmonic force
where fa is the amplitude and is the angular frequency with which the load excites the
system7 . Assuming uniform strain (= u/L) in the bar, we obtain the equation of motion
as
m + = Re{fa eit } with m = L2 (3.180)
It is convenient to solve (3.180) with f (t) replaced by the equivalent Fourier component
f F to obtain the corresponding F and F . Since the actual load is f (t) = Re{f F (t)}, we
obtain (t) = Re{F (t)}, etc. We thus obtain from (3.180)
m 2 () + () = (f ) with (f ) = fa (3.181)
fa
() = = a ei ; F (t) = a ei(t+ ) (3.182)
(R) m 2
7
For any variable u(t), we define its complex Fourier transform (u) ()
Z
(u) () = i u(t)eit dt = (u)R () + i(u)I ()
0
where (u)R and (u)I are real-valued functions. The corresponding Fourier component uF is defined as
where ua is the (real) amplitude and u is the phase angle given from
q
(u)I
ua = ((u)R )2 + ((u)I )2 , tan u =
(u)R
It appears that the Fourier transform is identical to the Laplace transform upon setting s = i. A more
explicit discussion of the use of complex variable technique for damped vibrations (including the Fourier
transform) is found in Akesson (1992).
L
fa cos t
u
L
where ! 12
2
fa 2
a = 1( ) + 2 , tan = (3.183)
(R)R ref 1 ( ref )2
In order to obtain the expressions in (3.183), we introduced the following representation
for (R) :
(R)I
(R) = (R)R (1 + i) = Ra eiE with = (3.184)
(R)R
where () is the loss factor. Moreover, we introduced the reference frequency ref ()
from the definition 12
(R)R
ref = (3.185)
m
and the amplitude Ra () as
21
Ra = (R)R 1 + 2 , tan E = (3.186)
We shall define E(0) = Ra (0) and E() = Ra () as the apparent elastic moduli for slow
( = 0) and rapid ( = ) loading, respectively.
Remark: For a general viscoelastic model, both (R)R and (R)I are frequency-dependent.
Hence, it is concluded that ref = ref (), and the subindex ref might seem awkward.
However, we shall see later that ref becomes the undamped eigenfrequency 0 > 0 for
the classical Kelvin model. 2
With (3.182) and (3.184), we may now obtain () as
a = Ra a , = + E (3.188)
The results in (3.182) and (3.187) are shown schematically in Figure 3.21.
(R)
(R)
Figure 3.21: Complex representation of Fourier transforms () and () due to the loading
(f ) = fa cos t.
where
F = a ei(t+ +E ) , F = iF = a ei(t+ + 2 ) (3.191)
p
where it was used that T = 2. However, since sin E = / 1 + 2 we obtain with
(3.186) and (3.188)
Wdiss = (R)R 2a = (R)I 2a (3.193)
With the choice tN = in the Linear Standard Model (by which solid behavior is
represented), we obtain from (3.39)
N 1
X E t i
(R) () = + EN = (R)R + i(R)I (3.194)
=1
1 + t i
where
N 1 N 1
X E (t )2 X E t
(R)R = + EN , (R)I = (3.195)
=1
1 + (t )2 =1
1 + (t )2
We note that
N
X
(R)R (0) = EN E(0) , (R)R () = E E() (3.196)
=1
and
(R)I (0) = (R)I () = 0 ; (0) = () = 0 (3.197)
which gives
21 N 1
! 21 12
E(0) X E()
ref (0) = (0) = , ref () = () = 2 + (0)
2
= (3.200)
m =1
m
The typical behavior of () is shown in Figure 3.22, which also shows the critical fre-
quency defined as
cr = arg[max ()] (3.201)
3
Maxwell
2.5
2 Kelvin
()/max
1.5
1
Linear Standard Solid (3-parameter)
0.5
0
0 1 2 3 4 5 6 7 8
t* / ( t*)
cr
Figure 3.22: Frequency-dependence of loss factor for the Three-parameter model (for
E() /E(0) = 4), the Kelvin model and the Maxwell model.
fa
a (0) = static
a = (3.202)
E(0)
and we may obtain the dynamic amplification factor A() as the ratio
" 12
2 #2
a E(0)
A() = 1 + 2 (3.203)
static
a (R)R ref
It appears that A(0) = 1 and A() = 0, and we show the characteristic behavior of A()
A
in Figure 3.23, where cr denotes the damped resonance frequency defined as
A
cr = arg[max A()] (3.204)
1.2
E() /E(0) (Kelvin)
1
E() /E(0) = 4/1 (3-parameter)
A() 0.8
0.6
0.4
0.2
0
0 20 40 60 80 100 120 140 160 180 200
[rad/s]
Remark: As pointed out previously, the situation E() = , i.e. infinite stiffness at
very rapid loading, represents an extreme situation of solid behavior. 2
We thus obtain the quite unrealistic result that grows linearly with without bound,
which is indicated in Figure 3.22. This behavior characterizes the classical notion of
p
viscous damping. Moreover, we conclude that ref = 0 = E/m, and we obtain from
(3.203) that
" 12
2 #2
A() = 1 + (t )2 (3.207)
0
Et i E(t )2 1
(R) () = ; (R)R = , = (3.208)
1 + t i 1 + (t )2 t
which gives
1
E(t )2 1 2
Ra = 1+ ; E(0) = 0, E() = E (3.209)
1 + (t )2 t
Hence, we obtain the peculiar result that is unlimited when = 0, as indicated in
Figure 3.22. Since the static solution is unbounded, the definition of A() does not make
any sense.
With the choice of parameters defined in (3.49), we obtain directly from (3.195) that
We obtain
21 " 12 12 #
1 E(0) 1 E(0) E(0)
cr = and max = (3.211)
t E() 2 E() E()
In the table below, the Laplace-Carson transform of some elementary (useful) functions
are listed. It is assumed that f (t) = g(t) = 0 when t = 0.
dn f
dtn
(t) sn (f ) (s)
H(t) 1
H(t t ), t > 0 est
f (t t )H(t t ), t > 0 (f ) (s)est
df
dt
(t) g(t) (f ) (s)(g) (s)
n! 1
tn , n 0 sn
(= s
for n = 1; = 1 for n = 0)
s
eat s+a
a
1 eat s+a
s2
cos t s2 + 2
s
sin t s2 + 2
s
f (t)eat (f ) (a + s) s+a
PLASTICITY
4.1 Introduction
The macroscopic theory of plasticity is probably the most important (and celebrated) the-
ory of inelastic response of engineering materials, when judged from its widespread use in
commercial FE-codes. The word plastic is a transliteration of the ancient Greek verb
that means to shape or form. Plasticity theory is traditionally associated with the
irreversible deformation of metals, viz. low-carbon steel, for which the inelastic deforma-
tion occurs mainly as distortion (shear), whereas the inelastic volume change is normally
negligible. However, plasticity theory has also won widespread use in the modeling of
non-metallic ductile materials, such as certain polymers and fine-grained soil (e.g. clay).
For these highly porous materials, the inelastic deformation has both distortional and
volumetric components.
The conceptual background of plastic (and viscoplastic) deformation in metals is plastic
slip along crystal planes in the direction of the largest resolved shear stress, or Schmid-
stress, and this slip is caused by the motion of dislocations of atom planes. In a perfect
crystal structure the plastic slip results in a macroscopic shear deformation without other
distortion of the lattice structure itself. This deformation is superposed by elastic defor-
mation, as illustrated in Figure 4.1.
p e
Figure 4.1: Microstructure of single crystal showing plastic deformation followed by elastic
deformation.
However, most metals are polycrystalline materials. This means that grains with differ-
ent crystallographic orientations and lattice structure (that represents different thermo-
dynamic phases) are interacting in the mesostructure, cf. Figure 4.2. If the distribution
Figure 4.2: Mesostructure of grains interacting via grain boundaries and possessing dif-
ferent crystal orientations.
Perfectly plastic behavior may be represented by the prototype model shown in Figure 4.3.
The frictional-plastic slider is inactive as long as || < y , where y is the yield stress.
=0 (L)
y
E
y =E (U)
(U)
=
e p p p e
y
(L)
(a) (b)
Figure 4.3: (a) Prototype model for elastic-(perfectly)-plastic material, (b) Stress-strain
relationship.
As the single internal variable we take the plastic strain p , and the expression for the
free energy is chosen as
1 1
= E(e )2 = E( p )2 (4.1)
2 2
where e = p is the elastic strain of the Hookean spring with modulus of elasticity E.
We then obtain the constitutive equation for the stress as
= = E( p ) (4.2)
and for the dissipative stress, that is conjugated to p , as
p = p
= E( p ) (4.3)
The yield criterion is = 0, where is chosen as
() = || y (4.4)
Since the magnitude of stress can never exceed the yield stress (in this simple prototype
model), it follows that the admissible stress range is defined as those stresses for which
0.
It is assumed that no plastic strain will be produced when < 0, i.e. when || < y .
The material response is then elastic and || < y thus defines the elastic stress range.
However, when = 0 plastic strain may be produced. The constitutive rate equation for
p is then postulated as the associative flow rule:
p = = (4.5)
||
where the plastic (Lagrangian) multiplier is a non-negative scalar variable. Combining
(4.5) with Hookes law expressed in (4.2), we obtain the differential equation for the stress
as
= E E (4.6)
||
The problem formulation is complemented by the so-called elastic-plastic loading criteria.
It follows from the aforesaid that the general format of the loading criteria is
0, () 0, () = 0 (4.7)
Plastic loading (L) is defined by the situation > 0 and = 0, in which case we may
solve for from (4.9) to obtain
= (4.10)
||
It follows that this is a valid solution only when (/||) > 0, which is the appropriate
loading criterion, that must be satisfied in order for plastic strain to evolve.
Elastic unloading (U) is defined by = 0 and 0, which is obtained whenever
(/||) 0.
Upon inserting the expression for given in (4.10), into (4.5), we obtain the rate equation
for the internal variable p in terms of the control variable as
By inserting this result in (4.6), we obtain the corresponding tangent stiffness relation as
The dissipation function D was defined in (2.3). For the formulation defined by (4.1) and
(4.4), we obtain for = 0, i.e. when || = y , the expression
D = p = || = y 0 (4.17)
Remark: It is interesting to note that D will not get the same value in the model
formulation that was defined in the previous Remark. For this formulation, defined by
(4.13) and (4.14), we obtain
D = p + k = y y = 0 (4.18)
Hardening plastic behavior is represented by the prototype model shown in Figure 4.4.
The frictional-plastic slider is now increasing its resistance due to the amount of slip
developed. More specifically, the excess stress over the initial yield stress is due to the
hardening spring with stiffness H that is related to the plastic strain. Upon unloading
and reloading, the slider will thus become inactive until the stress has resumed the previous
level during loading, i.e. as long as || < y + H|p |, where H > 0 is the (constant)
hardening modulus. This behavior is typical for hardening plasticity.
Apart from p , we now introduce the (isotropic) hardening variable k, such that the free
energy density is expressed as
1 1
= E( p )2 + Hk 2 (4.20)
2 2
From Colemans equations we still obtain
= = E( p ) (4.21)
Vol 0 March 7, 2006
4.3 Prototype model for hardening plastic behavior 83
H
. .
=Hp
E
y
y . .
=E
e p
(a) (b)
Figure 4.4: Prototype model for elastic-hardening-plastic material, (a) Rheological model,
(b) Stress-strain relationship.
p = (4.22)
p
whereas the dissipative stress , associated with k, is given as
= = Hk (4.23)
k
The yield function is now defined as
(, ) = || y (4.24)
Inelastic deformation can be produced when = 0. The associative flow and hardening
rules are then defined as
p = = (4.25)
||
k = = (4.26)
where the plastic multiplier is still defined by the loading criteria given in (4.7). The
pair (p , k) can be perceived as the outward pointing normal from the cone defined by
( , )=0
. .
.p .p ( p, )
=- =
0
-y y -y y
=0 =0
E
(a) (b)
Figure 4.5: (a) Associative flow rule for perfect plasticity, (b) Associative flow and hard-
ening rules for hardening plasticity.
E = E E (4.27)
||
= H (4.28)
0, (, ) 0, (, ) = 0 (4.29)
Remark: The loading criteria (4.29) are known as the Kuhn-Tucker complementary
conditions, which stem from constrained (convex) minimization. In fact, the associative
flow and hardening rules can be derived from a minimization principle. [The theoretical
basis is beyond the scope of this introductory treatment.] 2
Plastic loading (L), is defined as > 0 and = 0. Since h > 0, this situation is at hand
whenever (/||) > 0, and we obtain from (4.31) as
E
= (4.33)
h ||
On the other hand, elastic unloading (U), which is defined by = 0 and 0, is obtained
whenever (/||) 0. It is noted that the criteria for loading/unloading are exactly the
same as for perfect-plasticity.
We may now obtain
E
p = (L), p = 0 (U) (4.34)
h
and
= E ep (L), = E (U) (4.35)
where E ep is the elastic-plastic tangent stiffness modulus defined as
ep E E
E = 1 E= H (4.36)
h h
With a generalization of the hardening concept, the following situations are distinguished:
H = E ep = E (elastic) (4.40)
H = E (h = 0) E ep = (infinitely brittle) (4.41)
When = 0, where is given by (4.24), we obtain (by definition of D) with (4.25) and
(4.26) that the dissipation of energy is given as
D = p + k = (|| ) = y 0 (4.42)
It is noted that (in this particular case) the mechanical dissipation is larger than the total
dissipation.
where is the drag-stress due to the isotropic portion of hardening, whereas is the
back-stress due to the kinematic portion of hardening.
For the considered uniaxial stress state, the yield criterion is now defined as
(, , ) = 0 ( red , ) = 0
red
Figure 4.6: Illustration of mixed isotropic and kinematic hardening in stress space.
The associative flow and (linear) hardening rules are then defined as
red
p = = red (4.48)
| |
k = = (4.49)
red
a = = red = p (4.50)
| |
We may now combine these relations with (4.21), (4.45) and (4.46). The pertinent differ-
ential equations are then obtained as
red
= E( p ) = E E (4.51)
| red |
= rH = rH p (4.52)
red
= (1 r)H = (1 r)H p (4.53)
| red|
With the proper initial conditions, these equations taken together with the loading criteria
of type (4.29) define the elastic-plastic constitutive relations for the considered material
model.
The characteristic response is illustrated for pure isotropic, pure kinematic and mixed
hardening, respectively, in Figure 4.7. By introducing kinematic hardening, it is possible
to pick up the Bauschinger effect, i.e. that the yield stress in compression, upon reversed
loading from tension, is smaller than it was in tension.
