Finite Element Analysis

Contents
Truss Frame Analysis................................................................................................................................................2
Constant Strain Triangle...........................................................................................................................................3
Element Formulation................................................................................................................................................4
Linear Rectangular Element.................................................................................................................................4
Quadratic Rectangular Element...........................................................................................................................5
Axisymmetric Elements.......................................................................................................................................6
3-D Solid Elements (‘Brick’ Elements)..................................................................................................................6
Truss and Beam Elements....................................................................................................................................6
Truss Elements................................................................................................................................................6
Beam Elements................................................................................................................................................6
Membrane, Plate and Shell Elements..................................................................................................................7
Membrane Elements.......................................................................................................................................8
Plate Elements.................................................................................................................................................8
Shell Elements.................................................................................................................................................8
Membrane Element Characteristics................................................................................................................8
Plate Element Characteristics..........................................................................................................................8
Shell Element Characteristics..........................................................................................................................9
Isoparametric Elements.......................................................................................................................................9
Element Performance..........................................................................................................................................9
Elements are too stiff......................................................................................................................................9
Quadrilaterals work better than triangles.......................................................................................................9
Quadratic elements are better than linear ones...........................................................................................10
Elements work best when close to their unmapped shape..........................................................................10
Improving Element Performance......................................................................................................................10
Non-Conforming Elements............................................................................................................................10
Reduced Integration......................................................................................................................................10
The Element Library...............................................................................................................................................10
Line Elements.....................................................................................................................................................10
Area Elements for 2D Analysis...........................................................................................................................11
Area Elements for 3D Analysis...........................................................................................................................11
Volume Elements...............................................................................................................................................11
Springs and Links................................................................................................................................................11
Materials, Loads, Supports and Solution...............................................................................................................12
Material Properties............................................................................................................................................12
Geometric Properties........................................................................................................................................12
Applied Loads.....................................................................................................................................................13
Supports and Prescribed Displacements...........................................................................................................13
Global Stiffness Matrix and the Solution...........................................................................................................14
Other Problems..................................................................................................................................................15
Guide to Good Modelling.......................................................................................................................................15

Ian FEA 1
Truss Frame Analysis
Truss elements are used to model bars that are pin-jointed, therefore only take axial loads (no
moments), also, if the structure is in two dimensions, a 2D truss element is used. If stress is constant
throughout its length, single elements can be used to model each member. Simple shape functions
are used to describe the displacement anywhere within an element and then used to calculate
stresses and strains.

Node point variables are characteristic to an element type, if a beam element is used to model axial
and bending behaviour, then it must be connected to its neighbour in both translation and rotation,
therefore transmitting both forces and moments. In 3D analysis, this leads to 6DoFs at each node. 2D
trusses only have forces in two directions, therefore 2DoFs at each node.

For a simple truss element at an angle θ to the x-axis, u and v are displacements in the X and Y
directions, z is the distance along the element from the first node (i) to the second (j), w is
displacement in direction z:

w ( z )=C 1 +C2 z

w j−wi
At z=0 (node i), w=wi, therefore C 1=w i and at z=L and w=w j therefore c 2=
L

w i=ui cos θ+ v i cos θ

w j=u j cos θ+ v j sin θ

Next is to derive the element stiffness matrix [k]. To do this it is necessary to know the relationship
between displacements and forces at each node. It expresses the force in each DoF of the element
due to displacement in each DoF. It is square and the number of rows and columns is equal to the
total number of degrees of freedom of the element. For 2D truss there are two nodes and two DoFs
at each node, therefore the matrix is 4x4.

If P and Q are forces in the X and Y directions at nodes and T is the tension in the element, A is cross
sectional area and E is Young’s modulus:

Pi=−T cos θ , Qi=−T sin θ , P j =T cos θ , Q j =T sin θ

T
Stress=E × strain=
A

dw d w j−w i
Strain= = ( C 1 +C2 z )=C 2=
dz dz L

EA
T= ( u cos θ+ v j sinθ−ui cos θ−v i sin θ )
L j

Ian FEA 2
 ui
T=
EA
L
v
[ −c −s c s ] i
uj
vj
[]
1
where the second matrix is the matrix of displacements [d], [ −c −s c s ] is the dimension
L
vector B.

strain= [ e ] =[ B ][ d ]

T =EA [ B ][ d ]

The vector of forces, [R] at each node:

Pi

[ ][ ]
−c
Qi −s T t
[ R ]= = T =[ B ] TL=[ B ] EAL [ B ][ d ]
Pj c
Q s

[ R ] =[ k ][ d ]

[ k ]= [ B ] T EAL [ B ] =[ B ]T E [ B ] × volume of element

c2 cs −c s −cs

[ 2
[ k ]= c s2 s −cs
−c −cs c 2
−cs −s 2
cs
−s2 EA
cs L
s2
]
the stiffness matrix is always symmetric and the coefficients of the leading diagonal are always
positive or zero.

