Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

Lecture Notes

Master Course in Computational Mechanics

Computational Inelasticity (MEC 9)

Prof. Dr.-Ing. habil. Jörg Schröder


Universität Duisburg-Essen
Fakultät für Ingenieurwissenschaften, Abteilung Bauingenieurwesen
Institut für Mechanik

Winter term 2015/2016

Lecturer: A. Schwarz
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 2

Contents

1 Introduction and Motivation 4

2 Continuum Mechanics - Some Foundations 5


2.1 Kinematics and Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Definition of Configuration and Deformation. . . . . . . . . . . . . . 5
2.1.2 Motion of Solid Bodies. . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.3 Time Derivatives of Kinematical Quantities. . . . . . . . . . . . . . 9
2.1.4 Stress Tensors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Balance laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Global balance laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Local Balance Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Constitutive Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Material Frame Indifference. . . . . . . . . . . . . . . . . . . . . . . 15
2.3.2 Axiom of Thermomechanical Admissibility. . . . . . . . . . . . . . . 16
2.3.3 Axiom of Material Symmetry. . . . . . . . . . . . . . . . . . . . . . 17
2.3.4 Concept of Internal Variables. . . . . . . . . . . . . . . . . . . . . . 18

3 Invariant Theory and Representation of Tensor Functions 20


3.1 Characteristic Polynomial . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Spectral Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Cayley-Hamilton Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Matrix Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 Principal and Main Invariants . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 Representation of Tensor Functions - Introduction . . . . . . . . . . . . . . 33
3.7 Representation of Isotropic Tensor Functions H(B) . . . . . . . . . . . . . 33
3.8 Representation of Isotropic Tensor Functions H(A, B) . . . . . . . . . . . 36
3.9 Irreducible Sets of Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4 Isotropic Damage and Elasto-Plasticity at Small Strains 40


4.1 Non-Linear Pure Elasticity at Small Strains . . . . . . . . . . . . . . . . . 41
4.1.1 Model Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.2 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.3 Linearized Variational Formulation . . . . . . . . . . . . . . . . . . 43
4.1.4 Finite-Element-Discretization . . . . . . . . . . . . . . . . . . . . . 44
4.1.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 3

4.2 Isotropic Damage at Small Strains . . . . . . . . . . . . . . . . . . . . . . . 52


4.3 Isotropic Elasto-Plasticity at Small Strains . . . . . . . . . . . . . . . . . . 56
4.3.1 Constitutive Model: Associative Elasto-Plasticity . . . . . . . . . . 56
4.3.2 Elasto-Plastic Tangent Moduli . . . . . . . . . . . . . . . . . . . . . 58
4.3.3 Algorithmic Moduli for Volumetric-Deviatoric Decomposition . . . 58
4.3.4 Model Problem: v. Mises Elasto-Plasticity . . . . . . . . . . . . . . 59
4.3.5 Algorithmic Treatment of Elasto-Plasticity . . . . . . . . . . . . . . 60

Literatur 69
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 4

1 Introduction and Motivation

The aim of the lecture is a compact introduction into the concepts of the modeling of
irreversible material behavior. Therein discretizations in multiple dimensions, conceptual
formulations of constitutive equations for nonlinear material behavior and algorithms for
nonlinear discrete equations are treated.

Learning objectives:
Engineering applications require more and more the use of modern materials which are
characterized by nonlinear mechanical properties. For the simulation of such materials the
mathematical description of the material behavior is as much important as the numerical
implementation. The main goal of this module is to impart fundamental knowledge with
respect to nonlinear constitutive material equations as well as their numerical treatment.
The module explains standard properties as isotropic elasto-plasticity at small strains
and is completed by modern requirements to material modeling such as finite strains or
anisotropy. The student gains skills with a view to numerical material description and
gets to know possibilities and limitations of simulating modern materials.

Subject matter:
a) Motivation and overview
b) Foundations of continuum mechanics
c) Anisotropy
d) Damage at small strains
e) Elasto-plasticity at small strains
f) Hyperelasticity (finite strains)

Prerequisites
Preceding modules: Tensor Calculus, Continuum Mechanics,
Linear Finite Element Method

Examination for receipt of credits and grade points


- exercises and examination at the end of the semester.
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 5

2 Continuum Mechanics - Some Foundations

2.1 Kinematics and Stresses

In this section we summarize the basic elements of continuum mechanics which we need
for the nonlinear problems in the course Numerical Simulations in Mechanics II. For
more details we refer to the text books: Marsden & Hughes [1983], “Mathematical
Foundations of Elasticity”, Prentice-Hall and Eringen, A.C. [1971], “Continuum Physics
Vol. I, II”, Academic Press, New York.

