Sunteți pe pagina 1din 13

FINITE ELEMENTS METHOD

Purpose : To see how accurate FEM is, to find the errors inherent, and to provide suggestions.

Verification: First a simple model with one mesh will be solved by hand and by the MATLAB code
provided to see if the results verify each other. Plane Stress conditions are considered. Only crucial
steps are shown for the handmade solution.

Node 1 and Node 4 will be the fixed nodes and a load of 1000N will be acted on node 3 acting in the
– y direction. Young Modulus will be taken 10000 N/mm2 and Poisson’s ratio ν will be 0.

Solution by Hand:
x and y values are in meters here:
Node1 = ( 0,0 )
Node2 = ( 2,0 )
Node3 = ( 2,1 )
Node4 = ( 0,1 )

Shape Functions
N1 ( r,s ) = 0.25 × ( 1 – r ) × ( 1 – s )
N2 ( r,s ) = 0.25 × ( 1 + r ) × ( 1 – s )
N3 ( r,s ) = 0.25 × ( 1 + r ) × ( 1 + s )
N4 ( r,s ) = 0.25 × ( 1 – r ) × ( 1 + s )

Jacobian Matrix = [ 1,0 ; 0,0.5 ]


Inverse of Jacobian Matrix = [ 1,0 ; 0,2]
Determinant of Jacobian Matrix = 0.5
With Plane Stress Conditions the D Matrix is = [ 10,0,0 ; 0,10,0 ; 0,0,5 ]

The element matrices will be integrated using 2 × 2 Gauss quadrature with the following coordinates
in the parent element and weights :
r1 = –1 / ( 30.5 ), r2 = 1 / ( 30.5 )
s1 = –1 / ( 30.5 ), s2 = 1 / ( 30.5 )

The stiffness matrix given by:


K=K(1)=∑∑Wi ×Wj ×|Je( ri,si )| × BeT ( ri,si ) × De × Be ( ri,si )
Repeating the process for four times for each Gauss Points we get the element stiffness matrix,
which is the Global stiffness matrix for this structure which is given by:

5.0008 1.2500 -0.0008 -1.2500 -2.4992 -1.2500 -2.5008 1.2500


1.2500 7.5012 1.2500 2.4988 -1.2500 -3.7488 -1.2500 -6.2512
-0.0008 1.2500 5.0008 -1.2500 -2.5008 -1.2500 -2.4992 1.2500
-1.2500 2.4988 -1.2500 7.5012 1.2500 -6.2512 1.2500 -3.7488
-2.4992 -1.2500 -2.5008 1.2500 5.0008 1.2500 -0.0008 -1.2500
-1.2500 -3.7488 -1.2500 -6.2512 1.2500 7.5012 1.2500 2.4988
-2.5008 -1.2500 -2.4992 1.2500 -0.0008 1.2500 5.0008 -1.2500
1.2500 -6.2512 1.2500 -3.7488 -1.2500 2.4988 -1.2500 7.5012

The force matrix is = [ 0 ; 0 ; 0 ; 0 ; 0 ; -1 ; 0 ; 0 ]


The boundary conditions matrix is = [ 0 ; 0 ; ux ; uy ; ux ; uy ; 0 ; 0 ]
From 3rd row to 6th row and 3rd column to 6th column of the stiffness matrix is required.

Solving K × U = D by Matlab gives the displacements:


ans =

-0.3998
-1.1631
0.3998
-1.2359

corresponding to 2nd Node x displacement, 2nd node y displacement, 3rd node x displacement, 3rd row
y displacement in order.

The final displacement vector is = [ 0 ; 0 ; -0.3998 ; -1.1631 ; 0.3998 ; -1.2359 ; 0 ; 0 ]

Validation:
The results make sense since the displacement at node 2 and 3 in x directions are equal but in the
opposite direction, since our shape is linear, and displacement at node 3 at y direction is slightly
larger than the displacement in y direction at node 4 as expected.
Solving the same system with the MATLAB Codes provided for the coursework

Modified Inputs :
L = 2000; nx = 1; % dimension and number of elements in x-dir
H = 1000; ny = 1; % dimension and number of elements in y-dir
W = 1; % dimension in third direction
E = 10000;
v = 0;
fglobal(nodnum*2) = -1000;

And the displacements given by the code are :


displacements =
0 0 0 0 -0.4000 -1.1636 0.4000 -1.2364
Since the numbering is different, the displacement vectors are not exactly same. But corresponding
displacements are same so this verifies that we are on the right track.