Equations are then assembled for the whole structure to give [ F ]= [ K ][ δ ] and [ δ ] =[ K ]
−1
[F]

Constant Strain Triangle

This is the simplest continuum element and differs from a truss element that it models part of a
continuum rather than a 1D bar. It is a 2-demensional element and used to model plane strain, plane
stress or axi-symmetric geometry and indicates a constant rate of strain (and therefore stress) at all
locations.

In a pin-jointed par, the stress is constant along its length, however the stresses in a continuum may
vary in a complicated way, therefore there will be a degree of approximation in the representation.
This is easiest to see in a cantilever, as the assumption that stress is constant throughout is not
correct, as the bending stress will vary linearly in both the X and Y directions in the beam, and the
shear stress will vary parabolically in the x-direction. There will also be a concentration of stress
around the point of application of a load.

Ian FEA 3
The CST is rarely used nowadays as more sophisticated elements with higher order shape functions
that represent the varying stress within an element are preferred.

The derivation of the CST is seen in the notes, Section 3 pp2-9.

Element Formulation
It is vitally important to choose the right element type, for example continuum plane elements to
model plane stress, plane strain, and axi-symmetric; continuum solid elements for modelling 3D
solids; beam; and shell elements.

For more complicated elements it is difficult to find forces at nodes, therefore an energy method
called the principle of virtual work is used. This is simply

work done within an element=external work done by forces at nodes

Strain energy density can be calculated for any location in an element:

1
SED= ( σ xx e xx + σ yy e yy +σ xy e xy +σ yx e yz +σ xz e xz )
2

and in plane stress or strain reduces to

1
SED= ( σ xx e xx + σ yy e yy +σ xy e xy )
2

then the work done internally to the element is equal to the integral over the volume of the SED.
External work is done by forces at nodes and these can be equated:
n
1 1 T
External work = ∑ P n un +Qn v n = [ R ] [ d ]
2 node1 2

[ R ]T [ d ]= ∫ ( [ σ ] T [ e ] )= ∫ ( [ σ ] T [ B ][ d ] )
volume volume

this leads to

[ k ]= ∫ ([ B ] T [ D ][ B ] )
volume

Linear Rectangular Element

Used for plane strain, plane stress and axisymmetric models.

In elements, the displacements at any location are described by shape functions. Shape functions for
a linear rectangular element are:

1
N 1= ( a−x ) ( b− y )
4 ab

Ian FEA 4
 1
N 2= ( a+ x )( b− y )
4 ab

1
N 3= ( a+ x )( b+ y )
4 ab

1
N4= ( a−x )( b+ y )
4 ab

Shape functions should be 1 at the node that it corresponds to and zero at any other node.
Displacements are found from a big matrix on section 4 p4.

Expansion of the shape functions gives an (incomplete) quadratic expression:

u ( ¿ v )=C1 +C 2 x+ C3 y +C 4 xy

Strain is obtained by differentiating the shape functions. When the displacements (or [B] matrix)
varies according to location in the element, numerical integration is required to find [k]. This is done
using Gauss integration because the element strain functions are polynomials. Gauss integration
evaluates the integral of a polynomial by looking at the value of certain points (Gauss Points) and
putting them into function to calculate the integral:
4
T
[ k ]=ab ∑ ( [ B ] [ D ][ B ] )
GP=1

Once the solution of the global matrix has been found, the stresses and strains are calculated at the
Gauss points (they are more accurate at these locations) and then extrapolated from them if needed
elsewhere.

Quadratic Rectangular Element

An improvement to using linear shape functions is to use quadratic shape functions, again used for
plane strain, plane stress and axi-symmetric models. For these, more nodes are needed in order to
describe the quadratic variation of displacement:

Expansion of the shape functions gives incomplete cubic polynomial functions of the form:

u ( ¿ v )=C1 +C 2 x+ C3 y +C 4 xy +C 5 x 2 C6 y 2 +C7 x 2 y +C 8 x y 2

Ian FEA 5
Gauss integration of the stiffness integral is now exact (integrand is to the fourth power) if only four
points (two in each direction) are used. For exact integration it is necessary to use the next order of
Gauss integration where 3 coordinates are taken in each direction, therefore giving 9 Gauss points.
(3 points give accurate integration up to the 5 th order). For the different number of Gauss points,
there are specific locations and weighting factors.