Remark: This reduction in compressive yield strength should not be confused with
the softening phenomenon, which means that the yield strength is reduced in tension
(compression) whilst the material is actually loaded in tension (compression). 2
In order to obtain a more realistic response in cyclic loading, we should resort to nonlinear
laws of hardening. For example, it is of value to model the asymptotic case of perfect
plasticity for large plastic strains, which corresponds to saturation of dislocations. As
to the specific choice of flow and hardening rules, can not be used as the potential
function (for the hardening rules). Instead, we introduce the plastic potential =
6 of
the following form:
2 2
= + + (4.54)
2 2
In this fashion, the flow rule will still be of the associative type, while the hardening rules
for k and a are both non-associative. The positive constants and are saturation
values of the drag-stress and the back-stress , respectively. We thus obtain the flow
and hardening rules
red
p = = red (4.55)
| |
Figure 4.7: Uniaxial stress versus plastic strain characteristics for (a) Linear isotropic
hardening, (b) Linear kinematic hardening, (c) Mixed linear isotropic and linear kinematic
hardening.
k = = 1 (4.56)
red
p
a = = = + (4.57)
| red |
The pertinent differential equations are now obtained as the slightly adjusted versions of
those in (4.51) to (4.53) as:
red
= E( p ) = E E red (4.58)
| |
= rH 1 (4.59)
red
p p
= (1 r)H = (1 r)H | | (4.60)
| red |
The characteristic response is shown schematically in Figure 4.8.
Figure 4.8: Uniaxial stress versus plastic strain characteristics for mixed nonlinear
isotropic and kinematic hardening.
Next we consider the cases of loading with increasing plastic strain, p > 0, and reversed
loading with decreasing plastic strain, p < 0, in further detail. At this analysis we denote
the assumed initial state by p0 , 0 and 0 . First we note that p = when p > 0, whereas
p = when p < 0, which follows from (4.55) upon observing that 0.
Loading
= y + + when p = (4.65)
Reversed loading
From the solutions (4.61), (4.62), (4.66) and (4.67) we obtain simply that
Hence, in this situation we may use the triangle inequality to conclude that
|| | | + || y + + (4.69)
|| y = 0 (4.70)
represents the limit criterion that can never be violated for any amount of accumulated
plastic strain.
We shall next consider the situation of cyclic loading with constant stress amplitude, and
we intend to investigate whether shake-down or ratchetting is predicted by the nonlinear
mixed hardening model. To this end we consider the extremes of pure isotropic hardening
and pure kinematic hardening, respectively. However, we shall refer to Figure 4.9 that
depicts the typical behavior in reversed loading followed by renewed loading when both
isotropic and kinematic hardening effects are present.
For isotropic hardening (r = 1), we obtain shake-down upon renewed loading if the
material yielded plastically at reversed loading. This may be seen formally as follows:
During reversed loading from 1 to 2 it is obvious that will increase from the value 1
to the value 2 . The question is whether the maximum stress max at 3 will be reached
before yielding takes place. This is indeed the case, since
max = y + 1 y + 2 (4.71)
The value of remains constant upon renewed elastic loading up to 2, whereafter renewed
plastic loading takes place. Eqn. (4.62) now gives the increase of from the value 2 to
Figure 4.9: Results for uniaxial stress showing characteristic behavior of stress and back-
stress at reversed and renewed loading for nonlinear mixed hardening model.
Upon combining (4.72) and (4.73), while introducing the plastic range p3 p2 = p , we
obtain
p 2 2 2
= ln (4.74)
H ( 3 )( + 1 )
Now, since the yield criterion is satisfied at 1 as well as at 3, we obtain (since = 0)
1 = 3 = max y (4.75)
where it was used that max > 1 . Moreover, since the yield criterion is satisfied also at
2, and min < 2 , we obtain
2 = min + y (4.76)
Upon inserting the values of 1 , 2 and 3 from (4.75) and (4.76) into (4.74), we obtain
p 2 (min + y )2
= ln (4.77)
H 2 (max y )2
Hence, no ratchetting is obtained when the cyclic stress variation is symmetric in tension
and compression.
In the general situation of mixed hardening we expect reduced ratchetting, that gradually
decreases with the accumulation of plastic strain. However, since the analytic solution
becomes quite complex in this case, the pertinent model behavior is most conveniently
assessed from numerical integration of the constitutive equations.
y H
= 0.001, = 0.1, 0 r 1, = = ay (4.79)
E E
y
[2y , 2y ], where y =
E
where a is a parameter that controls the amount of nonlinear hardening. Linear hardening
is defined by a = , whereas the chosen nonlinear hardening model is characterized by
the choice a = 0.25. Kinematic hardening is defined by r = 0, whereas isotropic hardening
is defined by r = 1.
1.5 1.5
a b
1 1
0.5 0.5
stress
stress
0 0
0.5 0.5
1 1
1.5 1.5
2 2
2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5 2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5
strain 3 strain 3
x 10 x 10
1.5 1.5
c d
1 1
0.5 0.5
stress
stress
0 0
0.5 0.5
1 1
1.5 1.5
2 2
2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5 2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5
strain 3
x 10 strain 3
x 10
1.5 1.5
e f
1 1
0.5 0.5
stress
stress
0 0
0.5 0.5
1 1
1.5 1.5
2 2
2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5 2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5
strain 3
x 10 strain 3
x 10
Figure 4.10: Predicted results from cyclic straining of linear and nonlinear hardening
models. Results for linear hardening are shown in (a,c,e), whereas results for nonlinear
hardening are shown in (b,d,f).
Applying the fully implicit (Backward Euler) integration rule to (4.51), we obtain
n+1
n n+1 red def
=E n+1 red
with n+1 ered = |n+1 red |, = t (4.80)
t t t e
which can be rewritten as
n+1 n+1 tr E n+1 red
= n+1 red
(4.81)
e
n+1 tr
where is the elastic trial stress defined by
n+1 tr
= n + E (4.82)
We shall use implicit integration of (4.52) and (4.53) as well, whereby we obtain
n+1
= n + rH (4.83)
n+1 (1 r)H n+1 red
= n +
n+1 red
(4.84)
e
In principle, can be solved from (4.81), (4.83) and (4.84) together with the loading
conditions
n+1
0, 0, n+1 = 0 (4.85)
Upon combining (4.81) and (4.84), we obtain
[E + (1 r)H] n+1 red
1+ n+1 red
= n+1 red,tr with n+1 red,tr
= n+1 tr
n (4.86)
e
Taking the absolute values of both sides, we obtain
n+1 red n+1 red,tr n+1 red,tr
e = e [E + (1 r)H] with e = |n+1 red,tr | (4.87)
n+1 red,tr
and it is noted that e is a known quantity when has been prescribed. Now,
we may introduce the updated stresses from (4.83) and (4.87) into (4.47) to obtain the
updated yield function
n+1 n+1 red n+1 n+1
= e y = tr h 0 (4.88)
h=E+H (4.90)
Whether plastic loading or elastic unloading is at hand can be determined with use of the
complementary conditions in (4.85). We distinguish between the following conditions:
n+1
Loading (L) is defined by = 0 and > 0, which gives the solution
n+1
tr n+1
= > 0 when tr > 0 (4.91)
h
n+1
Unloading (U), on the other hand, is defined by 0 and = 0, which obviously
gives
n+1
= 0 when tr < 0 (4.92)
Updated solution
When has been determined, it is possible to compute the updated values of all state
variables of interest. First, we obtain from (4.86) that
n+1 red
n+1 n+1 e n+1 red,tr
= (4.93)
n+1 red,tr
e
which may be inserted into (4.81) and (4.84) to give the updated solution
n+1 E
= n+1 tr n+1 red,tr
= c1 n+1 tr + (1 c1 )n (4.94)
n+1 red,tr
e
n+1
= n + rH (4.95)
n+1 (1 r)H n+1 red,tr
= n + = c2 n+1 tr + (1 c2 ) n (4.96)
n+1 red,tr
e
where
E (1 r)H
c1 = 1 red,tr
, c2 = (4.97)
n+1 n+1 red,tr
e e
In the very simplest situation of perfect plasticity, defined by H = 0 and n = n = 0, we
obtain from (4.91), in the case of (L), that
1
= (|n+1 tr | y )
E
which may be inserted to (4.94) to give
n+1 tr
n+1
= y
|n+1 tr |
Remark: In the present case of linear hardening, it appears that the crucial equation
(4.88) is linear in . In fact, it can be shown that this linearity carries over to the
multiaxial situation. 2
Summary of algorithm
1. Given tr n+1 tr
= n
+ E
2. Check L/U
If n+1 tr 0, then = 0
n+1 tr
else = > 0, h = E + H
h
3. Update solution
n+1 E
= c1 n+1 tr + (1 c1 )n , c1 = 1
n+1 red,tr
e
n+1
= n + rH
n+1 (1 r)H
= c2 n+1 tr + (1 c2 ) n , c2 =
n+1 red,tr
e
Closest-point-projection-method (CPPM)
It appears that we may rewrite (4.81), (4.83) and (4.84) in a slightly more general fashion
as
n+1
n+1 n+1 tr
= E (4.98)
n+1
n+1 n
= rH (4.99)
n+1
n+1 n
= (1 r)H (4.100)
where we directly used the characteristics of the flow and the hardening rules. We also
repeat, for completeness, the loading conditions (4.85)
n+1
0, 0, n+1 = 0 (4.101)
Let us now introduce the convex set E of plastically admissible dissipative stresses as
E = {(, , ) | (, , ) 0} (4.102)
where we (implicitly) used that is the dissipative stress that is energy conjugated to p .
We are now in the position to establish the following important theorem:
Theorem: The updated solution (n+1 , n+1 , n+1 ) of (4.98) to (4.101) is also the solution
of the convex minimization problem as follows:
" #
Min
(n+1 , n+1 , n+1 ) = arg (, , ) (4.103)
(, , ) E
where is defined as
1 1 n+1 tr 2 1 n 2 1 n 2
(, , ) = ( ) + ( ) + ( ) (4.104)
2 E rH (1 r)H
which has a unique solution when H 0. It thus appears that Eqns. (4.98) to (4.101)
represent the Kuhn-Tucker problem corresponding to the minimization problem (4.103).
Proof: The KT-conditions of (4.103) are also the stationary conditions of the Lagrangian
function
n+1
(, , , ) = (, , ) + (, , ) (4.105)
The updated solution is illustrated in Figure 4.11 for the special case that r = 1, which
means that there is no kinematic hardening. Considering, for example, the situation that
n+1 tr
> 0, we may combine (4.94) and (4.95) to the vector equation in the (, )-space
as follows:
(n+1 , n+1
) = (n+1 tr , n ) (E, H) (4.108)
which relation is illustrated in Figure 4.11(b). We have thus established that (n+1 tr , n+1 , n+1 )
is the projection of (n+1 tr , n , n ) onto the convex set E in the particular metric defined
Figure 4.11: Updated solution (n+1 , n+1 ) for isotropic hardening, (a) Restriction to
- relation, (b) Space of dissipative stresses (, ).
by the norms defined in the quadratic form . This is the reason why the BE-method
applied to the plasticity problem is also known in the literature as the Closest-Point-
Projection-Method. The projection will also be denoted as the operator CPPM in the
mapping
(n+1 , n+1
, n+1
) = CPPM{n+1 tr , n , n ; E, rH, (1 r)H} (4.109)
Like for viscoelasticity, the behavior of an elastic-plastic structure will depend on its
statical (in)determinacy. We shall then first outline some general features, which will be
illustrated later in conjunction with the discussion of truss structures.
Isostatic structures
We recall that, for an isostatic structure, the stress distribution is uniquely determined
by the applied load. Considering, for example, a truss with bars of perfectly plastic
material, we conclude that the maximum load bearing capacity can simply be calculated
as the load for which the normal stresses in the most severely stressed bar has reached
the yield point. At this load level, the structural tangent stiffness matrix is zero and the
truss behaves like a mechanism, i.e. it can not sustain any further load increase without
excessive geometrical distortion. The considered load is thus the limit load.
Hyperstatic structures
We recall from Subsection 3.4.1 that, for a hyperstatic structure, the stress distribution
is not uniquely defined by the load but is determined also by the material properties. For
example, for the truss with perfectly plastic material in the bars, the statical redundancy
is reduced during the load increase each time a new bar is yielding plastically until the
structure (truss) has become isostatical. For further load increase, the discussion of the
isostatic truss (above) applies so that the limit load is achieved when a mechanism has
been obtained.
We shall analyze the truss in Figure 3.8 in the same fashion as for nonlinear viscoelastic
response, and we refer to Figure 3.8 for the introduced definitions and notation. Hence, the
where
(n)i
itr = n i + Ei ()i with ()i = (4.111)
Li
n+1 def
In analogy with (3.136), we now obtain the corresponding normal forces Ni = Ni as
Ni = Ai itr () (4.112)
which are collected in the column matrix N (n). The equilibrium and kinematic rela-
tions for the truss are still those of (3.138), i.e.
P = AT N and n = A p (4.113)
and iterations are required in order to find the solution p from (4.113)1. For given load
n+1
P , we may device the following Newton procedure (k being the iteration count):
p(k+1) = p(k) + p
(AT N )
S ep(k)
a p = (P AT N (k) ) with S ep
a = (4.114)
(p)
and with p(0) chosen as the converged value of p in the previous timestep. It thus fol-
lows that S ep
a is the ATS-matrix for elastic-plastic behavior, which is a nonlinear function
of the incremental solution p for a given timestep.
where, for each bar, the algorithmic tangent stiffness modulus Eaep is defined as d()/d().
Remark: For linear mixed hardening Eaep E ep , where E ep = EH/h is the tangent
stiffness modulus defined already in (4.36). Show this as homework! 2
In accordance with the discussion for a general structure, the cross-section moment for
which the most stressed fiber starts to yield plastically is denoted the elastic limit mo-
ment Mel . If the material is assumed to be perfectly plastic, then the limit moment Ml
corresponds to plastic yielding in the whole cross-section, i.e. || = y everywhere.
Subsequently, we consider (for the sake of simplicity) a double-symmetric cross-section.
Figure 4.12 shows the situation in the elastic stage, the elastic-plastic stage and the (fully)
plastic stage, respectively.
It follows trivially that
2I
Mel = W y with W = (4.119)
h
When the plastic zone has advanced to the location defined by z = zy , the corresponding
stress distribution is defined as
(
y zzy , 0 |z| zy
= z
(4.120)
y |z| , zy |z| h2
h
2 zy
zy=0
zy
h
2
In the fully plastic regime, i.e. when zy = 0, we have Iy = 0 and Zy = 0, which inserted
into (4.121) gives the limit moment
Ml = Zy (4.123)
This relation is shown in Figure 4.13 together with typical relations for other common
cross-sectional shapes. 2
Figure 4.13: Relation between moment and curvature for various cross-sections.
