Linear Statics Fem

Overview
Introduction
Linear elasticity problem
The Finite Element Method
Finite Element Method applied to linear statics of

deformable solids
Joel Cugnoni, LMAF / EPFL
March 10, 2010
Finite Element Method applied to linear statics of deformable s
Overview
Introduction
Introduction
Definitions
Some Maths
Linear elasticity

Equilibrium of a linear elastic solid
Differential Equation
Virtual Work Principle
Weak Form

Approximate solution of the Weak Form
Finite Element Method
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Definitions
Linear statics = statics of
linear elastic solids in
infinitesimally small
deformations
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Definitions
deformations
Elastic body =
represented in Lagrangian
form by its original, stress
free, configuration
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Definitions
deformations
Elastic body =
represented in Lagrangian
form by its original, stress
free, configuration
Boundary conditions:
Imposed displacement
u(x) on u
Imposed surface
traction t(x) on
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Some Maths
Symmetric 2nd order tensors written in vector forms:
= {11 , 22 , 33 , 23 , 31 , 12 }T
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Some Maths
Symmetric 2nd order tensors written in vector forms:
= {11 , 22 , 33 , 23 , 31 , 12 }T
Differential operator:
/x1
0
0
0
/x2
0
0
0
/x3
0
/x3 /x2
/x3
0
/x1
/x2 /x1
0
hence we have:
div () = T
and:
grad(u) = u
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Linear elasticity
The stress is defined by the symmetric Cauchy stress tensor
(x) = {11 , 22 , 33 , 23 , 31 , 12 }T
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Linear elasticity
(x) = {11 , 22 , 33 , 23 , 31 , 12 }T
For small deformations, the (infinitesimal) strain tensor
(x) = {11 , 22 , 33 , 23 , 31 , 12 }T is defined by:
(x) = u(x)
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Linear elasticity
(x) = {11 , 22 , 33 , 23 , 31 , 12 }T
For small deformations, the (infinitesimal) strain tensor
(x) = {11 , 22 , 33 , 23 , 31 , 12 }T is defined by:
(x) = u(x)
The constitutive stress - strain relationship of linear elasticity
is given by the Hooks law:
(x) = C(x) (x)
where C(x) is the 4th order elasticity tensor
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Elasticity tensor
For an isotropic material in 3D, the 4th order symmetric
elasticity tensor is defined by two constants: the Youngs
modulus E and the Poissons ration
Overview
Introduction
Definitions
Some Maths
Linear elasticity
Elasticity tensor
For an isotropic material in 3D, the 4th order symmetric
elasticity tensor is defined by two constants: the Youngs
modulus E and the Poissons ration
With the previous definitions, C can be written in matrix form:
0
0
0

0
0
0
1
0
0
0
C =
(12)
0
0
0
0
0
(12)
0
0
0
0
0
2
(12)
0
0
0
0
0
2
with: =
E
(1+)(12)
Overview
Introduction

Weak Form
Stress equilibrium
Overview
Introduction

Weak Form
Stress equilibrium
Internal Stress Equilibrium
Equilibrium is achieved if, at
any point x of :
div ((x)) + f(x) = 0
or
T
(x) + f(x) = 0
Overview
Introduction

Weak Form
Stress equilibrium
any point x of :
div ((x)) + f(x) = 0
or
T
(x) + f(x) = 0
where:
Overview
Introduction

Weak Form
Stress equilibrium
any point x of :
div ((x)) + f(x) = 0
or
T
(x) + f(x) = 0
where:
is the internal Cauchy
stress tensor
Overview
Introduction

Weak Form
Stress equilibrium
any point x of :
div ((x)) + f(x) = 0
or
T
(x) + f(x) = 0
where:
is the internal Cauchy
stress tensor
f is the body (volumic)
force vector
Overview
Introduction

Weak Form
Differential problem of linear statics

Using the constitutive relationship in the equlibrium equation, we
can formulate the linear static problem as:
find the displacement field u(x) such that :
T (C u) + f = 0 , x
Overview
Introduction

Weak Form
Differential problem of linear statics

Using the constitutive relationship in the equlibrium equation, we
can formulate the linear static problem as:
find the displacement field u(x) such that :
T (C u) + f = 0 , x
satisfying the boundary conditions:
imposed displacement:
u(x) = u
(x), x u
imposed surface traction:
NT C u(x) = t(x), x
Overview
Introduction

Weak Form

The weak form of the elastostatic problem can be obtained using
the Virtual Work Principle. Taking the differential equation, we
multiply it by an arbitrary admissible test function u and integrate
it over the domain .
Z
h
i
u T C u + f d = 0 , u
Note that the initial x of the local equilibrium diff. equation has
been replaced by u in the integral form.
Overview
Introduction

Weak Form
Virtual Work Principle, development..

