Sei sulla pagina 1di 62

Chapter 1 Mathematical preliminaries

1.1 Notation
The most common mathematical concepts used in mechanics are vectors and tensors. The modern literature in continuum mechanics (solids and uids) uses at least three types of notation [2, 3]: 1. Index notation: vectors and tensors are expressed in terms of their components on a specied basis. For instance, a vector has components xi and a second-rank tensor has components Ai j with i j 1 3 in three-dimensional space. 2. Matrix notation: vector and tensor components are organized in arrays. For instance, the vector and tensor components mentioned above can be arranged as x

to dene two arrays x and A, respectively. They are also sometimes written as x i and Ai j . 3. Direct notation: vectors and tensors are used as abstract geometrical entities. Their components with respect to a chosen basis are those mentioned above. In direct notation, vectors and tensors are usually printed bold face, i.e. x and A , in books and papers but are written as x and A on the blackboard. Here we will primarily use index and direct notation. Switching between them is done by means of the base vectors with respect to which the components are dened. To avoid mathematical complexities, and without loosing any generality, we conne attention to Cartesian coordinate systems, with orthonormal base vectors, i.e. orthogonal and of unit length. Denoting these base vectors with e i , a vector x and its components xi are connected via


xi e i

xi 1

x1 x2 x3

A11 A12 A13 A21 A22 A23 A31 A32 A33

x ei

(1.1)

The rst of these makes use of the Einstein summation convention: summation is implied over indices that appear twice. The operator in the second expression in (1.1) is the standard dot product of two vectors. For second-order tensors and their components, the corresponding connection reads

The symbol denotes the tensor product that turns two vectors into a second-order tensor 1 . In the left-hand equation, e i e j is the basis tensor with respect to which Ai j are the components of A . Higher-order tensors are dealt with in a similar manner.

1.2 Tensor Algebra


We will frequently make use of the following operations: 1. A mapping of one vector x into another vector y can be described in terms of a secondorder tensor: y Ax (1.3) If the components of all are taken with respect to the same set of base vectors, the mapping of components 2 reads y i Ai j x j (1.4) The second-order unit tensor I is such that

Ix

for any vector x . The components of I are the Kronecker delta

i j

1 if i 0 if i

2. A second-order tensor X is mapped onto another second-order tensor Y by means of a fourth-order tensor L via The fourth-order unit tensor I

Y
4

LX I
4

or Yi j

Li jkl Xkl
4 Ii jkl Skl

maps any symmetric second-order tensor S onto itself,


and its components can be expressed as


aware that some literature, which claims to use dyadic notation, does not use the but just writes x y instead of x y . 2 The operation in eq. (1.4) is often termed contraction over the index j. Note that there is no special symbol for contraction in the direct notation (1.3)

1 Be

or Si j

4 Ii jkl

1 ik jl 2

il jk

Ai j e i

ej

Ai j

e i Ae j

(1.2)

j; j.

(1.5)

3. The scalar product of two second-order tensors is also denoted with and is dened as

A B


Ai j B ji

Note the order of the contractions. Also note that without the , the product of two tensors is another second-order tensor:

4. The transpose A T of a tensor A is dened by




A second-order tensor A is called symmetric if


AT and skewsymmetric or anti-symmetric if

5. A second-order tensor Q is orthogonal if

6. The trace of second-order tensor is the scalar




tr A

I A

Akk

It characterizes the spherically symmetric part of a tensor. 7. The deviator of a second-order tensor, denoted with a is a tensor without its spherically symmetric part: 1 A A tr A I 3

Note that, by denition, tr A

0. 3

Proper orthogonal tensors are those for which in addition det Q represent rotations.

QT

AT

AT

for any two vectors x , y. With A

Ai j e i

e j one readily nds that A ji e i ej

x AT y


yA x

AB

Aik Bk j e i

ej

det Qi j

1. They

1.3 Tensor Calculus


Direct, coordinate-free notation of tensors and vectors emphasizes that they are geometric entities that exist in Euclidean space independent of any coordinate system in which they happen to be observed. Actual calculations, however, do require that a coordinate system be introduced to decompose vectors and tensors into a set of numbers, i.e. their components. A chart or coordinate system is introduced in space so that each point x is attributed a unique set of coordinates xi , i 1 3. The corresponding basis, g i , at this point is obtained from

gi

where x is the position vector of x with respect to some xed reference point O . This denition applies to general curvilinear coordinates, but for simplicity, we shall use only Cartesian coordinates. The corresponding base vectors are usually denoted as e i , and are independent of position. Also, they are orthonormal,

ei e j


so that vector and tensor components are identical to their physical components. In continuum mechanics, we will be dealing with vector and tensor elds, i.e. tensors that are a function of position. Such elds can be expressed as v x or as v xi . The following operations are essential:

1. The gradient 3 of a qth-order tensor eld A x , grad A , is a tensor of order q by A x ei grad A xi so that

For example, the gradient of a vector eld v is a second-order tensor, grad v , with coordinates vi grad vi j vi j x j Note the order of the indices.

2. The divergence of a qth-order tensor eld A x , div A , is a tensor of order q by Ai1 iq A div A grad A I or divA i1 iq 1 xiq
3 Be

aware that some authors use a different denition, where the order of the tensor product in (1.6) is interchanged.

A dA

x grad A dx

x x xi

i j

1, dened (1.6)

1, dened

(note that the operation here involves a double contraction, which reduces the rank of grad A by 2 down to q 1). In particular, for a vector eld v , the divergence div v is a scalar, div v vi i One can show (see Exer. 1.6) that


for any tensor A and vector b . 3. The divergence theorem is the generalization of Greens theorem. For vector elds, say v x , it reads
B B

while for a second-order tensor eld its generalization is

div A dV
B

A n dA

for any continuous tensor eld A dened in a region B of Euclidean space, having boundary B with unit outward normal vector n .

1.4 Exercises
Exercise 1.1 Show that


tr A B Exercise 1.2 Show that AB



T

A B BT AT

Exercise 1.3 Prove that the scalar product of a symmetric tensor and an anti-symmetric tensor vanishes. Exercise 1.4 Show that the unit tensor dened in (1.5) maps any anti-symmetric tensor onto 0 . (Hint: write the mapping in components, interchange the contracted indices and make use of (anti)symmetry.)

Exercise 1.5 Prove that the Cartesian components of grad A , with A of order q are given by
1

Exercise 1.6 Prove the chain rule (1.7).

grad A

i1

iq

Ai1 xiq

div v dV

v n dA

div A b

A grad b

div A b

(1.7)

(1.8)

(1.9)

iq

Chapter 2 Continuum Mechanics


2.1 Introduction
Continuum mechanics is a branch of mechanics which tries to explain the motion of solids, uids or gases in an idealised way. Rather than following the motion of their individual constituents molecules , a continuum point of view is taken. At the macroscopic scale of everyday life, this is how most people tend to look upon a piece of glass, metal or water: as stuff you can see by the naked eye. In fact, this is how scientists, like Gauss, Stokes and Maxwell, looked at matter until late in the 19th century before atoms had been discovered. Of course, we know that matter consists of molecules, and has structure on even much smaller scales. Continuum descriptions are not concerned with the motions of these entities but rather with the average of a huge number of molecules. The quantities we observe by the naked eye are quantities such as density, velocity and temperature. These quantities change on a much larger length and time scales than individual molecules, and can be considered to be continuous functions of position and time. This so-called continuum hypothesis can be exemplied by the example of measuring the density [4]. We consider a control volume V containing a large number of molecules, say N. The total mass M of all the molecules in V is given by M N 1 mk where mk is the mass of the kth molecule. The average density of the k molecules in V is given by

Due to the thermal motion the number of molecules N varies in time, and also the type of the molecules may vary. If we measure we will nd that it depends on V and the outcome of a possible experiment has been depicted schematically in Fig. 2.1. If V is very large, it contains all molecules of the sample and the density is just the overall density of the sample. If we make V smaller, we will reach a plateau, which we will denote as the local density . However, at some point the number of molecules in V becomes small enough to sense statistical uctuations. The smallest volume for which we cannot detect these uctuations using our instruments will be called the physical volume element V . The local density is now 7

M V

V *

Figure 2.1: Measurement of the density over a volume V . given by

In summary, we can introduce the local density if the plateau exists, which requires that 1. V is large enough, such that it contains a large number of molecules. 2. V is small enough, so that it is small with respect macroscopic spatial variations and thus can be considered to be innitely small. If these requirements are fullled we can speak of a continuous medium or continuum and we can use density distributions that are continuous functions of position and time, over time and 3 length scales that are larger than those at the molecular level. The length scale V is considered to be much smaller than the macroscopic length scale L of the body or of the wave length of the loading or motion ( L). Therefore, within the continuum theory we can safely use mathematical limits to zero instead of or V . Consequently, physical quantities are considered to be dened in a point that has no volume at all. In the previous discussion we have used the density as an example. Other quantities, such velocity, pressure and temperature, can also be dened as local values by statistical averaging. Similarly we end up with quantities that are continuous functions of position and time. We assume that these elds are differentiable as much as the theory needs them to be. It is emphasized that continuum theory cannot say anything about events on length scales below . 8

V V

lim

M V

(2.1)

2.2 Kinematics
Within the continuum description discussed above, we will deal with the conguration of a body as a region B of (three-dimensional) space. Material points are identied by their position x with respect to a point of reference O . As the body moves, its conguration changes. But there remains a one-to-one correspondence between the position of a material point in the current conguration (at time t) and its position X in the reference conguration (usually the initial conguration). In other words, the current position of any material point can be viewed as a function x x X t (2.2) The reference position X therefore serves well as a label to identify the material point. Moreover, the function x X t is invertible, i.e.

and differentiable. This is the central point of continuum mechanics and its importance cannot be overestimated. One consequence is that neighboring material points remain neighbors, no matter how much the current conguration differs from the reference. For a slightly deforming solid this is not hard to visualize, but for a uid or a gas it is more difcult. The key in understanding this is that a material point is not a particle of the actual matter. Rather it represents a small sample, containing sufciently many particles to be statistically meaningful. Neighboring particles will certainly not need to remain neighbors (and only occasionally do so, e.g. during elastic deformation), but on average they do. Instead of the function (2.2), it is sometimes convenient to work with the displacement vector of a material point, u x X (2.3) which, obviously, is a function of t and of X or x . The rate of change of the displacement is called the velocity, and is formally obtained by time differentiation 1 of u X t ,