2.1.1 Definition of Configuration and Deformation. The body of interest in the


reference configuration is denoted by B⊂IR3 , parametrized in X. The non-linear defor-
mation map
φt : B −→ S (2.1)
at time t ∈ IR+ maps points X ∈ B onto points x = φt (X) of the current configuration
S, i.e.
φt : X ∈ B 7→ x ∈ S. (2.2)
The deformation gradient F is defined by

F (X) := Gradφt (X); F = F a A g a ⊗ GA , (2.3)

∂xa (X)
with F a A = , and the Jacobian
∂X A
J(X) := detF (X) > 0. (2.4)

It should be noted that φt (X) is not a vector, but rather a point mapping of B to S.
Let TX B and Tx S be the tangent spaces with respect to the reference and the current
configuration and TX∗ B and Tx∗ S their duals. Thus

F (X) : TX B → Tx S (2.5)

is a linear transformation for each X ∈ B.

dX
F dx
X
x
E3 e3
B
S
E2 e2

E1 e1
Figure 1: Non-linear deformation map φt : B −→ S and linear tangent map F (X).
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 6

The inverse deformation gradient is defined as

∂X A
F −1 = gradX , {F −1 }A a = = X A ,a . (2.6)
∂xa
For the construction and geometrical interpretations of various functions we often use
expressions based on the mappings of the area and volume elements. Let N dA and n da
stand for the infinitesimal area vectors and dV and dv denote the infinitesimal volume
elements defined in the reference and the current configuration, respectively, then

n da = Cof[F ]N dA; daa = J{F } a dAA 
−1 A
(2.7)
dv = det[F ] dV 

holds with Cof[F ] = JF −T . Equation (2.7)1 is the well-known Nanson’s Formula. It


should be mentioned that the argument (F , AdjF , det F ), with AdjF = (CofF )T , plays
an important role in the definition of polyconvexity functions.
Remark. Let Pij : IR3×3 → IR2×2 be a projector and F ∈ IR3×3 , where IR3×3 and IR2×2
characterize the sets of real 3 × 3 and 2 × 2 matrices. Then

P ij [F ] = Aij ∈ IR2×2

is a matrix which remains if row i and column j are cancelled. The cofactor of F is defined
by

{Cof[F ]}ij := (−1)i+j det ( P ij [F ]).
In detail we obtain
 
F11 F12 F13
Cof[F ] = Cof  F21 F22 F23 
F31 F32 F33
 ( ) ( ) ( ) 
∗ ∗ ∗
 +det P 11 [F ] −det P 12 [F ] +det P 13 [F ] 
 ( ) ( ) ( ) 
 ∗ ∗ ∗ 
= 
 −det P 21 [F ] +det P 22 [F ] −det P 23 [F ] 

 ( ) ( ) ( ) 
 ∗ ∗ ∗ 
+det P 31 [F ] −det P 32 [F ] +det P 33 [F ]
 [ ] [ ] [ ] 
F22 F23 F21 F23 F21 F22
 +det F32 F33 −det +det
 F31 F33 F31 F32 
 [ ] [ ] [ ] 

 F12 F13 F11 F13 F11 F12 

=  −det +det −det 
F32 F33 F31 F33 F31 F32 
 [ ] [ ] [ ] 
 
 +det F 12 F 13
−det
F 11 F 13
+det
F 11 F 12 
F22 F23 F21 F23 F21 F22
 
F22 F33 − F32 F23 F31 F23 − F21 F33 F21 F32 − F31 F22
=  F32 F13 − F12 F33 F11 F33 − F31 F13 F31 F12 − F11 F32  .
F12 F23 − F22 F13 F21 F13 − F11 F23 F11 F22 − F21 F12
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 7