(The deformed shape above has a scale factor of 250.)

Here we have used a rectangular element with four nodes only. To use in our final conclusions about
finite elements method, the same system will be solved with the commercial software called
SAP2000, which uses high-order quadratic elements and the finite elements method.

The displacement at the right top corner will be used for comparison. From hereafter the code
min(min(displacements)) is added to the given Matlab Code to gather the top right displacement.

With the necessary modelling a displacement of -2.8212 millimetres is given at the top right corner.
(With the hand solution and the Matlab code given we had a displacement of -1.1636 millimetres
which is a difference of more than 100%)

By a little experiment, a mesh of 2 × 2 on the Matlab Code Provided we get a displacement of -2.4982
millimetres for the system for the node considered which is the top right one.

It is seen that with four – four noded quadratic elements, the displacement is closer to the one with
one mesh higher order element. It can be concluded from this little experiment that, the
disadvantage of four-noded element is that the stiffness is too high, so the displacement was too
low. The more meshes we got, the closer we got to the result we got with the higher order element.

The conclusions derived from this little experiment will be used later.
Cantilever Beam Solution

L = 6 meters, H = 0.4 meters, F = 10kN and E=200 × 103 MPa ν = 0. W=0.2 meters. (With the values
given G = Shear Modulus = 100 × 103 MPa)

Getting the displacement using the beam theory


P L E I G A
-10000 6000 200000 1067000000 100000 80000

(All the units are in N and millimetres.)


Assuming the cantilever to be 1D and beam theory gives the displacement of the node on the axes
of the element at far right side:
=P × L3 / 3 / E / I + 6/5 × P × L / A / G
=-3.384 millimetres.

Plane Stress and Plane Strain:

Plane stress solids are solids whose thickness in z direction is


very small compared with dimensions in the x and y directions.
External forces are applied only in the x – y plane and the
stresses in the z direction σzz, σxz and σyz are all zero. For plane
stress solids εzz is not zero.

Plane strain solids are solids with a very large thickness in the z
direction compared to x and y. When this assumption is made,
only a very small cross section is taken in account and solved.
With plane strain assumption, εzz is zero, but the stress along
the z direction is not.
Solving the Cantilever with Plane Stress Assumption

The given cantilever above in the figure in the previous page will be solved using the Matlab Code
given and SAP2000.
The table below gives the displacements at the top right corner in millimetres.

30 x 2 60 x 4 120 x 8 240 x 16
Displacement Displacement Displacement Displacement
SAP2000 -3.382 -3.384 -3.385 -3.386
MATLAB -3.007 -3.281 -3.357 -3.379

The meshes are chosen in this way, to keep the aspect ratio at 1:1 which gives the best results with
FEM.
The conclusions we can get from this is, with a 30 × 2 we get a close enough result, a result that we
have expected. But we also examine that the result is actually not converging, but increasing with
the more meshes we have.

To examine this situation and get a better understanding of the FEM, a different cantilever will be
solved and the results will be examined by using SAP2000 in detail. A cantilever with a same value of
Young’s Modulus, and a Poisson’s ratio ν = 0 will be used, but this time the length and the height of
the beam will be taken as 1000 millimetres and 500 millimetres. The thickness of the beam will be
50 millimetres and the loading at the tip will be 10 × 106 Newton in the –y direction.
The beam theory gives us a displacement of -36.8 millimetres under this loading condition.