A rigid body translation is characterized by u or v being independent of x . A rigid body rotation is characterized by a displacement eld of the form

with X 0 a stationary point (in the reference conguration) and Q a proper orthogonal tensor, which possibly is a function of time. The motion x X t contains information about the distortion of the material when it is not a rigid body translation or rotation. Locally, at each material point, the distortion is characterized
that it is important here that the argument is considered a function of X , instead of x since the latter is a function of t as well. This is called the material time derivative and is generally indicated by a superposed dot.
1 Note

Q X

u X t t

X0

X x t

by considering the change of innitesimal material line elements from the reference conguration, d X , into the current conguration, d x . Straightforward differentiation of (2.2) yields the following mapping 2 : dx Grad x d X (2.4) where the capital G in Grad emphasizes that the gradient is taken with respect to the reference coordinates Xi . The second-order tensor Grad x is called the deformation gradient tensor and written as (2.5) F Grad x

F with Cartesian components (F

Fi j e i

e j)

Fi j

xi X j

Note that F becomes

I in the reference conguration. In terms of the displacement eld, eq. (2.3), F

The deformation gradient eliminates rigid body translations, because then F I , while for rigid body rotations F I Q . General, nonlinear continuum mechanics allows the displacements to be arbitrarily large and to vary in a general, but continuous, manner in space (see, e.g., [1]). Thus, F can be much different from unity, I . In order to avoid the mathematical complexity that comes with it, we will here conne attention to situations where the deformation gradients are small. Hence, F I , so that according to (2.4) the difference between d X and d x is small. For all intents and purposes, this means that we can neglect the geometrical differences between the reference and the current conguration, so that all functions of the material point X can be regarded as equivalent functions of x . The gradient in (2.6) can thus be replaced by grad, i.e. differentiation with respect to xi , F I grad u or Fi j i j ui j (2.7)

2.3 Strain
As an introduction to strain let us have a look at the physical meaning of the deformation gradient and/or the displacement gradient, for simplicity in two dimensions. We consider, see Fig. 2.2, a rectangular sample spanned by two vectors d X dX e 1 and d Y dY e 2 in the reference conguration. In the current conguration, it has taken a parallelogram shape, spanned by d x and d y , respectively. From (2.4) and (2.7), the components of d x dxi e i are given by u2 1

2 The

x notation d x rather than dx is deliberate to emphasize that it is a vector.

10

dX

dx1 dx2

F11 F21

u1 1

dX

Grad u

(2.6)

e2

u 1, 2 dY u 2, 2 dY dY
X

dy

dx dX e1 u 2, 1 dX u 1, 1 dX

Figure 2.2: Two-dimensional illustration of the displacement gradient.

Similarly, for the components of d y

dyi e i ,

Figure 2.2 reveals that u1 1 gives rise to a stretch of the sample in the x1 -direction, while u2 2 gives a stretch in x2 . The displacement gradients u1 2 and u2 1 tend to shear the sample. If all ui j vanish except u1 2 , the deformation is referred to as simple shear in the x1 -direction. Strain in the intuitive sense of the word is elongation, change of length. The well-known uniaxial denition of strain is l l0 e (2.8) l0

i.e. the elongation l measure of strain is

l0 compared to the original length l0 . Although not often used, another

which is approximately equal to e when the elongation is small, see Exer. 2.2. This denition now is useful to dene strain in the general three-dimensional context. We dene a second-order strain tensor on the basis of the difference of the squared lengths, just as in (2.9),


dx2

dX 2

where dx and dX are the lengths of a line element in the current and reference conguration, i.e. dx d x d x and similarly for dX . Note that, by construction, is a symmetric tensor. Both sides of this denition can be expressed in terms of d X by making use of d x F d X according to (2.4), so that d dX F T F dX d X d X 2d X d X 11
 

2 l 2 l0 2 2l0

(2.9)

d 2d X d X 2dXi i j dX j

(2.10) (2.11)

dy1 dy2

F12 F22

dY

u1 2 1 u2 2

dY

Since we have not specied that this relation should hold for any specic line element, we require that the above equality holds for any d X to nd

When we substitute the expression (2.7), the strain tensor is expressed in terms of the displacement gradients, 1 grad u grad u T grad u T grad u 2 or 1 i j ui j u j i uk i uk j 2 All these expressions in fact hold for arbitrary displacement gradients 3 . For small displacement gradients, the strain tensor reduces to

In this form, and with reference to Fig. 2.2, its interpretation is quite straightforward. The normal component 11 u1 1 denotes the extension of the material in the x1 -direction; 22 u2 2 does so in the x2 -direction, etc. The off-diagonal, shear component 12 is the average of the changes in angle in the x1 x2 plane, 12 1 u1 2 u2 1 4 . 2 Notice from (2.13) that the strain tensor is just the symmetric part of the displacement gradient. The antisymmetric part,

is the rotation tensor. Its off-diagonal component 12 describes the rotation of the material element in the x1 x2 plane in the clockwise direction. One of the most important aspects of continuous displacement elds is that they ensure that the deformed body is compatible, loosely phrased this means that the deformed body ts together. The strain eld computed from the displacement eld using (2.13) inherits this property. On the other hand, when one gives a strain eld, it is not necessarily derivable from a continuous displacement eld. In that case, compatibility has to be enforced by an additional condition. The compatibility condition reads (cf. Exer. 2.3)

jl

and ensures that the strain eld is derivable from a displacement eld. Out of these 81 equations only six are independent, just enough for the six components of the strain tensor.
the nonlinear continuum theory, this is called the Green-Lagrange strain tensor. we will not employ it, it is useful to note that some older literature employs a different denition of shear strain, the so-called engineering shear strain, which is exactly double, e.g. 2 12 .
4 Although 3 In

12

i j kl

kl i j

ik

jl ik

1 grad u 2

grad u

or i j

1 ui 2

uj i

(2.14)

1 grad u 2

grad u

or i j

1 ui 2

1 T F F 2

or i j

1 Fki Fk j 2

i j

(2.12)

uj i

(2.13)

2.4 Stress
Point masses can be loaded by forces. Continua can also be loaded by forces, but then these will be distributed over the boundary B of the body. Force per unit area is called traction and is denoted here by t . The total force on the body is
B

and the moment with respect to a reference point x R is


B

into the body. If the body is undeformable 5 its behavior is governed solely by the classical laws of mechanics. In the case of quasi-static motions, i.e. in which inertia can be ignored, Newtons law become a force balance and a balance of momentum, which from (2.15) and (2.16) read
B B

If the body is deformable, these laws still apply but they are not sufcient to solve for the motion of all material points. It is intuitively clear that the complete solution depends on the behavior of the material, an issue we will return to later (in Sec. ??). In order to get access to a description of the material, we will need to be able to transfer information on the boundary of the body, i.e. boundary tractions and boundary velocities/displacements, to the interior of the body. We already have one entity that contains information about the internal state of the body, namely strain, and the other one will be stress. This concept can be introduced in several ways. The traditional one is to rst consider tractions on internal surfaces of the body obtained by cutting it up in more pieces. In the limit of an innitesimal piece of the material in the shape of a tetrahedron 6 the tractions of all faces can be shown to form a second-order tensor. We will here introduce stress via the concept of virtual work, since this is something that we will need anyway later on. As the term suggests, virtual work is not real work but the work done during a virtual motion, characterized by an arbitrary virtual displacement eld u x . This displacement eld is an instantaneous change of the current displacements, with the adjective instantaneous emphasizing that it does not depend on time 7 It is important to note that u is not the variation of
evidently is an idealization tetrahedron is the natural element of volume in three dimensions, just like a triangle is the natural surface element in a plane. 7 In fact, the virtual displacement eld is better thought of as a velocity eld, since this also gives the instantaneous change of the displacement eld (but now in time).
6A 5 which

13

t dA

0 and

Pex

The tractions insert a power

t v dA

x R dA

x R dA

t dA

(2.15)

(2.16)

(2.17)

(2.18)

the actual displacement eld u x t ; it is a completely imaginary eld. The virtual work, Wex , of the current tractions is Wex (2.19) t u dA
B

cf. with the expression for the power (2.17). The principle of virtual work states that the virtual work is zero as long as the body does not deform. The latter condition poses certain constraints on the form of the virtual displacement elds. Since we have strain as the measure of deformation, the deformation associated with an arbitrary virtual displacement eld u x is described by 1 grad u grad u T (2.20) 2 Hence, the principle of virtual work can be formally written as

Statements like this occur in physics quite frequently, and a natural way to incorporate the side condition 0 is to add it using Lagrange multipliers. Since is a symmetric second-order tensor, we need a symmetric second-order tensor as Lagrange multiplier. The rst reason is that the side condition actually entails six equations for the six strain components. A nonsymmetric second-order tensor as Lagrange multiplier would have nine components; three of these, corresponding with its antisymmetric part, would vanish because the scalar product of a symmetric and an antisymmetric tensor vanishes (see Exer. 1.3). When we denote this Langrange multiplier with and note that the side condition 0 is dened in every point in the interior, (2.21) can be re-phrased as
B

or

with the internal virtual work dened as


B B

In this form, we will use the principle of virtual work later to introduce the nite element method (e.g. Sec. 3.5). We will now see what the consequences are of (2.22). For this, let us rst substitute by its denition (2.20), grad u dV Win
B

14

u n dA


Win

div u

div u dV div u dV

where we have made use of the symmetry of to obtain grad u use of the chain rule (1.7) we can re-write the above expression as


Win

dV

i j i j dV

Wex

Win

t u dA


dV

Wex

u for which

(2.21)

u (2.22)

(2.23)

grad u . Making

where in deriving the second equality we have made use of the divergence theorem (1.9), n being the outward normal to the boundary B. Substituting the latter expression into (2.22) and re-arranging yields
B

Since this condition contains separate, irreducible integrals over the arbitrary volume and its boundary, we must require both integrands to vanish for arbitrary u . This leads to

t and

n or ti 0

i j n j

on V , 0 in V .