Hardening plasticity
n+1 def
In the case of linear hardening, = was defined in (4.94) as
= c1 tr + (1 c1 )n (4.131)
where
tr = n + E with = c z (4.132)
The Newton-type iteration procedure to calculate c for a given moment M at t = tn+1
reads
(k+1)
c = (k)
c + c (4.133)
d(M)
c = (Saep )1 (M M (k) ) with Saep = (4.134)
d(c )
(0)
and with c chosen as the converged value of c in the previous timestep. Hence, we
conclude that Saep becomes
Z h/2 nint
X
ep ep 2
Sa = Ea (z)z b(z)dz = 2 (Eaep )i Ii with Ii = zi2 (A)i (4.135)
h/2 i=1
Remark: To find the algorithmic bending stiffness Saep is straightforward for the case
of linear hardening, since it will coincide with the tangent bending stiffness S ep in this
particular case. 2
Z h+ nint
X
M= (z)zb(z)dz = i zi Ai (4.137)
h i=1
which must equilibrate the prescribed normal force N and moment M . We may compute
= ((z)) from (4.131) with
(z) = + zc (4.138)
N(, c ) = N (4.139)
M(, c ) = M (4.140)
h c
t>0 M
y
t=0 N
z
h+
z (z) = + c z
Z h+ nint
X
ep
Sa,NM = Eaep (z)zb(z)dz = (Eaep )i zi Ai (4.144)
h i=1
Z h+ nint
X
ep
Sa,MM = Eaep (z)z 2 b(z)dz = (Eaep )i Ii (4.145)
h i=1
Moreover, by assuming that the y-axis is located at the center of gravity and choosing
Eaep = E, we obtain
ep ep ep
Sa,NN = EA, Sa,NM = 0, Sa,MM = EI (4.146)
VISCOPLASTICITY
5.1 Introduction
The idea of overstress, i.e. there exists a threshold value of stress that must be exceeded
before time-dependent inelastic deformation will take place, was invented in the 1920s
by Bingham for metals (while employing von Mises quasistatic yield surface), and it was
generalized by Hohenemser & Prager to also include hardening of the quasistatic yield
surface. However, this is only a special case of the specific formulation of viscoplasticity of
Perzyna, which applies to general quasistatic yield criteria. In this way, it is possible to
treat the time-dependent response of cohesive, as well as frictional, materials in a unified
fashion. For example, viscoplasticity based on the Perzyna concept has been used with
considerable success to model creep (secondary consolidation) of soft soils, such as clay.
Very sophisticated models for simulating the rate-dependent response of metals, including
classical creep and relaxation situations, have been developed. Models that are able to
represent complex features such as time-recovery of back-stress, strain-range memorization
at cyclic loading, etc., have been proposed.
A special class of viscoplastic models do not possess any quasistatic yield surface at
all (which corresponds to zero threshold stress), i.e. the models are essentially of the
nonlinear viscoelasticity type, although they are termed viscoplastic in the literature. It
thus appears that viscoplasticity has much in common with nonlinear viscoelasticity; the
main difference being the concept of an elastic region in stress space (like in plasticity).
For example, it is possible to retrieve the simple Maxwell viscoelastic model as a special
case of viscoplasticity.
In recent years the format of viscoplasticity put forward by Duvaut & Lions has been
advocated. As it turns out, this formulation is the natural extension of the underlying
rate-independent plasticity problem when it is set in a incremental format (after integra-
tion using the Backward Euler rule). In fact, the Perzyna and Duvaut-Lions versions of
plasticity can be obtained as special subclasses within a more general framework.
As the prototype for perfectly viscoplastic behavior, we consider the rheological model
in Figure 5.1, which is often denoted the Bingham model. Like for rate-independent
plasticity, the plastic slider is inactive as long as || < y , where y is the quasistatic
yield stress. This indicates that y is the proper yield stress only at very slow loading.
Hence, the major difference in comparison to (rate-independent) plasticity is that the
stress is allowed to exceed the (quasistatic) yield stress, i.e. || > y , since stress can be
transferred to the viscous dashpot when the frictional resistance of the slider has been
exhausted. As the single internal variable we take the viscoplastic strain p in the slider
and the dashpot, and the expression for the free energy is chosen (like in the case of
plasticity) as
1 1
= E(e )2 = E( p )2 (5.1)
2 2
where e = p is the elastic strain of the Hookean spring with modulus of elasticity E.
We then obtain the constitutive equation for the stress as
= = E( p ) (5.2)
Vol 0 March 7, 2006
5.2 Prototype rheological model for perfectly viscoplastic behavior 111
t ,()
*
e p
= || y (5.4)
Creep
In the particular case when = 0 > 0 is held constant (after rapid loading), we obtain
the solution of (5.8) as
0 t
(t) = + (0 ) (5.9)
E t
which may be compared with the expression in (3.108) pertinent to the Norton law.
Relaxation
When (t) is a given function, we may solve for (t) directly from (5.8). As long as the
stress is small enough to satisfy y , then (5.8) has the trivial elastic solution = E.
Let us next consider the situation when the yield stress has been exceeded. In the relax-
ation situation, when = 0 > 0 is held constant (after rapid loading), then the stress
must be solved from
E
+ () = 0, (0) = 0 = E0 > 0 (5.10)
t
It appears from (5.10) that the stress is decreasing monotonically until the stage has been
reached (after long time) when = 0. This holds independently of the parameter values
E and t as well as of the explicit choice of the overstress function ().
The simplest possible model, that features viscoplastic behavior, is the Bingham model.
This model is defined by the choice
hi h|| y i
() = = (5.11)
E E
where hi is the McCauley bracket defined as
(
if > 0
hi = (5.12)
0 if 0
Creep
0 h0 y i t
(t) = + (5.14)
E E t
This solution is shown in Figure 5.2(a).
Relaxation
(2)
0 >y
(2)
(2)
0 -y
0 E (2)
0 (2)
1 0 >y
E
(1)
0 (1)
0 < y (1)
0 (1)
0 < y
E
t t
t t
* *
(a) (b)
Figure 5.2: (a) Creep and (b) Relaxation curves for a Bingham-material.
||
() = = (5.17)
E E
Upon introducing the viscosity parameter = Et , we may insert (5.17) into (5.13) to
obtain
E
+ = E (5.18)
A common choice of (), that defines the generalized Norton viscoplastic law, is given
as
n n
hi c h|| y i c nc
c
() = = and t = (5.19)
E E E
where c is the creep modulus, is the relaxation time and nc is the creep exponent (as
introduced in Chapter 3). The classical Norton creep law is retrieved from (5.19) when
y = 0, in which case we may insert (5.19) into (5.5) and (5.8) to obtain
nc nc
p 1 || E ||
= , + = E (5.20)
t E || t E ||
It is of some interest to assess the behavior of the prototype model when t = 0 and
t = , respectively:
In the case that t = 0, it can be shown from (5.8) that () = 0, i.e. = 0, which
means that the plastic consistency condition is satisfied at all times independent of the
loading. Hence, this viscoplastic model coincides with the plasticity model when t = 0.
This is the reason why the elastic-viscoplastic model can be considered as a regularization
of the corresponding elastic-plastic model.
In the case that t = , it follows directly from (5.8) that = E , i.e. we obtain the
elasticity solution. From the model point of view, this case corresponds to a rigid dashpot
(or slider with infinite yield stress).
1 1
= E( p )2 + Hk 2 (5.21)
2 2
which gives, once more, the constitutive relations
= = E( p ) = p (5.22)
= = Hk (5.23)
k
Vol 0 March 7, 2006
116 5 VISCOPLASTICITY
E y
t ,()
*
e p
(, ) = || y (5.24)
Like in the case of perfect viscoplasticity, we may solve for (t) and (t) when (t) is
a prescribed function (creep), or we may solve for (t) and (t) when (t) is prescribed
(relaxation). Solutions must in practice be obtained using numerical integration, except
in the particularly simple case when is a linear function, which defines the Bingham
model. This model is considered next.
Choosing () as in (5.11), we obtain from (5.27) and (5.28) the set of governing equations:
1
+ h|| y i = E (5.29)
t ||
H
h|| y i = 0 (5.30)
Et
Creep
Consider the creep situation, when = 0 > 0 is held constant (after rapid loading).
Upon using the initial conditions (0) = 0 /E and (0) = 0, we obtain the solutions of
(5.29) and (5.30) as
0 h0 y i H t
(t) = + 1 e E t (5.31)
E H
H t
(t) = h0 y i 1 e E t (5.32)
It is noted that (t) 0 when t . Hence, this model is less suitable for describing
real creep behavior, although it can be used to describe the primary stage (stage I) for
small times. In order to mimic the transient as well as the stationary stages (stages I and
II), we must resort to nonlinear hardening characteristics.
Relaxation
In the case of relaxation, when = 0 > 0, we obtain from (5.29) and (5.30) the problem
1
+ h y i = 0, (0) = 0 = E0 (5.33)
t
H
h y i = 0, (0) = 0 (5.34)
Et
(E/h) y + (H/h) 0
0 - y 0 - y
E H
0 1 y
E
t t
t t
* *
Stage I
(a) (b)
Figure 5.4: (a) Creep and (b) Relaxation solution for linear hardening of the Bingham
material when 0 > y .
As an alternative to the formulation of Perzyna, we may formulate the flow and hardening
rules (that are pertinent to the quasistatic yield surface) in the spirit of a formulation
first suggested by Duvaut and Lions (1972). With a slight generalization of their
formulation we propose
1 1
p = D () ( s ) (5.37)
t E
1 1
k = D () ( s ) (5.38)
t H
Since E and H are assumed to be constants, it may be shown that the optimality condi-
tions corresponding to the constrained minimization problem in (5.41) are:
s
s s s
= E = s E s (5.44)
| |
s
s s
= H = + s H (5.45)
where the Lagrangian multiplier s can be calculated if (5.44) and (5.45) are subjected
to the complementary conditions
s 0, s 0, s s = 0 (5.46)
The Bingham model is defined by choosing () as in (5.11). Upon inserting this ex-
pression in (5.49), we retrieve the same model within the framework of Duvaut-Lions
formulation by choosing
H
D () = 1 + (5.50)
E
This expression can be inserted into (5.42) and (5.43) to give the constitutive rate equa-
tions
1 H
+ 1+ ( s ) = E (5.51)
t E
1 H
+ 1+ ( s ) = 0 (5.52)
t E
Remark: The original form of the Duvaut-Lions formulation, cf. Simo & Hughes
(1988), seems to be
1
+ ( s ) = E (5.53)
t
1
+ ( s ) = 0 (5.54)
t
rather than (5.52) and (5.53). From the discussion above, it should be clear that the
relations (5.54) and (5.55) are equivalent to those of Perzyna only in the special case
when H = 0, in which case D () = 1. 2
In the case of uniaxial stress, the relevant quasistatic yield function was given in (4.47)
as:
= | red | y , red = (5.55)
and the constitutive differential equations according to the Perzyna formulation become
E red
+ () red = E (5.56)
t | |
rH
() = 0 (5.57)
t
(1 r)H red
() red = 0 (5.58)
t | |
These equations should be compared with those of rate-independent plasticity given in
(4.51) to (4.53), and it turns out that they are quite similar.
When the Backward Euler method is used to integrate Eqns. (5.56) to (5.58), we obtain
(formally) the same expressions for the updated stresses as in rate-independent plasticity
that were given in (4.94) to (4.97). The expressions are repeated here for completeness:
n+1 n+1 tr
= c1 + (1 c1 )n (5.59)
n+1
= n + rH (5.60)
n+1 n+1 tr
= c2 + (1 c2 )n (5.61)
where
E (1 r)H
c1 = 1 , c2 = (5.62)
n+1 red,tr n+1 red,tr
e e
n+1
t
y() = () =0 (5.63)
t
n+1
where () is given as
n+1
() =n+1 tr hep with hep = E + H (5.64)
t
(n+1tr hep ) = (5.65)
t
Let us next consider the simple model of Bingham, defined in (5.11). From (5.65) we
obtain
<> Et
() = = <n+1 tr hep > = (5.66)
E t
n+1
In the case of loading (L), defined as 0 and 0, we obtain from (5.66)
n+1 tr
n+1 tr
= > 0 when >0 (5.67)
hevp
Summary of algorithm
1. Given tr n+1 tr
= n
+ E
2. Check L/U
n+1
If tr 0, then = 0
n+1 tr
else = evp > 0
h
3. Update solution
n+1 n+1 tr E
= c1 + (1 c1 )n, c1 = 1
n+1 red,tr
e
n+1
= n + rH
n+1 n+1 tr (1 r)H
= c2 + (1 c2 )n, c2 =
n+1 red,tr
e
Box 5.1: Solution algorithm for linear mixed hardening in the Perzyna formulation of viscoplasticity.
The structural analysis is carried out in exactly the same fashion as for rate-independent
plasticity. For example, the analysis of a truss is the same as in Subsection 4.5.2 if only
the ATS-matrix S ep evp
a is replaced with S a . We shall then need the algorithmic stiffness
Eaevp , defined by the relation
d()
Eaevp = (5.69)
d()
It is then remarked that the corresponding CTS-relations in viscoplasticity (which would
be denoted E evp ) does not exist!
Remark: For linear mixed hardening of Perzyna viscoplasticity, we obtain
evp E
Ea = E 1 evp (5.70)
h
In this chapter we introduce the concept of distributed (or continuum) damage. A model
for damage coupled to elastic deformation is presented. We also touch upon the issue of
how to experimentally quantify damage.
Close to the state of failure the microstructure of any engineering material will start to
disintegrate or break up. The physical nature of this deterioration is, of course, not the
same for different materials. For example, in a metal the deformation accelerates due to
the simultaneous propagation of microcracks and growth of microcavities. These defects
initiate due to severe stress concentrations in the neighborhood of in-situ inclusions and
interfaces, e.g. along the grain boundaries. Eventually, these microcracks and microvoids
coalesce to form an incipient macroscopical crack. From the material point of view, failure
has already occurred at this point, which may be well in advance of the stage where the
crack can be observed.
The process of successive material degradation may be modelled by damage theory. New
(internal) damage variables are introduced to represent the degree to which the material
has degraded. Clearly, it would be desirable to use this concept throughout the deforma-
tion process up to and including the localized mode of failure that represents a partially
open crack. Then the validity of continuum theory could be extended up to complete fail-
ure. However, as pointed out already in Chapter 1, the traditional approach is to employ
classical linear fracture mechanics in order to assess the possibility for further propagation
of the (pre)existing crack. The main purpose is then to establish whether the crack will
propagate in a stable or unstable manner at increased loading, and the latter situation is
viewed as structural failure.
The development of damage may be linked to the development of inelastic (plastic, creep,
et.) strain, or it may be governed by a separate criterion (that is analogous to the yield
criterion). The first approach is feasible for the modeling of ductile failure and creep
failure, as demonstrated in Chapters 7 and 8, whereas the latter is useful in conjunction
with brittle behavior, which is shown more explicitly in this Chapter. The following
classification of damage-related phenomena is commonly made:
Brittle damage (also denoted elastic damage): Only elastic strains occur (in the
intact material), and damage develops with the total strain after a certain threshold
of strain has been exceeded.
Ductile damage: Damage develops with plastic strains after a threshold of plastic
strain has been exceeded. Plastic strains and damage may localize at incipient
failure, cf. Figure 1.2.
Creep damage: Damage dominates the tertiary creep phase, and is enhanced at
elevated temperature. Intergranular decohesion is pronounced, cf. Figure 1.7.
Low-cycle fatigue: Ductile failure terminates the cyclic loading process, cf. Fig-
ure 1.13.
