Finite Elements: Some Advanced Topics and Non-Linear (Implicit) Analysis

IFB
A.K. Pickett, 2013-2014

Institut fr Flugzeugbau, University Stuttgart
Composites modelling:
Advanced Finite Elements
Finite Elements: Some advanced topics and

non-linear (Implicit) analysis
Outline of lecture:
IFB
Assembly of the global stiffness matrix

Properties of the global stiffness matrix
Boundary conditions and constraints
Numerical Gaussian integration
Nonlinear p
problems: Geometric, contact and material
Solution strategies for nonlinear problems in Implicit codes
The geometric non-linear stiffness matrix
Materials laws and elasto-plasticity
Tutorial: Elasto-plastic and geometric nonlinear analysis of a

bonded joint
A.K. Pickett, 2013-2014

Finite Elements: Advanced topics

Contents:
Part 1
Gaussian integration and hourglassing
Part 2
Nonlinear analysis: Geometric and Material
Parts 3
Buckling analysis
IFB
A.K. Pickett, 2013-2014

Revision of the simple (1D) Bar element

Length L, Modulus E and sectional area A
Only axial loads Fi , Fj and axial displacements ui, uj are allowed
It is said to have 1 d.o.f. (degree of freedom) per node: Torsion, bending and transverse
shear are not permissible for this element type.
Sign convention
+ positive forces and displacement (to the right)
Case 1: ui>0 and uj=0

Fi = (EA/L)ui, Fj= -(EA/L)ui
Fi
uj= 0
ui
Case 2: ui=0 and uj>0

Fi(=- Fj)
Fi = -(EA/L)u
(EA/L)uj, Fj= (EA/L)uj
ui= 0
uj
For the general case ui0 and uj0

Fi = (EA/L)*( ui-uj)
Stiffness matrix
IFB
Fj
Fi EA / L EA / L ui

F j EA / L EA / L u j
F K d
In matrix
form
Fj = (EA/L)*(-ui+uj)
Fj(=-Fi)
Fi k

F j k
k ui

k u j
A.K. Pickett, 2013-2014

Connectivity of two elements (1/2)

L2
L1
1
E1, A1
E2, A2
k1 = E1A1/L1
f1
F1=- f11
u 1 , F1
F1
Force equilibrium
k2 = E2A2/L2
u2
u1
2,
F2
Element 1
Element 2
1 k1
2 k
1
2
2 k2
3 k
2
u3
2
3
3,
F3
dof
k1 u1 f

k1 u2 f
1
1
1
2
Internal
nodal forces
dof
k 2 u 2 f 22

k 2 u3 f 32
IFB
A.K. Pickett, 2013-2014

Connectivity of two elements (2/2)

The stiffness matrices for each element are:
u1
u2
u3
d.o.f.
0
0 0
0 k 2 k 2 Element 2 only
0 k 2 k 2
u1
u2
u3
k 1 k 1 0
Element 1 only k 1 k 1 0
0
0 0
ASSEMBLY
Global stiffness matrix of the assembly
K U F
1
k1
k1
k k k
1
2
1
0
k
IFB
Force equilibrium at nodes requires:

Nodes 1: forces=0 -f11
dof
= F1
Nodes 2: forces=0 -f21 f22 = F2
0 u1 F1

k 2 u2 F2
k 2 u3 F3
Nodes 3: forces=0 -f32
= F3
External loads
A.K. Pickett, 2013-2014

Solution for unknowns (application of Boundary conditions)

Matrix [K] is singular (it can have any number of solutions since the structure is unsupported
and can undergo arbitrary Rigid Body Motion (RBM). At least one support must be imposed.
To suppress RBM set, for example, u3=0: (u1=0 or u2=0 would also suppress RBM).
k1
k1
k k k
1
2
1
0
k 20
0
Operations done
to enforce u3 =0
= -P
0 0 u1 F1

k 20 u2 F2
=0
0
k 2 1 u3 F3
u2 = - P / k2
2
1
3 dof
2
u1 = -P / k1 - P / k2
u3 = 0
Stresses and strains could also be found using:

1 = E1 1 = E1 (u2-u1) /L1 etc.
Note:
Ditto for 2
Of course changing the position of the constraint to suppress RBM changes

the problem and the results.
Exercise: Impose the constraint at d.o.f 2; what are the expressions for displacement
6
and strain? (hopefully 2 = 0).
IFB
A.K. Pickett, 2013-2014

A more complex connectivity example

u2
u4
4
Connectivity
k2
k3
Element
Exercise: Show that the

assembled global
stiffness matrix
in terms of ks is:
IFB
k1
u1
k4
u5
u3
Node 1
Node 2
u1
u2
k4
k
4
K 0
0
0
u3
k4
k1 k 2 k 4
k2
k1
0
u4
u5
k2
k2 k3
k1
0
k1
0
k3
d.o.f.
0
0
k3
0
k 3
A.K. Pickett, 2013-2014

Assembly of the global stiffness matrix

Remember how the 2 simple in series bar elements were assembled:
1
d.o.f.
u1
u2
u3 u1
u2
u3
k 1 k 1 0
0
0 0
k 1 k 1 0 + 0 k 2 k 2
0
0 k 2 k 2
0 0
Element 1 only
Continuing this logic a

series of bars would have
this Global stiffness
matrix:
Note:
This numbering scheme
gives a so-called
banded matrix.
8
It is also symmetric.
Element 2 only
1
k1
k
1
0
0
0
k1
k
1
0
k1
k1
k1 k 2
0
k2
0
0
k2
0
0
k 2 k3
k3
0
k3
k3 k 4
k4
k1 k 2
k2
4
3 (dof)
0
k 2
k 2
4
0 u1 P1

0 u 2 P2

0 u3 P3
k 4 u 4 P4

k 4
u5 P5
IFB
A.K. Pickett, 2013-2014

Changing the numbering order changes the [K] matrix bandwidth
1
2
3
4
5
k1
0
0
k1
0
k4
0
0
0
k4
0
k4
0
k 2 k3
k3
k2
k3
k3 k 4
0
The numbers of these

two nodes are swapped
k1
0
k2
0
k1 k 2
P1 k1u1 k1u5
P2 k 4u 4 k 4u 2
P3 k 2u5 (k 2 k3 )u3 k3u 4
etc.
Note:
This poor numbering scheme will not change the final results. But the global stiffness
matrix structure is modified and it is no longer banded.
But it does remain symmetric.
9
IFB
A.K. Pickett, 2013-2014

Size and some properties of the Global stiffness matrix [K]