In order to get some insight in the interpretation of (2.24), consider the simple situation depicted in Fig. 2.3. A corner of an innitesimal material element is shown with traction vector
22 e2 n = e2 12 21 t ( 1) 11 n = e1 e1 t ( 2)

Figure 2.3: Two-dimensional illustration of the relation between traction and stress. on the two boundaries. Note that we consider here a single material point so that there is only one stress tensor, while there are, for the moment, two independent traction vectors t 1 and t 2 . First look at the boundary where the unit outward normal coincides with e 1 . The components of the traction vector t 1 become

Hence, in this example, 11 is the stress normal to the boundary n e1 and 21 is the shear stress in the x2 direction. Similarly for the other boundary in Fig. 2.3,

15

t1 2 t2

12 22

on n

e2.

t1 1 t2

11 21

div

or i j

n u dA

div u dV

(2.24) (2.25)

Here, we sample the second column of i j , so that 21 is the shear component and 22 is the normal component. Note also that because of symmetry of i j , we must have 12 21 so that the tangential components of the traction vector necessarily must satisfy

t1

t2

These interpretations are readily generalized by returning to (2.24). The diagonal components of the stress tensor represent normal components to planes with normals in one of the base vectors. The off-diagonal components are shear stresses: i j is the traction in the xi direction on planes with normal to the x j direction 8 . The condition (2.25) is the local force balance 9 as is illustrated in Fig. 2.4. It shows a
22 + 22, 2 dx 2 12 + 12, 2 dx 2 e2 11 dx 2 21 x 12 22 e1 dx 1 21 + 21, 1 dx 1 11 + 11, 1 dx 1

Figure 2.4: Two-dimensional illustration of the equilibrium condition (2.25). innitesimal sample of a two-dimensional body of dimensions dx 1 by dx2 located at the point x . The sample is considered to be cut away from the body and the interactions with the rest of the body represented as appropriate tractions on its boundaries. As we have seen above, these tractions are related to the stress components at the particular point. At x , the stress give n n traction components on the lower (n e 2 ) and left-hand sides (n e 1 in the negative x2 and 10 in the x direction on the lower and left-hand sides is x1 directions, respectively. The force 1

11 dx2

12 dx1

as illustrated in Fig. 2.4 as well. The force in the x1 direction on the top and right-hand sides is

Generically, shear stress if often denoted by the symbol opposed to the global force balance (2.18)a. 10 per unit thickness in the x direction 3
8 9 As

16

11

11 1 dx1 dx2

12

12 2 dx2 dx1

At x

d x , the stress components are i j x dx i j i j k dxk

so that the net force in x1 direction on the entire sample becomes


Since dx1 and dx2 have been chosen arbitrarily, force balance in this direction implies that

which is indeed what follows from condition (2.25) in two dimensions applied in the direction i 1. One can easily convince oneself that the other two conditions in (2.25) govern equilibrium in the x2 and x3 directions.

2.5 Transformations
In the previous sections, we have developed the basic theory as much as possible in direct notation. These results then are very general and apply to any set of base vectors. However, for illustration purposes we have used simple orthonormal base vectors and samples parallel to these vectors, cf. Figs. 2.2, 2.3 and 2.4. The latter is of course done for convenience and is advisable whenever possible. Sometimes, however, it is necessary to choose other directions, which will lead to other components of the same vector or tensor. The most frequent of such situations occurs when the original base vectors are rotated by a certain angle. In such a case, the original and rotated base vectors e i are related by

ei

Ti j e j

ei

e j T ji

Any vector or tensor can be decomposed on either e i or e i . The resulting components are then related to one another through the transformation matrix T . For a vector x , for instance, after substitution from (2.26)

and conversely, (note that the base vectors transform in the same way as the corresponding components). For the example in Fig. 2.5, one nds

17

x1 x2

cos sin

sin cos

xi

x j T ji

or x

xi e i

xi T ji e j

xje j

xi

Ti j x j TTx

or x

x1 x2

e1 e2

cos sin sin cos

e1 e2

Tx

with the matrix T of coefcients Ti j being proper orthogonal. For example, the shown in Fig. 2.5 are obtained from e i through a rotation by the angle ,

11 1

12 2

11 1

12 2 dx1 dx2

(2.26) base vectors

(2.27)

e* 2

e2

e* 1 e1
Figure 2.5: Two-dimensional illustration of a rotation of the vector basis. In a similar manner, any second-order tensor A can be decomposed on either basis. This leads to the transformation rules

and conversely Particularly useful for the purpose of solid mechanics are the transformation rules for symmetric tensors such as the strain tensor or the stress tensor . After some algebra (see Exer. 2.5), the stress components, for instance, are found to transform as

Ai j

Tki Akl Tl j

or A

11 12

11 cos2
2

22 sin2 11 sin2

212 sin cos 212 sin cos


12 cos

under the transformation of base vectors (2.27). Since the strain tensor is a symmetric secondorder tensor as well, its components transform in a completely equivalent manner. The transformation rules (2.29) can be conveniently re-written as

It is in this form that they are often used to assess the stress state on a surface at some angle compared to a reference place, see Fig. 2.6. Let the reference plane be such that the i j components are given, with 12 0. The normal stresses 11 and 22 are then called principal

18

12

sin 2

12 cos 2

22

11

1 11 22 2 1 11 22 2 1 11 22 2

22

22 cos2

sin
2

22

11 sin cos

1 11 2 1 11 2

22 cos 2 22 cos 2

Ai j

Tik Akl T jl

or A

TAT T TTA T

(2.28)

(2.29)

12 sin 2 12 sin 2

(2.30) (2.31) (2.32)

* 12 e* 2 e2 11 12 12 22

* 11

e* 1 e1

Figure 2.6: Two-dimensional illustration of the stress components on a plane inclined at from the reference plane.

* * 11, 12 2 2 * * 22, 12 1

Figure 2.7: Illustration of the transformation rules (2.30). See text for explanation. stresses (more about these subsequently) and denoted by 1 and 2 , respectively. The question as to what the normal and shear stress are on a plane inclined at an angle are now immediately given by (2.30). Plotted in shear stress () versus normal stress () space (Fig. 2.7), the normal stresses 11 and 22 lie on a circle with radius 1 1 2 centered at 1 1 2 2 2 (recall that 12 0). The stress state on planes rotated by away from these principal stress planes are found by tracing an angle 2 on this surface, see Fig. 2.7. This circle is very useful to remember the transformation rules (2.30), and is called Mohrs circle after the German civil engineer O.C. Mohr (18351918). 11 . Note that because of the double angles traced in the
textbooks on solid mechanics often derive Mohrs circle and the transformation rules (2.30) by considering triangular pieces cut out of an innitesimal rectangular sample and requiring equilibrium. This is slightly misleading in the sense that it would perhaps suggest that (2.30) and Mohrs circle rely on equilibrium. The
11 Elementary

19

circle, 22 is located opposite from 11 , in agreement with the fact the respective planes are at 90 with respect to each other. Also note that the maximum shear stress is found at planes with 4 from the principal stress planes. Principal stress states are very important for a number of physical mechanisms, for instance brittle fracture. By denition, principal stresses are normal stresses on the planes oriented such that the shear stress vanishes. As illustrated above, in two dimensions that can be easily found by rotating the planes and the associated base vectors. The notion can be readily extended to three dimensions by noting that, by denition, on a principal stress plane the traction vector is parallel to the normal on the plane and has a magnitude , i.e.

On the other hand, the traction components are in general related to n i via (2.24). By equating these two expressions for ti , we obtain the eigenvalue problem

We know from linear algebra that every symmetric 3 3 matrix has 3 real-valued eigenvalues, which tells us that there are in general three different principal stresses which are the eigenvalues of i j . It is a good practice to denote these principal stresses by i and to order them so that 1 2 3 .

Figure 2.8: The complete Mohrs circles for three dimensions. Now that we know this, we can go back to the idea of rotations of base vectors that led to Mohrs circle in (2.30). The two principal stress 1 and 2 in Fig. 2.7 are only two of the
present derivation, on the basis of changes of base vectors, show that they have nothing to do with equilibrium but, rather, are a direct outcome of the use of tensors. The approach has the additional advantage that one immediately sees that the transformation rules for strain components or any other symmetric second-order tensor components are directly similar.

ti

ni

i j

i j n j

(2.33)

20

three principal stresses and were obtained by performing rotations in the two-dimensional subspace with x3 const. In a similar fashion, one can perform rotations in the subspaces with x1 const and with x2 const. In each of these subspaces, one nds a Mohrs circle based on two principal stresses. Since these must be out of the set i , i 1 2 3, three of the total of six principal stresses must be identical, leading in summary to the three nested Mohrs circles shown in Fig. 2.8.

2.6 More about stress and strain


Being second-order tensors, stress and strain tensors contain all the information about the states of stress and strain in a material point. In general, this information comprises six numbers for both. There are several types of scalar measures of parts of this full information. The hydrostatic stress or mean stress is the spherically symmetric part of the stress tensor, dened by 1 1 m (2.34) tr kk 3 3 With the opposite sign it is the pressure p m . Similarly, the volumetric strain is dened as

tr

kk

With the spherically symmetric part being contained in m , the stress deviator dened by

contains information about the non-spherical part of stress. The strain deviator is dened in a similar measure in terms of v . Note that, by denition,

As a consequence, the internal power per unit volume, cf. (2.23), can be expressed as

Finally, it is sometimes useful to work with the invariants of stress and strain. These are, mathematically dened, scalar representations of parts of these tensors which are independent of the choice of base vectors (i.e. unlike the usual components, see Sec. 2.5). Every symmetric second-order tensor has three invariants, but they can be dened in different ways. We will here use the following ones, based on the trace (introduced for stress, and similar for strain): 1 1 2 2 ij ij 1 J3 3 i j jk ki The rst invariant, J1 , contains essentially the same information as the hydrostatic stress or pressure. The second and third invariants contain scalar information about the non-hydrostatic components of stress. 21

J2

J1

tr

m v

tr

tr

m I

or i j

i j

m i j

(2.35)

(2.36)

2.7 Exercises
Exercise 2.1 Consider the displacement eld

u1

x1 x2

ui

0 otherwise.

1. What are the physical dimensions of ? 2. Consider a small element like that in Fig. 2.2 and study how it deforms. What is the physical interpretation of this displacement eld? 3. What is the interpretation of ?

Exercise 2.3 Check the compatibility condition (2.14) by substitution of the straindisplacement expressions (2.13) and differentiation. Exercise 2.4 Calculate the net moment about the center of the sample in Fig. 2.4 in the spirit of (2.18)b. Which property of the stress tensor guarantees that the net moment vanishes?
N.B. Most textbooks argue in the opposite direction: they start out from force and moment equilibrium, and then derive this property by application to a sample like in Fig. 2.4. It is worthwhile to carefully think about the different manner in which this property has been derived in the virtual work approach used here.

Exercise 2.5 Derive the transformation rule (2.29) for symmetric second-order tensor components. Exercise 2.6 Check that 12 changes sign when is replaced by 2 by considering equilibrium of the triangular pieces of this simple square under uniaxial stress.