High-cycle fatigue: Brittle failure terminates the cyclic loading process, cf. Fig-
ure 1.9.
A major borderline can be drawn between the models in which damage is coupled to
the total strain (elastic damage) and those models in which damage is coupled to the
inelastic part of the strain (plastic and viscoplastic damgage including low-cycle-fatigue
and creep). The principal behavior is sketched in Figure 6.1.
Figure 6.1: Undamaged and damaged (a) Elastic material, (b) Elastic-plastic material.
The debonding mechanisms, that are represented as damage, are different in metals and
alloys, polymers, composites, ceramics, concrete and wood. It should be noted that
damage is not the same thing as deformation, although it may be coupled to deformation.
Let us, for example, consider a metal. Plastic slip, which is enhanced by a high dislocation
density, occurs without debonding. However, dislocation movements can be stopped by
an inclusion in the lattice in such a way that plastic deformation can not continue without
braking the atomic bonds, i.e. resulting in damage. In this way a microcrack is created.
Since the number of atomic bonds (which is a measure of the elastic stiffness) decreases
with damage development, it follows that the elastic modulus decreases with damage.
Hence, the development of damage in a metal is coupled to the development of inelastic
strain, but it is the reduced elasticity modulus that is a measure of damage, as will be
shown below.
The physical nature of damage for a variety of engineering materials are listed as follows:
Metals and alloys: Nucleation of microcracks at inclusions within the grains and
the matrix. Debonding between grains and matrix along weakened interfaces.
Fiber composites: Debonding between fibers and the polymeric or metallic ma-
trix.
Consider a unit area of the material which has been damaged so that the area portion
d is completely broken and can not sustain any stress. The scalar damage variable d is
the simplest form of representing this situation. It is clear that d = 1 corresponds to a
completely deteriorated material.
Remark: The material will reach its ultimate strength at a certain finite value dF , which
may be taken as the failure value. For larger damage, the stress-strain relation shows
softening. 2
The view of damage taken above leads to the Equivalent Strain Principle: The material
is considered as if it would consist of parallel fibers subjected to the same strain. At a
certain state the area portion d of the fibers are broken, and the remaining portion 1 d
is intact, as shown in Figure 6.2. For given applied nominal stress , the reduction in
stress-carrying area leads to a corresponding increase of stress in the undamaged portion.
This is the intrinsic or effective stress , which notion was first introduced by Kachanov
(1958). A simple equilibrium equation for the unit area of material (see Figure 6.2) gives
= (6.1)
1d
Figure 6.2: Elementary model of elastic damage based on the Equivalent Strain Principle.
Remark: This is a very special case of a two-phase material with parallel coupling of
the two phases, where the extreme view is taken that one of the phases has completely
deteriorated. 2
A consequence of the Equivalent Strain Principle is that any constitutive functional rela-
tionship expressing the behavior of undamaged material will also be valid for the damaged
material if only the nominal stress is replaced by the effective stress. For example, assum-
ing that the virgin (undamaged) material obeys the constitutive law
= f (, kd ) (6.2)
= f (, kd ) (6.3)
1
(, d) = = 2 (6.11)
2E(1 d)
Since /d is the release rate of elastic stress energy during damage growth at constant
nominal stress, the dissipative force is sometimes denoted the damage energy release
rate. This definition is analogous to the definition of the fracture energy release rate Gf
in fracture mechanics, cf. Chapter 9.
Damage criterion
Similarly to the situation in plasticity, we may introduce a damage criterion (rather than
a yield criterion) expressed as = 0, where (, d) is the damage function. In the case
that < 0 the response is elastic, whereas the case = 0 admits the possibility for
development of damage. A quite general class of damage criteria, that couples damage to
the total elastic strain, may be expressed in the form
= g(d) (6.13)
For reasons that will be evident later, we shall here consider the specific choice
23
m 2 E 6S
g(d) = gmax [1 (1 d) ] , gmax
3 = (6.15)
2 Em
where S is the damage modulus (dimension N/m2 ) and m is the damage exponent, both
of which govern the rate of damage development.
Stress-strain relation
Ef d
1
m=3 1
m=2
m=1
m=1 2 3
f f
1 1
(a) (b)
0, 0, = 0 (6.24)
We may use the definition of and the damage law in (6.23) to obtain
= + d = E + 0 (6.25)
d d
In order to conform the formulation with the elastic-plastic behavior (discussed in Chapter
4), we may rewrite (6.25) as
tr h 0 (6.26)
where we have introduced the elastic rate tr and the inelastic (damage) modulus h as
tr = E (6.27)
g 2m 1
h= = = gmax [1 (1 d)m ] 3 (1 d)m1 > 0 (6.28)
d d 3
Vol 0 March 7, 2006
134 6 DAMAGE AND FRACTURE THEORY
Ef d
1
m=3 1
m=2
m=1 3
m=1 2
f f
f 1 f 1
(a) (b)
> 0 (6.29)
E
= (6.30)
h
Upon inserting (6.30) into (6.23), we may express the damage law as
E E||
d = = = = m1
(6.31)
h h h || S(1 d) ||
The last expression of (6.31), which is obtained using the definition of h in (6.28) and
noting that E|| = (2E)1/2 , is quite interesting, since it reminds strongly about the
formulation of the damage law that is frequently used in the context of damage coupled
to plasticity, as will be discussed in some detail in Chapter 7.
Remark: The damage law in (6.31) can be taken as the basic law, rather than the
postulated damage criterion in (6.15). In fact, integration of the damage law will give
precisely the criterion in (6.15) and, as shown above, the stress-strain behavior that is
valid for loading. 2
It is now a simple matter to obtain the tangent relation by first differentiating (6.5) to
obtain
+ d = (1 d)E (6.32)
and then inserting the damage law in (6.32) to finally obtain the relation
1 2 ||
= (1 d)E = (1 d)E (6.33)
h S(1 d)m1
Finally, we conclude that elastic unloading without development of damage will take place
if
0 (6.34)
We thus obtain the tangent relation
= (1 d)E (6.35)
where the value d = d remains constant during unloading. It is simple to integrate (6.35)
to obtain the unloading relation
= (1 d) (6.36)
Ef f
It is noted that the unloading stress-strain path returns straight to the origin, i.e. (6.36)
represents a secant relation. This extreme situation is quite unrealistic for most materials,
but has nevertheless been used in modeling. For concrete, this behavior was denoted
elastic-fracturing by Dougill (1976). The more realistic behavior, that there is some
irreversible strain after unloading, is discussed in the following Chapters.
A review of the more significant methods to measure damage in the special case of uniaxial
stress is given as follows:
Degradation of elasticity
As stated above, damage can be measured from the unloading modulus at unloading,
following monotonic loading, as the relation
E
d=1 (6.37)
E
Vol 0 March 7, 2006
136 6 DAMAGE AND FRACTURE THEORY
If the stress state is uniaxial, there is a simple relation between the (axial) wave velocity
vL in the bar and its elasticity modulus. Provided that the density is unaffected by the
damage, we have the relations
E E
vL 2 = , vL2 = (6.38)
Using this expression in (6.8), we obtain the alternative measure of damage as
E v 2
d=1 = 1 L2 (6.39)
E vL
Estimates of damage from fatigue and creep failure tests are discussed in Chapters 10 and
8.
DAMAGE COUPLED TO
PLASTICITY
In order to describe ductile fracture under monotonic loading and fatigue failure under
cyclic loading, it is necessary to couple damage to the development of plastic deformation.
(The physical explanation for this assumption was given in Chapter 6). In particular, low-
cycle fatigue (LCF) is the consequence of damage development under cyclic loading, which
is discussed in further detail in Chapter 9.
The presence of material damage means that part of the material has lost its bearing
capacity, as illustrated schematically for the prototype model in Figure 7.1.
As a result, the free energy is reduced, and we may consider a prototype model based on
the following expression for (, p , d):
1
= (1 d)E( p )2 (7.1)
2
from which Colemans equations give
= = (1 d)E( p ) = (1 d) with = = E( p ) (7.2)
1d
Figure 7.1: Prototype model of elastic-plastic damage (based on the Equivalent Strain
Principle).
p = = (1 d)E( p ) (7.3)
p
1 1 2
= = E( p )2 = (7.4)
d 2 2E
It is noted that the expression = /d for the damage force is exactly the same as
for damage coupled to elastic deformation, that was discussed in the previous Chapter.
We now introduce the yield function = ()
() = || y (7.5)
which corresponds to a perfectly plastic behavior of the undamaged material. The plastic
potential = (, , d) is proposed as
2
= + 6= (7.6)
2S(1 d)m
The constitutive rate equations for p and d are obtained as the associative flow rule and
the non-associative damage rule as:
p = = = (7.7)
1 d 1 d ||
2 2 p
d = = m
= m1
0 (7.8)
(1 d) 2ES 2ES(1 d) ||
Remark: The last expression of (7.8), which was obtained upon using (7.7), is completely
equivalent to that obtained for elastic coupling in (6.30); the only difference is that is
now replaced by its plastic part p . 2
From (7.7), we obtain
= (1 d)pe with pe = |p | (7.9)
Together with (7.8), this equation is solved when subjected to the loading criteria (com-
plementarity conditions):
0, () 0, () = 0 (7.11)
= 0 (7.12)
Upon inserting (7.10) into (7.12), we obtain
= E h 0 (7.13)
||
E
h= (7.14)
1d
tr = E (7.15)
||
which is precisely the same loading criterion as for plasticity without damage. At loading
(L) we thus obtain as
tr
= = (1 d) > 0 (L) (7.16)
h ||
and we may insert this expression into (7.7) and (7.8) with (7.4) to obtain
p = (L) (7.17)
||
d = (L) (7.18)
2ES(1 d)m1
Finally, we obtain the tangent stiffness relation by combining (7.2), (7.10), (7.17) and
(7.18):
||3
= (1 d) d = (L) (7.19)
2ES(1 d)m1
Remark: We note that = 0 due to perfectly plastic response of the undamaged mate-
rial. Moreover, the strain can be considered as a fixed parameter during the process of
unloading. 2
Stress-strain relation
In this simple case it is possible to integrate (7.18) and (7.19). By using the yield criterion
(7.5), i.e. || = y at loading, we first obtain tangent relations
y 2
d = sign () (7.20)
2ES(1 d)m1
y 3
= (7.21)
2ES(1 d)m1
Next, we solve (7.20) and (7.21) for two different load cases representing tensile and
compressive states.
Load case I
Assume that = y > 0, > 0 ( tr > 0). The initial conditions d = 0 and = y , when
= y , give the solutions of (7.20) and (7.21):
m1
y 2 m
d=1 1 ( y ) , y (7.22)
2ES
m1
y 2 m
= y 1 ( y ) , y (7.23)
2ES
Load case II
Assume that = y < 0, < 0 ( tr > 0). The initial conditions d = 0 and = y ,
when = y , give the solutions of (7.20) and (7.21):
m1
y 2 m
d=1 1+ ( + y ) , y (7.24)
2ES
m1
y 2 m
= y 1+ ( + y ) , y (7.25)
2ES
Alternatively, we may eliminate the parameter S in terms of the fracture strain f . For
example, in Load Case I we obtain d = 1 when = f , which gives
S y 2
= (f y ) (7.26)
m 2E
where s is a non-dimensional ductility measure. For metals, s is in the range 100-200 for
ductile damage, while s may be as small as 1-2 for brittle damage. 2
Upon introducing the expression (7.27) into (7.22) and (7.23), we obtain
m1
f
d=1 , y (7.28)
f y
m1
f
= y , y (7.29)
f y
Let us next consider elastic unloading (E), which takes place while the damage is constant,
In this particular case, the damage value d corresponds to the strain via the
d = d.
relation (7.28). Hence, the stress rate is given by the rate law
m1
= f
= (1 d)E E (7.30)
f y
Characteristic stress-strain and damage-strain relationships are depicted in Figure 7.2 for
the choice m > 1. (The values m < 1 are less realistic). The influence of the value of m
on the characteristics of these relationships is demonstrated in Figure 7.3. Bilinear curves
are obtained in the special case that m = 1.
Figure 7.2: (a) Typical stress-strain and (b) Damage-strain relationships (for m > 1)
showing loading and unloading characteristics.
Let us consider the behavior when f . From (7.27) and (7.28) we obtain
m1 1
dd 1 f
= 0 (7.31)
d m(f y ) f y
m1 1
d y f
= 0 (7.32)
d m(f y ) f y
It is clear that both dd/d and d/d tend to zero when f in the case m < 1, whereas
they tend to infinity in the case m > 1. The latter behavior is more in accordance with
experimental evidence.
Figure 7.3: (a) Stress-strain and (b) Damage-strain relationships in loading showing the
influence of the exponent m.
D = p + d = + (7.33)
1 d || S(1 d)m
2
= y + 0
S(1 d)m
where it was used that || = y at plastic loading. Hence, the model is thermodynamically
consistent.
The rate of work W and the rate of plastic work D p are defined as
def def
W = , D p = p = y 0 (7.34)
We shall also introduce the fracture energy release gf (energy per unit volume of material),
defined as Z
gf = W dt (7.35)
0
where it is assumed that = 0 when t = (like in the prototype model above). Hence, gf
is the necessary consumed energy in order to reduce the stress to zero, which corresponds
to the fully fractured state. Thereby we consider the prototype model based on perfect
plasticity coupled to damage as well as the alternative model of softening plasticity without
damage. The latter model mimics the same (nonlinear) softening characteristics as does
the damage-based model when m < 1.
In order to unify the developments in an explicit fashion, we rederive the results of the
damage-based model while using stress control, i.e., is the control variable. The same
strategy is then used for the plasticity model. Moreover, we shall directly use the condition
> 0 in pure tension.
2 ||2 y2
= || y , =+ , = = (7.36)
2S(1 d)m 2E 2E
which gives
y2
p
= , d = (7.37)
1d 2ES(1 d)m
The consistency condition (at loading) reads
= = 0 (7.38)
However,
y
= ; = + d= + d (7.39)
1d 1d 1d 1d 1d
1 y3
= with h = (7.40)
h 2ES(1 d)m
2ES m1 2ES m
p = ; D p = p = (7.42)
(y )m+2 (y )m+2
and we have obtained one equation for determining the unknown parameters S and m.
The remaining equation is obtained from an experimentally given value of the fracture
strain f (or ductility measure s, as defined in (7.27)). To this end, we integrate (7.42)1
as follows:
Z 0
e p y 2ES y 2ES 2ES
f = f + f = m+2
m1 d = + 2
if y f (7.45)
E (y ) y E my my2
Remark: The relation (7.45)1 is identical to (7.27) expressed in terms of the ducility
measure s. 2
We may now solve for S and m from (7.44) and (7.45)2 to obtain
1 1
y gf gf gf gf
S= 1 , m= 1 (7.46)
2E f y f y f y
It appears that a downwards convex stress vs. strain relation, defined by m < 1, requires
that
gf 1
< (7.47)
f y 2
Softening plasticity
where it was used that the elastic part of the strain does not contribute to the dissipation
of energy in a closed stress cycle (when stress is increased to the peak stress and then
reduced to zero during the fracture process).