Let g a and g a be the covariant and contravariant base vectors at the current configuration
and GA and GA the co – and contravariant base vectors defined in the reference config-
uration. In the coordinate systems {xa |a = 1, 2, 3} and {X A |A = 1, 2, 3} the components
of the associated covariant metric tensors gab , GAB are defined by
gab := g a · g b , GAB := GA · GB . (2.8)
The inverse quantities are defined by
gab g bc = δa c , GAB GBC = δA C . (2.9)
g ab and GAB are denoted as the contravariant metric coefficients. With the metric tensor
we can raise and lower the indices of tensor quantities, e.g.:



b a
g a = gab g , g = g g bab 
(2.10)


GA = GAB G , G = GAB GB . 
B A

In direct notation the covariant metric tensors within the Lagrangian and Eulerian settings
are denoted by G and g, respectively. Furthermore their contravariant counterparts are
denoted by G−1 and g −1 . For the representations in Cartesian coordinates we arrive at the
simple expressions GAB = GAB = δAB for Lagrangian metric tensors and gab = g ab = δab
for the Eulerian metric tensors. The right Cauchy-Green tensor C is defined by
C(X) = F T (X)F (X) with CAB = F a A gab F b B . (2.11)
C transforms elements of TX B to elements of TX B, i.e.
C(X) : TX B → TX B. (2.12)
C is symmetric and positive-definite. The so-called Green-Lagrange strain tensor is given
by
1 1
E := (C − G) with EAB = (CAB − GAB ) . (2.13)
2 2

2.1.2 Motion of Solid Bodies. The motion of a solid body is a one-parameter family
of sequences of configurations as a function of time t. Let
[0, T ] ⊂ IR+ (2.14)
be the considered time interval. Then (1) and (2) describe the deformation of the body;
for simplicity we write
x = φ(X, t) = φt (X). (2.15)
The material velocity V (X, t) is defined as the time derivative of the motion, i.e.
φ(X, t + △t) − φ(X, t) ∂φ(X, t)
V (X, t) = lim = . (2.16)
△t→0 △t ∂t
The material acceleration A(X, t) is defined as the time derivative of the material velocity.
We arrive with (2.16) at
∂V (X, t) ∂ 2 φ(X, t)
A(X, t) = = . (2.17)
∂t ∂t2
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 8

It should be noted that the material velocity and material acceleration are parametrized
in the Lagrangian coordinates XA , so we also speak of a Lagrangian description of motion.
In index representation we write

∂xa (X, t) ∂ 2 xa (X, t)


V a (X, t) = and Aa (X, t) = . (2.18)
∂t ∂t2

We assume that the motion of B is regular (invertible), i.e. each φt (B) has an inverse
φ−1
t : φt (B) → B. To shorten notation we write

X = φ−1
t (x, t). (2.19)

Another possibility for the description of motion, in addition to the material (Lagrangian)
approach, is the spatial (Eulerian) description. Changing the independent variables in
V (X, t) and A(X, t) from material to spatial coordinates leads to


v = V ◦ φ−1
t ⇔ V = v ◦ φt 

. (2.20)

a=A◦ φ−1
t ⇔ A = a ◦ φt 

Equations (2.20)1,3 can be written as



( ) ( ) 

v(x, t) = V φ−1
t (x, t), t ; v a (x, t) = V a
φ−1
t (x, t), t

. (2.21)
( ) ( ) 

a(x, t) = A φ−1
t (x, t), t ; a
a (x, t) = A a
φ−1
t (x, t), t

For the material time derivatives of functions defined in spatial coordinates x we have to
take into account the time dependency of x = x(t). Consider

Dv (x(t), t) ∂v ∂v ∂x
a= = + · = v̇ (2.22)
Dt ∂t ∂x ∂t

∂xa
In index notation we arrive with = v a at
∂t

∂v a
aa = + v a ,b v b . (2.23)
∂t

The term v a ,b plays an important role in several applications; we define

∂v
= gradv =: l , v a ,b =: la b . (2.24)
∂x

Here l is denoted as the spatial velocity gradient. A useful identity is

l := gradv = Gradv F −1 = Ḟ F −1 . (2.25)


Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 9

2.1.3 Time Derivatives of Kinematical Quantities. We consider the material


time derivatives of the:
(i) infinitesimal line element
˙ = Ḟ dX = Ḟ F −1 dx = ldx ;
dx (2.26)