Exercise 2.7 Why is there a factor 1 3 in the denition (2.34) of the hydrostatic stress but not in that of the volumetric strain (2.35)? Exercise 2.8 Derive eq. (2.36). 22

Exercise 2.2 Show that e

according to (2.8)(2.9) if e

1.

Exercise 2.9 It is often convenient for calculations to gather the stress components i j in an array as follows: i j 11 22 33 12 13 23 T Show why

is the corresponding notation for the strain components. Exercise 2.10 The strain eld

is singular in the origin and there considered suspicious. Check if this eld is compatible.

23

x1

x2

x1

x2

13 x1 x2 x3

x2

23 x1 x2 x3

x1

i j

11 22 33 212 213 223

otherwise i j

24

Chapter 3 Elasticity
3.1 Elastic energy
Elasticity in every day life means: reversible. Elastic deformation is the kind of deformation that takes place in a reversible manner: the material deforms under stress, but when the stress is released, the body instantly returns to its undeformed conguration. Formally, elasticity is best introduced through energy considerations. The work by tractions on the boundary of the body is stored as internal energy U , and upon removal the loading, the internal energy is released. Internal energy can, evidently, be measured per unit volume, i.e.


with u the specic internal energy. The rate of change of the bodys internal energy, U, is equal to the external power given by (2.17). Making use of the the virtual work statement (2.22) (2.23) (and replacing the virtual quantities by rates), we nd

u dV
B

Since this must hold for any body (or part of that), we conclude from (3.1) that


saying that the power of stresses (given by the right-hand side) is instantly stored as specic internal energy. For elastic deformations, the specic internal energy, u, is a direct function of strain,

so that its value in the undeformed state, u 0 , is some reference value (often taken to be zero). As a consequence, u u 25

u dV
B

(3.1)

dV

(3.2)

(3.3)

Figure 3.1: Schematic representation of a polymeric chain. From [5]. and equating this to (3.2) yields the key result:

or i j

u i j

(3.4)

The right-hand sides of these expression is in general still a function of strain, so that they give a direct relation between stress and strain once the internal energy function is known. Physically, in many materials like metals and organic materials, elasticity arises from the stretching of bonds between atoms. The internal energy in such cases can be identied with the atomic interaction energy. This does not hold, however, for rubbers. The dominant internal energy that holds macromolecular chains together are strong covalent bonds which create the stiff backbone of molecules. Elastic deformation of polymers arises from changes in the conformation 1 of macromolecular chains by rotation of segments. This is much easier than stretching the backbone and it is this process that gives rise to rubber elasticity. However, this is not controled by internal energy. Instead, what changes during deformation is the entropy of the molecular network, s, as illustrated in Fig. 3.1. Therefore it is convenient to introduce the specic free energy for a continuum by the usual Legendre transformation

Ts

T being temperature. Hence, both classes of materials can be handled in terms of a free energy function and (3.4) is replaced by

or i j

i j

(3.5)

The precise response of an elastic material is thus incorporated through its free energy function. Once this function is dened the stressstrain equations follow from (3.5) 2 . It is important to note that the free energy can also depend on the temperature T .
is the physical outline of a molecule, which can only be altered by rotation of the chain atoms about single bonds. 2 Note that the stressstrain law is a set of six equations for the six components of stress and strain. But they are dened through only a single scalar-valued function.
1 Conformation

26

3.2 Linear elasticity


The free energy can in general be any complex function of strain. However, when one connes attention to small elastic strains, the function can be simplied. The key idea is to imagine that the free energy function is expressed as a series expansion in terms of high-order polynomials of . The rst term in the series expansion is a constant; i.e. the reference free energy and we will assume without loss of generality that it vanishes. The second term will be linear in , and according to (3.5) this will give rise to a constant stress independent of strain. This is physically not very relevant, so we continue to the next term, which is quadratic in strain. When leaving out the linear terms all together and terminate the expansion after the quadratic terms, we have

so that we arrive at linear stressstrain relations,


The L is called the stiffness tensor of the material. Its components on a Cartesian basis possess the following symmetries:

Li jkl

because of symmetry of the stress and the strain tensor, and

Li jkl

because they enter as a quadratic form in (3.6). In general, there are 21 independent coefcients (see Exer. 3.2). The number of independent coefcients reduce if the material has certain symmetries. Qubic crystals, for instance, have cubic symmetry. The elastic moduli for a cubic crystal should reect these symmetries, i.e. the tensor components should be invariant under symmetry operations that leave the crystal unchanged. Cubic symmetries are conveniently introduced by noting that 90 rotations about an edge of the unit cell (a 100 direction) leave the material unchanged. This can be incorporated by requiring that the components Li jkl on such rotated base vectors are identical to the original ones Li jkl . If the rotations are denoted by Ti j as in Chap. 2, this requirement reads Li jkl Tia T jb Labcd Tkc Tld (3.8) There are three of such rotations, but they do not all lead to independent conditions. Eventually, they lead to

27

L1111 L1122 L1212

L2222 L2233 L2323

L3333 L3311 L1313

1 L 2

1 i j Li jkl kl 2

(3.6)

or i j

Li jkl kl

(3.7)

L jikl

Li jlk

Lkli j

C11 C12 C44

(3.9)

etc. In addition to these rotational symmetries, cubic crystals have 180 rotational symmetry about their cubic axes, i.e. the 111 directions. Invariance under these transformations leads to the conclusion that the cross terms between normal and shear components vanish, e.g.

L1123

L2231

L3312

etc. The nal result is that there are only three independent moduli for cubic materials, namely C11 , C12 and C44 as dened in (3.9). An even more restrictive material symmetry is isotropy. This means that the material response is invariant under all rigid body rotations, i.e. (3.8) must hold for all proper orthogonal matrices Ti j . This is possible, but it is in fact easier to return to the free energy formulation (3.6). For an isotropic material, the strain dependence of must be expressed in terms of the strain invariants, since these are the only representations of the strain tensor that are invariant under any change of basis 3 If we again conne attention to quadratic terms in strain, can only depend on I1 2 and I2 ; we write

1 k I1 2

2I2

coefcients k and . Differentiation now proceeds as follows:

dI1 dI2 I1 d I2 d I 2 kI1

or in components 2i j 2 kk i j k 3

Evidently, the number of independent coefcients Li jkl is only two:

Notice the extended form of the rst term on the right-hand side. This has been done to express the necessary symmetries of Li jkl mentioned above. The interpretation of the physical meaning of k and is conveniently done by rst considering a pure shear strain situation in the x1 -x2 plane. First note that, like for all shear components, 12 12 . From (3.10), 12 212
scalar

3 The

free energy function i j must also be invariant under rotations of the base vectors, since it is just a

28

Li jkl

ik jl

il jk

2i j

i j

kkk i j

2 i j kl 3

(3.10)

(3.11)

which immediately demonstrates why is known as the shear modulus 4 . Next, consider straining such that 11 and 22 33 0. Straightforward substitution into (3.10) then gives

2 3 The rst of these is obvious: straining in one direction requires a stress in the same direction. The appearance of stress in the transverse directions may be somewhat unexpected at rst. However, it is quite natural. If one stretches a rubber band, one can easily see that the material contracts in the directions perpendicular to the straining direction. In the strain state assumed above, we prevented this contraction by imposing 22 33 0. In response, the material builds up a tensile stress. The contraction is seen more easily by inverting (3.10) to express strain as a result of stress; in general, from (3.7), or i j Mi jkl kl (3.12) M with M being called the compliance tensor. Inverting (3.10) most easily done by rst noting that 1 kk 3kkk so that kk kk 3k which demonstrates why k is called the compression or bulk modulus. Using the latter equality to eliminate kk from (3.10) and re-arranging, we nd

and

For shear stresses, this result is trivial. For normal stresses, one nds

while the corresponding results for the other two normal strain components are found by cyclic interchange of the indices 1, 2 and 3. In the last equality above, we have replaced k and by two other frequently used parameters, namely Youngs modulus E and Poissons ratio . Youngs modulus, according to the above expression, is the stiffness against uniaxial straining, 11 11 E when 22 33 0. The Poisson ratio is the ratio between the transverse strains and the strain in the stress direction, say 22 33 11 under uniaxial stress 11 . The complete correspondence between the two groups of coefcients is given in Table 3.1. With the
factor 2 has a historical origin. In the early days of elasticity, people used a different denition of shear strain, namely 212 .

4 The

29

11

1 k 2 3 11 11 22 33 2 3k 3k k 2 3 11 22 33 9k 6k 1 11 22 33 E

i j

1 i j 2

2 3 kk i j 3k

22

33

11

4 3 k

(3.13)

Table 3.1: Conversion between various isotropic elastic moduli. aid of this table it is easy to rewrite (3.10) as 1

and (3.11) as

which are to be supplemented with boundary conditions in terms of displacements, u i ui 0 , or tractions, i j n j ti 0 . In the cases of isotropic elasticity, so that the constitutive relation can be written as (3.10), the three equations inside the body can be combined to give a second-order differential equation in the displacement eld,

2 30

ui kk

uk ki

equilibrium: i j constitutive equation: i j

Li jkl kl

kinematics: i j

1 ui 2 0

uj i

1 1 1 2 ik jl il jk k i j kl 2 2 3k 3 Finally, we summarize the governing equations of linear elasticity:

Mi jkl

or

Mi jkl

1 ik jl E 2

il jk

i j kl

1 ik jl il jk 1 2 Finally, the compliance moduli can be cast in the form

Li jkl

i j

i j

1 2

kk i j

i j kl

9k 3k

3k 6k

2 2

3 1 2 k 21

21 31

31

2 k

3k E 6k

3Ek 9k E

E 9 3E 2

21

2 2

E 21

E 31 2

(3.14)

(3.15)

(3.16)

(3.17) (3.18) (3.19)

when the material is homogeneous. This set of three equations is known as Naviers equation 5 . Notice that the equations are independent of Youngs modulus, but that the solution may still depend on E because of the boundary conditions.