To represent the same type of behavior as the damage-based model, we propose the
nonlinear softening model defined by
||m
= || y , = + (7.49)
mym1
= E( p ) = E( ) (7.52)
= = 0 (7.53)
from which
= ; = y < 0 (7.54)
where it was used that = 0 when = y . Moreover, by inserting (7.51) and (7.52) into
(7.53), we obtain " m1 #
1
= with h = H0 1 1 (7.55)
h y
Now, upon combining (7.55) with (7.50)1, we obtain
1
p = ; Dp = m1 (7.56)
h
H0 1 1 y
y2
gf = (7.58)
2H
Case 2: m = 2 corresponds to the (classical) nonlinear saturation hardening, cf. Sec-
tion 4.4.4, and gives
y2
gf = (7.59)
H
These two cases are depicted in Figure 7.4.
y m=
m=2
H
1
y f
Damage coupled to (isotropic) hardening plasticity is defined by the free energy density
1 1
= (1 d)E( p )2 + Hk 2 (7.60)
2 2
From Colemans equations, we obtain
= = (1 d), p (7.61)
= = Hk (7.62)
k
1 1 2
= = E( p )2 = (7.63)
d 2 2E
The yield criterion is given as
(, ) = || y (7.64)
2
=+ (7.65)
2S(1 d)m
The constitutive rate equations for the internal variables are given as
p = = = (7.66)
1 d 1 d ||
k = = = (7.67)
2
d = = = (7.68)
(1 d)m S (1 d)m 2ES
It is noted that the flow and hardening rules are associated with , whereas this is not
the case for the damage rule. Upon inserting (7.66) and (7.67) into (7.61) and (7.62), we
obtain the constitutive rate equations
= E E (7.69)
1 d ||
= H (7.70)
Together with (7.64), these equations can be solved when they are subjected to the loading
condition
0, (, ) 0, (, ) = 0 (7.71)
It remains to establish the tangent stiffness relation.
From the consistency condition 0, valid at the plastic state ( = 0), we obtain
= + 0 (7.72)
Analogously with the situation of perfect plasticity, we obtain
= tr h 0 (7.73)
For a particular state during the development of damage and plastic deformation, we
obtain (for given H > Hcr ) that
1
m+1
(y + cr )3
E ep > 0 if d < dcr with dcr = 1 (7.83)
2ESH
where it is noted that = cr depends on the history of damage development (that must
be calculated while the constitutive relations are integrated numerically), as illustrated
in Figure 7.5.
where it was used that || = y at plastic loading. Hence, the model is thermody-
namically consistent.
The pertinent constitutive equations for damage coupled to linear mixed isotropic and
kinematic hardening plasticity in the case of uniaxial stress are summarized as follows:
The yield function of von Mises, accounting for damage, is defined by
red
= E E red and = (7.86)
1 d e 1d
= rH (7.87)
red
= (1 r)H (7.88)
ered
2
d = (7.89)
2ES(1 d)m
which can be solved when they are subjected to the complementary conditions
0, 0, = 0 (7.90)
Upon applying the Backward Euler (or fully implicit) rule for integrating the damage law
in (7.89), we obtain the relation
n+1 2
n+1
d = nd + (7.91)
2ES(1 n+1 d)m
n+1
We are thus seeking the solution d = d of the equation
n+1
(d)2
y(d) = d n d (d) =0 (7.92)
2ES(1 d)m
A convenient and robust (although not necessarily the most efficient) iterative technique
to obtain the root of y(d) = 0, if such a root exists, is based on a bisecting procedure within
the interval 0 d < 1, which will be discussed in detail below. For each iterative value of
d = n+1 d in this procedure, it is necessary to calculate the appropriate values of (d) and
n+1
(d). This can be done in a fashion that is almost identical to the solution procedure for
plasticity without damage (as described in the next subsubsection). The major difference
is that nominal stress quantities are replaced with the effective correspondents.
Following the procedure outlined in subsection 4.4.6, we arrive at the constitutive equation
n+1
where tr is given as
n+1
tr = n+1 red,tr
e y n with n+1 red,tr
e = |n+1 red,tr | , n+1 red,tr
= n+1 tr
n (7.94)
n+1
The generated plastic modulus h is slightly modified as compared to (4.90), i.e. it is
defined as
n+1 E
h= n+1 d
+H (7.95)
1
n+1
where we note the occurrence of the effective elastic modulus E/(1 d).
n+1 n+1
(0) = tr (0) = n+1
tr > 0 (7.96)
n+1
where tr was defined in (7.94). From (7.93), we obtain the solution
n+1
tr
= n+1 h
(7.97)
Remark: The loading condition need to be checked (within each load step) only at the
first iteration on the damage variable when d = n d. 2
The updated state becomes
n+1
= c1 n+1 tr + (1 c1 ) n (7.98)
n+1
= n + rH (7.99)
n+1
= c2 n+1 tr + (1 c2 ) n (7.100)
E (1 r)H
c1 = 1 , c2 = a (7.101)
n+1 red,tr (1 n+1 d) n+1 red,tr
e e
Computational algorithm
The complete computational algorithm for a time increment t is given in Box 7.1 and
in Figure 7.6.
Vol 0 March 7, 2006
154 7 DAMAGE COUPLED TO PLASTICITY
n+1
= n + , n+1 tr
= n + E
Stop
n+1
elseif tr > 0, then plastic loading
n+1 (k)
Integration ; , (k) ; y (k)
n+1 (k+1) 1
d = (dmin + dmax )
2
6. Check convergence
If | n+1 d(k+1) n+1 d(k) | < tol then Stop
If k > maxiter, then reduce t and goto 1
else goto 4 and continue iteration
Remark: As to the bisecting procedure, we remark that > 0 always, which means that
y(n d) < 0. Hence, one of the two situations depicted in Figure 7.6 is the relevant one.
Figure 7.6: Function y(d) used in bisectioning algorithm. (a) Solution exists, (b) Solution
does not exist since t is too large.
DAMAGE COUPLED TO
VISCOPLASTICITY
In order to describe creep rupture under constant or moderately varying load as well as
creep-fatigue phenomena under cyclic loading, it is necessary to couple damage to the
development of viscoplastic deformation.
Like in the case of damage coupled to rate-independent plasticity (that was discussed in
Chapter 7), we consider the simplest prototype model based on the following expression
for (, p , d):
1
= (1 d)E( p )2 (8.1)
2
Hence, Colemans equations are the same as those of damage coupled to rate-independent
plasticity:
= = (1 d)E( p ) = (1 d) with = E( p ) (8.2)
1d
p = p
= (1 d)E( p ) (8.3)
1 1 2
= = E( p )2 = (8.4)
d 2 2E
The perfectly plastic quasistatic yield function = () is chosen as
= || y (8.5)
2
=+ 6
= (8.6)
2S(1 d)m
where S is a material constant; dim(S) = dim(), and m is an exponent that governs
the rate of damage evolution. All these expressions coincide with their rate-independent
counterparts.
Like in the case of viscoplasticity without damage, cf. Chapter 5, we define the generalized
Norton viscoplastic law by the following choice of ():
nc nc
hi h|| y i
() = = (8.12)
E E
We shall only consider the classical Norton creep law, which is retrieved from (8.12) when
y = 0, in which case we may insert (8.12) into (8.7), (8.8) and (8.11) to obtain
nc nc
1 || 1 1 || 2
p
= , and d = 0 (8.13)
t (1 d) E || t (1 d)m E 2ES
nc
E ||
+ d + = E (8.14)
t E ||
It is assumed that the stress = 0 > 0 is applied suddenly, whereafter creep occurs for
a constant stress. This process is defined by the initial value problem
1 1 nc 2
0 0
d = m+n +2
, d(0) = 0 (8.15)
t (1 d) c E 2ES
0 d 1 1 nc
0 0
= 2
+ n +1
, (0) = (8.16)
E(1 d) t (1 d) c E E
It is possible to solve (8.15) first, which gives
1
d(t) = 1 p(t) m+nc +3 (8.17)
Upon introducing the solution for d(t) given in (8.17) into (8.16), we obtain, after some
elaboration, the solutions for (t) and (t) as
1 0 nc 0 3 m+nc +4 nc +1
m+n 1 0 nc nc +1
(t) = 2
p(t) m+nc +3 + p(t) c +3 p(t) m+nc +3 (8.19)
t E 2E S t E
0 2ES m+2
m+n1 +3
(t) = p(t) c + 1 p(t) m+n c +3 (8.20)
E (m + 2)0 2
0 2ES m+2
+ 1 p(t) m+n c +2
E (m + 2)0 2
The first term of the exact solution in (8.19) is due to the change of elastic stiffness when
the stress is held constant, i.e. = 0 . It appears that this term can be neglected in
comparison with the viscoplastic contribution only when the following criterion is satisfied:
0 3
k << 1 (8.21)
2E 2 S
For realistic values (0 200MPa, E = 200GPa, S 1MPa), we obtain k 0.0001,
which means that the indicated approximation is justified except when t is close to tR .
The approximation is frequently adopted in the literature, cf. Lemaitre (1992).
Remark: According to the exact solution, when t tR , whereas the introduced
approximation gives a finite value of when t = tR . 2
In the particular case that S = , we obtain
Let us next consider the situation when t = 0, at which moment we have p(0) = 1, and
(8.19) gives
1 0 nc 0 3
(0) = +1 (8.24)
t E 2E 2 S
where the first term in this expression is due to damage development.
Creep rupture
From (8.15) and (8.16) we note that and d when d 1. This situation
is achieved after the (finite) rupture time tR , which can be calculated from the condition
d(tR ) = 1. This condition is, obviously, equivalent to the condition p(tR ) = 0, and from
(8.18) we then obtain
tR 2ES nc
0
= 2
(8.25)
t (m + nc + 3)0 E
It is noted that the life-time is reduced by an increase of the stress level. Moreover, the
rate of damage development is increased by reducing the modulus S and raising the value
of the exponent m in the damage law. As a result the life time is reduced. The typical
behavior is shown in Figure 8.1.
Figure 8.1: Creep rupture behavior for the Norton-material in terms of (a) Creep strain
rate, (b) Creep strain.
Taking the logarithm of both sides of (8.25), we obtain the alternative expression
tR 2S 0
ln = ln (nc + 2) ln (8.26)
t (m + nc + 3)E E
This relation, which shows the influence of the creep exponent nc , is depicted in Figure 8.2.
Isotropic hardening of the quasistatic yield surface can be included in a fashion that is
similar to viscoplasticity without damage. We thus propose
1 1
= (1 d)E( p )2 + Hk 2 (8.27)
2 2
and Colemans equations give
= = (1 d), = E( p ), p (8.28)
= = Hk (8.29)
k
1 2
= = (8.30)
d 2E
The quasistatic yield function accounting for isotropic hardening is given as
(, ) = || y (8.31)
p 1 1 1
= () = () = () (8.32)
t t (1 d) t (1 d) ||
1 1
k = () = () (8.33)
t t
1 1 2
d = () = () 0 (8.34)
t t 2ES(1 d)m
By combining (8.32) and (8.33) with (8.28) and (8.29), we obtain the constitutive rate
equations
E
+ () = E (8.35)
t (1 d) ||
H
() = 0 (8.36)
t
Together with (8.34), these equations are sufficient to determine the time dependent
stress-strain relation under various loading conditions. In practice, it is necessary to
use numerical integration methods to solve even this (moderately complex) initial value
problem.
The general characteristics of the commonly observed behavior of metals and alloys under
creep conditions was discussed already in Chapter 1. The subsequent comments refer to
Figure 8.3.
After the transient (primary) phase, corresponding to saturation hardening of the qua-
sistatic yield surface, the behavior is virtually perfectly viscoplastic up to a certain level
of accumulated creep strain pe where damage starts to develop. This (secondary) phase of
stationary creep is ideally assumed to take place with a constant quasistatic yield surface,
although the transition from the transient to the stationary stage is smooth. In practice,
it seems reasonable to employ the nonlinear hardening model without damage (discussed
in Chapter 5) for these two first stages.
In order to model the tertiary stage of the creep process, which terminates in creep failure
after the time tR , it is necessary to modify the damage law given in (8.34) in such a fashion
that pe will serve as the threshold value, below which no damage develops. This value is
achieved after the the time t. It should be noted that t is not a material constant, since
it is dependent on the stress level. The damage law in (8.34) is thus reformulated as:
1 2
d = () m
H(pe pe ) (8.37)
t 2ES(1 d)
Due to the development of damage, the effective stress moves away from the quasistatic
yield surface (as shown in Figure 8.3), which results in increasing creep rate. Hence, it is
possible to unify the typical behavior of all creep stages within one single model concept.
I Saturation hardening
= fixed
2
= 0, t = t I
I II III t 1
tI t II tR
= 0, t = 0
II Perfect viscoplasticity
(stationary creep) = fixed
2
I II III t 1
tI t II tR
= 0, t t t
I II
(saturated hardening)
, t > t II
III Softening
2
fixed
I II III
-y
t 1
tI t II tR
= 0, t t t
II R
(a)
30
25
20
/ y 15
10
0 20 40 60 80
t/t
*
Figure 8.4: (a) Strain versus time, (b) Hardening stress and damage versus time.
FATIGUE PHENOMENON
AND ANALYSIS
The fatigue phenomenon, which is the most common reason for failure of engineering
components, is treated in this chapter. Classical approaches (which are based on empirical
relations) are reviewed. A damage mechanics approach is then discussed, whereby the
famous Manson-Coffin relation is derived. Finally, the growth of a macroscopic fatigue
crack, using the fracture mechanics approach, is discussed.
9.1 Background
9.1.1 Nomenclature
As alluded to already in the Chapter 1, fatigue is the accepted term for the damage and
eventual failure of a material that is subjected to cyclic variation of the loading. The basic
situation is defined as mechanical fatigue at ambient temperature, which occurs as a result
of rate-independent material behavior for cyclic variation of the externally applied loads
only. At elevated temperature, depending on the rate of loading, creep deformation may
take place in combination with mechanical LCF. This gives rise to creep-fatigue (CLCF),
which is of particular importance in hot engine parts, such as in jet turbines. Thermal
fatigue is encountered when the stress change is caused by cyclic variation of temperature
in a statically indeterminate structure. A typical situation is start-stop of an engine. The
combined effect of cyclic mechanical and thermal loading is denoted thermomechanical
fatigue.
Another fatigue phenomenon is the socalled corrosion fatigue. This is rather the gradual
deterioration of the microstructure due to chemically aggressive or embrittling environ-
ment.
From the microstructural point of view, the progression of fatigue for a metal may be
classified as follows:
3. Propagation of the macroscopic crack until its growth becomes unstable and struc-
tural failure occurs (fracture phase, Nf cycles).
The crack growth can be accelerated by stress-corrosion at the grain boundaries and by
the presence of embrittling substance, such as hydrogen and chromium.
Remark: The total fatigue life is the sum of the cycles in the damage and fracture phases.