(ii) infinitesimal vectorial surface element

da ˙ −T + J Ḟ −T )dA = divvda − lT da ;
˙ = (JF (2.27)

(iii) infinitesimal volume element


˙ = JdV
dv ˙ = ∂F J : Ḟ dV = JF −T : Ḟ dV = JdivvdV = divvdv . (2.28)

Let us now summarize some rates of deformation tensors, which we use in the following
sections. The material rates of the Green-Lagrange strain tensor E and the right Cauchy-
Green tensor are
1
Ė = Ċ. (2.29)
2
The spatial rates could be obtained by the transformation of the material rates to the
current configuration. These are the so-called push-forward operations, e.g.

d = F −T ĖF −1 with dab = {F −1 }A a ĖAB {F −1 }B b . (2.30)

The inverse transformations are denoted as pull-back operations. Furthermore, d is defined


as the symmetric part of the spatial velocity gradient l, i.e.
1
d = sym{l} = sym{Ḟ F −1 } = F −T ( Ċ)F −1 with dab = sym{gac lc b } . (2.31)
2

2.1.4 Stress Tensors. Let t be the traction vector acting on the surface ∂S of the
current configuration S. The fundamental theorem of Cauchy states the existence of the
Cauchy stress tensor σ (true stresses) via the mapping

t = σn , (2.32)

where n characterizes the outward unit normal in x ∈ ∂S. Furthermore, let T denote the
traction vector on ∂B. Starting from the identity

t da = T dA (2.33)

we arrive with (2.32) and (2.7)1 at

σn da = σda = σJF −T dA = P dA. (2.34)

P is the so-called First-Piola-Kirchhoff stress tensor given by

P := JσF −T = τ F −T . (2.35)

In (2.35) we introduced the Kirchhoff stress tensor τ . The (material) Second-Piola-


Kirchhoff stress tensor is defined as

S := F −1 P = F −1 τ F −T = JF −1 σF −T . (2.36)
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 10

2.2 Balance laws

In this section, some basic balance laws in the framework of continuum mechanics are
summarized.

2.2.1 Global balance laws


Axiom of Conservation of Mass. The total mass of a body is unchanged with motion.
Let ρ(x) be the mass density, then the total mass M is given by

M= ρ(x)dv . (2.37)
S

The law of conservation of mass states that


∫ ∫
ρ0 (X)dV = ρ(x)dv, (2.38)
B S

and with dv = JdV we arrive at



(ρ0 (X) − ρ(x)J(X)) dV = 0. (2.39)
B

We can also apply the material time derivative of (2.37); it follows that

d
ρ(x)dv = 0. (2.40)
dt S

Axiom of Balance of linear Momentum. The time rate of change of momentum is


equal to the resultant force F ext acting on the body.
With the external force vector and the total momentum of the body
I ∫ ∫
ext
F := t da + f̄ dv, I := ρv dv (2.41)
∂S S S

this axiom is expressed by


d
I = F ext . (2.42)
dt
Here f̄ denotes the body force vector per unit volume.

Axiom of Balance of Moment of Momentum. The time rate of change of moment of


momentum is equal to the resultant moment of all forces and couples acting on the body.
This is expressed with the resulting external moment and the moment of momentum
I ∫ ∫
ext
M := x × t da + x × f̄ dv, D := ρx × v dv , (2.43)
∂S S S

respectively, as
d
D = M ext . (2.44)
dt
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 11

Axiom of Conservation of Energy. The time rate of the sum of kinetic energy K and
internal energy E is equal to the sum of the rate of work of all forces and couples W and
all other Energies Uα which enter and leave the body per unit time.
With the assumption of the existence of the internal energy density e the total internal
energy is ∫
E= ρ(x) e(x) dv. (2.45)
S
The total kinetic energy of the body is

1
K= ρ(x) v(x) · v(x) dv (2.46)
2 S
and the power of the external loads is
I ∫
W= t · vda + ρf̃ · v dv , (2.47)
∂S S

where f̃ denotes the body force vector per unit mass. With this definition the axiom is
given by
d ∑
(K + E) = W + Uα . (2.48)
dt α

The energies Uα for α = 1, ...n are from thermal, electromagnetic, chemical or other
effects.