3.3 Planar problems


Three-dimensional solutions are quite hard to obtain, but there is a well-developed methodology for two-dimensional problems. We will give here a rst glance on the latter for isotropic materials. There are two limiting cases of two-dimensional problems: one being called plane strain and the other plane stress. Plane strain is characterized by the strain conditions

i3

123

(3.20)

when x1 x2 is the plane of interest. Plane stress conditions are similar but with strain replaced by stress. The governing equations for planar problems are conveniently written in index notation with Greek indices in the range 1 2 . Substituting the plane strain conditions into Hookes law (3.14), the planar version becomes

while the out-of-plane normal stress is given by 1

and the equilibrium conditions Evidently, one can combine them again to arrive at second-order differential equations in displacements, just like in the three-dimensional case discussed above. However, a more useful approach for many problems is to express all governing equations in terms of stress. To this end, one needs to invert (3.21) or work directly from (3.13) with the aid of (3.22). In this derivation it is useful to note 1

5 This

is the same Navier as the one in the more famous NavierStokes equations for viscous uids.

31

E 1

and 13 tions,

23

0. The governing equations are completed with the straindisplacement rela 1 u 2 u

33

E 1

2 3

1 2

(3.21)

(3.22)

(3.23)

The result is

(3.24) E From here on, the derivation is not trivial. If the strain satisfy this constitutive law, and the stresses satisfy equilibrium, there is one more set of equations to be satised, namely the kinematic relation between strain and displacement. From displacement to strain, this is evident, but the reverse route is less evident. What we need to require is that the strains from (3.24) can indeed be derived from a displacement eld this is the compatibility condition (2.14) discussed in the previous chapter. For planar problems we only need the subset

and simplify to get


The nal step is to multiply this by to get


or With this, (3.26) simplies to

These are called the BeltramiMichell equations 6 . These expressions are instrumental in deriving a solution procedure that is extremely powerful for 2D problems with predominantly traction boundry conditions. It is based on noting that the equilibrium conditions (3.23) have a particular divergence form often encountered in physics, namely

is the plane strain version. The full three-dimensional version is derived in a completely analogous fashion from the three-dimensional compatibility conditions (2.14). Contrary to the present two-dimensional version, the general three-dimensional expression depends on .

6 Evidently, this

32

xx x yx x

xy y yy y

0 0

0 (3.27)

Use equilibrium (3.23) to eliminate

0 and

0 so that (3.26)

Substitute Hookes law (3.24) into this for

: 0

(3.25)

(3.28) (3.29)

when written with x x1 and y x2 . One frequently succeeds in nding a solution in terms of scalar functions. Here we attempt a solution in terms of two functions and such that

For the solution of this equation, we attempt a similar idea and suppose that

Expressed in this function, the stresses are written as


and since alone gives the stresses it is called a stress function. Moreover, it is called the Airy stress function after the British astronomer Sir G.B. Airy (18011892) who rst suggested the use of a stress function. The Airy stress function ensures that the stresses according to (3.30) satisfy equilibrium. This is a necessary condition but not sufcient. should also ensure that compatibility and boundary conditions are met. We require compatibility through the condition (3.27), which in the present notation reads

After substitution of (3.30) we obtain

or when using the well-known 2 operator

33

2 x2

2 y2

2 x2

2 y2

2 x2

2 y2

xx

yy

xx

2 y2

xy

2 xy

2 x2

2 y2

so that equilibrium, (3.28), is always satised. Since xy

yx

xy

xx

y y

xy

x x yx , these functions must obey

yy

2 x2

(3.30)

xx

yy

(3.31)

(3.32)

y ds

dx ds dy

Figure 3.2: Denitions for the derivation of the boundary conditions in terms of the Airy stress function, (3.33). for Cartesian components. The power of the Airy stress function approach lies in the fact that there are many solutions of the biharmonic equation (3.31) known in mathematics. For example, known solutions include x x 2 x3 x3 y x 2 y xy y y 2 y3 y3 x y 2 x cos y cosh x y cos y cosh x cos x cosh y x cos x cosh y

and all linear combinations. The only thing left to do is to specify boundary conditions. We will conne attention to traction boundary conditions for which the stress function approach is most useful. Imagine a piece of boundary, see Fig. 3.2, at which the traction vector is given along the boundary with coordinate s. The components of the normal vector can be expressed in terms of s using the usual left-hand rule for the contour around the body:

With this, the condition that the traction, with components tx ty , match the stresses from (3.30) at the boundary becomes

Finally, the procedure for solving a plane strain problem with Airy stress functions is to nd the biharmonic function that satises the boundary conditions (3.30). Since the solution in of linear elasticity problems is unique, the solution one nds is the correct solution. Boundary value problems with prescribed displacements can be done in principle, but are more cumbersome because the boundary conditions have to be expressed in terms of through Hookes law and the strain-displacement expressions. 34

tx

d ds

ty

d ds

nx

cos

dy ds

ny

sin

dx ds

x4

y4 x2 y2

1 4 3y

(3.33)

e2

er r

e1

Figure 3.3: Denition of polar coordinate system and associated base vectors. We have derived the Airy stress function approach in terms of Cartesian components. The result, however, holds more general. This can be easily seen by noting that the 2 operator in (3.31) is dened for any coordinate system. In fact, 2 div grad . For a polar coordinate system r , with x r cos and y r sin (see Fig. 3.3), it reads

The stresses in the corresponding basis are found from

3.4 Cracks
The Airy stress function approach is one of the convenient methods to solve for the elds around the tip of a sharp crack. The analysis we present is an asymptotic one in the sense that we search for the solution close to the crack tip, i.e. as r 0, with r the distance from the tip, see Fig. 3.4. The length of the crack therefore is irrelevant for the solution. The boundary conditions for this problem are that the crack faces are traction free, i.e.

35

2 0 r2 1 r r

2 r2

rr

1 1 2 r r r2 2 1 r r

2 r2

1 r r

1 2 r2 2

at

(3.34)

Figure 3.4: Denition of the asymptotic analysis of near-tip elds in an isotropic linear elastic material. The solution is sought for by separation of variables,

where f can also depend on the parameter . Substitution into the biharmonic equation yields that the rdependence is always satised and that f must obey an ordinary differential equation (as is typical for separation of variables). The solution of the latter is of the form

with C1 to C4 integration constants (possibly depending on ). Substitution into (3.34) gives four equations for these four integration constants. These equations are homogeneous and have a non-zero solution of the Ci s if they are dependent. This yields an eigenvalue problem from which it follows that there only is a non-zero solution when satises

sin2 2

All roots of this equation are double (because of it being a quadratic condition) and read

With these roots, the rdependence of the stresses,

i j r goes as i j r

36

1 2

r0 r1

1 2

k 1

IN

C1 sin

C2 cos

C3 sin

C4 cos

(3.35)

(3.36)

a w

h
Figure 3.5: Denition of the asymptotic analysis of near-tip elds in an isotropic linear elastic material. in increasing order of the exponent. All these solutions are admissible 7 and the actual solution can be a linear combination of all. Neverheless, the one that dominates close to the tip is the 0. This singular term r 1 2 all other contributes remain nite or even decay to 0 when r means that to leading order, the stress eld near the tip of a crack takes the form

where fi j is some function that will follow from the function f above. The coefcient of the square-root singularity is generally denoted by K 2 although there is no reason for the 2. The parameter K determines how singular the eld is it is known as the stress intensity factor. Dimensional considerations of (3.37) reveal that the stress intensity factor has dimension length and therefore is typically expressed in terms of MPa m. stress The value of K does not result from the above asymptotic analysis (which is typical for eigenvalue problems). Its determination requires analysis of a real specimen or component containing the crack, as illustrated for instance in Fig. 3.5. It is key to recall that the asymptotic solution (3.37) is valid only at the very tip of the crack (r 0). The full solution has been obtained in the past by a variety of methods (increasingly, numerical methods) and the results have been tabulated in comprehensive books. Without performing the actual analysis, for the geometry sketched in Fig. 3.5, one can guess the dependence of K on the problem parameters just from dimensional considerations. Remember that K has dimensions stress length; this means that one must be able to express it in terms of another quantity with dimension stress times the square root of a length. The candidate for stress is obvious: the remotely applied stress . There are three lengths that determine the problem: the crack length a, the specimen width w and the height h. It is more or less physical intuition that the state near the crack tip is controlled most by a 8 so that we conclude that

ones with k 0 in (3.35) are not admissible because they would lead to an innite elastic enery. The square-root singularity is admissible since this leads to a nite energy, see Exer. 3.8. 8 Hint: consider the two limiting situations: a 0 and a w.

37

7 the

K a

i j r

K fi j r 2r

(3.37)

The non-dimensional coefcient here will depend on the dimensionless ratios a w and h w.

3.5 Finite element method


The nite element method is a numerical technique to obtain approximate solutions. There are different ways of introducing it and there exist numerous excellent textbooks, e.g. [6]. We will introduce it here on the basis of the principle of virtual work, which makes it easy to generalize the method in the next chapter for nonlinear problems. At the end we will show that for linear elastic problems, the method can also be obtained from a variational statement. The basic idea behind the nite element method is to try and nd approximate solutions to the governing equations (see (3.17) for linear elasticity) by chopping the body B up into small pieces the nite elements inside of which one seeks simple approximate algebraic solutions. When all the elements are then assembled to construct the discretized version of the actual sample of material, the mathematical problem becomes an algebraic one which provides an approximate solution to the problem. Adjacent nite elements are connected by sharing nodal points or nodes. The set of elements that discretize the material is the nite element mesh. Triangular elements for two-dimensional models, shown for example in Fig. 3.6c, have three nodes at their vertices, three-dimensional brick elements have eight nodes, but there are variants of these elements that have more nodes positioned along their edges. The latter, higher-order elements allow the element boundaries to be curved; this offers the possibility to described the shape of e.g. particles more closely, but a faceted approximation of a curve is often acceptable. In the most popular version of the nite element method, the displacements or the velocities at the nodes serve as the nal unknowns in the nal set of algebraic equations. Inside each element, the displacement eld u i x j is approximated by the values uI at the, say, N e nodes on the element (I 1 N e ) by relatively i I x , simple interpolation functions (known as shape functions) N j ui x j

I 1

For three-node elements in two dimensions, for instance, the displacements are interpolated linearly from the nodal values. The shape functions are dened so that the displacements between two nodes on the element boundary only depend on the nodes on this boundary; this guarantees displacement continuity across element boundaries. Boundary conditions are incorporated in the nite element method through the external virtual work statement (2.19). The mesh introduced inside the region induces a discretization of the boundary, so that the boundary integral becomes a sum of integrals over the n element boundaries Ae (e 1 n). Inserting the interpolation (3.38) for the virtual displacements in each element adjacent to the boundary, the virtual work of external loads becomes Wex

I 1e 1

Ae

38

ti ui dA

fieI uIi

e I fi

Ne

NI

Ne

x j uI i

(3.38)

Ti N I dA

(3.39)

Figure 3.6: The computation of stresses in an inhomogeneous material rst requires the actual microstructure (a) to be idealized as a material model (b) that can be analyzed numerically in terms of a nite element model (c).