Axisymmetric Elements
These are widely used as many structures have cylindrical form and axisymmetric elements
essentially allow a 3D geometry to be modelled in two dimensions. They can be used to model a
thick walled cylinder, with each 2D element lying in the RZ plane for cylindrical coordinates (RθZ).
Kinda represents a very thin slice of pie. For this to be possible the geometry and loading of the
structure must be axially symmetric (rotational symmetry).

Elements designed for plane stress or plane strain can be formulated to model asxiymmetry. You
need to convert to cylindrical coordinates and incorporate the relationship between radial
displacement and circumferential stress. This is done by differentiating the shape functions (which
are the same as for plane stress/plane strain) in a slightly different way and the [D] matrix is slightly
different too.

3-D Solid Elements (‘Brick’ Elements)

The same principles for the formulations of 2D continuum elements are used for 3D solid (or ‘brick’)
elements. Another dimension is added to the matrix equations. A linear brick element is a cuboid
with nodes at each of its 8 corners and displacement varies linearly in three coordinate directions.
Quadratic brick elements have midside nodes, giving a total of 20. Both have 3DoFs at each node,
therefore giving a total of 24 for linear and 60 for quadratic. Gauss integration is also extended tot
the third dimension so the integration is done by a three dimensional array within the brick. Linear
elements require 2 points in each direction, giving 2 3 (8) points and quadratic elements require 3
points in each direction, therefore 33 (27) points.

Truss and Beam Elements

These are line elements that are used predominantly to model members in frames. They are 1D in
shape, but are used in 2D or 3D models. Truss elements are pin-jointed, therefore describe axial
stress only, whereas beam elements describe axial, bending and shear stresses.

Truss Elements
2D truss elements were discussed earlier, where each node has 2DoFs corresponding to the two
coordinate directions. A 3D truss element is created in the same way, incorporating an extra axis.
There are now 3 displacements at each of the two nodes, making 6DoFs in total. The stiffness matrix
is therefore 6x6.

Beam Elements
These contain the same formulation as truss elements to describe axial stiffness, but also
incorporates bending and shear deformations. A 3D beam therefore has 6DoFs at each node; three
rotations and three displacements corresponding to three moments and three forces. 2D beams
work in one plane, therefore have 3DoFs; two displacements and one rotation.

Ian FEA 6
Beam elements are used to describe beams with built-in boundary conditions, for example
cantilevers, frameworks with rigid joints (transfer moments) and 3D beams also model torsion.

Bending stresses in a beam vary with axial distance and also lateral distance from the neutral axis.
The displacements along the beam require cubic shape functions, therefore the algebra is a bit
lengthy for the derivation of the stiffness matrix. Therefore standard beam deflection formulae are
used for each mode (deflection and rotation) at each node. When all the DoFs have been
superimposed, the stiffness matrix looks like:

12 6 L −12 6 L
EI
[2
[ k ]= 3 6 L 4 L −6 L 2 L
L −12 −6 L 12 −6 L
2

6 L 2 L2 −6 L 4 L2
]
Generally shear stiffness is negligible in beams, however in a short deep beam some deflection due
to shear loading will be present. This is just added to the displacement due to bending.

Axial stiffness is simple that of the truss element, however if the element lies parallel to the x-axis, it
becomes much simpler (all the angles go) and related only to ui∧v i. The full stiffness matrix is then
assembled:

A ( L2 +γ ) − A ( L2 + γ )

[ ]
0 0 0 0
Pi I I

[]
Qi 0 12 6L 0 −12 6L
2
Mi EI 0 6 L 4 L +γ 0 −6 L 2 L2−γ
= 3
Pj L + Lγ − A ( L2 +γ ) A ( L2 +γ )
0 0 0 0
Qj I I
Mj 0 −12 −6 L 0 12 −6 L
2
0 6 L 2 L −γ 0 −6 L 4 L2+ γ

The element is then orientated in any direction in the XY plane by:

[ k ' ]= [ t ] [ k ][ t ]T
c s 0 0 0 0
−s
t= 0
0
0
0
[ c
0
0
0
0
0 0 0
1 0 0
0 c s
0 −s c
0 0 0