The size of the global stiffness matrix [K] depends on element type (number of d.o.f
per node) and the number of nodes:
e.g. 2D planar elements have 2 dof per node
3D solids have 3 dof (translations) per node
Beams and shell elements have 6 dof per
node (3 translations and 3 rotations)
Overall size and some properties:

1. Size = Number of nodes * number of d.o.f per
node. A similar logic applies if mixed element types
are used in one model.
2. It is always square.
3. It is invariably symmetric (for special cases and
some non-linear problems it may not be).
10
* * * * * * 0 0 0 0 0 0
* * * * * 0 0 0 0 0 0
* * * * * * 0 0 0 0
* * * * * 0 0 0 0
* * * * * * 0 0
* * * * * 0 0
* * * * * *
* * * * *
Symmetric
* * * *
* * *
* *
IFB
A.K. Pickett, 2013-2014

Good and bad bandwidth for 2D (and 3D) problems

Consider 2D triangular elements (2 dof per node) with 2 different numbering schemes:
2
Good
numbering
1
5
Half bandwidth = 6
Good numbering
scheme (optimal): The
global stiffness matrix
looks like this. It is 12*12
(=6 nodes *2 dof per
node) and well banded.
FE codes will only store
the banded part of one
symmetric half of the
matrix for inversion.
11
IFB
Bad
numbering
1
* * * * * * 0 0 0 0 0 0
* * * * * 0 0 0 0 0 0
* * * * * * 0 0 0 0
* * * * * 0 0 0 0
* * * * * * 0 0
* * * * * 0 0
* * * * * *
* * * * *
Symmetric
* * * *
* * *
* *
A.K. Pickett, 2013-2014

3
Bad numbering
scheme: If node 1 is
connected to node 6
then tems appear
here, for example.
The half bandwidth is
now 12 and the
matrix is sparsely
populated. It will
require much more
CPU and storage.
Some final comments on stiffness matrix [K] properties (Implicit codes)
Advantages can be made for reduced storage and easier inversion of [K] from:
1. Matrix symmetry,
2. Matrix sparcity (the large number of zero terms),
3. The fact that it is usually banded, or can be made to be banded.
Inversion procedures for [K] (e.g. Gaussian elimination) can be designed to
operate only on the banded part of the matrix; unnecessary operations with
zeros are avoided (saving CPU and storage).
All commercial FE codes will automatically renumber the nodes internally to try
and minimise bandwidth (e.g. renumber in the width wise direction of the
structure); you do not really see this as input and output nodal numbers remain
unchanged and refer to your original model.
Note: Renumbering is done with in Implicit codes for the above reasons.
12
IFB
A.K. Pickett, 2013-2014

Example information from PAM- Implicit

on bandwidth minimisation
NUMBER OF NODAL POINTS.........
NUMBER OF 3-D ELEMENTS..........
5192
3780
SOLVING MECHANICAL SYSTEM OF DIMENSION, NDOF .......
15576
TOTAL MECHANICAL MATRIX NON-ZERO TERMS .............

TOTAL MECHANICAL MATRIX MEMORY SIZE (Mbytes) .......
SPARSITY OF MECHANICAL MATRIX ......................
521427
16.28
4.30E-03
= 5192 nodes * 3 dof per node

(Potential total matrix size
15576 * 15576 = 242.5 million)
0.5 million (the sym, banded part)
START SYSTEM SOLVE

TOTAL SYSTEM SOLVE USED MEMORY (Mbytes)
TOTAL SYSTEM SOLVE USED MEMORY (Mbytes)
SPARSITY OF FACTOR MATRIX
RATIO OF FACTOR/MECHANICAL
:
:
:
:
16.35 (INTERNAL)
52.00 (EXTERNAL)
2.81E-02
6.54
Sparsity of the stiffness matrix

(ratio of non-zeros to full matrix)
= 0.5214 / (242.5/2). This is low
showing it is an efficient
structure to renumber for band
width minimisation.
Some final memory requirements
and information of array sizes for
inversion calculations.
13
IFB
A.K. Pickett, 2013-2014

Solution for unknowns (application of boundary conditions)

F1 k

P k
k u1

k u 2
F1 ku 1 ku 2
P ku 1 ku 2
Rigid Body Motion (RBM) can be shown using the

simple bar element with load P at one node.
F1
Following the calculation we get that the reaction

force must balance the applied force
F1 P
But the two equations are the same; meaning that

there is no unique solution (the matrix determinant is
zero and it has no inverse).
P ku 1 ku 2
P ku 1 ku 2
Implicit
p
FE models must have sufficient boundaryy conditions to suppress
pp
RBM
In 1D analysis 1 constraint
y
x
In 2D analysis 3 constraints (x, y and rotation about z)

In 3D analysis 6 constraints (x, y, z and 3 rotations)
They should also be physically meaningful !!
Note: This is for Implicit codes only the problems is not relevant to dynamic Explicit
14 codes; although you probably want some boundary conditions to represent correct loading.
IFB
A.K. Pickett, 2013-2014

Boundary conditions (must represent physical constraints and/or

suppress rigid body modes)
For Implicit solutions (NOT Explicit):
The solution of [K] -1{P} = {D} is singular (no unique solution)
unless sufficient boundary conditions are introduced to suppress
all structure Rigid Body Modes (RBM).
It is important to understand the possible rigid body modes of
the structure which must be constrained.
A constraint simply imposes zero displacement at a dof: If
possible it should also represent the real support conditions!
E.G. for a 2D plane loaded structure appropriate bcs (to suppress RBM) could be:
Constrained
Constrained in
x only
Constrained
Constrained in
x and y
15
IFB
Rule for req.

constraints
Constrained
y
x
This structure has 3 RBM (x,y,z). The number constraints needed to

suppress these Rigid Body Modes = (2 Number dofTranslation 1)
A.K. Pickett, 2013-2014

Example of necessary contraints (boundary conditions) to suppress

RBMs) for a 3D solid problem
Number constraints needed to suppress the Rigid
Body Modes (RBM) = (2 Number dofTranslation 1) = 5
E.G. for a 3D solid there are 6 RBMs.
The number of constraints needed to stop these 6 RBMs = 5
x constrained
Note this stops rotation

about y and also rotation
about x
loading
x constrained
y
x, y, z
constrained
x
z
16
IFB
A.K. Pickett, 2013-2014