39

The e fiI is the work-equivalent force vector acting on node I due to the traction distribution along element boundary e. According to (3.39), a node in between two elements gets two contributions to its nodal force, one from each element. The total force at a node, f iI , is the sum of the contributions of the element boundaries sharing node I, and the total external virtual work can be written as
I 1

where N now denotes the total number of nodes in the entire mesh. A signicant number of all nodes will not lie on the boundary and will not get a contribution from the boundary tractions. Implicit in (3.40) is that the nodal force on internal nodes is zero after all, there is nothing to pull on an internal node. According to this energetically consistent method, the prescribed distributed loading is represented in the nite element model by the corresponding nodal forces. Prescribed displacements along other parts of the boundary are simply incorporated by prescribing the displacements at the nodes on the boundary. Note that all nodes on the boundary must either have prescribed nodal displacements or prescribed nodal forces. Next we turn to the internal virtual work (2.23). Again we split up the integral in integrals over elements, i.e.
B
e e 1 B

where Be is the volume of element e 1 n, and n now is the number of volume elements. In preparation for the calculation of the element integrals, we rst substitute the elastic constitutive law (eq:linseps)b, so that the element virtual work becomes

e Win

Be

i j i j dV

Be

i j Li jkl kl dV

The (virtual) strains can be evaluated inside each element from the interpolated (virtual) displacement eld expressed through (eq:dispintpol) using the denition (eq:epsugrad). Differentiation involves the interpolation functions N I x j now, and the strains can be expressed as

I 1

BIi j puIp

Ne

in terms of the nodal displacements. The last equality is just the short-hand denition of B I jk . i Notice that this is a constant for linear interpolation functions, but is in general a polynomial over the element to which node I belongs. Substituting this expression and the similar one for the virtual strains into (3.41), we obtain

e Win

40

I 1J 1

uIp

Ne Ne

Be

BI j p Li jkl BJ dV uJ i q klq

i j

1 N N I uI 2 I j i 1

N Ii uIj

Win

i j i j dV

Wex

fiI uIi

(3.40)

i j i j dV

(3.41) (3.42)

(3.43) (3.44)

after appropriate re-ordering of terms. The matrix of coefcients

IJ K pq

is called the element stiffness matrix and allows us to write


e Win

I 1J 1

Recall that nodes I and J are in general shared by a number of elements. This means that in the total internal virtual work,
e 1

there are several element contributions that involve the same (virtual) nodal displacement u J . q By adding them all up, the total internal virtual work can be written as Win with

I 1J 1

where e runs over all elements sharing nodes I and J. According to the principle of virtual work, the internal work (3.46) and the external work (3.40) must be equal for all nodal virtual displacements. This leads to the following set of equations
J 1

Note on the left-hand side that there is an explicit summation over the node index J as well as implicit summation over the displacement component q. In practice, the nodal displacements uIp are ordered in an array u as

node1

node2

nodeN

while the nodal forces are gathered in a similar way into an array f , so that the total external IJ virtual work can also be written Wex f T u. The coefcients K pq then get ordered into a 3N 3N matrix, K, so that (3.47) takes the form

Ku

41

u 1 u1 u1 u2 u2 u2 1 2 3 1 2 3

KiIJ uJj j

fiI

KiIJ j

e KiIJ j

uN uN uN 1 2 3

uIi KiIJ uJj j

Win

e Win

Be

BI j p Li jkl BJ dV i klq

(3.45)

uIi eKiIJ uJj j

Ne Ne

(3.46)

(3.47)

(3.48)

This is evidently a large set of equations, but the solution is aided by the fact that it is sparse when the nodes are ordered in an efcient way9 . For such systems of equations there are efcient solvers available these days (e.g. in libraries shipped with compilers) and also the amount of storage is limited. Once the system of equations (3.48) has been solved for the unknown nodal displacements I , all other quantities can be computed. Strain elds inside an element are obtained from up (3.43) and, subsequently, the stress elds are directly obtained by substitution into the constitutive equations (3.7)b. It is important to notice that the displacement elds in a nite element discretization are continuous across element boundaries, whereas strain and stress elds, by construction, are only continuous inside the element with discontinuous jumps in between. Stresses and strains are usually (e.g. in commercial programs) calculated only in special points, the integration points. These points are introduced during the evaluation of the element stiffness matrices according to (3.45). The integrals involved are typically integrated numerically, usually by Gaussian quadrature [6]. This process introduces integration points that are usually located inside an element. The number of such points depend on the type of element and on the quadrature rule that is being used. Although stresses and strains are typically only computed in the integration points, the entire elds within an element can be constructed from the integration point values in the element by interpolation and extrapolation. Even though stress and strain elds are only piecewise continuous, the distributions of stresses and strains in the entire material sample being analyzed are usually provided after smoothing. This is typically done by rst computing for each element the nodal values from the integration point values, followed by averaging the value at each node over all elements connected to that node. This smoothing process is allowed when the stress or strain jumps across element boundaries are sufciently small, which will be the case when the mesh is well designed. Note that smoothing in a multiphase material is best performed separately for each phase, since some components of the actual stress and strain elds may jump discontinuously across phase boundaries.

3.6 Exercises
Exercise 3.1 Write the matrix L of elastic moduli, cf. (3.7), in the notation of Exer. 2.9. Explain why it is useful to name the cubic moduli according to (3.9). Exercise 3.2 Show that there are only 21 independent coefcients in the fourth-order modulus tensor Li jkl in (3.7). Exercise 3.3 Prove (3.11) on the basis of (3.10). Exercise 3.4 Show that a cubic crystal, with elasticity coefcients C11 , C12 , C44 , is isotropic when C11 C12 2C44 (3.49)
9 Efcient

node numbering is such that the numbers of nodes on any element differ as little as possible.

42

Exercise 3.5 What value of Poissons ratio characterizes a material that is incompressible? Exercise 3.6 Derive (3.24). Exercise 3.7 Consider a homogeneous stress and strain state as sketched in the gure above,
e2

e1

The same situation can also be seen as a combination of pure stretch and compression in the principal directions. 1. Calculate the principal stresses and strains, as well as the corresponding principal directions. 2. Employ Hookes law in the form (3.14) to relate the principal stresses and strains in terms of E and . 3. Make use of the equivalence of both views to prove the relationship

see Table 3.1. Does it make any difference whether the situation is plane strain or plane stress? Exercise 3.8 The stress eld in the direct neighbourhood of a crack tip, as given by the expression (3.37), is singular (if the material is regarded as behaving in a linear elastic fashion). The corresponding strain eld is also singular, which raises the question about the near-tip elastic energy. 43

E 21

and i3

i3

0 (i

1 2 3). According to Hookes law (3.10) (3.51)

1 2

0 1 1 0

i.e.

0 1 1 0

(3.50)

(3.52)

1. Show that the specic free energy (i.e. per unit of volume) can be expressed in terms of the stresses as 1 i j Mi jkl kl 2 2. Calculate the elastic energy W in a cilindrical area with radius R around the crack tip in Fig. 3.5. The exact form of fi j does not need to be taken into account there. 3. Show which other possible Airy stress function according to (3.35)(3.36) guarantees a nite energy.

44

Chapter 4 Crystal Plasticity


Crystal plasticity is the common term for models describing the plastic response of individual crystals (or grains or crystallites) or that of a polycrystal. The latter polycrystal plasticity models are used to describe the plastic deformation of, mostly, metals taking into account that they consists of distinct crystals, each with its own response. For this reason they are potentially capable of capturing texture development during, for example, forming processes. Single crystal models can be seen as the basic building block of polycrystal models. It describes plastic deformation of a crystal taking into account that it is a highly anisotropic material, in which plastic deformation can take place only along well-specied slip systems. Asaro [9] gives an excellent overview of the basics of the subject. Computational aspects can be found in [10], while a modern account for applications can be found, for example, in [15]. Applications in the prediction of forming operations involve large to very large strains. The nite strain formulation of crystal plasticity introduces some extra mathematical complexities that we wish to avoid here. This chapter will therefore continue to assume small strains, just as before. The nite strain version has been treated last year in the Micromechanics course.

4.1 Single crystal plasticity


The formulation of the theory requires a description of the geometry of the slip systems, the kinematics of deformation due to slip and constitutive laws. The latter express, in some phenomenological manner, the effect of physical mechanisms at a smaller length scale (dislocations).

4.1.1 Geometry
Material science (e.g. [5]) tells us that slip occurs by virtue of dislocations moving over certain planes and in certain directions. The combination is called a slip system. The slip systems are determined by the crystal structure, e.g. face-centered cubic (fcc), body-centered cubic (bcc) or hexagonal closed-packed (hcp). In fcc crystals, slip may occur on the close-packed planes and in the close-packed directions (Fig. 4.1). This leads to a total of 12 slip systems with slip 45

plane normals of the type 111 and slip directions of the type 110 , using the standard Miller indices. The shorthand notation for these slip systems is 111 110 . Bcc crystals have many more slip systems; at least 24, but even 48 according to some authors.
[ 110 ]

( 111 )

e1

Figure 4.1: Unit cell and one of the 12 slip systems for an fcc crystal. Using Miller indices, the slip plane is 111 and the indicated slip direction 110 is one of the three equivalent directions of the type 110 . For the mathematical formulation of plasticity theories the slip plane normals are denoted by m and the slip directions by s , where runs from 1 to the number of slip systems. During large plastic strains, the orientations of the slip systems may rotate, leading to texture development, but this is neglected in this chapter. Hence, m and s are xed vectors.