Applying integration by part, and imposing that u is continuous,
differentiable and satisfies the homogeneous form of the
displacement boundary condition:
u(x) = 0 , x u
Overview
Introduction

Weak Form

u(x) = 0 , x u
We obtain:
Z
Z
(u)T C u d+
Z
uT t d()+ uT f d = 0 , u
Overview
Introduction

Weak Form

u(x) = 0 , x u
We obtain:
Z
Z
(u)T C u d+
Z
uT t d()+ uT f d = 0 , u
which terms corresponds to the virtual work of internal stresses,

surface tractions and body forces.
Overview
Introduction

Weak Form
Comments on the function space

From the previous assumptions, the function space of the
displacement field u(x) is :

Ui = ui (x) | ui (x) H 1 (); ui (x) = ui (x), x u , i = (1, 2, 3)
Overview
Introduction

Weak Form


And the function space of the displacement field u(x) is :

Vi = ui (x) | ui (x) H 1 (); ui (x) = 0, x u , i = (1, 2, 3)
Overview
Introduction

Weak Form


And the function space of the displacement field u(x) is :

Vi = ui (x) | ui (x) H 1 (); ui (x) = 0, x u , i = (1, 2, 3)
where H 1 represents the 1st order Sobolev space of functions on
satisfying:
Z
(u)T u + uT u d <
Overview
Introduction

Weak Form
Weak form of linear statics

Linear elasticity problem in Weak Form
find the displacement field u(x) U such that :
Overview
Introduction

Weak Form

Z
Z
Z
T
T
(u) C u d =
u t d() +
uT f d ,
Overview
Introduction

Weak Form

Z
Z
Z
T
T
(u) C u d =
u t d() +
uT f d ,
u V
Overview
Introduction

Weak Form

Z
Z
Z
T
T
(u) C u d =
u t d() +
uT f d ,
u V
where:

Ui = ui (x) | ui (x) H 1 (); ui (x) = ui (x), x u
Vi = ui (x) | ui (x) H 1 (); ui (x) = 0, x u
Overview
Introduction

Weak Form
Comments
The displacement boundary conditions have been moved in
the class of functions U. This type of boundary condition is
called essential because it is related to the essence of the
problem = the displacement field!!
Overview
Introduction

Weak Form
Comments
The surface traction boundary conditions t on are now
explicitly introduced in the weak form. This type of boundary
condition is called natural because it falls naturally in the
weak form of the problem!!
Overview
Introduction

Weak Form
Comments
The surface traction boundary conditions t on are now
explicitly introduced in the weak form. This type of boundary
condition is called natural because it falls naturally in the
weak form of the problem!!
The name Weak Form comes from the fact that the
requirements of differentiability / continuity of the fields u
and u are relaxed compared to the initial differential equation
form. Note that the functions do not necessarilly need to be
continuous!
Overview
Introduction


We are looking for an approximate solution of the Weak Form
problem. We give us two subsets U h and V h of the function spaces
U and V to form the basis of the approximations:
uh u , uh u
uh U h U , uh V h V
Overview
Introduction


uh u , uh u
uh U h U , uh V h V
By considering an approximation U h based on a linear
combination of a finite number of basis functions, we can write:
uh (x) = H(x) q
where H(x) is the 3xn matrix of shape functions and q is the
vector of displacement components.
Overview
Introduction


uh u , uh u
uh U h U , uh V h V
By considering an approximation U h based on a linear
combination of a finite number of basis functions, we can write:
uh (x) = H(x) q
where H(x) is the 3xn matrix of shape functions and q is the
vector of displacement components. Using Galerkin method, we
choose the same base function space for u:
uh (x) = H(x) q
Overview
Introduction