Beware the number of RBM can also be code dependent

From the previous bar tutorial example:
Theoretically, the simple bar element
has 1 dof per node.
B
Butt in
i a 2D structure
t t
((after
ft
transformations) the structure has 2
dof per node: Displacements u and v in
the x and y directions. This would imply
this structure has:
(2*2 ) - 1 = 3 RBM
But many codes treat this as a general
3D problem (dof x,y,z) meaning:
(2*3) - 1 = 5 RBM
Fixing the 2 wall nodes constrains:
x, y, z and z
but y is still a free !! The obvious extra

boundary condition is z at the loaded
node.
17
IFB
A.K. Pickett, 2013-2014

Slidelines and contact

So far the structure was complete, or joined via some elements. Contact is a method to
analyse two (or more) separate structures that interact via a contact interface. Impact
problems often involve large areas of multiple contact. Contact examples include:
Projectile impact, crashworthiness, metal stamping, bolt and joint contact, etc.
E
Essentially
ti ll there
th
are two
t
types
t
off contact:
t t
Type 1 (Constraint equations) Special constraint equations and
interpolation functions enforce compatibility of free nodes to the coarse
mesh.
Type 2 (Temporary spring elements) Temporary spring elements
are introduced as contact occurs and removed when it disappears (=
penalty force method in Explicit codes). Fairly easy in Explicit codes, but
complex in Implicit codes as a new stiffness matrix must be formulated
with each new contact its highly
g y non-linear.
Master side
Slave side
18
Type 1 - Tied slidelines are a convenient way

to tie dissimilar meshes. Essentially constraint
equations force compatibility of displacements
on either side. It can be a convenient way to
avoid mesh grading.
Type 2 - Friction slidelines allow relative
movement, contact and friction. The contact
zones change with loading (non-linear).
IFB
A.K. Pickett, 2013-2014

Applications of constraint equations

A useful option to tie dissimilar mesh types (e.g. shells to solids) and for local mesh
refinement.
Shells to solids; or
beams to shells
shells....
Mesh refinement in stress concentrations

Note: Mixing of element types with different dof
per node can present problems. E.g. the shells
have translational and rotational dof, whereas the
solids only have translational. The rotations must
be imposed over a number of solids.
19
IFB
Moments
free
If the 6 dof beam joins a single node of the 3 dof solid the
rotational moments are not transferred it acts as a pin joint.
Of course this free joint could be required in some cases!
Moments
transferred
A constraint must be defined to extend over one (or more)

solids for moment transfer, i.e. the rotation is converted to
solid element nodal translationals.
A.K. Pickett, 2013-2014

Gaussian Integration (Gauss quadrature): Background

Remember the stiffness matrix [k] for the simple triangular element was evaluated in the
local x,y element frame,
B E B t dxdy
T
Stiffness matrix [k]

T
y23 0 y31 0 y12 0

y23 0 y31 0 y12 0
1
1
[k ] t A 0 x32 0 x13 0 x21 E 0 x32 0 x13 0 x21
2A
2A
x32 y23 x13 y31 x21 y12
x32 y23 x13 y31 x21 y12
Area
Elasticity law e.g. for 2D plain stress:
x
E

y
2
1
xy
20
1
1
0 0
x

y
1
2 xy
0
0
We derived the stiffness matrix by hand

for this simple element. The integration
over the area (or volume) is easy and
can also be computed by hand.
But for complex elements the integration
is difficult. So isoparametric formulations
and numerical integration are used.
10
IFB
A.K. Pickett, 2013-2014

Gaussian Integration (Gauss quadrature): Background

The stiffness matrix [k] can be converted from a local x,y element frame, to a natural
coordinate frame with limits 1. The two are equivalent due to the scaling factor (det[J]),
B E B t dxdy
B E B (detJ ) t dd
T
1 1
Stiffness matrix [k]
isoparametric element formulations
This makes numerical integration possible; e.g. Gauss quadrature which requires the
integration limits to be 01 (or -1 +1).
Example shape functions N for the linear quadrilateral element
Interpolation in natural coordinates

y(v)
3
1
1 1
N 1
=+1
1
4
1
4
1 1
N 2
4
1
1 1
3
4
1
1 1
N 4
4
The same method as for the triangle element is used
to get the [B] matrix, except now in terms of and
=+1
=-1
1
=-1
x(u)
21
IFB
A.K. Pickett, 2013-2014

Gaussian Integration (Gauss quadrature)

Gauss quadrature for 1D integration replaces the integral as follows,
F ( , )
1 1
d d
W W F ( ,
i 1
j1
In our case
F( i , j )
is the
expression for stiffness matrix
F ( i , i ) B ( i , i ) EB ( i , i )
T
where
h
Wi and
d Wj are weighting
i hti
f t
factors
and
d F(i , i ) are values
l
off the
th integral
i t
l
function computed at n Gauss points. The following slides demonstrates the principle for
n=1, 2 and 3.
22
11
IFB
A.K. Pickett, 2013-2014

With usual area integration methods: n=1, 2, 3 ...

n=1
F(X)
n=2
F(X)
00
0.0
Simpsons Rule,
etc
(X2-X1)
(X2-X1)
X
X1
X2
X1
X2
Area [F ( X1) F ( X 2)]
Area F (0) * ( X 2 X 1)
* ( X 2 X1) / 2
Increasing accuracy
23
IFB
A.K. Pickett, 2013-2014

With Gauss Quadrature: n=1, 2, 3 ...

n=1
F()
Increasing accuracy
n=2
F()
0.774
0.0
0.0
-0.577
-0.774
8/9
1.0
2.0
-1
n=3
F()
0.577
1.0
-1
5/9
-1
Tables of the gauss

point locations and
corresponding
weighting functions
are available in text
books.
24
12
IFB
A.K. Pickett, 2013-2014

Some comments on Gaussian Integration

Gauss quadrature gives exact integral value for a polynomial of degree 2n-1 when using n
points. The accuracy (integration of the element) increases with n to the exact value.
Gauss quadrature is easily extended and applied to 1D, 2D or 3D elements.
U
Usually
ll 2*2 reduced
d d (less
(l
th
than
exact)
t) integration
i t
ti
i sufficient
is
ffi i t for
f mostt 2D elements;
l
t
(2*2*2 for solids). In fact, the reduced accuracy does compensate for the over stiffening
of finite elements leading to better results! It is also cheaper for CPU and storage.
=0.577
Plane
=-0.577
Solid
element
element
=-0.577
=0.577
Usually codes allow the user control over the number of integration points per element.
Sometimes only n=1 is advised for certain solid elements to avoid locking problems.
25
IFB
Do not confuse the number of element nodes (which decides the strain-displacement
field) with the number of integration points (for accuracy of the element integration).
A.K. Pickett, 2013-2014

Hourglassing: A problem of using to few integration points!