4.1.2 Kinematics
The total deformation of a crystallite is the result of two distinct physical mechanisms: crystallographic slip due to dislocation motion on the active slip systems, and elastic lattice distortion. Plastic deformation of the crystal is envisaged to occur as a set of plastic simple shears along s the various slip systems, leaving the lattice and the slip systems vectors (s , m ) undistorted (and unrotated within our small strain assumption). Next, the material and lattice are considered to deform elastically from the plastically deformed state to the current conguration. Thus, the total strain tensor is decomposed additively in a plastic and an elastic part:

This decomposition can also be understood through the following thought experiment. Imagine a crystal being deformed elastoplastically in a homogeneous manner until a certain stress level and then being de-stressed. This de-stressing is elastic and thus part of the strain, namely e is recovered. The remaining part is remaining strain, in other words: plastic strain, i.e. p e . This decomposition is illustrated in Fig. 4.2. 46

e3 e2

(4.1)

m s

m s

Figure 4.2: Decomposition of the total deformation in a plastically deforming crystal into a plastic part and a part due to lattice strain. By denition, p describes the plastic deformation caused by a set of plastic simple shears along the various slip systems. With the shear on system , the plastic strain tensor p can be constructed by making reference to the simple shear process discussed in Sec. 2.3. Lets rst consider single slip, i.e. plastic deformation on only a single slip system with slip plane normal m and slip direction s . Let the base vectors coincide with the slip plane direction and normal as s e 1 , m e 2 ; from Sec. 2.3 we then know that single slip is described by a displacement gradient with nonzero components 12 only. If we denote this component by , the displacement e s gradient tensor is e 1 e 2 s m . The strain tensor is the symmetric part of this, i.e.

1 s 2

for single slip. In case of multiple slip, each slip system can generate a contribution like this to the plastic strain, so that in total 1 p (4.2) P 2 with the orientation tensor P dened through

4.1.3 Constitutive laws


The elastic deformation, i.e. the distortion of the lattice, is still governed by Hookes law as discussed in Sec. 3.2. With full account of the elastic anisotropy of the crystal, the elastic constitutive law reads (4.4) L e with the total strain in (3.7 being replaced by the elastic strain e . The stiffness tensor L is based on the anisotropic elastic constants. For an fcc crystal they thus exhibit the appropriate cubic symmetry with respect to the cube axes; the components are given in (3.9) when the base vectors are taken along the cube axes. We know from material science textbooks such as [5] that slip occurs by the motion of dislocations on a smaller length scale. Moreover, dislocations have long-range interactions, 47

or Pi j

mi s j

1 s 2

1 s mj 2 i

(4.3)

leading to patterning and junction formation. The latter gives rise to hindrance of dislocation motion, which we observe as hardening. All this renders plastic deformation to be inherently history dependent. As a consequence, constitutive models for it are must be formulated in terms of rates. The plastic strain rate is immediately obtained by differentiation of (4.2) as

since we neglect changes of the crystal orientation. Suppose now that at any instant we know the slip rates and therefore the instantaneous plastic strain rate. Making use then of the rate form of the decomposition (4.1), we can eliminate the elastic strain rate from (4.4) and obtain

with

With the afore-mentioned presumption, we also know 0 in (4.6) so that we can compute the change in stress for a given applied strain rate from (4.6). However, we do not know the values of the slip rates and their dependence on the state of the material. This is governed by the physics of dislocations which determines how slip in the crystal evolves. Since these are phenomena taking place below the current scale of observation, this cannot be described by the present theory but needs to be added through constitutive laws 1 . The most important ones are (i) the yield criterion and (ii) the hardening law. Yield condition Crystallographic slip is primarily determined by the so-called Schmid stress or the resolved shear stress. For any slip system , this is dened by

so that from (4.5) one has that the plastic work rate can be expressed as : D p . is conjugate to , the condition for slip on any slip Consistent with the observation that system to occur is formulated in terms of the resolved shear stress. The simplest form of a yield criterion then reads that slip commences when

is similar in spirit as the description of elasticity in the previous chapter. In continuum elasticity, one averages out over individual atoms or molecules, while in continuum plasticity one averages out over dislocations as well. This immediately suggests that continuum elasticity may work well down to smaller length scales than continuum plasticity.

1 This

48

:P

L 0

1 P 2

(4.5)

(4.6)

LP

(4.7)

(4.8)

(4.9)

where g is the current yield strength of slip system . The latter need not be a constant, but can evolve with continued plastic deformation (see Sec. 25). Just like in classical phenomeno G . The slip system is then said logical plasticity, continued plastic ow requires that the system is inactive. G to be active. When Not all of the n possible slip systems need to be active, depending on the deformation process. If fact, for any general three-dimensional plastic strain rate, only ve slip systems need to be active 2 . The determination of the active slip systems is not trivial, however, and quite advanced procedures have been developed for this purpose (see e.g. [14, 12]). A computationally convenient way around this is to use a viscoplastic formulation instead of the above time-independent condition. Asaro and Needleman [10] proposed the ow rule

For large values of the strain-rate sensitivity m, say m 100, the stress dependence of the shear rate is so highly nonlinear that (4.10) practically implies a boundary between elasticity and plastic ow just like in the rate-independent criterion (4.9). It should be noted that there are physical reasons for plastic ow being rate sensitive indeed. However, fcc crystals are practically rate-insensitive at room temperature; bcc crystals are more rate sensitive [9]. The form (4.10) is computationally convenient, for it can be simply substituted into (4.6)(4.7). Hardening models The shear strength, or hardness, g , of a slip system will evolve with plastic deformation and is mainly controlled by the underlying dislocation structure. This, in turn, is determined by the slip activity on all slip systems. Consistent with this, slip system hardening is generally expressed by the evolution law

in terms of a hardening matrix h . The initial value of g is taken to be a constant 0 for each slip system. The key phenomenon for hardening of slip system is that the dislocations on other slip systems hinder the motion of dislocations on slip system . This so-called latent hardening is a complex and not fully understood/modeled process, and therefore hardening models are currently phenomenological to some extent. We discuss two: rst one that is purely phenomenological, and then a model which is at least motivated by the physics. Asaro and Needleman [10] proposed the following form of the hardening matrix:

origin of this number is as follows. Strain is a symmetric second-order tensor and therefore is characterized by six independent components. Plastic strain preserves volume, i.e. tr p 0, which eliminates one of them. This leaves only ve independent variables.

2 This

49

q h

no sum on

0 g g

m 1

(4.10)

(4.11)

(4.12)

where h is a single slip hardening rate, and q is the matrix describing the latent hardening behaviour of the crystallite. For fcc crystals with 12 slip systems, q is given by q

where q is the ratio of the latent hardening rate to self hardening rate, and A is a 3 3 matrix fully populated by ones. In the above, slip systems 1 2 3 are coplanar, as are systems 4 5 6 , 7 8 9 and 10 11 12 . Thus the ratio of the latent hardening rate to the self hardening rate for coplanar systems is taken as unity. The value of q is typically taken as 1 4. Various expressions for the single slip hardening rate have been proposed in the literature. For instance [10] h0 hs h hs h0 hs sech2 a (4.14) s 0 where a is the total accumulated slip,

Another often used expression is the power law

An example of a hardening theory which is more physically motivated is the one by Bassani [11]. It relies rst of all on experimental observations of various stages of hardening, as illustrated in Fig. 4.3. After the regime of easy glide (stage I), hardening increases rapidly in stage II due to the interaction with forest dislocations. At even larger strains, the hardening is determined mainly by the dislocation cell structures. The second observation is that latent hardening interpreted as the hardening of, say, slip system 2 by slip system 1 occurs in the form of very rapid hardening once the second slip system is activated, Fig. 4.4. This led Bassani [11] to suggest that q 0 in (4.13). The remaining diagonal part he then suggested to be of the form

with the self-hardening term specied as

and the interactive hardening term according to

50

tanh

hs

h0

hs sech2

h0 hs I 0

h G

h0 1

g s

dt

A A A A qA qA qA A A A qA A qA qA A A A qA qA A qA A A A qA qA qA A

(4.13)

(4.15)

(4.16)

(4.17)

(4.18)

(4.19)

Figure 4.3: Typical curve for a fcc single crystal oriented initially for single slip. Point B indicates when secondary slip commences. From [11].

Figure 4.4: Typical curve for a fcc single crystal during a latent hardening experiment. From [11].

51

The amplitude factors f depend on the type of dislocation junction between slip systems and and can be found in [11]. The I in (4.18) is the so-called stage I stress and h0 is the initial hardening rate; hs is assumed to depend on the total accumulated slip a on all slip systems:

where hI and hIII are the hardening rates during the stage I and stage III, respectively, and III is s s 0 approximately the accumulated slip at the onset of stage III.

4.2 Numerical solution


Because of the complexity of the constitutive equations of crystal, and the complexity of the associated stress and strain elds, it is practically impossible to nd closed-form analytical solutions except for a few simple cases. A numerical solution technique is unavoidable. The approach takes two steps related to the fact that the governing equations are partial differential equations in space (equilibrium and kinematics) and in time (crystal plasticity constitutive equations). We will discuss here only the implementation of the rate sensitive version based on the viscoplastic ow rule eq. (4.10). This is essentially the same as for any viscoplastic model, and is based on the observation that in such a case the tensor 0 in the right-hand side of (4.6) is known in any given state. What is not known in this constitutive law is the strain rate it is to be determined so that all governing equations are satised. One of these is the equilibrium condition or, equivalently, the principle of virtual work. However, the constitutive equation here is not in terms of stress but in terms of stress rate. But if we require rate equilibrium at all times, then equilibrium will also be satised. In a small strain formulation as we maintain here, rate equilibrium is expressed simply by the time derivative of the equilibrium conditions, or of the principle of virtual work 3 . The latter becomes, from (2.22),
B

The next step, as in Sec. 3.5, is to substitute the constitutive equation (4.6). After re-arrangement, this gives i 0j i j dV i j Li jkl kl dV ti ui dA (4.21)
B B B

The rst term is exactly the same as in the elastic nite element method, except from the fact that it is expressed in terms of strain rate instead of strain. Its discretization proceeds in directly similarity with that in Sec. 3.5, eq. (3.46). In the same way, the rst term in the right-hand side is the rate counterpart of that in (3.39), which led to (3.40), but now in terms of the nodal force rates fiI . Plasticity enters in the last term in the right-hand side and this is the only term
taking the time derivative of the virtual work statement, one needs to be aware of the fact that the virtual displacement eld is instantaneous by denition. Its derivative vanishes!
3 In

52

ti ui dA

hs

hI s

hIII s

hI tanh s

i j i j dV

a III 0

(4.20)

that needs separate attention. Its discretization is carried out in a similar manner as that of the left-hand side. First, the integral is written as a sum of integrals per element,

Ne

i 0j i j

dV

e 1

and then the virtual strain is expressed in terms of the virtual nodal displacements, similar to (3.43). Substitution and re-arrangement yields

i 0j i j dV

I 1e 1

uIp

The element integral here has physical dimensions of force; for this reason we introduce the short-hand notation e 0I i i fp 0j BI j p dV (4.22)

Be

so that

Summing all the contributions in the right-hand side to each node, we can write

i 0j i j

dV

I 1

fi0I uIi

similar to (3.40). Now, we can return to (4.21) and by requiring that it should hold for all virtual nodal displacements, we nd
J 1