Approximation of the Weak Form

Using the approximations uh = H q and uh = H q, the weak
form can be rewritten as:
Z

Z
Z
T
T
T
T
q
(H) C H q d
H f d
H t d() = 0
Overview
Introduction


Z

Z
Z
T
T
T
T
q
(H) C H q d
H f d
H t d() = 0
This relation must hold uh V h and thus q, thus:

Z
Z
Z
BT CB q d =
HT f d +
HT t d()
Overview
Introduction


Z

Z
Z
T
T
T
T
q
(H) C H q d
H f d
H t d() = 0
This relation must hold uh V h and thus q, thus:

Z
Z
Z
BT CB q d =
HT f d +
HT t d()
where B = H is called the displacement-strain interpolation

matrix because = (Hq) = H q = Bq
Overview
Introduction

Approximate linear statics problem

find the displacement components q such that :
Kq = r
Overview
Introduction


Kq = r
and:
, x u
uh (x) = H(x) q = u
Overview
Introduction


Kq = r
and:
, x u
uh (x) = H(x) q = u
with:
Z
K=
BT CB d
Z
r=
H f d +
HT t d()
Overview
Introduction

Comments
K is called the stiffness matrix of the system, r is the force
vector containing contributions of both surface tractions and
body forces
Overview
Introduction

Comments
body forces
By choosing a finite subset of basis functions to build the
approximation of the weak, we have turned the algebraic
differential equation problem into a linear system of equations:
all terms of K and r are constant, hence the name linear
static problem.
Overview
Introduction

Comments
body forces
static problem.
As the displacement is a linear combination of functions
defined over , enforcing exact displacement boundary
is not trivial except for zero imposed
conditions uh (x) = u
displacement.
Overview
Introduction

Comments
body forces
static problem.
As the displacement is a linear combination of functions
defined over , enforcing exact displacement boundary
is not trivial except for zero imposed
conditions uh (x) = u
displacement.
Integrating the matrices & force vector over an arbitrary
domain can become very complex, and thus numerical
integration methods may be used.
Overview
Introduction


In some sense, the Finite Element Method is one way of defining a
systematic approximate function space with the aim to:
1
build approximate solutions over arbitrarily complex domains
Overview
Introduction


1
simplify the treatment of essential boundary conditions
Overview
Introduction


1
simplify the integration of the global system matrice(s) and

load vector(s)
Overview
Introduction


1

load vector(s)
automate the resolution procedure: integration of the linear

system & solution
Overview
Introduction


1

load vector(s)
automate the resolution procedure: integration of the linear

system & solution
Now lets see how this can be achieved !!
Overview
Introduction

The key ideas behind the Finite Element Method
How to process complex domain?
Overview
Introduction


By discretizing the domain in a set of smaller & simpler domains.
Overview
Introduction


The discretization h of is defined by a set of discrete points
called nodes and a set of sub-domains of basic topology called
elements, which forms what we call a mesh.
h , h = 1 2 ... n
Overview
Introduction


The discretization h of is defined by a set of discrete points
called nodes and a set of sub-domains of basic topology called
elements, which forms what we call a mesh.
h , h = 1 2 ... n
The j-th node is represented by its coordinates xj and the each
element is defined by its connectivities: an ordered list of nodes
which serve as support for the element.
Overview
Introduction

A simple mesh made of 16 elements and 25 nodes
Overview
Introduction

An much more complex mesh of an Hydroptere

Overview
Introduction


How to simplify the treatment of essential BC?
Overview
Introduction


Use discrete node-based shape functions on h : each node of
coordinates xi is associated to its own shape function hi . The
approximation uh is then written as :
uh = H q with H = [h1 I, h2 I, ..., hn I]

T
and q = q1 , q2 , ..., qp
Moreover, the shape functions are required to have compact
support (Kronecker property):
hi (xj ) = ij
xj h
Overview
Introduction

An illustration of h5 (x) showing the compact support property

Overview
Introduction

Overview
Introduction


Thanks to the Kronecker property, the approximate displacement
uh on nodes xj is then:
uh (xj ) =
p
X
hi (xj ) qi = qj
i=1
In other words, the vector q has now the physical meaning: it

represents the unknown displacements at nodes!
Thus the essential boundary conditions can be simply rewritten as:
(xj )
u(xj ) = u
(xj ) = q
j
xj u = qj = u
j | xj u
Overview
Introduction