Membrane, Plate and Shell Elements

0
0
0
0
1
]
Examples of things made from plates or shells are pressure vessels, oil storage tanks, ships, car
bodies and concrete floor slabs. Whilst these are 3D, simple but accurate calculations can be
performed to calculate displacement and stresses without resorting to 3D analysis. E.g. membrane

Ian FEA 7
behaviour (in-plane stretching) of a thin-walled cylinder with internal pressure is calculated by
simple formulae:

Pr Pr
Axial stress= , Circumferential stress=
2t t

these are possible by assuming very simple behaviour in the through-thickness direction of the
structure – above the stresses are assumed to be constant through the thickness of the shell. Finite
elements that model these structures take advantage of this, therefore are able to model using 2D
shapes. The elements are positioned in the mid-plane of the plate or shell.

Structures could be modelled using brick elements, however a plate or shell element only requires
nodes on the mid-plane and therefore it is more efficient to build the model. They will also give
better results unless lots of bricks are used, as bricks need the sides to be equal-ish lengths to
perform well

Membrane Elements
These only model the in-plane stiffness (and stresses) of a structure. The stresses are constant
through the thickness of the structure – analogous to truss element. Could be used to model thin-
walled cylinder (above).

Plate Elements
These model bending behaviour of a structure. The in-plane stresses are assumed to vary linearly
through the thickness of the plate. Also, in-plane stresses are 0 on the midplane (neutral plane in
bending) so the element does not model membrane behaviour. Analogous to bending and shear
parts of a beam element, without the axial part. Thin plates model bending only, where thicker
plates model both bending and transverse shear. Used to study concrete floor slab supported at
edges and loaded with heavy equipment – the slab deforms in bending and transvers shear but
membrane deformation is negligible.

Shell Elements
These are general-purpose elements that incorporate the features of both membranes and plates
(in-plane stresses, bending stress and optionally transverse shear (need thick shells for this)). Shell
elements can be used to model thin-walled cylinder subjected to a local radial force – under this
loading the cylinder experiences both membrane and bending stresses.

When modelling using these elements, it is important to identify the type of behaviour expected.
Shells are safe as they model both but have 6DoFs at each node and are therefore expensive in
computer resources. It is also important to choose between thick and thin elements – if the
thickness of the structure is small compared to its length then thin elements will suffice, if not then
you need thick elements that also take into account transverse shear. It is important to realise when
these elements CANNOT be used – e.g. thick cylinders as stress varies according to non-linear
functions through the wall thickness (Lame).

Membrane Element Characteristics

These need only two in-plane displacement DoFs at each node. Can be set up in a local coordinate
axis system where it is in the XY plane and displacements at each node are u and v. The element
models σxx, σyy, and σxy and all other stresses are zero. This behaviour is described by plane stress

Ian FEA 8
elements discussed earlier (CST, and quadrilaterals) – the plane stress element is the membrane
element, however membranes are usually valid for 3D space by using coordinate transforms to
create elements in any orientation – this introduces a third coordinate so that the element has 3
displacement DoFs at each node.

Plate Element Characteristics

These assume the same coordinate system as the membrane element, there are now three DoFs at
each node – w ,θu∧θv (similar to setting up bending stiffness in beam element). The element
models bending and transverse shear of the plate. Bending assumes linear variation of bending
stress through the thickness. Corresponding to the displacement degrees of freedom, it has a lateral
force and two bending moments at each node. σxx and σyy are bending stresses at the two surfaces of
the plate and the transverse shear stresses σxz and σyz are averaged through the thickness of the
plate. Again, coordinate transformations are used and this results in an extension of the number of
DoFs – a plate in an arbitrary orientation in 3D space requires all 6DoFs.

Shell Element Characteristics

These have 5 DoFs – u , v , w , θu∧θv though when coordinate transformations are used we need all
6DoFs. If the shell is flat, it can be formulated by addition of a plate element to a membrane because
membrane behaviour is completely independent to bending and transverse shear. A stiffness matrix
with all 6 DoFs is assembled adding the coefficients from each element type. However if the shell is
curved the formulation must also include coupling between membrane and bending/shear
behaviour – a force applied to a curved shell will generate bending moments. It is important to
remember that a shell has no stiffness in the rotation about its normal plane or θw direction –
moments cannot be applied about the normal to the plane.

