Linear Statics Fem Hand Out

Overview
Introduction
Linear elasticity problem
The Finite Element Method
Finite Element Method applied to linear statics of
deformable solids
Joel Cugnoni, LMAF / EPFL
February 20, 2013
Joel Cugnoni, LMAF / EPFL Finite Element Method applied to linear statics of deformable solids
Overview
Introduction
1
Introduction
Denitions
Some Maths
Linear elasticity
2
Equilibrium of a linear elastic solid
Dierential Equation
Virtual Work Principle
Weak Form
3
Approximate solution of the Weak Form
Finite Element Method
Overview
Introduction
Denitions
Some Maths
Linear elasticity
Denitions
Linear statics = statics of
linear elastic solids in
innitesimally small
deformations
Elastic body =
represented in Lagrangian
form by its original, stress
free, conguration
Boundary conditions:
Imposed displacement
u(x) on
u
Imposed surface
traction t(x) on
Overview
Introduction
Denitions
Some Maths
Linear elasticity
Some Maths
Symmetric 2nd order tensors written in vector forms:
= {
11
,
22
,
33
,
23
,
31
,
12
}
T
Dierential operator:
=
_
_
/x
1
0 0
0 /x
2
0
0 0 /x
3
0 /x
3
/x
2
/x
3
0 /x
1
/x
2
/x
1
0
_
_
hence we have:
div() =
T
and:
grad(u) = u
Overview
Introduction
Denitions
Some Maths
Linear elasticity
Linear elasticity
The stress is dened by the symmetric Cauchy stress tensor
(x) = {
11
,
22
,
33
,
23
,
31
,
12
}
T
For small deformations, the (innitesimal) strain tensor
(x) = {
11
,
22
,
33
,
23
,
31
,
12
}
T
is dened 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
Denitions
Some Maths
Linear elasticity
Elasticity tensor
For an isotropic material in 3D, the 4th order symmetric
elasticity tensor is dened by two constants: the Youngs
modulus E and the Poissons ration
With the previous denitions, C can be written in matrix form:
C =
_
_
1 0 0 0
1 0 0 0
1 0 0 0
0 0 0
(12)
2
0 0
0 0 0 0
(12)
2
0
0 0 0 0 0
(12)
2
_
_
with: =
E
(1+)(12)
Overview
Introduction
Dierential Equation
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
where:
is the internal Cauchy
stress tensor
f is the body (volumic)
force vector
Overview
Introduction
Dierential Equation
Weak Form
Dierential problem of linear statics
Using the constitutive relationship in the equlibrium equation, we
can formulate the linear static problem as:
nd the displacement eld 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:
N
T
C u(x) = t(x), x
Overview
Introduction
Dierential Equation
Weak Form
The weak form of the elastostatic problem can be obtained using
the Virtual Work Principle. Taking the dierential equation, we
multiply it by an arbitrary admissible test function u and integrate
it over the domain .
_
u
_
T
C u +f
_
d = 0 , u
Note that the initial x of the local equilibrium di. equation has
been replaced by u in the integral form.
Overview
Introduction
Dierential Equation
Weak Form
Virtual Work Principle, development..
Applying integration by part, and imposing that u is continuous,
dierentiable and satises the homogeneous form of the
displacement boundary condition:
u(x) = 0 , x
u
We obtain:
(u)
T
C u d+
_
u
T
t d()+
_
u
T
f d = 0 , u
which terms corresponds to the virtual work of internal stresses,
surface tractions and body forces.
Overview
Introduction
Dierential Equation
Weak Form
Comments on the function space
From the previous assumptions, the function space of the