Using reduced integrations has some advantages avoids locking in some elements and
most importantly saves CPU and storage costs.
BUT it can cause hourglassing; for example, see PAM-CRASH tutorial example 1. This can
be a problem
p
in Explicit
p
FE codes that often use one point
p
integration
g
elements.
An hourglass mode
For this deformation mode the integration process will

predict zero change in area at the central integration point.
Therefore the element has no strain and computes zero
internal stress. It is obviously wrong!!!
26
Some (Explicit) codes have hourglass control algorithms to

detect this and correct it (so 1 pt integration can be used).
Or higher order integration (eg 2*2) must be used.
13
IFB
A.K. Pickett, 2013-2014

Hourglass control
In Explicit FE codes we like to try and use 1 point integration elements

1. CPU efficient
2. Minimum storage
BUT Hourglassing is possible.
This is controlled by (usually) very small Hourglass forces which are generated and try
to resist development of these modes.
Resistive
hourglass forces
1p
point integration
g
If it happens:
Reduced integration
1. First the results (stresses, strains and deformation)

may not be as bad as they look!
2. Try increasing hourglass forces (via penalty factors)
Full integration
3. Or use higher order integration elements
27
IFB
A.K. Pickett, 2013-2014

Example: A problem having hourglassing and its consequence on results

No Hourglassing
Hourglassing
Hourglassing
28
Above: Using reduced integration

hourglassing disappears giving sensible
loading and failure sequence.
Above: with 1 point integration

hourglasing has occurred. It
can be a problem in reguar
meshes (especially solids) and
with certain loading
14
IFB
A.K. Pickett, 2013-2014


Contents:
Part 1
Part 2
Parts 3
Buckling analysis
29
IFB
A.K. Pickett, 2013-2014

Non-linear analysis
Linear analysis assumes:
1. Strains are small (usually 1-2%) and strains are linearly related to deformations, =du/dx.
2. Stresses are linearly related to strains, =E; E=constant.
3. Geometric changes
g are small and applied
pp
loads are proportional
p p
to stresses.
Geometric Nonlinearity
P
compression
Material Nonlinearity; eg plasticity, creep...

Linear
part
Non-linear part
Non-linear
part (strain - (failure, strainsoftening, damage)
hardening)
tension
P
Non-linear analysis is needed

if this region is entered
30
Note: The above non-linearities are often combined. For an Explicit code these are easily
handled; for an implicit code special analysis methods are needed.
15
IFB
A.K. Pickett, 2013-2014

Examples of Geometric nonlinear problems

Analysis of a plate: If the loads and displacements
are small then bending effects dominate. As loading
increases membrane effects become important. A
geometric nonlinear analysis may be necessary to
correctly predict the coupled membrane/bending
behaviour if deflections exceed 50% of the plate
thickness (Solid, shell or thick shell elements, for
example, should then be used!).
Snap-through analysis of an Arch: This class of

problem could also include problems such as thin
walled pressure vessels. The geometric changes that
take place greatly effect the internal stresses and a
non-linear analysis is needed.
The snap-though at bifurcation is particularly difficult
to treat for an IMPLICIT code (see Riks or modified
Riks methods); with a dynamic Explicit code it is
easy!
31
IFB
A.K. Pickett, 2013-2014

Examples of non-linearity: Contact and material (usually Implicit analysis

is difficult and Explicit easy)
For Implicit codes applied loading changes the regions of contact which in turn change the
internal stress distributions and structure stiffness matrix. The regions of contact cannot
(usually) be predicted in advance and special non-linear analysis methods are needed.
shaft
Material
nonlinearity
lug
g
Region of contact depends on load
directions and the magnitude.
One surface is called SLAVE the other
is MASTER it doesnt matter which
is which they are interchangeable.
Note the orientations are important.
32
If these peak stress levels are seen to exceed

the material yield stresses then an elastoplastic non-linear analysis will also be needed.
Similarly, if the lug was made of a plastic that
had a time dependent behaviour (creep) then
loading and deformations vary with time and
a non-linear creep analysis is required.
16
IFB
A.K. Pickett, 2013-2014

Example of a contact problem: Pin-to-lug contact

The pin loading could approximately be represented with a constant face load, or
more accurately with a physical pin; in this case a contact interface is needed.
With simple force loading (no plasticity) the problem is linear {P}=[K] {u}
The global stiffness matrix is linear; therefore {u} is easily found by inversion and is linearly
related to the load {P}
With the Pin the area of contact is unknown in advance (it depends on load) and [K]
changes with loading A nonlinear iterative solution is required! [K] = f (P)
Klug
u
Kpin
33
IFB
Klug and Kpin are both constant, but the overlap changes with load.
A.K. Pickett, 2013-2014

Example of load applications

in non-linear analysis
P
P=2e08N
In LUSAS (and other) FE codes the user has control

over initial, increment, and final loads to be applied.
The increments should not be too large or the
solution will have problems (or fail) to converge.
Load P
Sought solution
Iterations
Max load = 2*2e08
Typical load increment

Max change in load = 0.5*2e08
Initial load = 0.001*2e08

34
Displacement u
17
IFB
A.K. Pickett, 2013-2014

Geometric nonlinear problems

The force-deflection response of geometrically nonlinear structures is non-linear:
For cases where there is a linear relationship between applied loading and nodal
displacements computation of the conventional stiffness matrix [Ko] is straightforward.
For a geometrically nonlinear spring we need to find an equivalent element stiffness matrix,
which is now a function of displacements.
displacements We also need a solution strategy.
strategy
Linear spring
u
[P]
P=[K] u
Nonlinear spring P= [K] u
where [K] = [KO]

where [K] = [KO]+[KG]
= linear stiffness plus a
geometric stiffness that depends
on the displacements.
[KG], like [KO], can be derived for an element: This

will be shown later (assume it is known for now).
Slope [K]
=constant=[KO]
The linear and nonlinear responses are different

and the nonlinear response cannot be directly
obtained from solution of P= [KO + KG] u, since KG
is a function of displacements.
Slope [K] = [KO]+[KG]

where [KG] varies with
u (u)
35
IFB
A.K. Pickett, 2013-2014

Solution strategies: The

modified Newton-Raphson
method.
PB
The modified Newton-Raphson (NR) method:
1) Assume the start is known at loads PA with known