However, in the engineering approach discussed in this chapter, no distinction is made
between the different phases. 2
The most famous investigations of fatigue failure (in railroad axles) were carried out
by Wohler (1860), although it was not the first study of fatigue. Based on various
experimental findings, an empirical relation between stress amplitude and the number of
cycles to failure was proposed 50 years later. Of importance for the understanding of
LCF is also the discovery by Bauschinger that the current yield stress upon reversed
loading is smaller. From a Swedish perspective, we may mention the concept of linear
damage accumulation for assessing the combined effects of a series of block-loadings with
different, but constant, amplitudes, which was proposed by Palmgren (1924).
A design criterion for LCF was established quite late. Coffin (1954) and Manson
(1954) independently proposed an empirical relation between plastic strain amplitude
and number of cycles to failure.
It is wideley accepted that the early phase of fatigue is characterized by slow growth of
damage. However, a rational theory was suggested rather recently by French scientists, see
Lemaitre (1992). Based on the concept of an evolution law for continuum damage in
addition to plastic deformation, the Basquin and Coffin-Manson relations can be derived,
which is also shown below in Section 9.2. The analysis is still based on simplifying
assumptions. For completely arbitrary variable amplitude loadings, it is necessary to
adopt a more complete constitutive theory in order to predict the fatigue life. This is also
shown later in this Chapter for LCF. A corresponding theory for HCF has been suggested
by Ottosen (1995).
After the initiation of a macroscopic fatigue crack, its propagation can be considered
within the realm of fracture mechanics. Paris & al. (1961) proposed an evolution law
for the fatigue crack such that is advancement per load cycle is related to the amplitude
of the stress intensity factor at constant amplitude loading.
The basic mode of fatigue testing involves a loading program with constant amplitude
stress or strain cycles. Under rather ideal conditions the cyclic response will result in
stabilized hysteretic loops after a number of cycles. This is particularly true for smooth-
surfaced specimens without notches and excessive surface roughness (that act as stress
concentrators), whereby material deterioration does not start until after quite many load
cycles. Under the further idealized assumption that the stabilized hysteretic loops are
symmetrical, it is sometimes useful (in conjunction with fatigue life predictions) to re-
late the strain amplitude a to the stress amplitude a of the stabilized loops, as shown
in Figure 9.1. It is noted that this stabilized cyclic stress-strain curve, can be located
above (cyclic hardening) or below (cyclic softening) the corresponding monotonic stress-
strain curve. Generally speaking, well-annealed metals and alloys exhibit cyclic hardening,
whereas work-hardened materials undergo cyclic softening.
In the case that plastic strains are produced during a typical cycle, then a commonly used
idealization of the stabilized curve is the power-law expression
n1
a a cyc
a = ea + pa = +
(9.1)
E cyc
Remark: The expression (9.1) is valid only if the yield stress was exceeded in the first
cycle, which pertains to LCF. If the macroscopic stress is purely elastic, which pertains
to HCF, then the expression (9.1) has no meaning. 2
As to the parameters, E is the elasticity modulus, whereas cyc and ncyc are material
parameters to be determined via a suitable curve-fitting procedure. In the literature
cyc is denoted the cyclic strength coefficient, whereas ncyc is the cyclic strain hardening
coefficient. For most metals, ncyc is in the range 0.1-0.2.
Remark: Do not confuse the cyclic strain hardening modulus ncyc with the creep expo-
nent in the Norton law c . 2
a
a
a
a a
9.2.1 Basquin-relation
In the approach for assessing the fatigue life that was suggested by Wohler (1860),
smooth (unnotched) test specimens are fatigue-tested in plane bending, rotating bending
or uniaxial tension-compression. The most basic loading modes are defined as reversed
stress (m = 0) and pulsating stress (m = a ), which are shown in Figure 9.2.
Let us first consider the case of fully reversed cyclic loading, in which case m = 0. The
typical result of fatigue tests are shown in Figure 9.3, where a is plotted against NR (the
max
2 a m
t
min
(a)
m =0
a
(b)
m = a
2 a
(c)
Figure 9.2: (a) Constant stress amplitude cyclic loading, (b) Reversed stress loading, (c)
Pulsating stress loading.
where f is the fatigue strength coefficient, which (as a rule of thumb) can be taken as
the fracture strength as observed in a monotonic tensile test. Moreover, b is the socalled
Basquin exponent, whose value is in the range 0.005 to 0.012 for most metals and
alloys.
Remark: Eqn. (9.2) is supposed to be valid for both HCF and LCF, although the
physical mechanisms of damage are different. For example, in LCF plastic strains will
a ln a
fl > 0 ln a = ln f + b ln ( 2 N R )
fl = 0
b<0
1
fl
NR ln ( 2 N R )
(a) (b)
Figure 9.3: (a) Wohler-curves (S-N-curves) for material with fl > 0 and with fl = 0, (b)
Basquin relation (fl = 0).
a fl = f (2NR )b (9.3)
However, for many high strength steels and aluminum alloys, no fatigue limit exists, i.e.
fl = 0.
It is emphasized that NR represents the total fatigue life, i.e. NR includes the initial
damage phase as well as the subsequent fracture phase (as already discussed in Chapter
1). The damage portion may vary from 0 % for structures containing severe stress con-
centrations or surface defects to 80 % in very smooth surfaced, defect-free materials of
high purity.
The mid stress m has a significant influence on the fatigue life such that the Wohler
curves are lowered for increasing m , as shown in Figure 9.4(a). In other words, for
given life NR , it appears that a is reduced with increasing m . The resulting socalled
Haigh-diagram is shown in Figure 9.4(b).
a
a represents
(1) (2)
m < m pulsating load
ar
ap
(2)
m
NR m
mr = 0 mp = ap y u
(a) (b)
Figure 9.4: (a) Wohler-curves for different constant m , (b) Typical Haigh-diagram show-
ing possible range of a versus m for given life NR .
where ar is the amplitude for reversed loading (mr = 0). From this relation we may
calculate (a =)ap , which is the pertinent amplitude for pulsating loading, by setting
(m =)mp = ap .
For hand calculation purposes, the relation may be further simplified by a bilinear re-
lationship, as shown in Figure 9.5(a). An alternative, but equivalent, representation is
given by the socalled Goodman-diagram, Figure 9.5(b), which shows the stress range
(m + a , m a ) as a function of m .
Stress concentrations at holes and notches will have an influence on the fatigue life, al-
though in a manner that is different from that of stress concentrations for monotonic
loading. In the case of monotonic loading, we recall that it is possible to relate the local
stress concentration to the far-field stress by the stress concentration factor Kt , which is
calculated for the appropriate constitutive relation. For example, for a hole in an infinite
plate with isotropic elastic material behavior, we obtain Kt = 3.
In order to account for the effect of stress concentrations on the fatigue life, we introduce
(admittedly in a somewhat ad-hoc manner) the fatigue notch factor Kf via the empirical
= m + a
= m
u
2 ap
a
a = m
ar ar
ap
m m
ap u ap u
(a) (b)
ar
Figure 9.5: (a) Haigh-diagram and (b) Goodman-diagram suitable for design.
relation
Kf = 1 + q(Kt 1), 0 q 1, Kf Kt (9.5)
where q is the socalled notch sensitivity index. For a notch with root radius , it has been
suggested that q is calculated according to the empirical relation:
s !1
An
q = 1+ (9.6)
where An is a constant that depends on the strength and ductility of the material.
(An 0.0250.25 mm for steel, where the lower limit is for very high strength steel,
whereas the upper limit is for well-annealed steel.) In brief, we note that An = 0 (q = 1)
for very brittle material response, while An is large for ductile materials corresponding to
a smaller value of q. Obviously, the larger value of An , the smaller sensitivity to a notch.
It is noted that
= 0 q = 0, = q = 1 (9.7)
In practice, the stress concentration factor Kt , and therefore also Kf , are always calculated
under the assumption that the material is isotropic elastic. This is valid with good
approximation only for HCF, in which situation Kf is not affected by the actual stress
level.
The design procedure is then as follows: Use the relations for the nominal (far-field) stress
by making the substitutions
m Kt m , a Kf a (9.8)
Then use these values (with the appropriate safety factors) in the design situation.
LCF will occur in the inelastic regime, whereby the stress concentration due to a notch
will be less than that for the elastic response. Moreover, the corresponding concentration
of strain at the notch root will be different than that of stress. To this end, we introduce
the notation
a
a = K a , a = K a = K , K K , K Kf (9.9)
E
where K and K are the stress and strain concentration factors, respectively, and where
a and a are the nominal (far-field) stress and strain values, which are calculated under
the assumption of elastic response. Clearly, if the material was elastic (HCF), then it
follows trivially that K = K = Kf , which is illustrated in Figure 9.6 (dotted line).
For LCF, the actual values of K (and K ) are obtained from the following approximate
method, denoted Neubers method: It is assumed that the stress and strain states at the
notch satisfy the socalled Neuber hyperbola, which is satisfied by the (fictitious) elastic
state (Kf a , Kf a ), as shown in Figure 9.6. This means that the Neuber hyperbola has
the equation
Kf2 a2
= with a = Ea (9.10)
E
Let us next assume that the actual stabilized cyclic stress-strain curve is given by the
function
a = f (a ) (9.11)
For each stress level a , it is assumed that the state (a , a ) at the notch satisfies the
hyperbola (9.10), in addition to the cyclic relation (9.11). We thus obtain
a
a = f (a ) K a = f K (9.12)
E
Moreover, introducing (a , a ) from (9.9) into (9.10) gives the relation between K and
K :
K K = Kf2 (9.13)
elastic
a
K f a
a = f ( a )
K a
a
= const. ( Neuber hyperbola )
a
a Kf a K a
Combining (9.12) and (9.13), we may solve for K (and then for K ) for given a from
the relation
Kf2
K a = f a (9.14)
K E
Finally, the design procedure is defined as follows: Use the relations for the far-field stress
by making the substitutions
m Kt m , a K a (9.15)
Design Safety
So far, we have given explicit consideration to the influence of mid stress and stress
concentration, which both reduce the fatigue strength. However, in reality there are a
number of other effects that reduce the fatigue strength, but which are difficult to quantify:
To account for these effects the calculated stress is amplified by a suitable load factor, or
safety factor, which may be taken differently depending on whether the safety is related
to m , a , or a combination of m and a . Usually, the following situations are considered:
a
m
= const Design stresses: (Fam Kt m , Fam K a )
This is shown schematically in Figure 9.7. Note that K = Kf in the case of HCF.
where Ni is the number of cycles in the i:th loading block with constant amplitude a,i ,
and NR,i is the fatigue life at a,i , as shown in Figure 9.8. The total number of blocks in
the sequence is n. Fatigue failure then corresponds to d = 1.
The following assumptions are implicit in (9.16):
The same amount of damage is caused by each cycle for a given amplitude (inde-
pendent of whether it occurs at the beginning or end of the fatigue life).
The order of the blocks of different amplitude does not affect the damage develop-
ment and the total fatigue life.
Neither of these assumptions are particularly realistic. Later in this chapter, we examine
in further detail the assumptions above in conjunction with the derivation of fatigue-life
curves based on a rational damage mechanics approach. Despite its weak theoretical basis,
the Palmgren-Miner rule is widely used in a variety of contexts in fatigue life analysis.
ar ( K t m , F aK f a )
( K t m , K f a )
m
u
(a)
ar
( F m K t m , K f a )
( K t m , K f a )
m
u
(b)
a
ar
( F am K t m , F am K f a )
( K t m , K f a )
u m
(c)
Figure 9.7: Design diagram for (a) Constant m , (b) Constant a , (c) Constant a /m .
a1 a 2
NR
N1 N2
Figure 9.8: Loading consisting of sequence of blocks with constant stress amplitude (at
reversed cyclic loading).
So far we have discussed fatigue at uniaxial stress. However, in most situations in en-
gineering practice, the stress state is multiaxial, and it is not obvious how to generalize
the analysis for uniaxial stress to this more general situation. In order to simplify the
discussion, we shall first assume that all components of the stress tensor are periodic with
the same frequency (period T ). The only difference is the phase angle, i.e. we can write
where (m )ij is the mid stress and (a )ij is the amplitude defined as
1
(m )ij = max ij (t) + min ij (t)
2 0tT 0tT
1
(a )ij = max ij (t) min ij (t) (9.18)
2 0tT 0tT
defined as !1/2
3
3X 2 2
a,e (t) = ( ) A (t + i ) (9.20)
2 i=1 a i
where a prime denotes deviator, i.e. i = i 13 (1 + 2 + 3 ).
The maximum equivalent stress over one period T is defined as
a , e ( t )
a
t max
t
t =0
Remark: When all stress components vary cyclically in-phase, i.e. i = 0 for i = 1, 2, 3,
we obtain from (9.20) and (9.21) that
3
!1/2
3X
a,ph = (a )2i (9.22)
2 i=1
where it was used that max || = 1. In all other situations than in-phase changes, we have
a a,ph 2 (9.23)
The equivalent mid stress m is defined in precisely the same way as a , but simply by
replacing (a )i with (m
)i in (9.22). Hence, we obtain
3
!1/2
3X 2
m = m,e with m,e = ( ) (9.24)
2 i=1 m i
a = a , m = m with a = (a )1 , m = (m )1 (9.25)
are retrieved.
The simplest way of extending the analysis for the uniaxial stress state to the multiaxial
stress state is to replace a by a and m by m and to use the Haigh diagram and the
Basquin equation
a fl = f (2NR )b (9.26)
For example, for b = 0.12, which is a reasonable value for a ductile material, and
ncyc = 0.2, we obtain c = 0.6. These values of b and c are quite often adopted for repre-
senting the universal Manson-Coffin relation, since it has turned out that these exponent
values are remarkably insensitive to variations of the material composition (for the same
temperature). However, all material coefficients are strongly temperature-dependent.
ln a
ln a versus ln ( 2 N R )
e f
1 ln a = ln + b ln ( 2 N R )
E
1
p
2 ln a = ln f + c ln ( 2 N R )
b
1 c
1 2
ln ( 2 N R )
From (9.28), and with values of b and c given above, it may be concluded that pa is
the dominant term for short fatigue lives (small values of NR ), whereas ea is the more
significant term for long fatigue lives (large values of NR ).
Effect of mid-strain
A prescribed tensile mid strain m reduces the fatigue life (in much the same fashion
as does a mid stress m for stress-controlled fatigue tests.) This will lead to mid stress
relaxation, i.e. m 0, which is a gradual reduction of mid stress with cycling. Such mid
stress relaxation under strain-controlled fatigue loading is the counterpart of ratchetting
(or cyclic creep), that takes place under stress-controlled loading.