Thermal energy: Let r(x) be the heat source per unit mass and q the heat flux,
∫ ∫
U1 := Uthermal = (−q) · n da + ρ r dv. (2.49)
∂S S

Here q is directed outward from ∂S. Now (2.48) appears in the form
∫ ( ) ∫ ∫
d 1
ρe + ρv · v dv = (t · v − q · n)da + (ρf̃ · v + ρr)dv . (2.50)
dt S 2 ∂S S

The spatial heatflux vector q is the so-called true heatflux. Based on the assumption of the
equality of the heatflux with respect to the spatial and material configuration qda = QdA
we get with Nanson’s Formula

q da = q · da = q · Cof[F ]dA =: Qheat · dA = Q dA, (2.51)

where Qheat represents the material heatflux vector.

Axiom of Entropy-Production. The time rate of change of the total entropy H is


never less than the sum of the influx of entropy q in through the surface of the body and
the entropy B supplied by the body sources.
This law is postulated to hold for all independent processes. Let η(x) be the entropy
density and b(x) the entropy source in the body, so we define
∫ ∫
H := ρ(x)η(x) dv and B := ρ(x)b(x) dv. (2.52)
S S
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 12
Qheat q
heat
Q
n q
N
φt (X)

dA
F da

B
S
Figure 2: Heatflux H and h with respect to the reference and current configuration.

With these definitions in hand the total entropy production Γ is characterized by


I
dH
Γ := −B− q in · n da ≥ 0. (2.53)
dt ∂S

The entropy flux is q in = −q/θ and the entropy source is b = r/θ, both facts following
form the axiom of admissibility. Thus we arrive at
∫ I ∫
d 1 1
ρη dv + q · n da − ρ · r dv ≥ 0. (2.54)
dt S ∂S θ S θ

θ denotes the absolute temperature, with


θ > 0 and inf [θ] = 0. (2.55)

2.2.2 Local Balance Laws


For the derivation of local balance laws the following integral theorems are fundamental.
Let S be a material volume intersected by a discontinuity surface Γ̂S (t), which moves with
the velocity v ΓS .

∂S +

n
v Γs
S +

S

S

∂S −
Figure 3: Material volume S with and without discontinuity ∂S
surface ΓS .

The material derivative of a volume integral of a tensor field ϕ over S − Γ̂S (t) is given by
∫ ∫ [ ] ∫
d ∂ϕ
ϕ dv = + div(ϕv) dv + [[ϕ(v − v ΓS )]] · n da. (2.56)
dt S−ΓS S−ΓS ∂t ΓS
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 13

The generalized Green-Gauss theorem for a tensor field σ̃ over a surface ∂S − ΓS of the
body S − ΓS states that
I ∫ ∫
σ̃ · n da = div σ̃ dv + [[σ̃]] · n da. (2.57)
∂S−ΓS S−ΓS ΓS

Here S − ΓS means the volume of S of the body excluding the material points located on
the discontinuity surface ΓS . Furthermore, ∂S − ΓS means the surface which excludes the
line of intersection of ΓS with ∂S, i.e.
∪ ∪
S − ΓS := S + S − , ∂S − ΓS := ∂S + ∂S − . (2.58)

The bracket [[ • ]] characterizes the jump of the quantity (•) across ΓS , i.e.

[[ • ]] := (•)+ − (•)− , (2.59)

and n in (2.57) is the unit normal on ΓS oriented from S − to S + .

Conservation of Mass. Setting ϕ(x) := ρ(x) in (2.56) leads to


∫ [ ] ∫
∂ρ
+ div[ρv] dv + [[ρ(v − v ΓS )]] · n da = 0. (2.60)
S−ΓS ∂t ΓS

Postulating the validity of the balance law for every part of the body and the discontinuity
of the surface yields 
∂ρ
+ div[ρv] = 0 in S − ΓS 
∂t . (2.61)
[[ρ(v − v ΓS )]] · n = 0 on ΓS 

Balance of Momentum. The equation of global balance appears in the form


∫ I ∫
d
ρv dv = t da + f̄ dv . (2.62)
dt S−ΓS ∂S−ΓS S−ΓS

Setting ϕ = ρv (2.56) and σ̃ = σ in (2.57) leads with Cauchy’s theorem t = σn to


∫ [ ] 
∂(ρv) 
+ div[ρv ⊗ v] − divσ − f̄ dv 

∫ S
S−Γ ∂t
. (2.63)