displacements uA; this could be the undeformed state.
2) Apply a load increment P to get a new load PB;
PB=PA+ P
3) The easily computed and known [KO] matrix allows the
displacements u1 to be found,
P1
F1
Consequently, an iterative solution strategy must

be used for geometric non-linear problems.
u1=uA + [KO] -1 P
4) The force in the bar F1 for the displacements (u1) can
be computed
F1 = ([KO]+[KG] )u1
Slope [Ko]
5) The out of balance force P1 = (PB - F1) is computed

and used to get new displacements u2,
PA
u2=u1 + [KO] -1 P1
u1 u2
uA
u3
uB
6) Steps 3-5 repeat until the out of balance forces, or the

displacement increment, is sufficiently small.
Note: Each step u1 to u2, u2 to u3 etc., is an iteration in the load increment (PA to PB)
In highly geometrically non-linear problems the load increments P should be small to
ensure convergence, i.e. use many smaller load increments to reach the desired final load.
36
Convergence is when the norm of the out-of-balance forces and/or the norm of the
displacement increments, are sufficiently small. uT u
PT P
18
IFB
A.K. Pickett, 2013-2014

Solution strategies: The (Standard) Newton-Raphson method

The standard Newton-Raphson method is
similar to the modified NR method just
presented. Except now the global geometric
nonlinear
li
stiffness
tiff
matrix
t i [K] (=[K
( [KO] +[K
[KG])
is recomputed at each iteration and used to
provide improved (faster) updating of the
nodal displacements.
This requires assembly and inversion of [K]
at each iteration. It is usually CPU more
expensive, but can give more reliable
convergence in highly non-linear problems.
Note:
Slope [K]1 = [KO] +[KG]1

PB
F1
Slope [K]A = [KO] +[KG]U1
PA
uA
u1 u2
uB
Usually the modified NR method is CPU faster and preferred but it may have difficulties
to converge.
Some codes allow a mixture of the two methods with periodic updating, or updating only
when convergence gets slow and difficult. The user can usually select different options.
37
IFB
A.K. Pickett, 2013-2014

Solution strategies: Some

others aspects
Geometric nonlinear structures tend to be
either hardening or softening, i.e. they
become harder or softer with continued
loading. It can be important to appreciate
which class you are dealing with before
choosing the solution strategy. E.g. the
modified NR can struggle for a hardening
type structure!
At bifurcation points convergence is difficult
or impossible.
p
In this case techniques
q
such
as Riks or Modified-Riks must be used to
control load increments.
38
A geometric hardening
structure [KG] > 0
A geometric
linear structure
[KG] = 0
A geometric softening
structure [KG] < 0
Load
increment
Treating the snap-through also requires

special methods, e.g. displacement control
instead of load control, or dynamic methods
e.g. swap to a dynamic explicit code.
19
IFB
A.K. Pickett, 2013-2014

Revision: The relationship between displacements and material strains

deformed element
Strain in x (=x)
= change in length (OA-OA)
original length (OA)
y(v)
V+
Ditto for strain in y (=y)
Shear strain is the change in

angle AOB to AOB
Note if this was an element

each corner must move in x,y
for the 3 strain components
Original element
x(u)
( )
xy
u
dx u
u
x
x
dx
ditto
v
u
v
1
dx
x
x
x

....
u
1 u 1
dx
dx
39
IFB
Neglect second order terms

for small deflection theory
v
y
... x
v
x
u
x
x ... v u
x
2
To get unity in the denominator
A.K. Pickett, 2013-2014

Principles for computing the [KG] matrix
dx
w
z (w)
w
dx
x
Lateral
loading
dx
Inplane loading
The total membrane strain is
And in full 3D we
have the GreenLagrange strains
40
x(u )
The main strains are inplane, but the

transverse (w) deformations cause additional
strains; length dx stretches to dx,
2
1 w 2
w
dx dx 1 dx 1 ...
x
2 x
Series expansion: This part represents additional

membrane stretch due to a deflection wx
u 1 w

x 2 x
which includes the extra term for

coupled transverse stretching
2
2
2
u 1 u v w

x 2 x x x
2
2
2
v 1 u v w

y 2 y y y
xy
Similar expressions
are developed for,
z , yz , zx
u v 1 u u v v w w

.....
y x 2 x y x y x y
20
IFB
A.K. Pickett, 2013-2014

Implications of small and large deformation strains
1 u v w

2 x x x
ditto for y , z
2
u
x
Lo
Additional second order

terms due to coupling
In Plane
terms
x
L>Lo
Cantilever: Small strains/displacements
u v
1 u u v v w w
y x
2 x y x y x y
xy
Requires only [KO] and a linear analysis.

The strain-displacements are uncoupled and
the tip position is constant in x. The beam
length increases!
Valid for small strains and displacements.
ditto for yz , zx
This part assumes a linear straindisplacement relationship to compute [KO].
If only these terms are considered a linear
analysis is sufficient since [KO] is a
constant (independent of displacements).
L=Lo
Cantilever: Large strains/displacements
This part accounts for coupling effects and

large strains. It is used to develop the [KG]
matrix. A nonlinear analysis is required
since [KG] varies with displacements. [KG]
41 is formulated in the same way as [KO].
IFB
Requires [KO+ KG] and a NL analysis.

The strain-displacements are coupled and
the beam length now remains constant.
Valid for large strains and displacements.
A.K. Pickett, 2013-2014

The [B] matrices for small and large strains

For matrix [BO] to compute [KO]
x
v3
u3
(x3,y3)
y (v)
xy
v
u
(x,y)
v1
u1
v2
(x1,y1)
u
x
(x2,y2)
u ( x, y ) c1 c2 x c3 y
v( x, y ) c4 c5 x c6 y
u
v
c2 y
c6
x
y
u v
c3 c5
y x
x
xy
42
1 u v w

2 x x x
ditto for y , z
2
u

x x
y23
v 1
y
0
y 2 A x
32
xy
u v

y x
[BG]
Slope [K] = [KO] +[KG]
EBG dV
Linear straindisplacement matrix [BO]

for the constant strain
trianglar element
0
x32
y23
EBO dV
[KG]
u v
1 u u v v w w
y x
2 x y x y x y
ditto for yz , zx
u2
x (u)
For matrix [BG] to compute [KG]
u1
v
y31 0 y12 0 1
u
0 x13 0 x21 2
v
x13 y31 x21 y12 2
u 3

xij xi x j
v3
yij yi y j
[Bo] depends on differences of

nodal coordinates (not
displacements). It is a constant.
21
IFB
A.K. Pickett, 2013-2014

