MECH3300 Finite element Methods

LECTURE 2
- Commercial packages and how they work
Commercial finite element packages
History - initially punched card input and printer output.
The original codes are now the solver, but have graphical
preprocessing and post-processing added.
Geometry can be imported from CAD packages - a standard
format such as IGES may be needed for this. However CAD
geometry is often unsuitable for analysis, as additional
geometric approximations must be made. More later
Appropriate geometry can also be constructed within the finite
element preprocessor (a specialized CAD package).
This typically involves creation of geometry (points, lines,
surfaces and solids), then meshing - creating the actual finite
element model (nodes, elements, restraints, loads, properties
etc).
For analysis of machine components, a finite element package
can be embedded within a solid modeling CAD package, to
automatically mesh CAD solids. This is easy to use, but has
limitations ...
Typical output from a finite element package
Output can be displayed as scaled deflected shapes, contour plots of
stress, animation of deformation, xy plots etc.
Linear v nonlinear analysis
For elastic behaviour and small deflections stress strain and
force deflection. The result for static analysis is a set of linear
equations F=[K]u which can be set up and solved once.

However, if non-linearities are present, an iterative solution is
needed, involving a series of linear approximations. This may
involve 100s of updates to correct for
finite deformation causing shape change (load not deflection),
plastic deformation (stress not strain), creep or
finite sliding in contact or loss of contact (stiffness changes) etc.
Hence nonlinear analysis is much less efficient, and it may
not converge to a solution at all.

Most commercial packages have severe limitations when
applied to nonlinear problems.
Examples of commercial packages
STRAND7 - Australian package mainly for linear structural
analysis - good for beam models - will import solid models but
cannot generate its own solid geometry.
NASTRAN for Windows - mainly for linear structural analysis -
can generate its own solid CAD model.
ABAQUS - written for nonlinear analysis of structures
(nonlinearities include finite deformation, plastic deformation,
sliding contact, loss of contact etc.)
ANSYS - A versatile package that includes capabilities for other
field problems (eg electric and magnetic fields).
LS-DYNA - a package for short duration dynamics such as
impacts - used to simulate car smashes etc.
COSMOS - mainly for linear structural analysis, including
structural optimization. Available embedded in Solid Works.
ADINA - can model fluid-structure interaction, metal forming etc.

All these have some capability to solve dynamics problems.
Researchers
Commercial codes are often developed from research codes

NASTRAN was written for NASA at Univ. of Georgia. It has
been commercialised several times, most successfully by MSC
Software.
DYNA was written by Dr Hallquist and colleagues at University of
California, Lawerence Livermore doing military-funded research.
STRAND started as code developed at Univ. of Sydney and
Univ. of NSW.
Researchers continue to push the limits of what
problems can be solved. Eg at Northwestern
University, Professor Belytschko has recently
been growing cracks through finite element
meshes.

Demonstration versions
Most commercial packages have demonstration versions that are crippled
in various ways to prevent commercial use (eg node limits apply).

STRAND7 demo: available as a .zip file at www.strand7.net.au
It refuses to save a model, but results can be printed.

ABAQUS is available in a 1000 node student version for US$89 with 5%
discount for 10, 10% discount for 20 etc. http://www.worleyparsons.com
50 can be obtained for AUD$97 each including postage.

NASTRAN for Windows demo: available on CD. It will save models and
save results. It requires a 3 month license authorization file to be
downloaded by sending an email from the machine on which Nastran is
installed, to n4w.license@mscsoftware.com
The reply to this message is a file license.dat, which must be placed in the
/solver/conf folder created when Nastran is installed. You need to check that the
.rcf file in this folder contains the line $auth=c:\mscvn4~1\solver\conf\license.dat

What do solvers do?
What happens when you press the solve option on a menu.

SUBTOPICS

1. THE DIRECT STIFFNESS METHOD

2. ASSEMBLY OF ELEMENT MATRICES

3. DEGREES OF FREEDOM IN A STRUCTURE

4. RESTRAINTS [or constraints or freedom conditions]

5. THE SOLUTION PROCESS
The direct stiffness method
Most structures are statically indeterminate, so that loads are shared
between members in accordance with their stiffness. Hence we cant
just use statics.
The direct stiffness method refers to writing a large stiffness matrix to
describe a structure. It was invented as a method of hand calculation,
but only became popular once computers could solve the equations.
Structural equations can be written as displacement = f(loads) giving
compliance or flexibility terms or as loads = f(displacements) giving
stiffness terms.
The advantage of the latter is that it leads to a sparse stiffness matrix,
with non-zero terms only where local interactions between members
or elements occur.

Assembly of element matrices
Imagine expanding each element matrix with rows/columns of
zeroes so that it refers to all displacements at nodes and all
forces at nodes, giving [K
e1
], [K
e2
] etc..

Forces on element 1 are, F
1
= [K
e1
]u where u is the full
displacement vector. F
1
is a column vector of the same size as
u but with zeroes corresponding to nodes not on element 1.
Similarly for element 2, F
2
= [K
e2
]u etc.

