January 2012

Finite element analysis example.

We set up a triangle with 6 internal nodes,as shown.

Then we have three dierent triangular elements. One is the isosceles right
triangle treated by Jackson (Fig 2.16)

We also have a right triangle with sides 1,2 and

form this triangle, and also nodes 5,12,13.)

5; and area 1. ( Nodes 6,7,8

For this triangle we have

a1

b1

c1

a2

b2

c2

a3

b3

c3

1
1
(x2 y3 x3 y2 ) = (0 2) = 1
2A
2
1
1
1
(y2 y3 ) = (1 0) =
2A
2
2
1
1
(x2 x3 ) =
(0 2) = 1
2A
2
1
1
(x2 y1 x1 y3 ) = (0 0) = 0
2A
2
1
1
(y3 y1 ) = (0 0) = 0
2A
2
1
1
(x3 x1 ) =
(2 0) = 1
2A
2
1
1
(x1 y2 x2 y1 ) = (0 0) = 0
2A
2
1
1
1
(y1 y2 ) = (0 1) =
2A
2
2
1
1
(x1 x2 ) =
(0 0) = 0
2A
2
2

P
Note: Jackson claims that ai = 1; but this result depends on the orientation of
the triangle, which explains why I have 1. But the kij all depend quadratically
on the bs and cs, and so are independent of their signs.
And then
k11 = A(b21 + c21 ) =
1
2

k12 = A (b1 b2 + c1 c2 ) =
k13 = A (b1 b3 + c1 c3 ) =

1
2

1
5
+1=
4
4
(0) + 1
1
2

( 1) =

+1

0=

1
1
4

k22 = A(b22 + c22 ) = 0 + 1 = 1

k23 = A (b2 b3 + c2 c3 ) = 0 + 0 = 0
k33 = A(b23 + c23 ) =

1
1
+0=
4
4

Thus we have:

The third triangle is formed by nodes 5,13,14 and also 6,7,16 and looks like
this:

This triangle also has area 1 and its shape functions have coe cients:
a1

b1

c1

1
1
(x2 y3 x3 y2 ) = (0 2 1) =
2A
2
1
1
1
(y2 y3 ) = (1 2) =
2A
2
2
1
1
(x2 x3 ) =
(0 2) = 1
2A
2

1
1
(x2 y1 x1 y3 ) = (0 0) = 0
2A
2
1
1
(y3 y1 ) = (2 0) = 1
2A
2
1
1
(x3 x1 ) =
(2 0) = 1
2A
2

a2

b2

c2

a3

b3

c3

1
1
(x1 y2 x2 y1 ) = (0 0) = 0
2A
2
1
1
1
(y1 y2 ) = (0 1) =
2A
2
2
1
1
(x1 x2 ) =
(0 0) = 0
2A
2

And then
k11 = A(b21 + c21 ) =
1
2
1
2

k12 = A (b1 b2 + c1 c2 ) =
k13 = A (b1 b3 + c1 c3 ) =

k23

1
5
+1=
4
4
(1) + 1
1
2

+1

k22 = A(b22 + c22 ) = 1 + 1 = 2

1
= A (b2 b3 + c2 c3 ) = 1
+0=
2
1
1
k33 = A(b23 + c23 ) = + 0 =
4
4

3
2
1
0=
4

( 1) =

1
2

Now we have to compute the matrix coe cients for the six internal nodes. Node
1 is part of 8 identical triangles, so
1
2

K11 = 8

=4

Similarly we have
K22 = 4 (1) = 4 = K33
K55 = K66 = 2

1+2+1=5

Line 12 is a side of two triangles, so

1
2

K12 = 2

1 = K13 = K14

Similarly
1
2

K45 = K36 = 2
and these0are the only
4
1
B 1 4
B
B 1 0
K=B
B 1 0
B
@ 0
0
0
0
0
inverse: K

OR
0
B
B
B
B
B
B
@

0:310 2
7:755 10
8:163 10
8:163 10
1:633 10
1:633 10

B
B
B
=B
B
B
@
2
2
2
2
2

non-zero
1
1
0
0
4
0
0
4
0
1
1 0
76
245
19
245
4
49
4
49
4
245
4
245

19
245
66
245
1
49
1
49
1
245
1
245

7:755 10
0:269 39
2:041 10
2:041 10
4:082 10
4:082 10

elements. 1
Thus
0
0
0
0 C
C
0
1 C
C,
1 0 C
C
5
0 A
0
5
4
49
1
49
265
931
20
931
4
931
53
931