In the picture the displacement at y direction counters on the area objects are shown, with the key
given at the right hand side of the picture. It can be clearly seen that, the displacements are on the
left side where the cantilever is fixed is zero, but there is something wrong going on at the top right
corner as we get smaller meshes.
Also in this picture it is clearly seen that there is also something terribly going wrong at the top right
corner, in the means of stresses in y – y direction.
From these two pictures, we can also see that the displacement counters are linear, where as the
stress counters are constant. To this will be returned in the conclusion.
We will examine the top middle node, top right node, middle right node in detail in the means of
displacements and stress in Y direction. The stress values are given in kN/mm and the displacements
are in mm.
Top Middle Node Top Right Node Middle Right Node
σy-y Displacement σy-y Displacement σy-y Displacement
Element 1 11.8355 -10.6714 -50.3999 -34.9637 -33.2614 -33.7037
Element 2 -4.3064 -11.9362 -120.1124 -38.1282 -52.6211 -35.6778
Element 3 0.3881 -12.2146 -246.5981 -40.0747 -60.1088 -36.1768
Element 4 0.1314 -12.3018 -494.1331 -41.6944 -62.5381 -36.2983

If this was a real situation we would be interested in the displacement of the tip, the shear failure,
and of course, we would like to know if our beam could support the moment on it, which is
maximum at the fixed end. So before making any conclusions, the stresses in x – x direction will be
taken into account too, at the fixed ends, which eventually will give us the moments.
Here is what the stresses look like in x – x direction, units being in kN/mm.
The detailed table in the means of stresses in the x – x direction and the displacements for the fixed
nodes is :
Top Left Node Middle Left Node Bottom Left Node
σ x-x σ x-x σ x-x Moment
Element 1 180.1991 -0.1991 -179.8001 7.50 × 106
Element 2 217.1891 0.4502 -217.3285 9.04 × 106
Element 3 240.2552 0.0562 -240.3025 10.04 × 106
Element 4 255.5249 0.04085 -256.6101 10.64 × 106

Now let us try to visualise these values we have. We will first start with the displacements:

Diplacements
44

42

40
Middle Right Node

38 Exact Solution
Top Right Node
36

34

32
2x2 4x4 8x8 16 x 16

A 64 × 64 mesh gives a displacement of 44 millimetres at the top right node and a 36.40 millimetres
in the middle right node which is not shown before.

Moment Values at the Fixed End


11
10.5
10
9.5
Moment Values
9
Exact Solution
8.5
8
7.5
7
2x2 4x4 8x8 16 x 16

A 64 × 64 mesh gives a moment value of 11.25 ×106 N.m where as we have an exact value of
10.00×106.
Conclusions and an Understanding of the FEM

From this second cantilever example we can see the problems in a much bigger detail. Now the
errors inherent in FEM will tried to be explained.
From the strength of materials, it is known that the first derivate of the displacements gives us the
relative displacements, ε and the derivate of the ε, gives us the stresses, σ. (We have examined the
results in our SAP2000 outputs, where as our displacement vectors were linear, but our stresses
were constant lines.)
In the finite elements method, we basically know the material properties, and where our forces are,
and where our structure cannot move, i.e. has restraints. These are called boundary conditions.
What we do not know is the displacements we have. Once we find the displacements, we can find
the stresses.
For finding the displacements, comes the “shape functions” to help. A structure, (explanation will be
kept in 2D for simplicity) just like the first example solved in this paper will be considered. The shape
function of a particular structure gives us the displacements (or any other function) in the area of
our structure, in respect to the nodes on it.
In the first example, we saw that using a higher-order element, i.e. a structure with a shape function
contributed by more nodes, gives us a much better solution. We got the same answer with a 4 mesh
solution, which we got with one element with a higher order.
Also, even if we had the perfect shape functions, it is still impossible to solve the differential
equations, that is why these equations are solved numerically. This is a great opportunity for non
100% accurate solutions too.
Solving the cantilever examples, it is seen that the displacements at the nodes we have taken into
account, (which are the important ones) tend to go to infinity. Also, the stresses in y-y direction at
these nodes, at the stress values we got at the fixed ends tend to go to infinity too.
So if we had a very powerful computer, and had 1000x1000 mashes, we would probably have an
incredible amount of a displacement at the top right corner, which would result in a shear failure,
and a very large moment value which should not be.
But how is this possible?
This is because the smaller meshes we have, the greater stresses we have in that particular mesh.
Let us just visualise the mesh element we have at the top right corner only. We have a point load at
the top right node, and (roughly speaking, simplifying the FEM procedure greatly) that load must
somehow be transported to the nodes on that particular element, and then to the nodes of the
meshes those nodes are on until they the force vectors find a node where they can get the reaction
so that they have nowhere else to go anymore. With this bearing in mind, it is now easier to
understand, that with the smaller mesh we have, with the load value on that particular node not
changing, the stress will always increase and increase, and the deflection will follow the behaviour of
stress.
One possible way to avoid this can be not using only one point load, which causes this problem in
the first place. Although this procedure is based on simplification mainly, we never have point loads
in real life either. So say, if we were to design a load standing on a cantilever beam, instead of using
a point load at the end of the beam, we might as well find the base length of the load, and spread
the load on the nodes, on the meshes we have.
Also as can be seen from the note below the last table, even though we had a 64 x 64 mesh and an
extreme displacement at the top right corner, the displacement on the axes, at the right point never
passed the exact solution. So instead of using that node, the node at the axes could be used.
Coursework Solution