+ [[ρv ⊗ (v − v ΓS ) − σ]] · n da = 0 
ΓS

Considering

∂(ρv) ∂ρ ∂v 

+ div[ρv ⊗ v] = v+ρ + div[ρv] v + ρgrad[v] · v
∂t ∂t ∂t , (2.64)
∂ρ
= ( + div[ρv])v + ρv̇ 

∂t
with (2.61)1 we arrive at
∂(ρv) ∂v
+ div[ρv ⊗ v] = ρv̇ with v̇ := + grad[v] · v . (2.65)
∂t ∂t
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 14

Applying the same localization argument as before and exploitation of (2.64) yields with
σ = σ T (see later) to

divσ + f̄ = ρv̇ in S − ΓS 
. (2.66)
[[ρv ⊗ (v − v ) − σ]] = 0 on Γ 
ΓS S

(2.66)1 is known as Cauchy’s first law and (2.66)2 is the corresponding jump condition
across a singular surface ΓS .

Balance of Moment of Momentum. This leads to the symmetry of the stress tensor,
i.e.
σ = σT in S − ΓS . (2.67)
It is clear that a necessary and sufficient condition for satisfying the local balance condition
for the moment of momentum is the symmetry of the stress tensor.

Balance of Energy. With the integral theorems, as well as the local balance equations
of mass and momenta we obtain in a manner similar to before
}
ρė = σ : l − divq + ρ r in S − ΓS
1 . (2.68)
[[(ρe + v · v)(v − v ΓS ) − σ · v − q]] · n = 0 on ΓS
2

Entropy Inequality. Using the same arguments as before we can express the local
entropy inequality as
q r 
ρη̇ + div[ ] − ρ ≥ 0 in S − ΓS 
θ θ . (2.69)
1 
[[ρη(v − v ΓS ) + q]] · n ≥ 0 on ΓS
θ

2.3 Constitutive Equations

The local balance laws introduced above are valid for all bodies. They constitute five
differential equations: mass (1), momentum (3), and energy (1). In order to determine
the processes we need additional equations for the remaining unknowns appearing in the
balance laws. For this we define the material behaviour by constitutive equations. They
present restrictions on the admissible processes in a body.
Example: An elastic material with heat conduction is characterized by the four response
functions e, θ, σ, q. A formulation in terms of the Helmholtz free energy
ψ := e − θη with ψ = ψ̂(F , θ, gradθ) (2.70)
is often more convenient than using the internal energy function e. With this in hand we
arrive at 
ψ = ψ̂(F , θ, gradθ) 



η = η̂(F , θ, gradθ) 
. (2.71)
σ = σ̂(F , θ, gradθ) 



q = q̂(F , θ, gradθ) 
To construct the necessary constitutive equations and work out several restrictions we
apply some general principles. Several of them are discussed in the following.
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 15

2.3.1 Material Frame Indifference. The principle of material frame indifference re-
quires the invariance of the constitutive equation under superimposed rigid body motions
onto the current configuration. For example under the mapping x → Qx the condition
ψ(F ) = ψ(QF ) has to be fulfilled ∀ Q ∈ SO(3), see e.g. Fig. 4.

S+
B
x+

X
Q

φ(X) x
S
Figure 4: Superimposed rigid body motions onto the current configuration.

For the change of frame we consider a transformation with a time-dependent orthogonal


tensor Q(t). The scalars ψ, θ, η remain uneffected by a change of frame. The tensorial
quantities F , σ, q, gradθ transform as follows:

F −→ QF 



σ −→ QσQT
∀ Q ∈ SO(3) . (2.72)
q −→ Qq 



gradθ −→ Qgradθ

The principle of material frame indifference is satisfied if the response functions obey

ψ̂(F , θ, gradθ) = ψ̂(QF , θ, Qgradθ)  



η̂(F , θ, gradθ) = η̂(QF , θ, Qgradθ) 
∀ Q ∈ SO(3) (2.73)
Qσ̂(F , θ, gradθ)QT = σ̂(QF , θ, Qgradθ)  



