Вы находитесь на странице: 1из 15

MECH3300 - Lecture 4

Introduction to 2D finite elements


Displacement interpolation
In 2 or 3D, in order to write equations at nodes, we must
approximate boundary conditions between the nodes somehow.
Suppose we assume a linear displacement variation between
the values at 2 corner nodes. This implies a particular variation
of load per length along that edge which makes this happen.
While we dont have to, it is usual to relate the displacement
anywhere over the whole area (in 2D) or volume (in 3D) of an
element, to that at the nodes.
This implies the strain is everywhere related to the nodal
displacements.
Hence by integrating over an element its strain energy can be
found.
Strain energy written in terms of displacements at nodes leads
to estimated stiffness values.
Eg 1/2 k u
2
for a spring
Meshing in 2D
A mesh can be created in several ways.
A surface mapped from a rectangle may be defined first and
then subdivided into n by m elements, giving a mapped mesh.
A surface of more complex shape may be automatically subdivided
into triangles of similar size, using the Delaunay algorithm.
A row of 1D elements (beams) can be extruded to give 2D elements.
The constant strain triangle
The simplest possible 2D finite element is a triangle with
constant strain.

The state of strain in any small area becomes more constant as
the size of the area is reduced. Hence this element is capable
of getting the right answers if enough are used.

Consider arbitrary linear displacements in x and y directions, u
and v.

u = Ax + By + C v = D + Ex + Fy

There are 6 constants. These can be expressed in terms of 6
nodal displacement components at the 3 corner nodes. Hence
we can relate displacement anywhere to the nodal values.
Strain-displacement relations
Direct strain is change in length per length. For small strain this is
displacement/(increment in length). That is




Shear strain is change in a right angle. This is approximated as
(x displacement)/(increment in y) + (y displacement/increment in x).
y
v
x
u
y x
c
c
=
c
c
= c c
Ax
u
u+Au
Av
Au
Ax
Ay
x
v
y
u
xy
c
c
+
c
c
=
Why strain needs relating to displacement
Using u = Ax + By + C and v = D + Ex + Fy the strains are
c
x
= A, c
y
= F
xy
= B +E ie the strains are constant.

Strain-displacement relations enable us to write strain energy in
terms of the nodal displacements. This enables stiffness terms
to be found, as strain energy of of element is (1/2)u
T
[K
e
]u

Once the displacements have been found, then the strains in an
element can be found from them.





Types of finite elements
2D elements are normally triangles or quadrilaterals.
They typically use either linear interpolation with corner nodes only or
quadratic interpolation with corner and mid-side nodes. (A quadratic
curve fits through 3 points along a side)
3D elements are tetrahedra, pentahedra (wedges), and hexahedra
(bricks), also with corner or corner plus mid-side nodes.
h v p finite element methods
The conventional approach to obtaining better accuracy is to
refine the mesh, using the same polynomial order.
This is called the h finite element method as h is a symbol that
has been used to represent the side length of an element.

An alternative is to increase the polynomial order used to
interpolate displacement in each element, avoiding remeshing.
This is the p (for polynomial) finite element method.
The p method is easily automated, with a local measure of
accuracy being used to decide where to increase the polynomial
order. It has been embedded in some CAD software.

A p mesh looks coarse, but the elements have curved surfaces,
so that accurate geometry is captured with the coarse mesh.
This requires more data to describe each element.
Accuracy of answers
Clearly, with the constant strain triangle, the stresses change at
element boundaries. The accuracy is less where these changes
are larger.

Other elements permit linear or quadratic strain variations but
still show jumps in stress at element boundaries.

To improve accuracy, and give unique nodal values of stress,
stresses are usually averaged between elements at nodes when
plotting contours. This averaging has less effect at edges where
fewer elements meet.

The most accurate stresses are those estimated at element
centroids, away from the discontinuities at the edges.
Averaged v unaveraged nodal stresses
Averaged x stress Unaveraged x stress
These contour plots are generated by first finding nodal
stress values and element centre stress values. Each
quadrilateral is divided into 4 triangles and contours are
drawn as series of straight lines, each crossing one of
these subtriangles.
Obtaining an approximate stiffness matrix
An approximate element stiffness matrix can be derived in a
number of ways.
One way is to differentiate the elastic strain energy V of a
structure wrt each nodal displacement, each time generating an
equation for the resultant force on a node.

This is a special case of what is done to find equations of motion
using Lagranges equations, as each nodal displacement is a
generalized coordinate.

Eg for a spring joining node 1 to node 2, the energy stored is
V = 1/2 k (u
2
- u
1
)
2


In matrix notation
i
i
F
u
V
=
c
c
2 1 2
2
2 1 1
1
ku ku F
u
V
ku ku F
u
V
+ = =
c
c
= =
c
c
(

=
(

2
1
2
1
u
u
k k
k k
F
F
Interpolation functions
Instead of writing polynomials in terms of coefficients to be
determined, we can write displacement at an arbitrary location
(x,y) within an element directly in terms of nodal values.

For linear interpolation over a triangle with nodal x-
displacements, u
1
, u
2
and u
3
this is
u(x,y) = N
1
(x,y) u
1
+ N
2
(x,y) u
2
+ N
3
(x,y) u
3


N
1
, N
2
and N
3
are called interpolation functions. They evaluate
to 1 at one node and to zero at the others, so that u at node 1 is
u
1
etc.

N
1

Node 1 Node 2
Node 3
1
x
y
Properties of interpolation functions
The functions N
i
must be a partition of unity. That is they must
sum to 1 at ant point in the element.

This is necessary to represent rigid body motion, as for instance
with a 3 node triangular element, if all nodes move a distance of
1 in the x direction (a rigid body mode), u = N
1
+ N
2
+ N
3






They must also capture states of constant strain correctly, as an
element tends to constant strain as it gets smaller. This means
the displacements must capture an arbitrary linear variation in
all coordinates. The functions for the constant strain triangle
are the simplest ones to meet this requirement.

Weighted residual formulation
A more general view of how the approximate nodal equations
are obtained, which is applicable to any set of partial differential
equations, is as follows.
Consider a scalar function of position (say). Say temperature T
is a function of position in space T(x,y,z).
A PDE like Laplaces equation can be written generally as
A(T,x,y,z) = 0

Instead of insisting that the PDE A be true exactly, we
approximate the spatial variation of T with interpolation functions
N
i
and nodal values T
i
(T = E N
i
T
i
).

We evaluate A approximately, then multiply A by a weighting
function w(x,y,z) and integrate over the volume of one finite
element, then set the result to zero.
This makes the equation true in an approximate weighted
residual sense. Ie the error in wA integrates to zero.
The Galerkin method
By choosing different w functions, we can penalize the error in
different locations in an element. For instance if we make w =
N
1
the error is penalized near node 1, where w is a maximum.

By choosing w to be each interpolation function in turn, we
generate equations for each node (eg for a triangle we get 3
equations relating nodal temperatures T
1
, T
2
and T
3
).

Applying this approach to elastic problems, the same equations
result that we obtain by differentiating energy.

This choice of making the weighting function the same as the
interpolation function is called the Galerkin method of
approximation. The finite element method is a type of Galerkin
method.

Вам также может понравиться