Model Used:
Since we are free to choose dimensions, I will be solving a cantilever beam with properties of:
L = 3000 mm
H = 500 mm
W = 200 mm
I = 2.083 ×109m4
A = 100000 mm2
E = 200000 N/mm2
Beam Theory Gives The Values:
Displacement at the tip : -2.1996 mm
Moment Value at the fixed end : 3 × 107 N.cm

A Matlab Code has been written to get the Moment Values at the fixed end:
tugayx=L/nx
tugayy=H/ny
xdisplacement=displacements(ny*2+3)
strain=xdisplacement/tugayx
sigma=strain*E*W
moment=sigma*H/2*H/2*2/3

Also because with the beam theory we get the displacement at the centre, but not at the right below
corner a code has been written to get the displacement at the y direction at that node: (The node
that at H/2 at the very right corner) Coordinates = (300,25)

if mod(ny,2) == 0
tugaydisp=displacements(((ny+1)*(nx+1)*2-ny)+0)
else
tugaydisp=displacements(((ny+1)*(nx+1)*2-ny)+1)
end
First Try:
With a 8 x 4 Mesh we get the results:
Tip Displacement = -1.711 mm
Moment Value = -2.20 × 107 N.cm
With a 8 x 4 mesh we are too far from the exact result.
Three improved meshes are designed and results are as follows:
Mesh Exact Solution 8x4 12x2 24x4 48x8
Displacement 2.20 1.71 1.95 2.12 2.18
Moment 3.00 2.20 2.55 2.88 3.02
Exact Displacement / FEM 1.00 1.29 1.13 1.04 1.01
Exact Moment / FEM 1.00 1.36 1.18 1.04 0.99
The displacement values are in millimetres and the Moment values are × 107 N.cm.
The relevant graphs are as follows:

Displacements
(mm)
2.30

2.20

2.10

2.00

1.90

1.80

1.70

1.60
8x4 12x2 24x4 48x8
Displacement Exact Solution

Moment
(107 N.cm)
3.50

3.00

2.50

2.00

1.50
8x4 12x2 24x4 48x8
Moment Exact Solution
As can be seen, a 48x8 mesh where the moment value starts exceeding the exact value we have.
This is where stresses start to increase more than they should. It is just 1% over so we are on the
safe side, so the perfect meshing for this situation can be chosen as 48x8.

The deformed shapes of 48x8 mesh. Deformed shape has a scale factor of 50.

3D Numerical Analysis
Since we have deformations on the y direction, and no stresses in the z direction, when doing the 3D
analysis, the most important meshes will be the ones diving the length of the beam. Meshing will be
chosen on this fact.
The diving at parallel to x direction does not help as much as the meshing parallel to y direction. Also
meshing along the z axis is not as important for us too. Using 100 elements best results can be
obtained by: ( A mesh of 25 x 2 x 2 )

We get a displacement of 2.1375 millimetres, which is very close to we got with the 2D Numerical
Analysis with the 24x4 meshes.
Effect of the Poisson’s Ratio

A mesh of 25 x 2 x 2 for 3D and a 25 x 2 for 2D (both plane stress and plane stress) will be used for
comparison.