Qq̂(F , θ, gradθ) = q̂(QF , θ, Qgradθ) 
for all tensorial and scalar arguments. We are able to derive so-called reduced forms of
the constitutive equations which automatically fulfill these invariance conditions. They
are expressed as 
ψ = ψ̂(C, θ, Gradθ) 





η = η̂(C, θ, Gradθ)
. (2.74)
S = Ŝ(C, θ, Gradθ) 




= Q̂ (C, θ, Gradθ) 
heat heat
Q
In (2.74), we have used the transformations
Gradθ = F T gradθ with θ,A = F a A θ,a . (2.75)
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 16

We arrive at expressions for the primary quantities via



ρ 

σ ab
= S AB (C, θ, Gradθ)F a A F b B 
ρ0 . (2.76)
ρ 

q a
= Q(heat)A (C, θ, Gradθ)F a A 
ρ0

2.3.2 Axiom of Thermomechanical Admissibility. Constitutive equations must


obey the balance laws and entropy inequality. The entropy inequality must be satisfied
for all independent processes. From ψ̇ = ė − θ̇η − θη̇ we obtain an expression for η̇ as a
function of ė. Based on that expression we obtain with (2.68)1 the entropy inequality in
the form
1 1
{−ρ (ψ̇ + θ̇η) + σ : l − q · gradθ} ≥ 0 in S − ΓS . (2.77)
θ θ
This is the so-called Clausius-Duhem inequality. Making use of (2.74) and (2.76) we arrive
at
( ) ( )
ρ0 ∂ψ ρ0 ∂ψ 1 ∂ψ
− + η θ̇ − θ̇,A + S − 2ρ0
AB
ĊAB ≥ 0 in S − ΓS . (2.78)
θ ∂θ θ ∂θ,A 2θ ∂CAB

Here we have used the Piola identity Div(JF −T ) = 0. The equations are linear in
θ̇, gradθ̇, ĊAB . For arbitrary variations of these quantities the inequality holds only if

∂ψ  
AB 
∂CAB 
S = 2ρ0



∂ψ
=0 . (2.79)
∂θ,A 



∂ψ 

η=− 
∂θ
(2.79)2 is satisfied since ψ is independent of Gradθ.
Note: It is reasonable to assume that for rigid deformations and vanishing thermal gradi-
ents there can be no heat conduction, i.e.

Q(heat) (C, θ, Gradθ) = 0 for C = 1 and Gradθ = 0. (2.80)

Theorem: A thermoelastic solid is thermodynamically admissible if the constitutive equa-


tions for t, q, ε, and η are of the form



∂ψ T ∂ψ 

e=ψ−θ 

σ = 2ρF
∂C
F ,
∂θ 
(2.81)


∂ψ  
η=− , 
ρ
q = F Q(heat) (C, θ, Gradθ) , 

ρ0 ∂θ

where ψ = ψ(C, θ) is the free energy and Q(heat) is subject to conditions (62) and
1 (heat)
Q Gradθ ≥ 0 . (2.82)
θ2
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 17

2.3.3 Axiom of Material Symmetry. Constitutive equations must be form-


invariant with respect to a group GK of orthogonal transformations Q ∈ GK of the ma-
terial coordinates. These restrictions are the result of the material symmetry conditions
characterized by GK in the material frame of reference, see e.g. Fig. 5.

B
φ(X)
X

QT x
S

φ(X + )
+
X

B+
Figure 5: Superimposed rotations onto the reference configuration.

Mathematically, the response functions will be form-invariant under all transformations


of the form
X + = QT X (2.83)

subject to
QQT = QT Q = 1, detQ = ±1 (2.84)

for all members of GK . These conditions express geometrical symmetries represented by


GK at X, in the thermomechanical properties of the body. The group GK is a subgroup
of the full orthogonal transformations. When GK is the full orthogonal group, the solid is
called isotropic; when it is the proper orthogonal group, it is called hemitropic. Otherwise,
the material is known as anisotropic. When the response functions do not depend on
translations of the material coordinates it is called homogeneous; otherwise, the solid is
inhomogeneous.
From the free energy function and the heat vector the material symmetry restrictions read



ψ(C, θ) = ψ(QT CQ, θ) 
∀ Q ∈ Gk . (2.85)