Derivation of [BG] for the simple 3 node triangular element
2
2
v
1 u v

y
2 y y
u u v v
u v
y x
x y x y
x
v
xy
u
x
1 u v

2 x x
2
xxL
x
L yyL 1 0
2
xyL
u
0
u
y
u
x
v
x
0
v
y
u
x
0 u

v y

y v
v x
x v

y
In matrix notation
x
1
0
2
u
43
IFB
0
u
y
u
x
v
x
0
v
y
L D.
u
x
0 u

v y

y v
v x
x v

y
Taking the variation for strain increments
d

d L D . dD .
2
2
Note: We work in increments of strain

and increments of du/dx, since the
problem is nonlinear.
Since [D].{d} = {}.[dD]
d L D . d
A.K. Pickett, 2013-2014

Derivation of [BG] for the simple 3 node triangular element (continued)
d L D . d
For the constant strain triangle element
d G d e
u
x
u
y23 0 y31 0 y12

y 1
These 2 rows need
d
computing !!
v
2A
x
0 x32 0 x13 0
v

y
d L D G d e
44
u1
0 v1
u
2

v2
x21 u3

v3
where
xij xi x j
yij yi y j
= [BG] Note this is a function of the element

dimensions and nodal displacement; making
the problem non-linear; it is a function of u.
22
IFB
A.K. Pickett, 2013-2014

Example of a geometrical non-linear analysis
LOAD CASE =
Loadcase 1
RESULT S FILE =
DISPLACEM ENT
Cantilever beam 100mm*7mm using plane stress quadrilateral

elements. The material is elastic Aluminium (E=70GPa, =0.3) with a
vertical concentrated load of 0.1kN at the tip.
CONT OURS OF DY
LOAD CASE =
Loadcase 1
RESULT S FILE =
DISPLACEM ENT
The linear analysis gives a vertical

displacement Dy of 16.6mm, but the
inplane displacement Dx is 0mm at
the centreline! This cannot be correct.
1
1
-0.872737
-0.763645
-0.654553
-0.545461
Linear:
Displacement
Dx
With the linear analysis the displacements will increase linearly with
applied loading
loading.
-0.436369
-0.327277
-0.218184
-0.109092
0
0.109092
Linear:
Displacement Dy
-6.24267
-5.20223
-4.16178
-3.12134
-2.08089
-1.04045
M ax 0.0000E+00 at Node 2
M LOAD
in -16.65
at Node
CASE
= 1 1
Increm ent 1 Load Factor = 0.100E+01
RESULT S FILE =
1
-0.34932
Nonlinear:
Displacement
Dx
45
-12.4853
-11.4449
-10.4045
-9.36401
-8.32357
-7.28312
DISPLACEMENT
CONT OURS OF DX
0.218184
0.327277
0.436369
0.545461
06 4 3
0.654553
0.763645
M ax 0.8727 at Node 1
M in -0.8727 at Node 50
For the Geometric Non-linear

analysis the centreline displacements
are correctly reduced, Dx =1.39mm
in this case.
-15.6067
-14.5662
-13.5258
CONT OURS OF DX
-16.6471
-0.17466
0 17466
0
0.17466
0.34932
0.52398
0.69864
0.873301
1.04796
1.22262
1.39728
1.57194
1.7466
1.92126
2.09592
2.27058
Max 2.434 at Node 1
Min -0.3602 at Node 63
IFB
A.K. Pickett, 2013-2014

Comparison of linear and geometric non-linear analysis for very large

deformation analysis of a point loaded cantilever beam
LOAD CASE
RESULT S FILE =
Linear analysis with large

deformation (large load)
Loadcase 1
1
DISPLACEM ENT
CONT OURS OF DX
-4 37679
-4.37679
-3.82969
-3.28259
-2.73549
LOAD CASE
-2.18839
Increm ent 1
-1.6413
-1.0942
RESULT S FILE =
Y
DISPLACEM ENT
-0.547099
CONT OURS OF DX
Z
X
0
0.547099
1.62563
1.0942
3.25127
1.6413
4.8769
2.18839
6.50253
2.73549
8.12816
3.28259
Large deformation
geometric nonli
linear
analysis
l i
9.7538
3.82969
11.3794
13.0051
M i n -4.388 at Node 18
14.6307
16 2563
16.2563
17.882
19.5076
21.1332
22.7589
24.3845
M in -0.7337 at Node 23
Wrong!
46
Correct (probably)
23
IFB
A.K. Pickett, 2013-2014

PAM-Implicit output file information on convergence progress

For each iteration the norms of nodal
displacement increments and out-ofbalance forces are computed. When
the convergence
o
g
criteria
a are
a met the
loads are incremented, or solution
stopped.
The norms for nodal displacement

increments and out-of-balance
(residual) forces are,
DPNRM
RDNRM
uT u
F T F
Individual maximums
47
IFB
A.K. Pickett, 2013-2014

Time dependent (creep) behaviour: Some

plastics, soils, etc. The solution can be in the
time domain as the material properties change
with time.
stress
Elasto-plasticity: Metals (and Plastics). The

models usually assume elasticity (Hookes laws)
up to a yield point, thereafter hardening occurs.
Different plasticity models (e.g. Von Mises...)
are available to approximate the real behaviour
of metals to fracture. Then there can be
specialised damage laws for the fracturing and
failure process (e.g. Gursons model).
strain
Yield point;
thereafter elastoplastic hardening
Instability
(necking)
leading to
damage and
failure
strain
stress
Nonlinear elasticity: E.g. rubber, plastics,

foams, etc. The unloading response could
be along the original loading curve (zero
hysterisis), or along a different curve for
materials
t i l with
ith hysterisis.
h t i i
stress
Examples of material non-linearity
time
48
24
IFB
A.K. Pickett, 2013-2014

Simple example of elasto-plasticity for a 1D case

Once the material behaviour is nonlinear then the stiffness matrix [K] must be computed
using the nonlinear stress-strain material law (E constant). The Newton Raphson, or
modified NR method is then used to iterate for a solution.
ET
P1
Simple 1D plasticity: Basic concepts
The material is linear elastic = EO for < y ,

Thereafter the material has a tangent stiffness ET
which varies with strain. Both y and the hardening
curve must be determined experimentally.
In the plastic zone the two strain contributions are:
And three definitions for material moduli are:
ES
P0
dp de
EO=/
EO
dp de
d = de + dp ,
d
d
d
EO
, EP
, ES
d E
d P
d E d P
Used to compute [k] for the load increment PO to P1
Solution method:
1. For an imposed load increment the corresponding stress increment d is found assuming
linear elasticity (e.g. for a bar = (P1 P0)/A).
2. From the experimental stress-strain curve the corresponding strain increment d is found.
49
3. The element secant stiffness matrix is simply the stiffness matrix EO scaled by ES/EO.
IFB
A.K. Pickett, 2013-2014