The effect of a strain concentration at LCF is treated in the same fashion as the case of
stress-controlled testing, that was discussed in Section 9.2. The only difference is that K
is now the quantity of interest (rather than K ). We thus obtain from (9.12)
a = f (a ) ; K Ea = f (K a ) (9.29)
Kf2
Ea = f (K a ) (9.30)
K
m Kt m , a K a (9.31)
The simplest loading program that involves the combined effects of LCF and creep, so-
called creep-fatigue, is defined as cyclic loading with a prescribed constant strain ampli-
tude that has a certain duration (hold time) th in each cycle. This is shown schematically
in Figure 9.11(a). A typical stabilized cyclic stress-strain relationship (which is valid prior
to excessive development of damage) is shown in Figure 9.11(b). The hold time th cor-
responds to the portion pc p pf
a of the total inelastic strain amplitude a , whereas a is the
time-independent LCF-portion of pa , as also shown in Figure 9.11(b).
In order for creep effects to become significant, the temperature must be sufficiently high.
In the simplest case, the temperature is constant in time, whereby the failure is denoted
creep-fatigue (CLCF). If, on the other hand, the mechanical loading is held constant in
time while the temperature is varying cyclically with the hold time th , then we speak of
thermal-fatigue (TLCF). In the more general situation both the mechanical and thermal
loadings will vary in a cyclic fashion, which is denoted thermomechanical-fatigue. This
variation may be in-phase or out-of-phase, while the same frequency is retained. In the
most general case, the mechanical loading and the temperature will not vary with the
same frequency.
creep fatigue
2 3 2 3
a
p
t
1 4 1 4
5 6
6 5
th
pf
2 a
pc
2a
(a) (b)
Figure 9.11: Creep-fatigue, (a) Loading program, (b) Stabilized stress-strain cycle.
Let us consider the case of thermomechanical fatigue with in-phase variation of temper-
ature. It is then common to extend the Palmgren-Miner rule to combine the effects of
creep and fatigue as follows:
NR tR
+ c =1 (9.32)
NRf tR
Upon introducing the relations
This relation is known as the Taira rule of linear creep-fatigue damage accumulation. The
similarity with the Palmgren-Miner rule is striking. In fact, it is possible to extend the
rule to the case of sequential loading with piecewise constant strain amplitudes to obtain
n
! n
!1
X Ni Ni X Ni fc 1 1
f
+ c = 1, or fc
= 1, NR,i = f
+ c (9.35)
i=1
N R,i N R,i i=1
N R,i NR,i NR,i
It must be realized that the theoretical basis for the assumption about linear damage
accumulation is very weak. Hence, it can not be expected that the agreement with real
behavior is good. Nor does the simplified rule necessarily give acceptable agreement with
a more rigorous analytical prediction. In fact, it is known that creep-fatigue interaction is
quite nonlinear in such a fashion that, unfortunately, (9.34) may overestimate the number
of cycles to failure. Nevertheless, since a more accurate analysis is quite involved, the Taira
rule has been used in practice, at least as a rule of thumb.
f
a = (2NR )b + f (2NR )c (9.38)
E
Remark: Because of mid stress relaxation, it is often claimed that the corresponding
mid strain m has a negligible effect on the fatigue life. 2
All fatigue analyses aim at predicting the lifetime of a component subjected to cyclic (me-
chanical and/or thermal) loading. To this end, a number of different prediction strategies
are possible, corresponding to the acceptable level of complexity and realism of analysis.
Comprehensive reviews are found in Skelton (ed.) (1987) and Riedel (1987). In
most strategies, the concept of a loading (stress or strain) cycle plays an important role.
lifetime prediction strategies. This concept relies heavily on the presumed existence
of a stabilized cyclic response, which is predicted without any provision for ma-
terial deterioration (in terms of damage accumulation or crack propagation). The
analysis of damage accumulation is then carried out in a post-processing step, based
on the assumption that the stabilized response is relevant up to failure.
A rational strategy for lifetime prediction should consist of the following three main steps:
Upon parameterizing the applied mechanical and thermal loading, we may characterize
each loading as a point in the design parameter space P = (p1 , p2 , . . .). For a turbine
blade the loading may conveniently be defined by the angular frequency (t) and the
temperature (t) in the surrounding fluid. Typically, for in-phase thermomechanical cyclic
loading, the relevant parameters are the working frequency amplitude a , the temperature
amplitude a , and the hold time th in each cycle (with duration tc ), which is shown in
Figure 9.12. This gives P = (a , a , th ). Corresponding to each point in P, there is a
critical number of cycles to failure, NR (obtained in Step 3 below). Hence, the failure
state is defined by a surface in the space (P, N).
Compute the stress, strain and temperature fields (in space and time) during a few cycles
of the (thermomechanical) fatigue process until reasonably stabilized cyclic behavior has
been obtained. In this step, a (more or less) elaborate material law without damage is em-
ployed, and it is calibrated to fit the stabilized response. The underlying key assumption
is that the damage development will not significantly affect the stiffness variation with
time (until close to global failure), such that the computed stress and strain fields will
be representative throughout the major part of the lifetime. In particular, if the design
of the component under consideration is such that the stress field is (almost) statically
determinate, it is obvious that the stress field should be insensitive to the approximation
employed in the constitutive relations. The corresponding strains may, however, be quite
th
a , a
tc
This is the key step in the strategy. First, the critical regions (or points) for the damage
accumulation analysis are selected. A critical location may, for example, be characterized
by maximum effective stress or maximum damage stress (elastic energy density release
rate), as defined below. For given stress and temperature histories3 from Step 2, compute
the damage accumulation. The aim is to obtain the lifetime NR satisfying
d(NR tc ) = dR 1 (9.39)
where dR is the (chosen) critical damage value, and where tc is the duration of a single
cycle. The value dR may be selected to represent a maximal allowable surface density
of microcracks. Alternatively, dR is computed as the result of a localization analysis to
determine the onset of a macroscopic crack.
As to the computation of damage accumulation, the following two different approaches
are considered:
3
The reason for choosing the stress field (rather than the strain field) as input was discussed above.
Assumptions
A simplified analysis of LCF in the case of fully reversed strain and stress loadings, which
leads to the Mansion-Coffin and Basquin relations, is outlined below. The necessary
simplifying assumptions are:
Stabilization under cyclic hardening occurs rapidly so that only stabilized cycles
need to be considered. These loops are assumed to correspond to perfectly plastic
response (in terms of the effective stress).
The stabilized cyclic hardening relation is given by (9.1) for prescribed strain as well
as prescribed stress amplitude. Most importantly, pa is connected to the effective
stress amplitude a via the Ramberg - Osgood relation in (9.1).
The change of damage is associated with increasing number of cycles; i.e. the
damage variable is taken as a constant during each cycle.
We first consider cyclic loading with constant strain amplitude a and fully reversed cyclic
strain.
Accounting for damage, we can rewrite the cyclic hardening relation (9.1) as
1
a p p a ncyc
a = + a , a =
(9.40)
E cyc
Now, we may solve for a in terms of pa and insert this expression into (9.40), i.e.
cyc
a = cyc (pa )ncyc ; a = (p )ncyc + pa (9.41)
E a
Since a remains a constant amount of strain during the cyclic process, it follows from
(9.40) that a is constant. Hence, pa is also constant.
The change of accumulated strain pe after each cycle is 4pa (corresponding to 4 quarter-
cycles), i.e.
dpe
= 4pa (9.42)
dN
Hence, after N cycles, the accumulated strain is
The threshold value N for the initiation of damage is obtained from (9.43) simply as
pe
N = p (9.44)
4a
Let us now consider the damage law, given as
2
d = pe (9.45)
2ES(1 d)m1
Since we have assumed perfect plasticity and no damage development during an individual
cycle, we may formally integrate (9.45) during a single cycle to obtain the rate of change
of damage per cycle as
dd a2 p 1
= 4a = k F , m1 (9.46)
dN 2ES(1 d)m1 (1 d)m1
where the coefficient kF is given as
2(cyc )2 (pa )2ncyc +1
kF (pa ) = (9.47)
ES
Vol 0 March 7, 2006
9.5 Damage mechanics approach to LCF 191
where 2n 1
ESfm (d) cyc +1
C (d) = )2
(9.51)
2(cyc
It is noted that fm (0) = 0. The number of cycles to failure is NR corresponding to the
situation when d = dR . From (9.50) we obtain
1
2n
pa = CR (NR N ) cyc +1 with CR = C (dR ) (9.52)
which is precisely the relation (9.28). Secondly, we may express f in terms of f as follows:
f = 2c CR f = 2b cyc
(CR )ncyc = cyc
(f )ncyc (9.55)
where (9.54) was used. Hence, we conclude that the two expressions of the Manson-Coffin
relation in (9.27) and (9.53) are completely equivalent.
Remark: It would be possible to plot fatigue curves for values of d < dR by replacing
CR with C (d). Since C (d) CR , we note that such curves are located below the fatigue
curve defined by CR . 2
We may also solve for d as a function of N N from the expressions in (9.50) and (9.52).
The pertinent relationship becomes
m1
N N
d=1 1 mfm (dR ) , N N (9.56)
NR N
which is shown in Figure 9.13 in the special case that dR = 1. It is noted that the ratio
N /NR decreases with a for given value of pe . In fact, from (9.44) and (9.52) it is obtained
that
N 1 4
= p 2ncyc +2 with c = p (9.57)
NR 1 + c(a ) e (CR )2ncyc +1
The interested reader should show this as homework!
Figure 9.13: Damage development for LCF for constant amplitude loading.
It is possible (although less usual in practice) to carry out LCF-tests with constant stress
amplitude a . The relationship in (9.40) for stabilized cyclic loops is still employed. The
threshold value N is now obtained, by using (9.40) into (9.44), as
1
(cyc ) ncyc pe
N = 1 (9.58)
4(a ) ncyc
where it was also used that d = 0 when N N.
Let us, again, consider the damage law in (9.46), which can be expressed in terms of the
stress amplitude as
dd 1
= kF 1 , m1 (9.59)
dN (1 d)
m+1+ ncyc
gm (d)
kF (a ) = (9.61)
N N
where we introduced the function
where
1 ! 2nncyc
cyc
+1
ES(cyc ) ncyc gm (d)
C (d) = (9.64)
2
The number of cycles to failure is obtained from (9.63) according to the Basquin relation
ncyc
2n
a = CR (NR N ) cyc +1 with CR = C (dR ) (9.65)
Hence, the Basquin relation may be derived in a fashion that is similar to the Manson-
Coffin relation.
The relation between d and N is now obtained from (9.63) and (9.65) as
ncyc
1+(m+2)n
N N 1 cyc
d=1 1 (m + 2 + )gm (dR ) , N N (9.66)
NR N ncyc
which is quite analogous to the relation (9.56) for prescribed strain amplitude.
Let us consider how damage accumulates if the cyclic loading history consists of n blocks
with constant amplitude, as was shown schematically in Figure 9.8. Since d is a nonlinear
function in N in general, it is clear that it is not possible to establish any simple accumu-
lation rule for the composite loading. Hence, we restrict our attention to the simple case
of linear accumulation of damage, which is obtained when the exponent is unity in (9.56)
or (9.66). For example, in the case of prescribed strain amplitude, defined by (9.56), we
must require that m = 1. For simplicity, we shall also assume that dR = 1.
Let each constant amplitude loading block of duration Ni cycles be associated with the
fatigue life NR,i and with the threshold value Ni (if it was acting alone). The contribution
to the damage from each such sequence is then obtained from (9.56) or (9.66) as
N1 N1
d1 =
NR,1 N1
Ni
di = , 2in (9.67)
NR,i Ni
It was tacitly assumed that the first applied sequence always will have a duration that
exceeds the corresponding threshold value N1 . At failure we must have
n n
X X Ni N1
1= di = (9.68)
i=1 i=1
NR,i Ni NR,1 N1
and we obtain the accumulation rule
n
X NR,i Ni NR,1
= (9.69)
i=1
NR,i Ni NR,i NR,1 N1
A particularly simple situation is obtained if we set
Ni
= constant, i = 1, 2, . . . , n (9.70)
NR,i
since (9.69) is then reduced to
n
X Ni
=1 (9.71)
i=1
NR,i
We have thus obtained the Palmgren-Miner rule of linear damage accumulation. However,
due to the severe restrictions imposed in order to motivate its validity, the Palmgren-Miner
rule should always be used with suspicion.
strain, where damage starts to develop due to nucleation of microcracks. For this first
phase of the deformation process, it is possible to employ the nonlinear hardening concept
without damage, as described in detail in Subsection 4.3. In the absence of damage
development, it is possible to show that the hardening (drag) stress can be expressed
for a completely arbitrary loading path, as
p
rH
K e
= 1 e (9.72)
pe = ln(1 x) (9.73)
rH
2
d = H(p
e p
e ) = p
e H(p
e p
e ) with = (9.74)
S(1 d)m S(1 d)m1 2E
This behavior is illustrated in Figure 9.14 for the case of uniaxial monotonic loading.
The corresponding scenario at cyclic loading with prescribed constant strain amplitude
leading to LCF is shown in Figure 9.15.
It may be remarked that the shape of the strain-cycles (in time) is completely irrelevant,
since it was assumed that the material is rate-independent.
The (expected) cyclic behavior is shown in Figure 9.16, which was obtained for the fol-
Figure 9.14: Delayed development of damage at uniaxial loading defined by the threshold
value p .
lowing data:
y H
= 0.001, = 0.02, r = 0.5, = = 0.25y , (9.76)
E E
S
= 0.5 106 , m = 2,
E
y
[2y , 2y ] , where y = ,
E
20 y
pe = ,
E
Remark: In order to take into account the effect of deactivation of damage under com-
pressive stress (often referred to as the Microcrack-Closure-Reopening effect), the results
in Figure 9.16 were computed with the (refined) effective stress relation
3 (1 + 2H())d 1 H() d3
= = (9.77)
3(1 d)(1 H()d) 1d
Hence, we obtain
= when > 0
1d
(1 d3 )
= when 0 (9.78)
1d
N N
1 2
N NR
a
Figure 9.15: LCF for constant strain amplitude loading with threshold for damage devel-
opment.
Repeating the analysis for different strain amplitudes, we may construct a typical Manson-
Coffin relation, as shown in Figure 9.17. It is noted that no damage can occur if the
strain amplitude is less than the yield strain. Hence, the yield strain may be taken as the
theoretical fatigue-limit in the context of LCF.
Assumptions
A simplified analysis of CLCF in the case of fully reversed stress loading under constant
temperature is outlined below. The necessary simplifying assumptions are (cf. those in
Subsection 9.5.1):
Stabilization under cyclic hardening occurs rapidly, and the stabilized loops core-
spond to perfectly plastic response (in terms of the effective stress).
The inelastic strain during a cycle is decomposed into one portion from fatigue, pf
a ,
pc pf
and another from creep, a . The rate-independent part (a ) is connected to the
effective stress amplitude (a ) via the Ramberg-Osgood relation in (9.1)1 .4
4
Such a decomposition is not consistent with the unified approach, in which pf pc
a and a are insepa-
1.2
0.8
0.6
0.4
0.2
0
0 1 2
10 10 10
N
1
0.8
0.6
damage
0.4
0.2
0
0 1 2
10 10 10
N
Figure 9.16: Predicted LCF-result for prescribed strain amplitude, (a) a versus N and
(b) d versus N.