In order to obtain the moment values we need to find the stresses in x direction in 3D analysis is to
which is given by =
σx-x = [ E / ( 1 + ν ) / ( 1 – ν ) ]× [ ( 1 – ν ) × εx-x + ν × εy-y + ν × εz-z )
With the meshing we have the values :
dx = 12
dy = 25
dz = 10
An Excel Spreadsheet has been used for the formulation and the tables are as follows:
ν dispxx1 dispyy1 dispzz1 dispxx2 dispyy2 dispzz2 dizpxx3 dispyy3 dispzz3
0.10 -0.0021 -0.0007 -0.0002 -0.002 -0.0007 0 -0.0021 -0.0007 0.0002
0.15 -0.002 -0.0008 -0.0003 -0.002 -0.0008 0 -0.002 -0.0008 0.0003
0.20 -0.002 -0.0008 -0.0004 -0.0019 -0.0009 0 -0.002 -0.0008 0.0004
0.25 -0.002 -0.0008 -0.0004 -0.0019 -0.0009 0 -0.002 -0.0008 0.0004
0.30 -0.0018 -0.0008 -0.0006 -0.0017 -0.0009 0 -0.0018 -0.0008 0.0006

ν εx-x εy-y εz-z


0.10 -0.000175 -0.000028 -0.00002
0.15 -0.000166667 -0.000032 -0.00003
0.20 -0.000166667 -0.000032 -0.00004
0.25 -0.000166667 -0.000032 -0.00004
0.30 -0.00015 -0.000032 -0.00006

ν σx-x
0.10 -3278.79
0.15 -3088.83
0.20 -3077.78
0.25 -3050.67
0.30 -2914.29

ν Moment ( x109 N.cm) Tip Displacement (cm)


0.10 -2.730 -0.2129
0.15 -2.574 -0.2123
0.20 -2.565 -0.2111
0.25 -2.542 -0.2110
0.30 -2.429 -0.2065

The nodes 1,2,3 mentioned here are the nodes on the below surface next to where the fixed ends
are. Node1=(12,0,0) Node2=(12,0,10) Node3=(12,0,20)
A mesh of 25 x 2 will be used for 2D analysis. Both plane stress and plane strain assumptions will be
made. Moment values shown here are in x109 N.cm. The code was already written to get the
moments and the tip displacement before.
Plane Stress Plane Strain
v Tip Disp Moment Tip Disp Moment
0.10 -0.2130 -2.8473 -0.2108 -2.8115
0.15 -0.2126 -2.8284 -0.2075 -2.7514
0.20 -0.2118 -2.8000 -0.2023 -2.6524
0.25 -0.2107 -2.7629 -0.1950 -2.5119
0.30 -0.2092 -2.7161 -0.1850 -2.3199

A General Comparison:

Moment Values (x109 N.cm)


Poisson 0.10 0.15 0.20 0.25 0.30
3D -2.730 -2.570 -2.560 -2.540 -2.420
Plane Stress -2.840 -2.820 -2.800 -2.763 -2.716
Plane Strain -2.812 -2.751 -2.652 -2.512 -2.320

Tip Displacements (cm)


Poisson 0.100 0.150 0.200 0.250 0.300
3D -0.2129 -0.2123 -0.2111 -0.2110 -0.2065
Plane Stress -0.2130 -0.2126 -0.2118 -0.2107 -0.2092
Plane Strain -0.2108 -0.2075 -0.2023 -0.1950 -0.1850

First thing we notice is that the displacement in the z direction on the axis of the beam is zero. That
is an expected conclusion , since the load is at the centre. Also that the displacements relative to x-y
plane are symmetrical. The Poisson’s ratio’s effect can clearly be seen in the 3D analysis, with an
increasing displacement of the nodes at the side in the z direction.
The second observation that can be made easily is that, with an increasing Poisson’s ratio the
displacements and the moment values are decreasing.
This is due to the energy stored in the structure. The energy stored in the structure due to strain has
increased with a greater Poisson’s ratio.

S-ar putea să vă placă și