2D-3D Plasticity: Basic principles

For 2D and 3D stress a good (basic) model to plastic behaviour is Von Mises law. This
requires the stresses in a 3D element to be separated into hydrostatic and deviatoric
parts:
1 2 3
1 2
1 3
3
2 3
2 1
3
TOTAL
3 1
3
For metals it is assumed that Hydrostatic stress does not contribute to

permanent plastic strains. But for
foams, plastics, etc., plasticity is usually
influenced by the Hydrostatic stresses.
3 2
3
Deviatoric (shear)
Hydrostatic
This part is the mean stress acting on

all faces which produces only
Volumetric strains.
These are the shear stresses which cause plastic

straining The Von Mises criteria uses these shear
straining.
stresses to develop a yield criteria; in the principal
stress space the Von Mises criteria at yielding is,
F
1 2 2 2 3 2 3 1 2
2
where Y is the uni-axial yield stress. For F<Y the

material is elastic and for F>Y its elasto-plastic.
50
25
IFB
A.K. Pickett, 2013-2014

2D-3D Plasticity: Basic principles

Hardening f(p)
Y and curve as
measured from a
uniaxial test
For the material law you
must specify elastic and
plastic data (various
methods are possible)
The Von Mises criteria in 3D stress space is an infinitely

long
g cylinder
y
whose axis is the Hydrostatic
y
stress and
radius the deviatoric stress (inside the cylinder the
material is elastic and on the surface plastic flow occurs).
During load increments a state of stress outside the
envelop may occur the solution strategy must correctly
bring these incorrect stresses back to the yield surface.
NB the surface may be expanding due to strain hardening,
or due to strain rate effects in dynamic problems.
IFB
A.K. Pickett, 2013-2014

F=Y
Elasticity condition F < Y

2
51
Yield condition
Hardening f(p): Expand

Von Mises envelope
1. Elastic load path

(Hookes law applies).
2. Elastic prediction during
a load increment: Check
yield condition.
3. Iterate back to yield
surface if plastic; or use
elasticity law if elastic.
F Y
1 2 3
3
An Elasto-plastic analysis
Preparing an elasto-plastic analysis
A linear elastic analysis using elastic material
behaviour is first undertaken.
undertaken Inspection of
the maximum stresses for the applied
loading allows an estimate of the load
needed to reach the yield point.
The elasto-plastic material law is then
defined and assigned.
Also specify the necessary controls to
perform
f
a non-linear
l
elasto-plastic
l
l
analysis.
l
The analysis is undertaken with several load
increments to a specified maximum load.
52
26
IFB
A.K. Pickett, 2013-2014

Typical load controls (for the FE program LUSAS)

Allows manual or automatic control
over load increments
Norms used to decide convergence
Indicates the load increment of the
linear load to start the NL analysis
Number of iterations to be used for
each load increment (max = 12)
Arc length controls are needed if the

deformed geometry stiffness drops
significantly (K/KO < 0.4)
The termination criteria chosen is
point 1, displacement V (y-direction),
having a max value = 10mm
More controls for the load increments
53
IFB
A.K. Pickett, 2013-2014

Example PAM- controls for nonlinear Implicit analysis
Static loading
nonlinear
l
analysis
l
Standard Newton
Raphson
User selected
convergence criteria
54
27
IFB
A.K. Pickett, 2013-2014

Example PAM- outputs for nonlinear Implicit loading
Iteration
Load increment
Iteration number
Load
increment
Criteria for norm of

residual forces not met
More iterations
needed before the
next load increment
Criteria for norm of

residual forces met
NEW LOAD INCREMENT
55
IFB
A.K. Pickett, 2013-2014


Contents:
Part 1
Part 2
Parts 3
Buckling analysis
56
28
IFB
A.K. Pickett, 2013-2014

Buckling analysis
Buckling is instability when the stored

membrane energy is transferred into
bending energy without any change in
the external loading.
Eg. for a simple cantilever column the
critical load is given by the Euler load,
Pcr
( EI )
2
2
e
Primary (theoretical)
load path
P=Pcr
P
L
a
b
c
d
Bifurcation point e
and actual load path f
b
a
where Le 2 L
Buckling of complex structures can be predicted using Finite Elements and the total
geometric nonlinear stiffness matrix ([K] = [KO]+ [KG]); NB the [KG] matrix is nonlinear
since it is a function of the applied loads P.
57
IFB
A.K. Pickett, 2013-2014

Buckling - a definition:
At buckling a load is reached where an applied virtual displace-ment {d} causes the loss in
membrane energy (due to this displacement) to equal the gain in bending energy without any
change in the external loading. The structure behaviour is unstable and there are different
possible deformations for the same applied loading. Mathematically the deformations {d}
causes zero
e o change
c a ge in tthe
ee
external
te a loading,
oad g,
([kO] +[kG]){d} = {dR} = 0 where {d} corresponds to the buckling mode.
P
Perfectly
straight
column
behaviour
Imperfect column
behaviour
2nd buckling mode
Pcrit 4 EI / L
2
More modes which occur at

higher loads are feasible
Important: A FE model will need a node near

here in order to capture this mode! You cannot
get all the buckling modes from a coarse mesh.
mesh
1st buckling mode (most likely): The
dashed line is an equally possible mode.
Pcrit 2 EI / L2
Note: There are as many possible buckling modes as unconstrained dof in the FE model.
58
The problem is actually formulated and solved as an eigenvalue problem.
29
IFB
A.K. Pickett, 2013-2014

Some basic Euler buckling loads

These usually apply to slender columns in compression. The Euler load is the maximum load
at which buckling occurs; in short (thick) columns other forms of compressive failure may occur
first. In practice both compressive and Euler buckling loads must be checked and designed for.
The formula (derived in many standard text books) for Euler buckling is,
Pcr
2 ( EI )
L2e
The Euler load Pcr depends on the bending stiffness (EI), the length (L) and the end
constraints (free, pinned or clamped). Depending on the constraints the column effective
length Le (=KL) is modified as below.
Le
Le
Deformation for a freefree column (Le = 1.0 L)
59
IFB
End restraint
Le=KL
pinned-pinned
i
d i
d
10*L
1.0
clamped-clamped
0.5 * L
pinned-clamped
0.7 * L
clamped-free
2.0 * L
Deformation for a clampedclamped column (Le = 0.5 L)
A.K. Pickett, 2013-2014

Preliminary design: Buckling

Thin walled structures are likely to suffer both global (the whole structure) and local
buckling (parts of it). This is most likely in areas of compression, but shear loaded parts
can also buckle as shown below.
P
Possible buckling
g in the side flanges
g (under
(
the point load). Simple Euler buckling is also
possible.
Possible buckling on the lower compression flange.