2
2
3
3

4
49
1
49
20
931
265
931
53
931
4
931

4
245
1
245
4
931
53
931
984
4655
4
4655

8:163 10
2:041 10
0:284 6
2:148 10
4:296 10
5:693 10

4
245
1
245
53
931
4
931
4
4655
984
4655
2
2

2
3
2

1
C
C
C
C
C
C
A

8:163 10
2:041 10
2:148 10
0:284 64
5:693 10
4:296 10

2
2
2

2
3

1:633 10
4:082 10
4:296 10
5:693 10
0:211 39
8:593 10

Now we set up boundary conditions with the sloping side grounded and the
other two sides at unit potential. Then the source vector includes the nodes on
the boundaries:
G1 =

16
X

K1j

= K1;9

+ K1;11

11

j=7

+ K1;14

14

+ K1;15

15

+ K1;16

16

2
3
3
2

1:633 10
4:082 10
5:693 10
4:296 10
8:593 10
0:211 39

2
3
2
3
4

1
C
C
C
C
C
C
A

The only nonzero K is K1;15 but

G2

G3

=
=
=
G5

15

K2;9

= 0; so G1 = 0:

+ K2;10 10 + K2;11
1
) 1= 3
2(
2
9

K3;9 9 + K38 8 + K3;16

1
1
) 1+
+0
2(
2
2
1
3
1
=
= G4
2
2
=
=
=

K5;12
1
2
2=

12

+ K5;13

11

16

1+

1
+0
2

+ K5;14 14
1
1+ 0
1+0
2

13

G6

Then
we multiply 1 0
0 to76solve19the problem:,
4
4
4
4

1
1 0
0:542 86
0
C B 3 C B 0:885 71 C
B
C
C B
CB
B
C B 1:5 C B 0:642 86 C
B
C
C=B
CB
B
C B 1:5 C B 0:642 86 C
B
C
C B
CB
B
A @ 2 A @ 0:528 57 A
@
0:528 57
2
The major uncertainty has to do with the value of the potential in the corners
(nodes 13 and 7) where the insulating strip runs. Should we use zero or 1, or
perhaps 1/2? If we use zero, we get dierent values for G5 = G6 :
245
19
245
4
49
4
49
4
245
4
245

245
66
245
1
49
1
49
1
245
1
245

49
1
49
265
931
20
931
4
931
53
931

G5

49
1
49
20
931
265
931
53
931
4
931

=
=
=

B
B
B
and thenB
B
B
@

76
245
19
245
4
49
4
49
4
245
4
245

19
245
66
245
1
49
1
49
1
245
1
245

or choosing 1/2
G5

4
49
1
49
265
931
20
931
4
931
53
931

245
1
245
4
931
53
931
984
4655
4
4655

245
1
245
53
931
4
931
4
4655
984
4655

K5;12 12 + K5;13 13 + K5;14 14

1
1
1
0+ 0
1+0
2
2
1
= G6
2
10
1 0
4
4
4
0
49
245
245
1
1
1
CB
C B
49
245
245 C B 3 C
B
20
4
53 C B
C B
931
931
931 C B 1:5 C = B
265
53
4
CB
C B
931
931
931 C B 1:5 C
B
53
984
4
A @ :5 A @
931
4655
4655
4
4
984
:5
931
4655
4655

= K5;12 12 + K5;13 13 + K5;14 14

1
1
1
1
+ 0
1+0
=
2
2
2
5
=
= G6
4
6

0:493 88
0:873 47
0:551 02
0:551 02
0:210 2
0:210 2

1
C
C
C
C
C
C
A

1
1 0
0:518 37
0
B
C B 3 C B 0:879 59 C
C
C B
B
CB
B
C B 1:5 C B 0:596 94 C
C
C=B
B
CB
B
C B 1:5 C B 0:596 94 C
C
C B
B
CB
@
A @ 1:25 A @ 0:369 39 A
0:369 39
1:25
The last choice gives results that come closest to the values from the relaxation
used as a0check. 1
1
0
0 method I 1
0:654 37
0:515
0:518 37
B 0:524 57 C
B 0:875 C
B 0:879 59 C
C
B
C
B
C
B
B 0:596 94 C
C
B
B 0:608 C
C or a dierence ofB 1: 819 1 C %
C versusB
B
B 1: 819 1 C
B 0:608 C
B 0:596 94 C
B
C
B
C
B
C
@ 8: 111 9 A
@ 0:402 A
@ 0:369 39 A
8: 111 9
0:402
0:369 39
0

76
245
19
245
4
49
4
49
4
245
4
245

19
245
66
245
1
49
1
49
1
245
1
245

4
49
1
49
265
931
20
931
4
931
53
931

4
49
1
49
20
931
265
931
53
931
4
931

4
245
1
245
4
931
53
931
984
4655
4
4655

4
245
1
245
53
931
4
931
4
4655
984
4655

10