It is emphasized that in this viscoplastic formulation, the effect of plasticity only enters through the additional term fi0I in the right-hand side. In practice, certainly when plastic deformation is non-uniform, the integrals in (4.22) need to be evaluated numerically. As mentioned in Sec. 3.5 this is typically done by Gauss quadrature, involving evaluation of the integrand in a small number of integration points inside each element. Once the nite element equations are solved for the nodal velocities u J , we are done with j the spatial part of governing equations. From the nodal velocities, we can nd the velocity eld inside all elements from the interpolation

ui x j

I 1

NI

Ne

x j uI i

cf. (3.38), while the strain rates are obtained via

i j

I 1

BIi j puIp 53

Ne

KiIJ uJj j

fiI

i 0j i j

dV

I 1e 1

e fi0I uIi

Ne

Be

i 0j i j dV

Be

i i 0j BI j p dV

fi0I

(4.23)

Figure 4.5: Optical micrograph of polycrystalline aluminum showing patchy slip. From [9]. i cf. (3.43). The plastic strain rates, or the plastic stress rates 0j , were already known so that we can compute the stress rates i j from the constitutive equations (4.6). Also we can then compute the rates of change entering in the hardening model, eqs. (4.11)(4.20). The nal step is to perform integration in time. The most straightforward manner is to use Euler integration. That is, at time t, where we know the entire state of the material (collectively denoted by q), we compute all rates as discussed above and estimate the state at t t from

Since one simply works with increments of quantities, q, this procedure is also known as an incremental method. When all quantities are updated in this manner, one calculates the new slip i rates at t t from (4.10), the plastic stress rate 0j from (4.7), and sets up the new right-hand side in the nite element equations (4.23). This closes the incremental loop in time. The above-mentioned incremental approach is straightforward and easily implemented, but it is not very stable numerically. The reason for this is the large exponent m in (4.10): small variations in give rise to large variations in . A relatively simple improvement has been proposed by Asaro and Needleman [10], the basic idea of which is to estimate the slip rate in between that at time t and at t t. The details are not discussed here.

4.3 Polycrystal plasticity


Neighboring grains in a polycrystalline aggregate have different orientations and therefore different orientation of the slip systems relative to the stress axis. Under macroscopically homogeneous states of macroscopic stress, different grains will start to yield earlier than others, leading to heterogeneous plastic ow in the aggregate (see Fig. 4.5). Whether or not this heterogeneity is relevant depends on the application. There are examples in the literature where detailed nite element representations of the individual grains in an 54

qt

qt

q t t

aggregate are used, such as [13], so that the heterogeneity is fully represented in the continuum plasticity sense. On a somewhat coarser level of description, grains are represented by single elements; this describes the heterogeneity among different grains, but not inside grains [12, 15]. The Taylor model is an even coarser description, based on the assumption that the total strains inside all grains are identical, i.e.

k 1

where the superscript k identies grain k 1 N. Similar to the Kelvin idea in rheology, the overall stress is then simply computed as the weighted average of the stress k inside each grain,
k 1

This implies that the constitutive equations for the aggregate are obtained directly by the sum of the constitutive equations (4.6) per crystal. The validity of the Taylor assumption for a few deformation processes has been investigated in e.g. [13]. Grain interactions are not represented well in the Taylor model, and equilibrium between grains is not guaranteed. The opposite assumption is the so-called Sachs model, similar to the Voigt model in rheology, which assumes that all grains are subject to the same stress, while the overall strain rate is the sum of the strain rate in each grain. The Taylor and the Sachs model represent two extreme homogenization approaches. There are also models that attempt to relax the constraints implied by either one of these extremes. They are referred to as, for instance, the relaxed constraints model, the pancake model, the lath model and the lamel model, and apply in particular to materials in which the grains have, for example, elongated shapes due to previous processing (e.g. [15, 14]).

4.4 Exercises
Exercise 4.1 Show that eq. (4.9) can also be written as



Exercise 4.2 1. Show that according to crystal plasticity, i.e. eqs. (4.2)(4.3), the plastic strain tensor satises p tr p kk 0 This means that there is no plastic volume change. Give the physical explanation. 2. Consider a two-dimensional single crystal with three sip systems with glide planes at angles of 1 30 , 2 90 and 3 30 with respect to the x1 -axis in the gure below. Determine which glide system (or systems) gets activated for slip rst when the yield stressfor all systems is the same, i.e. g 1 g2 g3 0 , and when the crystal is subjected to a monotonically increasing homogeneous shear stress 12 . At what value of 12 does slip start? 55

m s

(4.24)

(4.25)

e2

( 3)

( 2)

( 1)

e1

3. Is it possible that other slip systems (or system) get activated later on?

56

Chapter 5 Answers to Selected Exercises

which is indeed the same as according to (2.9). 3.7 (a) The principal stress is determined by the eigenvalue problem

This only has a solution when the determinant vanishes, i.e.

apart from an unknown pre-factor. This proves that the principal directions are at 45 from the shear directions. Completely similar results follow for the principal strains, 1 2 1 . 2

57

u1 u2

1 1

u1 u2

1 1

which implies that the two principal stresses are 1 and 2 directions are found by back-substitution into (5.1), giving

0 . The principal

0 1 1 0

1 0 0 1

Since e

1, one can add l 2 to the numerator, after which rearrangement yields l0 l 2 2 2l0
2 l0

2 l0

l0 . Multiply e by 2l0

2.2 Write l

2 2l0 and add l0

2 l0 to the numerator, to get

2l0 l 2 2l0

2 l0

u1 u2

(5.1)

while

Hence, 3 0 for plane stress implies that in this case 3 0 (plane strain). The reason for this is that there is no hydrostatic stress in this problem. (c) Expressing the principal stresses in terms of from (a) and the principal strains in , we obtain from the result in (b) that

which was to be shown. 3.8 (a) According to eqs. (3.6) and (3.7)

E 21

1 i j i j 2 Making use of (3.12) to eliminate i j , we nd the requested expression. (b) Expressed in terms of , we have for W
2 0 0 R

since the solution is written in cylindrical coordinates. With (a),


0 0

The interesting term here, obviously, is the last integral, which simply evaluates to R. 58

fi j Mi jkl fkl d

r0 dr

where the stress solution takes the form i j fi j r



1 2.

Mi jkl

i j kl drdr

drdr

kk

1 2

and the same for 2 in terms of 2 by taking i j the same under plane stress, since for i j 3,

Substituting i

1 into (3.14), we get 1 E 1 1 2 (signs cancel). The result is kk

(b) First assume plane strain, 3

0. Then, from (a), kk 1 2 3 0

Hence,

(c) Go back to (3.36) and redo the integrals to nd that the integral over r becomes

R 0

r2

dr

1 2 R 2 0. This is indeed the condition used

which remains nite for R to go from (3.35) to (3.36). 4.2

0 as long as

(a) Take the trace of the plastic strain from (4.2)


and substitute Pi j from (4.3) and contract




Pkk

sk m k

It is obvious from the last that tr P 0 since s and m are normal, i.e. s m 0. The physical origin for tr p 0 is that slip only gives rise to a change of shape, not to a change of volume. (b) The basic idea is to calculate the shear stress on all slip systems by eq. (4.9) with P given by (4.3); in components



An alternative, fast, approach is to make use of the result of Exer. 4.1; in components

For the present situation, in two dimensions,


si

mi

For the given stress state, the resolved shear stress is then obtained from the matrix vector relation

which, after a bit of algebra, yields


For the three given slip system orientations, the maximum value (in absolute sense) of cos 2 occurs for 2 where cos 2 1. Thus, slip starts on slip system 2 when 12 0 . 59

12 cos 2

sin cos

0 12 12 0

cos sin

cos sin

i j Pi j

m i i j s j

or tr P

s m

tr p

1 tr P 2

1 Pkk 2

sin cos

(c) Slip systems 1 and 3 have cos 2 1 2, so that 1 3 12 2. When 1 , slip will be activated on system 1 as 12 increases to such a value that 12 2g well (and similarly for slip system 3). Whether 12 can raise this high depends on the loading program and on the slip system hardening. When the loading is stress controlled, 12 can in principle become high enough. When the loading is strain controlled, the value of 12 itself depends on the hardening on the initially activated system 2. This can only be determined once the hardening law has been selected.

60

Bibliography
[1] P. Chadwick, Continuum Mechanics Concise theory and problems, Dover Publ., Inc., Mineola (NY), 1976. [2] Y.C. Fung, First Course in Continuum Mechanics, Prentice Hall, Englewood Cliffs (NJ), 3rd edition, 1993. [3] Y.C. Fung, Foundations of Solid Mechanics, Prentice Hall, Englewood Cliffs (NJ), 3rd edition, 1965. [4] G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press, Cambridge, 1967. [5] W.D. Callister, Materials Science and Engineering: An Introduction, 5th ed., J. Wiley & Sons, 2000. [6] O.L. Zienkiewicz, R.L. Taylor, The Finite Element Method, (2 Vols.), 4th Ed. McGraw-Hill, London, 1989. [7] J. Lemaitre (ed.), Handbook of Materials Behavior Models, Academic Press, San Diego (CA), 2001. [8] T.L. Anderson, Fracture Mechanics 2nd ed. CRC Press, Boca Raton, 1995. [9] R.J. Asaro, Micromechanics of crystals and polycrystals. Adv. Appl. Mech. 23, 1115 (1983). [10] R.J. Asaro and A. Needleman, Texture development and strain hardening in rate dependent polycrystals. Acta Metall. 33, 923953 (1985). [11] J.L. Bassani, Plastic ow of crystals. Adv. Appl. Mech. 30, 191-258 (1993). [12] L. Anand and M. Kothari, A computational procedure for rate-independent plasticity. J. Mech. Phys. Solids 44, 525558 (1996). [13] S.V. Harren and R.J. Asaro, Nonuniform deformations in polycrystals and aspects of the validity of the Taylor model. J. Mech. Phys. Solids 37, 191232 (1989). 61

[14] P. Van Houtte, A comprehensive mathematical formulation of an extended Taylor-BishopHill model featuring relaxed constraints, the Renouard-Wintenberger theory and a strain rate sensitivity model. Textures and Microstructures 8, 9, 313350 (1988). [15] P.R. Dawson and E.B. Marin, Computational mechanics for metal deformation processes using polycrystal plasticity. Adv. Appl. Mech. 34, 77169 (1998).

62

Potrebbero piacerti anche