displacement eld u(x) is :
U
i
=
_
u
i
(x) | u
i
(x) H
1
(); u
i
(x) = u
i
(x), x
u
_
, i = (1, 2, 3)
And the function space of the displacement eld u(x) is :
V
i
=
_
u
i
(x) | u
i
(x) H
1
(); u
i
(x) = 0, x
u
_
, i = (1, 2, 3)
where H
1
represents the 1st order Sobolev space of functions on
satisfying:
_
(u)
T
u +u
T
u d <
Overview
Introduction
Dierential Equation
Weak Form
Weak form of linear statics
Linear elasticity problem in Weak Form
nd the displacement eld u(x) U such that :
_
(u)
T
C u d =
_
u
T
t d() +
_
u
T
f d ,
u V
where:
U
i
=
_
u
i
(x) | u
i
(x) H
1
(); u
i
(x) = u
i
(x), x
u
_
V
i
=
_
u
i
(x) | u
i
(x) H
1
(); u
i
(x) = 0, x
u
_
Overview
Introduction
Dierential Equation
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 eld!!
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 dierentiability / continuity of the elds u
and u are relaxed compared to the initial dierential 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:
u
h
u , u
h
u
u
h
U
h
U , u
h
V
h
V
By considering an approximation U
h
based on a linear
combination of a nite number of basis functions, we can write:
u
h
(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:
u
h
(x) = H(x) q
Overview
Introduction
Approximation of the Weak Form
Using the approximations u
h
= Hq and u
h
= Hq, the weak
form can be rewritten as:
q
T
__
(H)
T
CHqd
_
H
T
f d
_
H
T
t d()
_
= 0
This relation must hold u
h
V
h
and thus q, thus:
_
B
T
CBqd =
_
H
T
f d +
_
H
T
t d()
where B = H is called the displacement-strain interpolation
matrix because = (Hq) = Hq = Bq
Overview
Introduction
Approximate linear statics problem
Approximate linear statics problem
nd the displacement components q such that :
Kq = r
and:
u
h
(x) = H(x) q = u , x
u
with:
K =
_
B
T
CB d
r =
_
H
T
f d +
_
H
T
t d()
Overview
Introduction
Comments
K is called the stiness matrix of the system, r is the force
vector containing contributions of both surface tractions and
body forces
By choosing a nite subset of basis functions to build the
approximation of the weak, we have turned the algebraic
dierential equation problem into a linear system of equations:
all terms of K and r are constant, hence the name linear
static problem.
As the displacement is a linear combination of functions
dened over , enforcing exact displacement boundary
conditions u
h
(x) = u is not trivial except for zero imposed
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 dening a
systematic approximate function space with the aim to:
1
build approximate solutions over arbitrarily complex domains
2
simplify the treatment of essential boundary conditions
3
simplify the integration of the global system matrice(s) and
load vector(s)
4
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?
By discretizing the domain in a set of smaller & simpler domains.
The discretization
h
of is dened 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 x
j
and the each
element is dened 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?
Use discrete node-based shape functions on
h
: each node of
coordinates x
i
is associated to its own shape function h
i
. The
approximation u
h
is then written as :
u
h
= Hq with H = [h
1
I, h
2
I, ..., h
n
I]
and q =
_
q
1
, q
2
, ..., q
p
_
T
Moreover, the shape functions are required to have compact
support (Kronecker property):
h
i
(x
j
) =
ij
x
j