Adding forces at nodes EF = {[K
e1
] + [K
e2
] + }u = [K]u

This is a special type of addition called assembly, as the
element matrices are expanded to full size before being added.

Example of assembly- 2 springs in series
Let spring 1 have axial displacements u
1
at node 1 and u
2
at node 2. Its
element equations are



(

=
(

2
1
2
1
u
u
k k
k k
F
F
Ie the force in the spring is its stiffness k times the difference in
displacement over its length. To add another spring in series, add
another node with F
3
and u
3
associated with it. The equations for
spring 1 are now
k k
u
3
u
1
u
2

(
(
(

(
(
(

=
(
(
(

3
2
1
3
2
1
0 0 0
0
0
u
u
u
k k
k k
F
F
F
Spring 1 Spring 2
Example of assembly - 2
For the second spring, its expanded stiffness matrix is that in the eqns.
Adding the forces at nodes
(
(
(

(
(
(

=
(
(
(

3
2
1
32
22
12
0
0
0 0 0
u
u
u
k k
k k
F
F
F
(
(
(

(
(
(

=
(
(
(

+
(
(
(

3
2
1
32
22
12
3
2
1
0
2
0
u
u
u
k k
k k k
k k
F
F
F
F
F
F
Hence the assembled stiffness matrix is
(
(
(

k k
k k k
k k
0
2
0
Note this summation assumes at all elements at a node have the same
displacement - for beams and plates rotations are matched as well, so
this means a joint is always treated as a rigid joint by default, not as a pin
joint.
A row of springs in series
The process of expanding element matrices with zeroes and
adding them can be continued. For instance, for 5 equal springs
end to end, [K] is
(
(
(
(
(
(
(

k k
k k k
k k k
k k k
k k k
k k
0 0 0 0
2 0 0 0
0 2 0 0
0 0 2 0
0 0 0 2
0 0 0 0
Note this matrix is banded - the non-zero terms cluster around the
diagonal and it is symmetric. These properties typically apply to
matrices of structural problems and are exploited in equation
solvers.
Degrees of freedom
Each movement that can occur independently is a degree of
freedom. The total number of degrees of freedom is the number
of equations that must be solved and the size of [K].

For beam or plate models there are potentially 6 degrees of
freedom per node.

For solid models there are potentially 3 degrees of freedom per
node, as no rotations are used to describe the deformation.

The number of degrees of freedom is reduced by applying
restraints or constraints, that stop certain displacements.

At least 6 restraints must be present in 3D to prevent 6 possible
rigid body modes of deformation, as there is no unique solution
to the displacements of an unrestrained structure.
Restraints (or constraints)
The effect of fixing displacement component number i is to
remove row i and column i from the stiffness matrix describing a
structure.

The row is removed as it contains coefficients of a redundant
equation for an unknown reaction to the constraint.

The column is removed as its terms are multiplied by zero.

The effect of these changes is to maintain a symmetric stiffness
matrix.

3 by 3 example - next slide.
3 by 3 example of applying a restraint
Consider the 3 by 3 set of equations below
(
(
(

=
(
(
(

(
(
(

3
2
1
3
2
1
6 5 3
5 4 2
3 2 1
F
F
F
u
u
u
k k k
k k k
k k k
If u
2
= 0 then column 2 of [K] is multiplied by zero, and F
2
becomes
an unknown reaction. Hence we can solve for u
1
and u
3
by solving
(

=
(

3
1
3
1
6 3
3 1
F
F
u
u
k k
k k
Having found u
1
and u
3
we can then find F
2
= k
2
u
1
+ k
5
u
3
if the reaction
is of interest. Finding a reaction may need to be done to check that
the reaction is zero, in a case where the restraint is an artificial one,
applied just to stop rigid body motion.

The solution process
1.Find element stiffness matrix for each member, linking
forces/moments to displacements/rotations at joints. Write
these in the global coordinate system.

2. Add (or assemble) these to give a large matrix [K], by expanding
each member matrix with rows/columns of zeroes.

3. Remove rows and columns of [K] corresponding to any
constrained displacements or rotations. This must be done to
at least prevent any rigid body motion or mechanism.

4. Solve [K] u = F for joint displacements or rotations u.

5. Use each element matrix times relevant displacements to find the
forces/moments on nodes of that element. These imply
internal loadings (eg bending moment), and internal loadings
enable stresses to be found.
Results for a beam model
A beam element loaded at each end only will have linear variation of
bending moment over its length in each principal plane.
Gravity or inertia loading can give a uniform distributed loading. This
leads to quadratic variations of bending moment.
The sign of a moment at a node may be different to that of the
bending moment in the beam there.
Results must be reported in local axes, aligned with the principal
axes of a beam, not in the global axes used to describe the whole
model, in order to distinguish loading causing bending from that
causing twisting or axial loading.

Bending
moment
Local axis directions are determined
by the reference node position.