How to simplify the integration?
Overview
Introduction


Split integration in each element..
Z
d =
Z
d() =
n Z
X
i=1
k Z
X
i=1
d()
Note that local support means that, when integrating hi , only

contributions of element containing node i must be processed.
Overview
Introduction


Z
Z
Z
Z
d =
d +
d +
d +
1
Overview
Introduction


How to automate the process?
Overview
Introduction


By treating the problem per element : localization !
Overview
Introduction


Local displacement approximation in the e-th element as :
e h
u =
H eq
Overview
Introduction


Local displacement approximation in the e-th element as :
e h
u =
H eq
Where:
e
H = [e h1 I, e h2 I, ..., e hp I]
And:
e
q=
Lq
eL
is called the binary localization matrix representing the global local numbering relationship (related to the element connectivity
table)
Overview
Introduction

An illustration of the global shape function h5 (x)

Overview
Introduction

The local shape functions corresponding to h5 (x)

Overview
Introduction


Connectivity table : Local - Global node numbering
e
1
2
3
4
1
1
2
5
4
2
2
3
6
5
3
4
5
8
7
4
5
6
9
8
Note:
The connectivity list of each element correspond to a column
of the above table.
The localization operator e L is constructed from the
connectivity table.
Overview
Introduction


Integration of the system of equation
Using the localized quantities e uh and e H, we can expand the
integration:

Z
Z
q
X
T
e T
e T e
e
B C B d =
L
B C B d e L
K=
h
e=1
where e B = e H is the local displacement - strain matrix

and e C is the element elasticity matrix (strain - stress
relationship).
We then define the elementary stiffness matrix
Z
e
e T e
K=
B C e B d
e K:
e
Overview
Introduction


Assembly operation
We define the so-called Assembly operator
p
]
{} =
e=1
p
X
as:
L {}e L
e T
e=1
With this operator:

K=
p
X
e Te
e=1
K L=
p
]
e=1
Overview
Introduction


Automating the integration / standardizing the shape
functions
To automate the integration and simplify the definition of shape
functions, we transform each distorted elementary domain e
into a reference domain a where we can apply standard numerical
integration procedures.
The coordinate transformation
e
T : a e | x()
maps any point of coordinate = {1 , 2 , 3 }T in the master

domain a to its (single) corresponding point of coordinate
x = x() in the elementary domain e (bijective application).
Overview
Introduction

An illustration of the coordinate transform
eT
Overview
Introduction


Master element & shape functions
With the coordinate transform e T and e T 1 , the shape functions
e h (x) can be mapped to master shape functions a h () defined in
i
i
the master element space a :
e
H(x) =
H(x()) =
H()
Such that now, in each element e:

e h
u [x()] =
H() e q
Thanks to that transformation, we only need to derive once the

shape function formulations a H() in the normalized master
element space a Standardized shape functions !!
Overview
Introduction

Element & Master shape functions
Overview
Introduction

How to choose a simple form of coordinate transformation ?
Overview
Introduction


Simply by using the shape functions to write the coordinate
transform:
e
T : x = x() = a H() e x
Overview
Introduction


Simply by using the shape functions to write the coordinate
transform:
e
T : x = x() = a H() e x
Which means that inside each element e, the coordinates are
interpolated as a linear combination of the shape functions and
nodal coordinates e x.
With the Kronecker property e hi (x j ) = ij a hi ( j ) = ij , this
ensures that each node of the master element corresponds to a
node in the deformed element e.
Overview
Introduction

Coordinate transformation
Overview
Introduction


Automating the integration
Using the coordinate transform e T and master shape functions,
the integrals can be rewritten to be carried out directly on a
standard domain a :
Z

e
K=
{ e B(x())}T C { e B(x())} e j da
a
Where
ej
is the determinant of the jacobian matrix

e
eJ
j = det( e J)
with the definition:

e
Jij =
xi
j
Overview
Introduction


Master element & derivatives
The spatial derivative operator is defined in the global
coordinate system x1 , x2 , x3 . Applying the coordinate transform
e T , we can then extend it to be applied on the master element a .
Using the standard derivation rules:
/xi = /j j /xi
or in matrix - vector form:
/x =
e 1
With this definition, the elementary strain-displacement matrix e B

can be directly derived from the master shape functions a H and
the integration is completely carried out in the master domain a :
e
B = a H()
Overview
Introduction