Q QT (heat)
(C, θ, Gradθ) = Q (heat)
(Q CQ, θ, Q Gradθ) 
T T 

It should be noted, that if Gk = SO(3) the material is isotropic. There are further material
symmetries which are finite sub–groups of SO(3), for the different crystal classes, see e.g.
Smith, Smith & Rivlin [1963]: “Integrity basis for a symmetric tensor and a vector. The
crystal classes.”, Arch. Rational Mech. Anal., 12, 93–133 and Spencer [1971]: “Theory
of Invariants”, in: Eringen, A.C. (Editor), Continuum Physics Vol. 1, Academic Press,
New York, 239–353, and the references therein.
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 18

2.3.4 Concept of Internal Variables. The principle of thermodynamically com-


patible determinism postulates that for a simple material the local history of F , θ and
gradθ suffices to determine the history of the constitutive behaviour. Thus the quantities
(2.71) could be described by functionals in terms of the history of the process variables.
For practical applications this framework is far to general. An alternative to this is the
so-called thermodynamics with internal variables. The starting point of this concept is
the hypothesis that any thermodynamical process, defined by

{σ, ψ, η, q} ∀x ∈ S . (2.86)

can be completely determined by a finite number of state variables α. The state variables
can be interpreted as a parametrization of the history of the process which approximate
the complex constitutive history functionals by a finite number of variables. Let the set

{F , θ, gradθ, α} (2.87)

determine the thermodynamic state, where {F , θ, gradθ} are instantaneous values and

α = (α1 , ...αk ) (2.88)

are internal variables associated to dissipative mechanisms. Now the specific free energy
is assumed to have the form
ψ = ψ̂(F , θ, α) . (2.89)
The rate of change of ψ is

∂ψ ∂ψ ∑ ∂ψ k
ψ̇ = : Ḟ + θ̇ + α̇i . (2.90)
∂F ∂θ i=1
∂α i

With the stress power


ρ
σ:d= P : Ḟ (2.91)
ρ0
we obtain for the Clausius Duhem inequality

∂ψ ∂ψ ∑ ∂ψ k
ρ0
(P − ρ0 ) : Ḟ − ρ0 (η + )θ̇ − ρ0 α̇i − q · gradθ ≥ 0 . (2.92)
∂F ∂θ i=1
∂αi ρθ

With the standard argument of rational continuum thermodynamics we arrive at

P = ρ 0 ∂F ψ and η = −ψ,θ . (2.93)

Let Ai be the conjugated thermodynamical force to αi , i.e.


∂ψ
Ai := ρ0 for i = 1, ...k , (2.94)
∂αi
then the Clausius Duhem inequality can be rewritten as


k
ρ0
− Ai α̇i − q · gradθ ≥ 0 . (2.95)
i=1
ρθ
Computational Inelasticity, WS 2015/2016, ⃝
c J. Schröder 19

In order to complete the constitutive models we need complementary functions for the
1
description of the dissipative mechanisms. Equations for the heatflux q and the evolution
θ
of the internal variables α̇i for i = 1, ...k are required. An effective way of ensuring that
the Clausius Duhem inequality is satisfied is based on the postulate of the existence of a
scalar-valued dissipation potential (pseudo potential), i.e.

ψ D := ψ̂ D (Ai |i = 1...k, gradθ) . (2.96)

For simplicity assume ψ D to be convex with respect to Ai |i = 1, ...k and gradθ. The initial
conditions are

Ai (t = t0 ) = 0 for i = 1...k and gradθ(t = t0 ) = 0 . (2.97)

Furthermore, we introduce the hypothesis of normal dissipation which leads to

∂ψ D 1 ∂ψ D
α̇i = − | for i = 1...k and q=− . (2.98)
∂Ai θ ∂gradθ

Φ∗ Φ
σ̃ s = konst.
Φ∗ (ε̇r )
Φ(σ̃)

Sup(σ̃ s : ε̇r − Φ∗ ) Ende

ε̇r σ̃ s σ̃
: ε̇ r
s
σ̃
=
Φ∗

Start
(σ̃ : ε̇r − Φ∗ )
Figure 6: Construction of ψ D using the Legendre-Fenchel-Transformation.

You might also like