Isoparametric Elements
So far we have assumed regular shapes, however it is often required to have irregular shapes such as
curved surfaces or edges not at 90°. Regular elements can be made to accommodate irregularity by
a mapping process. The normal rule for this is that the shape of the element takes the form as the
shape of the displacements given by shape functions. This means that shape functions describe both
the displacements within the element and the shape of the element and such elements are called
isoparametric. E.g. linear rectangular elements can be specified in a model as straight-sided
quadrilaterals, quadratic rectangles can be specified as quadrilaterals with quadratic shaped edges –
these are then mapped to squares. This means that curved surfaces can be modelled with quadratic
elements by appropriate positioning of midside nodes. If a curve being modelled is not quadratic,
fine mesh is needed to represent it accurately (however the error due to approximation of curvature
is never significant). More importantly the accuracy of an element is reduced if an edge is curved too
much – an upper limit of 30° of an arc of a circle is often used – therefore a plane stress/strain model
of a plate with a hole can represent the hole with 12 elements around the circumference. Mapping is
done using a “Jacobian” matrix.

Element Performance
Elements are too stiff
Shape functions approximate the displacement field using polynomial coefficients – higher
coefficients needed to describe the field perfectly are truncated – therefore displacements tend to

Ian FEA 9
be underestimated. As a general rule both displacements and stresses predicted by an FE model are
underestimated (although not universally true if correction factors or fixes are incorporated).

As element size is reduced, the error terms are reduced and its representation of the field is
improved. This is known as ‘convergence’ and with a sufficiently fine mesh the FE results should be
very close to the true solution. Confirmation of convergence requires a second analysis to
demonstrate that the results are the same (ish) when the mesh is refined.

Quadrilaterals work better than triangles

Linear rectangles have some variation of strain over the area of the element, whereas the CST did
not, therefore the rectangle is better. Triangles are used principally to occupy spaces where
quadrilaterals can’t fit such as acute angles or regions between course and fine parts of a mesh –
here it is better to use triangles than badly shaped quadrilaterals.

Quadratic elements are better than linear ones

Quadratic have at least twice as many nodes as linear elements, increasing the computation effort,
however with their increased accuracy you don’t need as many of them. There are some
circumstances where linear might be better, but it is best to start with a preference for quadratic
elements. E.g. a beam in bending deforms to a shape represented by quadratics, therefore is linear
elements are used (even lots of them) it will not be as good and the linear model will be far too stiff.

Elements work best when close to their unmapped shape

Accuracy of all elements is diminished when they are distorted therefore it is best to keep them as
close to their undistorted shape if possible. Most programs give warnings for excessively deformed
elements. Most elements are fairly good, but avoid thin elements with aspect ratios of more than 3
and angular distortions more than 30°. Errors are generally more evident in stresses rather than
displacements, therefore you can get away with it a bit if you don’t need stresses.

Improving Element Performance

Non-Conforming Elements
These are quadratic elements where the midside nodes are not connected. This improves
performance within an element, however the smaller number of DoFs is retained. They are a
compromise, and will lead to gaps and overlaps as they are not corrected attached to their
neighbours. This is bad, but non-conforming elements are seen as a compromise between linear and
quadratic elements. Lots of programs use this.

Reduced Integration
Elements can be artificially softened by not performing full integration when calculating stiffness.
This is done sometimes by using a Gauss integration scheme without the full complement of points.
Quadratic elements that should be integrated by 3 points in each direction are integrated using two
points. The effect is not large but nearly always beneficial.

The Element Library

Line Elements
Element Type Description Nod DoFs Geometr Allowable Stress/Strain

Ian FEA 10
 es y Loads
Pin-jointed with constant X, Y, Z (X, Y Nodal Force, xx, constant for
Truss 2 Area (A)
axial stress for 2D) T, body force element
Nodal force,
Straight Beam with all DoFs X,Y,Z, RX, xx ave, xx bending
Straight A, I (r & t distributed
at both nodes. Shortcut 2 RY, RZ (X,Y, about y and z (linear)
Beam if tube) force, T, body
spec for use as a tube RZ for 2D) torsion,
force
Similar to straight beam but Nodal force,
X,Y,Z, RX,
quadratic curved shape – A, I (r & t distributed xx ave, xx bending
Curved Beam 3 RY, RZ (X,Y,
used on edge of shell e.g. if tube) force, T, body about y and z, torsion,
RZ for 2D)
modelling stiffening rib force
Model axial stress and R, Z at each Membrane average,
Axisymmetri Thicknes Nodal force,
bending stress about 2 node, Rθ at bending about θ axis
c thin shell s T, body force
normal to plane each node linearly along element

Area Elements for 2D Analysis

All of the following model plane strain, plane stress or axisymmetric solids.