Buckling is also possible in the shear webs, especially if they are deep.
Compression may
cause a shear
h
buckling mode!
In each case different buckling modes are possible depending on the loading, geometry
and the interaction with the remaining structure.
Each possible mode will have a critical load to initiate it and usually the lowest will occur.
The structure must be designed to have sufficient reserve of strength before buckling.
60
30
IFB
A.K. Pickett, 2013-2014

LUSAS analysis of a composite box beam for the first 5 buckling modes
Geometry L=2000, b=94, h=188 and t=4.7.
Fully constrain one end and loaded at the other
with 1000N; E=32.8GPa, Gxy=4.8GPa, =0.3.
Z
MODE
FREQUENCY
ERROR NORM
53.5783
EIGENVALUE
1.16497
0.606236E-12
57.7202
1.20916
0.458210E-12
-57.7215
0.000000E+00 0.477816E-12
-62 1734
-62.1734
0 000000E+00 0.404732E-12
0.000000E+00
0 404732E-12
62.1760
1.25496
0.926765E-12
63.1296
1.26455
0.190411E-10
Mode 1: Pcr=1000 * 53.57

Local end collapse
Some results for the FE Buckling

analysis of the end loaded
(P=1000N) cantilever beam
Mode 2: Pcr=1000 * 57.72

Compressive lower face
Z
Y
Mode 3: Pcr=1000 * 57.72

Side walls due to shear
61
IFB
Mode 5: Pcr=1000 * 62.17

Higher energy mode on lower
face?
Note: A negative eigenvalue indicates a

reversal of the applied load is needed for
the shown buckling mode.
A.K. Pickett, 2013-2014

Frequency analysis: Summary
mass M
stiffness K
The dynamic equation of motion is,
M D C D K D R
damping
p gC
applied
load R
where [M], [C] and [K] are the nodal mass matrix, and the element damping and stiffness
, D and D the nodal accelerations,
matrices. {R} is the applied nodal loads and D
velocities and displacements.
Ignoring nodal loading and damping, and assuming harmonic motion with all particles in
phase with one another,
D D sin t
and
where D represents amplitudes of the d.o.f,

combining 1 and 2 and ignoring damping and
external forces gives the linear eigenvalue problem,
D D sin t
K M D 0
2
62
31
IFB
A.K. Pickett, 2013-2014

Frequency analysis:
Basic theory
M1
For undamped vibration without

external forces the displacements
x1, x2, , xn at a given instance of
time t for n masses (=dof ) are
given by the equations of motion
M 1
x1
k1
M2
x2
k2
M3
x3
M d
dt
x
2
k3
K x 0
M4
x4
d x1
k11 x1 k12 x2 k13 x3 .... k1n xn 0
dt 2
2
d x2
M2
k 21 x1 k 22 x2 k 23 x3 .... k 2 n xn 0
dt 2
.
.
M1
x1
x
2
.
x
.
.

xn
d 2 xn
k n1 x1 k n 2 x2 k n 3 x3 .... k nn xn 0
2
63 dt
Mn
IFB
k11
k
21
.
K
.
.
k n1
d 2 x1
k11 x1 k12 x2 k13 x3 .... k1n xn 0
dt 2
d 2 x2
M2
k 21 x1 k 22 x2 k 23 x3 .... k 2 n xn 0
dt 2
.
2
Mn
d xn
k n1 x1 k n 2 x2 k n3 x3 .... k nn xn 0
dt 2
M3
0
.
.
.
kn 2
d 2 x1
2
dt2
d x2
2
d 2 x dt
2
dt
2.
d xn
dt 2
k1n
k 2 n
Siffness
matrix
k nn
Displacement
and
acceleration
vectors
A.K. Pickett, 2013-2014

M1
k12
k 22
Mass
matrix
M n
M2
The problem is to find a set of possible

displacements x1, x2, .., xn and accelerations
that satisfy these equations.
But the equations are coupled with to many
unknowns and the solution is non-trivial.
1
It can be assumed that all masses

undergo a common harmonic motion
xi ei sin(t ) for i 1,2,3....n)
e4
2
x1
e1
xi = displacement of node i at time t.
64
x4
x2
e2
x3
e3
e5
x5
ei = amplitude of vibration of mass at node i.
The modified governing equations are:
= angular velocity (radian per second).
2 M 1e1 k11e1 k12 e2 k13e3 .... k1n en 0
/2 = The fequency of vibration (cycles per

second) and is the same for all points.
2 M 2e2 k 21e1 k 22e2 k 23e3 .... k 2 n en 0

.
2 M n en k n1e1 k n 2e2 k n3e3 .... k nn en 0
32
IFB
A.K. Pickett, 2013-2014

2 M 1e1 k11e1 k12 e2 k13e3 .... k1n en 0

2 M 2 e2 k 21e1 k 22 e2 k 23e3 .... k 2 n en 0
.
2 M n en k n1e1 k n 2e2 k n 3e3 .... k nn en 0
2 M e K e 0 or
K M e 0
2
where {e} is the vector

{e1,e2,e3en} of the amplitudes of
vibration of the n masses
This is an eigenvalue problem and for non-zero solutions the determinant of this system of
equations must be zero.
K 2 M
There are n eigenvalues of 2 each of which have a corresponding set of amplitudes (=

eigenvectors) {e1,e2,e3en}. The n eigenvalues are:
12 , 2 2 , 3 2 ,......, n 2 0 where 12 2 2 3 2 ...... n 2

Each corresponds to a natural frequency (/2); (1/2) is the first natural frequency .
65
33

Finite Elements: Some Advanced Topics and Non-Linear (Implicit) Analysis

Uploaded by

Copyright:

Available Formats

Finite Elements: Some Advanced Topics and Non-Linear (Implicit) Analysis

Uploaded by

Document Information

Original Description:

Original Title

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

Finite Elements: Some Advanced Topics and Non-Linear (Implicit) Analysis

Uploaded by

Copyright:

Available Formats

IFB

A.K. Pickett, 2013-2014