h
Overview
Introduction
An illustration of h
5
(x) showing the compact support property
Overview
Introduction
How to simplify the treatment of essential BC?
Thanks to the Kronecker property, the approximate displacement
u
h
on nodes x
j
is then:
u
h
(x
j
) =
p
i =1
h
i
(x
j
) q
i
= q
j
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:
u(x
j
) = u(x
j
) x
j

u
= q
j
= u(x
j
) = q
j
j | x
j

u
Overview
Introduction
How to simplify the integration?
Split integration in each element..
_
d =
n
i =1
_
i
d
_
d() =
k
i =1
_
i
d()
Note that local support means that, when integrating h
i
, only
contributions of element containing node i must be processed.
Overview
Introduction
How to simplify the integration?
_
d =
_
1
d +
_
2
d +
_
3
d +
_
4
d
Overview
Introduction
How to automate the process?
By treating the problem per element : localization !
Local displacement approximation in the e-th element as :
e
u
h
=
e
H
e
q
Where:
e
H = [
e
h
1
I,
e
h
2
I, ...,
e
h
p
I]
And:
e
q =
e
L q
e
L 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 h
5
(x)
The local shape functions corresponding to h
5
(x)
Overview
Introduction
Connectivity table : Local - Global node numbering
e

1
4
1 1 2 4 5
2 2 3 5 6
3 5 6 8 9
4 4 5 7 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
u
h
and
e
H, we can expand the
integration:
K =
_
h
B
T
C B d =
q
e=1
e
L
T
__
e
e
B
T e
C
e
B d
_
e
L
where
e
B =
e
H is the local displacement - strain matrix
and
e
C is the element elasticity matrix (strain - stress
relationship).
We then dene the elementary stiness matrix
e
K:
e
K =
_
e
e
B
T e
C
e
B d
Overview
Introduction
Assembly operation
We dene the so-called Assembly operator

as:
p
e=1
{} =
p
e=1
e
L
T
{}
e
L
With this operator:
K =
p
e=1
e
L
Te
K
e
L =
p
e=1
e
K
Overview
Introduction
Automating the integration / standardizing the shape
functions
To automate the integration and simplify the denition 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
e
T
Overview
Introduction
Master element & shape functions
With the coordinate transform
e
T and
e
T
1
, the shape functions
e
h
i
(x) can be mapped to master shape functions
a
h
i
() dened in
the master element space
a
:
e
H(x) =
e
H(x()) =
a
H()
Such that now, in each element e:
e
u
h
[x()] =
a
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 ?
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
h
i
(x
j
) =
ij

a
h
i
(
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
:
e
K =
_
a
_
{
e
B(x())}
T
C {
e
B(x())}
_
e
j d
a
Where
e
j is the determinant of the jacobian matrix
e
J
e
j = det(
e
J)
with the denition:
e
J
ij
=
x
i
j
Overview
Introduction
Master element & derivatives
The spatial derivative operator is dened in the global
coordinate system x
1
, x
2
, x
3
. Applying the coordinate transform
e
T, we can then extend it to be applied on the master element
a
.
Using the standard derivation rules:
/x
i
= /
j

j
/x
i
or in matrix - vector form:
/x =
e
J
1
/
With this denition, 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 stiness matrix
e
K and load vector. For each
dimension, the integrals are transformed into a weighted sum:
_
+1
1
() d
j

r
i =1
i
j
() |
j
=
i
j
where the points of coordinates
i
j
are called Gauss points or
integration points and
i
j
are the associated weights.
Using the integration schemes in the three directions:
_
+1
1
_
+1
1
_
+1
1
() d
3
d
2
d
1

r
i =1
s
j =1
t
k=1
i
1
j
2
k
3
() |
1
=
i
1
,
2
=
j
2
,
3
=
k
3
e
K =
_
a
_
{
e
B(x())}
T
C {
e
B(x())}
_
e
j d
a Joel Cugnoni, LMAF / EPFL Finite Element Method applied to linear statics of deformable solids
Overview
Introduction
Numerical integration
Using the integration schemes in the three directions:
_
+1
1
_
+1
1
_
+1
1
() d
3
d
2
d
1

r
i =1
s
j =1
t
k=1
i
1
j
2
k
3
() |
1
=
i
1
,
2
=
j
2
,
3
=
k
3
Thus, the elementary stiness matrix
e
K can be computed
numerically as:
e
K
r
i =1
s
j =1
t
k=1
i
1
j
2
k
3
_
(
a
H)
T
C (
a
H)
e
j
_
1
=
i
1
,
2
=
j
2
,
3
=
k
3
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 nally, what is a Finite Element ?
Using these principles, a standard, displacement based, Finite
Element formulation is dened by:
1
The basic topology of its master element
a
2
The coordinate transform
e
T mapping the master domain
a
to the real element domain
e
3
The shape functions
a
H and their derivatives h
i
/
j
in the
master element
4
The numerical integration scheme (Gauss points and weights)
in the master element.
Overview
Introduction
So nally, what is a Finite Element Model?
A Finite Element Model is thus dened by :
1
The geometrical mesh
h
consisting of all elementary domains
e
(dened by nodes and connectivity table)
2
The denition of the problem to be solved: K q = r
3
A list of Finite Elements (formulations) supported by
geometrical cells of the mesh
4
Boundary conditions: imposed nodal displacement q on
u
and imposed surface tractions on
5
Material properties (
e
C ) assigned to each Finite Element
Overview
Introduction
Shape functions
Requirements
1
Continuity: to have a continuous solution, shape functions
must be continuous along the interfaces between elements
2
Completeness: to be able to represent uniform strain &
displacement, shape functions must at least be ane functions
3
Dierentiability: shape functions & derivatives must be
suciently regular to be square integrable over each element
Consequences
Partition of unity: at each point x,
i
h
i
(x) = 1
Basic choice of shape function: piecewise polynomials, at
least piecewise trilinear functions in 3D

Linear Statics Fem Hand Out

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Linear Statics Fem Hand Out

Hochgeladen von

Copyright:

Verfügbare Formate

Overview

Das könnte Ihnen auch gefallen