Nod DoF Inegrati Geom

Element Type Description Allowable Loads Stress/Strain
es s on etry
2x2 xx,yy,xy pl stress
Linear In-plane nodal force,
X,Y Gauss t if pl xx,yy,zz,xy stresses
Quadrilateral (Pl Linear SFs 4 body force,
stress xx,yy,xy strain for
Stress/ Strain) distributed edge force
pl strain
Linear 2x2
In-plane nodal, body,
Quadrilateral Linear SFs 4 R,Z Gauss none RR,ZZ,RZ,θθ
Pressure to edge
(axisymmetric)
Straight sided but with
Non-conforming
quadratic shape 4/8
quad
functions.
Linear SFs, constant S/S X,Y Explicit t if pl See linear See linear
CST 3
pl S/S or axisymm R,Z stress quadrilateral quadrilateral
3x3
Quadratic X,Y Gauss t if pl See linear See linear
Quadratic SFs 8
Quadrilateral R,Z (2x2 for stress quadrilateral quadrilateral
softer)
Single
Quadratic X,Y T if pl See linear See linear
Quadratic SFs 6 3 for
Triangle R,Z stress quadrilateral quadrilateral
softer

Area Elements for 3D Analysis

Element Geomet
Description DoFs Allowable Loads Stress/Strain
Type ry
No variation through
In-plane nodal force,
thickness, in-plane only. X,Y,Z xx,yy,xy pl stress
Membrane t body force,
Same as pl stress elements
distributed edge force
but orientated arbitrarily
Bending and through-
In-plane nodal, body,
thickness shear only. xx & yy bending (linear)
Plate Z,Rx,Ry t Pressure normal to
Variation through xy & yz shear (average)
plane
thickness is linear
X,Y,Z, RX,RY,RZ
Membrane and plate
Shell (though RZ has t All of above All of above
behaviour
no stiffness)

Volume Elements
Element Type Description Nodes DoFs Allowable Loads Stress/Strain
Linear Brick Cuboid w/ 8 X,X,Z Nodal Force, body xx,yy,zz,xy,xz,yz

Ian FEA 11
 linear SFs force, pressure
Cuboid w/ 20 Nodal Force, body
Quad Brick X,X,Z xx,yy,zz,xy,xz,yz
quadratic SFs force, pressure

Springs and Links

Nodes Stress/
Element Type Description DoFs Properties
Strain
Axial Spring Linear Spring 2 X,Y,Z (X,Y if 2D) User sets stiffness None
Rotational Spring Torsion 2 RX,RY,RZ (RX,RY if 2D) User sets stiffness None
Rigid pin-jointed connection – 2
Rigid Link X,Y,Z (X,Y if 2D) None None
keeps nodes same distance apart

Materials, Loads, Supports and Solution

 Structure – This involves creating a mesh and specifying material properties and geometric
properties (e.g. thickness for shell elements).
 Loads, Supports and other Constraints – Load types include nodal forces, pressure and
thermal loads. Eventually, all loads must be expressed internally as nodal forces. Structural
supports must be applied to prevent rigid body motion and also to impose symmetry
conditions, non-zero displacements can also be prescribed. Other forms of constraints can
also be applied, such as rigid connections between nodes.
 Solution – This is what takes the most computational effort and the time taken is very
dependent on the sequence in which global equations are assembled – governed by the
sequence used when numbering nodes.
 Post-Processing – This includes calculation, printing and plotting of displacements and
stresses.

Material Properties
Material properties are used in the calculation of the stiffness matrix and may be different in
different regions. The constitutive equations for a 3-D isotropic elastic solid are:

e xx 1 −v −v 0 0 0 σ xx

[] [ ][ ]
e yy −v 1 −v 0 0 0 σ yy
e zz 1 −v −v 1 0 0 0 σ
= = zz
e xy E 0 0 0 2 ( 1+ v ) 0 0 σ xy
e yz 0 0 0 0 2 ( 1+ v ) 0 σ yz
e xz 0 0 0 0 0 2 ( 1+ v ) σ xz

D is the inverse of the above matrix ([σ=[D][e])

Ian FEA 12
 1−v v v 0 0 0

[ ]
v 1−v v 0 0 0
v v 1−v 0 0 0
1
E 0 0 0 (1−2 v ) 0 0
[ D ]= 2
( 1+ v ) (1−2 v ) 1
0 0 0 0 ( 1−2 v ) 0
2
1
0 0 0 0 0 ( 1−2 v )
2

For a 3D solid, this matrix is assembled in this form. For area and line elements it is reduced
according to the stress and strain components that are needed. E and v must be specified by the
user.