1
10
0
10
1 2 3
10 10 10
N
During the hold time, th , Norton creep is assumed. Assuming constant temperature
and ignoring any Microcrack-Closure-Reopening effects, the creep strain pc
a is equal
in tension and compression.
The change of damage is associated with increasing number of cycles, i.e. the
damage variable is taken as constant during each cycle.
Because of the assumptions above, we may integrate the Norton-type creep law during a
quarter of the hold time to obtain
n1 nc
a cyc th a
pa = pf
a + pc
a =
+ (9.79)
cyc 4t E
rable from a conceptual viewpoint and their sum pa is rather treated as an internal variable.
2
d = pe (9.80)
2ES(1 d)m1
Upon integrating (9.80) during one single cycle, while observing (9.79), we obtain
dd a2 1 1
= 4pa m1
= kF 1 + kC (9.81)
dN 2ES(1 d) (1 d)
m+1+ ncyc (1 d)m+1+nc
which must be solved by numerical integration in the general situation (for given ampli-
tude a ) due to the coupling between the fatigue (kF ) and creep (kC ) effects. However,
analytical solutions are easily obtained when decoupling is assumed. For given NR , cor-
responding to given dR , we may compute kF = kF,R when kC = 0 and kC = kC,R when
kF = 0. Evaluating (9.83) in these two special cases, we obtain
(9.85)
where we introduced the non-dimensional variables
kF kC
kF = , kC = (9.86)
kF,R kC,R
kF + kC = 1 (9.87)
which is the result of assumed independent damage accumulation form fatigue and creep,
respectively. This is Tairas rule of linear fatigue-creep interaction (as discussed in Sub-
section 9.3.2).
Remark: For the special case that nc = 1/ncyc , the coupling between fatigue and creep
disappears and Tairas rule is recovered. 2
It is also possible to construct Wohler (or S -N) curves, for given values of th (which
serves as a parameter). For each choice of a , we first compute kF and kC from (9.81),
and inserting the result in (9.82) we may compute NR . Typically, these curves look like
in Figure 9.19.
f (kF,kC)=1
0.8 rule)
kF+kC=1 (Tairas
0.6
kF
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
k
C
Figure 9.18: CLCF interaction diagram for nonlinear and linear interaction of fatigue and
creep. (dR = 1, m = 2, ncyc = 0.2, nc = 8).
th = 0
400 th > 0
350
a
300
250
0 2 4 6 8 10
Ncrit 4
x 10
Figure 9.19: Wohler curves for given holdtime th . (dR = 1, m = 2, ncyc = 0.2, nc = 8, E =
200e3, cyc = 3000, t = 1e 14).
9.7.1 Preliminaries
As discussed in Subsection 9.1.1, the fracture phase follows the damage phase of the
fatigue life upon the occurrence of a macroscopic crack. It is then important to predict
the number of cycles from the initiation of the crack via the process of stable crack
propagation up to the final structural failure. Assuming that the crack length is a(N),
such failure is defined by unstable crack propagation, i.e.
da
when N NR (9.88)
dN
Remark: It is usual in the fracture mechanics approach to fatigue to assume that the
crack has a given initial length irrespective of the preceding loading. In this way, no link
exists to the preceding damage phase of the fatigue life. 2
The typical crack growth behavior for constant far-field stress amplitude loading, that
can be observed experimentally, is shown in Figure 9.20.
a
(2) (1) (2)
a a < a
da
dN
ai 1
Figure 9.20: Typical crack growth behavior in constant far-field stress amplitude loading.
Remark: The successive advancement of the crack front in each cycle leads to the typical
striations in the fractured surface (which can be observed experimentally after complete
failure). 2
The rate of crack propagation is influenced by a variety of factors (apart from the far
field stress amplitude a that is applied). These are the mid stress m , the cyclic load
frequency, the temperature and other environmental conditions.
The following empirical relation for the fatigue crack growth rate was suggested by Paris
& al. (1961):
da
= C (2Ka )mf (9.89)
dN
where C and mf are scaling parameters, whereas Ka is the stress intensity factor ampli-
tude. This amplitude can be calculated from a according to equation (10.10) with the
appropriate geometric configuration factor f . If nothing else is stated, tensile fatigue is
considered, i.e. Ka refers to mode I. To be more precise, we should write KIa . Clearly,
KIIa and KIIIa may be introduced in a similar fashion to characterize fatigue crack growth
in mode II and mode III, respectively.
Remark: The relation (9.89) is somewhat awkward, since the actual dimension of C
depends on the value of the exponent mf . Typically, mf 2 4 for ductile alloys. 2
It is noted that the mid stress m has a significant influence on the crack growth, since
the driving force for crack growth is essentially the tensile stress range. In the ideal
situation of elastic material response, the crack remains fully open along its entire length
if min 0, i.e. when m a . For min 0, the crack faces are in contact and transmit
compressive stresses. The pertinent modification of the Paris law to accommodate such
crack closure effects is discussed below.
The effect of m stems from the fact that the crack has a tendency to close at unloading
already for a tensile stress, which effect is due to plastic yielding. It is obvious that if m
is large, then the tendency of crack-closure is smaller. In order to quantify this effect, we
use the stress ratio R = min /max = Kmin /Kmax .
The far-field stress for which the crack starts to close upon stress reversal in a cycle
corresponds to the value Kopen . Clearly, in the ideal situation that the crack closes at
= 0, we would have Kopen = 0. However, experimental investigations have shown that
the crack starts to close already for > 0. Hence, stress reversal below Kopen > 0 will
not contribute to crack growth. A simple empirical expression is
(
0.25 + 0.5R + 0.25R2 , 1 R 1
Kopen = (R)Kmax with (R) = (9.90)
0 , R 1
K open K min
= (R) =R
K max K max
R
1 1
Remark: Since Kmin = RKmax , we conclude that Kopen Kmin with (R) given in
(9.90). 2
The strategy is now to replace Ka in Paris law (9.89), which is valid for a specific value
of R, with the effective stress intensity factor amplitude Kaeff as follows:
1 1 1 (R)
Kaeff = (Kmax Kopen ) = (1 (R))Kmax = Ka (9.91)
2 2 1R
A plot of da/dN against Ka in log-scale reveals that this relationship is roughly linear
only for intermediate values of Ka . In fact, there is a lower limit Kal below which virtually
no crack propagation takes place. On the other hand, there is an upper limit Kau , above
which the crack growth rate becomes very large. These limit values may be defined by
the corresponding values
where al is a scalar which for metals are in the order of 0.01. In terms of Ka = (1
R)Kmax /2, we thus obtain
(1 R) 1R
Kal = al KIc , Kau = KIc (9.94)
2 2
ln ( da
dN ) R
(2)
R
(1) (2)
< R
1
m
ln K a
(1) (1)
K al K au
valid range for
Paris law
The fatigue life Nf = NR (N +Nd ) is calculated by integrating Paris law (9.92) from the
initial crack length ai up to the failure crack length af . The geometric configuration factor
f is thereby treated as a constant in order to simplify the integration. (For example, we
may choose the value of f for the initial crack length ai in order to avoid an iterative
procedure.) We thus obtain
1
CR f 2 (2a )2
ln aafi if mf = 2
Nf = m m
(9.95)
2 ( 2f 1) ( 2f 1)
(mf 2)CR f mf (2a )mf mf /2
ai af if mf 3
Like in the engineering approach to fatigue analysis, it is necessary to cope with situ-
ations when the loading is irregular and has variable amplitude. Such loading programs
may, in the simplest case, be composed of a sequence of blocks of constant amplitude
stress (or strain), as discussed in Subsection 9.2.2. In the more general situation, quite
arbitrary spectrum loads (with random character) are encountered. Vibrations in vehicles
due to uneven road surface is one typical example.
For the practical analysis, a few basic approaches may be identified, and they all involve
the use of Paris law.
When the loading program consists of a sequence of different blocks of constant amplitude,
it is common to use the Palmgren-Miner rule in much the same fashion as described in
Subsection 9.2.2. Paris law is used to calculate the appropriate life Nf,k for the k:th stress
amplitude. Each such calculation is done completely independently under the assumption
that the crack starts from its initial length ai . Clearly, this means that the influence of
the order, in which the loading blocks occur, is ignored.
In the more complex situation of random amplitude loading the main difficulty is to de-
termine what a cycle is. To this end, it is possible to use a variety of cycle counting
techniques, among which the most widely used is the socalled Rain-Flow-Count method;
see e.g. Suresh (1991), p502. More recently, a method based on statistical repre-
sentations, the Level-Crossing method, was proposed by Holm and deMare (1992).
(Neither of these methods will be further discussed here.)
Cycle-by-cycle method
A quite basic method for an arbitrary loading program is to express the variation of
amplitude as a (continuous) function of N, i.e. we first determine Kaeff (N). Paris law is
then used in straight-forward fashion as
da
= C(2Kaeff (N))mf (9.96)
dN
In practice, this relation has to be integrated numerically.
Sometimes the random loading program consists of a sequence of (nearly) identical blocks,
each of which consists of n cycles with different amplitudes, which is shown in Figure 9.23.
The strategy is then to define an equivalent stress intensity factor Kaeff , that represents
the block loading, such that the crack advance caused by the n cycles in each block, with
eff
amplitudes Ka,k , k = 1, 2, . . . , n, is the same as the crack advance of n cycles of amplitudes
eff
Ka . Thereby, it is assumed that the effect of each cycle is independent. This can be
expressed as follows:
The crack advance per cycle of the original cycles is
da eff mf
= C(2Ka,k ) k = 1, 2, . . . , n (9.97)
dN k
We thus obtain the relation
n n n
! m1
X da X
eff mf 1X f
Paris law is now used in straightforward fashion if only Kaeff is replaced by Kaeff , i.e.
n
da 1X
= C(2Kaeff )mf = C eff mf
(2Ka,k ) (9.99)
dN n
k=1
It is clear that also this approach is based on the assumption that it does not make any
difference in which order the different amplitudes occur.
N
{
n cycles
Figure 9.23: Sequence of identical loading blocks used as basis for the Equivalent Ampli-
tude method.
[1] Cannmo, P., Runesson, K., and Ristinmaa, M., 1995, Modelling of Plasticity and
Damage in a Polycrystalline Microstructure, International Journal of Plasticity,
Vol. 11, pp. 949970
[2] Coffin, L.F., 1954, A Study of the Effects of Cyclic Thermal Stresses on a Ductile
Metal, Transactions of the ASME, Vol. 76, pp. 931950
[3] Crandall, S.H., 1970, The Role of Damping in Vibration Theory, Journal of Sound
and Vibration, Vol. 11, pp. 318
[4] Dougill, J., 1976, On Static Progressively Fracturing Solids, Journal of Applied
Mathematics and Physics (ZAMP), Vol. 27, pp. 423437
[5] Duvaut, G., and Lions, J.L., 1972, Les Inequations en Mecanique et en Physique,
Dunod, Paris
[6] Ekh, M., Grymer, M., Runesson, K., Svedberg, T., 2006, Gradient crystal plasticity
and the modeling of polycrystalline metals, To be published.
[7] Ekh M., Lillbacka R., Runesson K., 2004, A model framework for anisotropic dam-
age coupled to crystal (visco)plasticity, International Journal of Plasticity, Vol. 20,
pp. 21432259
[8] Griffith, A.A., 1921, The Phenomenon of Rupture and Flow in Solids, Phil. Trans-
actions of the Royal Society, London, Vol. A221, pp. 163197
[9] Holm, S., and de Mare, J., 1988, A Simple Model for Fatigue Life, IEEE Transac-
tions on Reliability, Vol. 37, pp. 314322
[10] Hult, J., 1975, BaraBrista, Grundkurs i hallfasthetslara, (in Swedish), AWE/Gebers,
432 p.
[11] Irwin, G.R., 1957, Analysis of Stresses and Strains Near the End of a Crack
Transversing a Plate, Journal of Applied Mechanics, Vol. 24, pp. 361364
[12] Kachanov, L.M., 1958, Time of the Rupture Process under Creep Conditions, Izv.
Akad., S.S.S.R. Otd. Tech. Nauk., Vol 8, pp. 26-31
[13] Lemaitre, J., and Chaboche, J.-L., 1990, Mechanics of Solid Materials, Cambridge
University Press, 556 p.
[14] Lemaitre, J., 1992, A Course on Damage Mechanics, Springer Verlag, 204 p.
[15] Manson, S.S., 1954, Behavior of Materials under Conditions of Thermal Stress,
National Advisory Commision on Aeronautics: Report 1170, Cleveland Lewis Flight
Propulsion Laboratory
[16] Miehe C., 1996, Exponential map algorithm for stress updates in anisotropic multi-
plicative elastoplasticity for single crystals, Int. J. Numer. Methods Engng. Vol. 39,
pp. 33673390
[17] Nemat-Nasser, S., and Hori, M., 1993, Micromechanics: Overall Properties of Het-
erogeneous Materials, North-Holland Series in Applied Mathematics and Mechanics,
687 p.
[18] Norton, F.H., 1929, The Creep of Steel at High Temperature, McGraw-Hill
[19] Ottosen, N.S., and Stenstrom, R., 1995, Unified Approach for Modelling High-Cycle
Fatigue, Dept. of Solid Mechanics, Lund University of Technology, Lund
[20] Palmgren, A., 1924, Die Lebensdaur von Kugellagern, Zeitschrift des Vereins
Deutcher Ingenieure, Vol. 68, pp. 339341
[21] Paris, P.C., Gomez, M.P. and Andersson, W.P., 1961, A Rational Analytic Theory
of Fatigue, The Trend in Engineering, Vol. 13, pp. 914
[23] Runesson, K., Steinman, P., Ekh, M. and Menzel, A., Constitutive Modeling of Engi-
neering Materials - Theory and Computation, Vol I: General Concepts and Inelastic-
ity, Department of Applied Mechanics, Chalmers University of Technology, Goteborg,
(to be published)
[24] Runesson, K., Steinman, P., Ekh, M. and Menzel, A., Constitutive Modeling of En-
gineering Materials - Theory and Computation, Vol II: Damage and Failure, Depart-
ment of Applied Mechanics, Chalmers University of Technology, Goteborg, (to be
published)
[25] Simo, J.C., and Hughes, T.J.R., 1988, Elastoplasticity and Viscoplasticity, Compu-
tational Aspects, Springer (draft edition)
[26] Suresh, S., 1991, Fatigue of Materials, Cambridge University Press, 617 p.
[27] Talbot, A., 1979, The Accurate Numerical Inversion of Laplace Transforms, Jour-
nal of the Institute for Mathematical Applications, Vol. 23, pp. 97120
[28] Wohler, A., 1860, Versuche uber die Festigkeit der Eisenbahnwagenachsen,
Zeitschrift fur Bauwesen, Vol. 10
[29] Akesson, B., 1992, Komplexa tal och funktioner i svangningslaran, (in Swedish), Di-
vision of Solid Mechanics, Chalmers University of Technology, Goteborg, Publication
U56
[30] Skelton, R.P., 1987, Fatigue at High Temperature: Properties and Prediction (ed.),
Elsevier Applied Science.