Numerical integration
On the master element, standard Gauss-Legendre numerical
integration schemes can be used to automate the calculation of the
elementary stiffness matrix e K and load vector. For each
dimension, the integrals are transformed into a weighted sum:
Z +1
r
X
ji () |j =i
() dj
1
i=1
where the points of coordinates ji are called Gauss points or

integration points and ji are the associated weights.
Using the integration schemes in the three directions:
Z +1 Z +1 Z +1
r X
s X
t
X
() d3 d2 d1
1i 2j 3k () |1 =i ,2 =j ,3 =k
1
i=1 j=1 k=1
Joel Cugnoni,e LMAF / EPFL T
Finite Element
Method applied
e
e to linear
a statics of deformable s
Overview
Introduction


Numerical integration
Using the integration schemes in the three directions:
Z
+1 Z +1 Z +1
() d3 d2 d1
1
r X
s X
t
X
1i 2j 3k () |1 =i ,2 =j ,3 =k
1
i=1 j=1 k=1
Thus, the elementary stiffness matrix e K can be computed

numerically as:
e
r X
s X
t
X
1i 2j 3k
(a H)T C (a H) e j
i=1 j=1 k=1
i
1 =1i ,2 =2j ,3 =3k
Note that the internal terms (a H)T C (a H) j are evaluated

directly in the master element coordinate system 1 , 2 , 3 .
Overview
Introduction

Gauss points & numerical integration

Overview
Introduction


So finally, what is a Finite Element ?
Using these principles, a standard, displacement based, Finite
Element formulation is defined by:
Overview
Introduction


1
The basic topology of its master element a
Overview
Introduction


1
2

The coordinate transform e T mapping the master domain a
to the real element domain e
Overview
Introduction


1
2

The shape functions
master element
aH
and their derivatives hi /j in the
Overview
Introduction


1
2

aH
The shape functions

master element
and their derivatives hi /j in the
The numerical integration scheme (Gauss points and weights)

in the master element.
Overview
Introduction


So finally, what is a Finite Element Model?
A Finite Element Model is thus defined by :
Overview
Introduction


1
The geometrical mesh h consisting of all elementary domains

e (defined by nodes and connectivity table)
Overview
Introduction


1

The definition of the problem to be solved: K q = r
Overview
Introduction


1

A list of Finite Elements (formulations) supported by

geometrical cells of the mesh
Overview
Introduction


1


on u
Boundary conditions: imposed nodal displacement q
and imposed surface tractions on
Overview
Introduction


1


on u
Boundary conditions: imposed nodal displacement q
and imposed surface tractions on
Material properties (
eC
) assigned to each Finite Element
Overview
Introduction

Shape functions
Requirements
Overview
Introduction

Shape functions
Requirements
1
Continuity: to have a continuous solution, shape functions

must be continuous along the interfaces between elements
Overview
Introduction

Shape functions
Requirements
1

Completeness: to be able to represent uniform strain &

displacement, shape functions must at least be affine functions
Overview
Introduction

Shape functions
Requirements
1


Differentiability: shape functions & derivatives must be

sufficiently regular to be square integrable over each element
Overview
Introduction

Shape functions
Requirements
1



Consequences
Partition of unity: at each point x,
hi (x) = 1
Overview
Introduction

Shape functions
Requirements
1



Consequences
Partition of unity: at each point x,
hi (x) = 1
Basic choice of shape function: piecewise polynomials, at

least piecewise trilinear functions in 3D

Linear Statics Fem

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Linear Statics Fem

Uploaded by

Copyright:

Available Formats

Overview

Finite Element Method applied to linear statics of

March 10, 2010

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Linear elasticity problem

The Finite Element Method

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Differential problem of linear statics

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Differential problem of linear statics

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Virtual Work Principle

Virtual Work Principle

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Virtual Work Principle, development..

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Virtual Work Principle, development..

Joel Cugnoni, LMAF / EPFL

Finite Element Method applied to linear statics of deformable s

Equilibrium of a linear elastic solid

Virtual Work Principle, development..

which terms corresponds to the virtual work of internal stresses,