It is also possible to model anisotropic material and the user must specify coefficients for each
direction. The matrix is always symmetric, and there are up to 21 coefficients for 6 stress and strains
and 6 coefficients for plane stress or strain. Other properties that are nice are density and thermal
expansion.

Geometric Properties
These are needed for elements that aren’t 3D solids. The tables above show what is needed for
different elements.

Applied Loads
The global matrix equation related nodal forces to nodal displacements, therefore all forms of
loading need to be transferred to nodal forces (and/or moments). Modern programs simplify this
process by converting different loads to nodal forces automatically.

Nodal forces and moments are normally specified per unit thickness when applied to plain strain
elements and per radian of circumference for axisymmetric elements. Forces and moments can only
be applied in the degrees of freedom that exist at the node – it is not valid to apply moments at a
node that only has translational degrees of freedom.

Pressures can be applied to a face of a 3D solid or edge of an axisymmetric element or the surface of
a plate or shell.

Distributed loads can be applied to beam elements and to the edges of plane stress, plane strain and
shell elements. These are similar to pressure loads, however they are applied to a line and not an
area therefore their units are load per unit length.

Body forces such as gravity or centrifugal force are simple to specify. The user gives the density and
the value of acceleration and the program integrates the product of mass and acceleration over the
element and calculated equivalent nodal forces.

To apply thermal loads the user specifies the change of temperature at all nodes and the coefficient
of thermal expansion for the material(s). The conversion is then performed using an ‘Initial Strains’
method – strains are calculated and loads that would cause that strain are calculated, then when

Ian FEA 13
calculating element quantities after solution, the thermal strains are subtracted from the calculated
strains leaving just the elastic strains.

Supports and Prescribed Displacements

Supports are defined to represent fixed boundaries of a structure. At these boundaries, the relevant
degrees of freedoms (translation or rotation) are fixed. Supports are also used to stop the model
moving as a rigid body.

Supports are also used to introduce symmetric or anti-symmetric boundaries so that the domain can
be reduced and the solution time decreased. This should be done if possible. If a structure has
symmetry in both structural form and loading, symmetry can be applied. If it is structurally
symmetric but the loading has opposite symmetry then anti-symmetry can be applied.

e.g. cantilever with axial load (in plane strain or stress). The structure and load are symmetric about
the horizontal line at mid-depth, therefore symmetry conditions can be chosen. There are two DoFs
to consider – X and Y displacements. Under axial loading the nodes along the line of symmetry move
in the X direction but not in the Y direction, therefore Y displacements (v) can be fixed at all nodes on
this line. The structural support is also specified by fixing nodes on the wall end of the cantilever in
both X and Y directions.

If a similar cantilever but with a shear load is modelled, it will be anti-symmetric. In this case the
nodes along the mid-line will move in the Y direction (up and down due to shear) but not in the X
direction.

Sometimes by applying symmetry, the new boundary is ‘skewed’ so the nodes are now in a new
coordinate system (e.g. cannot fix Y displacements if the line is at an angle to the Y direction). Once
skewed, a node’s DoFs are permanently changed to the chosen axis system affecting all supports and
loads applied to it. Nodes can only be skewed once.

To skew something:

[ u' ]=[ t ][ u ]

t= cosθ sinθ
[
−sinθ cosθ ]
[ f ' ] =[ t ] [ f ]
f =[ k ][ u ]

[ f ' ]=[ t ] [ k ][ t ]T [ u ' ]

[ k ' ]= [ t ] [ k ][ t ]T
This transformation is made to all stiffness coefficients relating to the node. It can be done either
when the stiffness of each adjoining element is calculated or directly on the global stiffness matrix.

Ian FEA 14
Global Stiffness Matrix and the Solution
The solution to the global stiffness equation [ F ]= [ K ][ δ ] notionally requires the inversion of the
global stiffness matrix. Because [K] can be very large (thousands of rows and columns), this is difficult
computationally. In reality, full inversion is not done, however the process is still very
computationally intense. Matrices are solved using Gaussian Elimination and the number of
operations performed on and NxN matrix is to the order of N 3.

This number can be substantially reduced for [K] matrices encountered in FE models. The matrix is
always symmetric and is only sparsely populated and for a good model will be banded (non-zero
coefficients are banded on the leading diagonal). The semi-bandwidth, B, defines the width of half
the band in the upper triangle or the maximum number of columns from the leading diagonal to the
extreme non-zero coefficient. Elements outside of the band can be omitted from the calculations,
reducing the number of operations from order N 3 to NB2 and reduces array storage requirement.

The semi-bandwidth can be found from the maximum difference of freedon number on all the
elements. For elements with the same number of DoFs at each node, the difference of freedom
number on an element is the difference between max and min node number multiplied by DoFs at
each node.

Belement =( max node number−min node number+1 ) × ( number of DoFsat each node )

The best policy is by numbering across the smallest divisions first.

Direct solvers can be used to find displacements for more than one load case, using the same
stiffness matrix, for an increase of about 10% computational time for each additional load case.

Other Problems
Dynamics – modelling structural dynamics problems requires additional matrices representing mass
and damping of elements. In general they are banded matrices of the same size as [K]. Alternatively
lumped mass models can be used and eigenvalues used to find natural frequencies of the different
degrees of freedom of the model.

Plasticity – materials loaded beyond yield stress show reduced stiffness, therefore the model is non-
linear. These problems are modelled by incremental analysis where load is applied gradually and
then at each load the current state of stiffness calculated…a certain number of iterations is then
performed until convergence to the non-linear equations.

Large deflections – when structures deform lots their stiffness can change. In linear analysis this is
ignored as original geometry is used to define stiffness. Again, gradual loads can be applied and the
position of nodes ‘updated’ each time.

Contact problems – Making or breaking contacts affects stiffness, therefore making a non-linear
problem. Again incremental analysis is used. Special elements are normally used to define the
contact and can include stick-slip friction conditions.

Fracture mechanics – linear elastic and yielding fracture mechanics problems can be solved. Special
elements can represent crack tips and energy-based fracture criteria found by studying the field near
the crack.

Ian FEA 15
Guide to Good Modelling
1. Identify the spatial domain of the solution – the first important decision is to decide on the
region of the structure to be modelled and the geometry by which it is to be represented. All
structures are 3D in reality, but it might be possible to model them in 2D by making simplifying
assumptions such as plane stress, plane strain or axisymmetric. A member of a space frame may
be modelled as a single beam element or in detail using many shell elements depending on the
purpose of the model.

The shape of the structure and how it is loaded and supported is important. Need to keep things
as simple as possible depending on what you want from the model – while analysing a space
frame, if only the load path is needed then use beam elements but if the loads between a flange
and the web of the beam is needed then use more complicated shell elements. Also, you could
use two models – one for forces and moments in members and another using these forces and
moments applied to a detailed shell model of one beam.

2. Select the elements – need to choose an element type once special domain (plane stress/strain
or axisymmetric etc.) has been chosen. Quadratics are better than linear and quadrilaterals
better than triangles. Also need elements that are capable of matching the structure – if the
boundary is curved it is necessary to use quadratic elements to represent the shape.

Different elements can be mixed, but at a node only the degrees of freedom that are in common
are connected. If a 2D beam element (X,Y RZ) is attached to a node of a plane stress element
(X,Y) then only the X and Y degrees of freedom will be connected and the beam element would
be pin-jointed to the plane stress element. Curved beam elements can share the same nodes as
shell elements to model stiffeners on shell structures.

3. Create the mesh – use a fine mesh for rapidly changing stress or complicated stress variation.
Each element can only represent a simple stress field (constant or linearly varying), therefore if
variation is complicated then you need lots of them. If you use low order elements (linear shape
functions) then you need a much finer mesh. You need to use a fine mesh if you want detailed
results. Sudden changes in mesh refinement are bad – using big elements next to small ones can
lead to ill-conditioning of the global stiffness matrix due to large differences in adjacent stiffness
coefficients. A good rule is not more than twice as long as its neighbour. Avoid high aspect ratios
in rectangles – limit it to about 3. Avoid bad angular distortion – keep angles in quadrilaterals
close to 90° and not more than 30° from parent shape, if you need funny shapes then use
triangles (which should be as close to 60° as possible). Keep midside nodes near the midside –
with 10% of the length of the side to the mid-point. Limit element curvature (again 30° max).
Element shape is most important in areas where you want results – bad shapes introduce errors
but they tend to be localised.

4. Define material and geometric properties – these are straightforward but use consistent units

Ian FEA 16