MANE 4240 & CIVL 4240 Introduction to Finite Elements

Prof. Suvranu De

Reading assignment: Logan 10.2 + Lecture notes Summary: Computation of shape functions for 4-noded quad Special case: rectangular element Properties of shape functions Computation of strain-displacement matrix Example problem Hint at how to generate shape functions of higher order (Lagrange) elements

Four-noded rectangular element

Finite element formulation for 2D: Step 1: Divide the body into finite elements connected to each other through special points (nodes) py v3 3 px 4 3 u3 u 1 v4 2 v v2 Element e v 1 1 4 u 2 u u4 ST u2 v1 2 v 2 y d= x y u 3 Su u1 v 3 1 v x x u 4 u v 4

Summary: For each element Displacement approximation in terms of shape functions

u=Nd
Strain approximation in terms of strain-displacement matrix
=Bd

Stress approximation = DB d Element stiffness matrix

k = e B D B dV
T V

Element nodal load vector

f = e N X dV + e N T S dS
T T V ST f
b

Constant Strain Triangle (CST) : Simplest 2D finite element v1

1 (x1,y1)

Formula for the shape functions are v1 v3 1 u1

(x1,y1)

u1 v u u2
x

v3
(x3,y3) 3

v2

(x3,y3)

(x,y) 2 (x2,y2) x

u u2

u3

v2

u3

where

a1 + b1x + c1 y 2A a + b x + c2 y N2 = 2 2 2A a + b x + c3 y N3 = 3 3 2A N1 =

(x,y) 2 (x2,y2)

u (x,y) N1(x,y)u1 + N2(x,y)u2 + N3(x,y)u3 v (x,y) N1(x,y)v1 + N2(x,y)v2 + N3(x,y) v3

3 nodes per element 2 dofs per node (each node can move in x- and y- directions) Hence 6 dofs per element

1 x 1 1 A = area of triangle = det 1 x 2 2 1 x 3 a1 = x2 y3 x3 y 2 a2 = x3 y1 x1 y3 a3 = x1 y 2 x2 y1 b1 = y 2 y3 b2 = y3 y1 b3 = y1 y 2

y1 y2 y3

c1 = x3 x2 c2 = x1 x3 c3 = x2 x1

Approximation of displacements

u=Nd
u (x, y) N1 u= = v (x, y) 0 0 N1 N2 0 0 N2 N3 0

Approximation of the strains

x = y xy u x v = B d y u v y + x

u1 v 1 0 u 2 N 3 v 2 u 3 v3

Element stiffness matrix

k = e B D B dV
T V

Since B is constant

k = B D B e dV = B D B At
T T V

t=thickness of the element A=surface area of the element

N1(x,y) N2(x,y) 0 x x N1(x,y) 0 B= 0 y N (x,y) N (x,y) N (x,y) 1 2 1 x y y

b1 0 b2 0 b3 0 N2(x,y) N3(x,y) 1 0 = 0 c1 0 c2 0 c3 y y 2A c1 b1 c2 b2 c3 b3 N2(x,y) N3(x,y) N3(x,y) x y x 0 N3(x,y) x 0

Element nodal load vector

f = e N X dV + e N T S dS
T T V ST f
b

Class exercise
For the CST shown below, compute the vector of nodal loads due to surface traction

Class exercise
1 y

f S = t

l13e

T along 2 3

T S dS

= e N T S dS
T ST

fS2y
2 1 y

fS3y fS2x
3 py=-1

0 TS = 1

fS3x
x

The only nonzero nodal loads are

fS2y
2 (0,0)

fS3y fS2x
3 py=-1 (1,0)

f S = t

l13e

T along 2 3

T S dS

f S3 y = t

f S2 y = t

x =0 1

N 2 along 2 3 p y dx N 3 along 23 p y dx

x =0

fS3x
x

a2 + b2 x (x3 y1 x1 y3 ) + ( y3 y1 )x a + b x + c2 y N 2 along 23 = 2 2 = 2A = 2A 2A y =0 y1 y1 x y1 (1 x) y (1 x) = = 1 1 x1 y1 1 x1 y1 y1 ( x3 x2 ) det 1 x 2 y2 det 1 x 2 0 1 x 3 0 1 x 3 y3 (can you derive this simpler?) = 1 x =

4-noded rectangular element with edges parallel to the coordinate axes:

f S2 y = t
1 x=0

N 2 along 23 p y dx

(x4,y4)

4 v
(x,y)

3 (x3,y3)
u (x,y) Ni (x,y)ui
4

= t (1 x)(1) dx
x =0

t 2
y

2b

v (x, y) N i (x, y)v i

i =1

i =1

Now compute

f S3 y = t

(x1,y1)

2a
x

2 (x2,y2)

x =0

N 3 along 23 p y dx

4 nodes per element 2 dofs per node (each node can move in x- and y- directions) 8 dofs per element

Generation of N1:
y

At node 1
l1(y)

l1 ( x ) =

x x2 x1 x 2

Using similar arguments, choose

1 (x x 2 )( y y 4 ) 4 ab 1 (x x1 )( y y 3 ) N2 = 4 ab 1 (x x 4 )( y y 2 ) N3 = 4 ab 1 (x x 3 )( y y1 ) N4 = 4 ab N1 =

has the property 4 3 2b

N1

l1 ( x1 ) = 1 l1 ( x 2 ) = 0

Similarly 1
x

l1 ( y ) =

y y4 y1 y 4

1 2a
l1(x)

has the property

l1 ( y 1 ) = 1 l1 ( y 4 ) = 0

Hence choose the shape function at node 1 as

x x2 N 1 = l1 ( x ) l1 ( y ) = x x 1 2 y y 4 y y 1 4 1 = 4 ab ( x x 2 )( y y 4 )

Properties of the shape functions: 1. The shape functions N1, N2 , N3 and N4 are bilinear functions of x and y 2. Kronecker delta property

3. Along lines parallel to the x- or y-axes, the shape functions are linear. But along any other line they are nonlinear. 4. An element shape function related to a specific nodal point is zero along element boundaries not containing the nodal point. 5. The displacement field is continuous across elements 6. The strains and stresses are not constant within an element nor are they continuous across element boundaries.

1 at node ' i ' N i ( x, y ) = 0 at other nodes

3. Completeness
4

4 4

i =1

Ni =1 N i xi = x N i yi = y

i =1

i =1

The strain-displacement relationship

x = y xy N1(x, y) 0 x N1(x, y) = 0 y N (x, y) N (x, y) 1 1 x y N 2 (x, y) x 0 N 2 (x, y) y u1 v 1 N3(x, y) N 4 (x, y) 0 0 u 2 x x N3(x, y) N 4 (x, y)v 2 0 0 y y u 3 N3(x, y) N3(x, y) N 4 (x, y) N 4 (x, y)v 3 y x y x u 4 v 4

Computation of the terms in the stiffness matrix of 2D elements (recap)

v4 v3 4 u4

The B-matrix (strain-displacement) corresponding to this element is

u3 u1 v1
0 N1 (x,y) y N1 (x,y) x

v y v1 (x,y) 1 u1 x u v2

u2
N 2 (x,y) x 0 N 2 (x,y) y

v2
0 N2 (x,y) y N2 (x,y) x

u3
N 3 (x,y) x 0 N 3 (x,y) y

v3
0 N 3 (x,y) y N 3 (x,y) x

u4
N4 (x,y) x 0 N4 (x,y) y

v4
0 N4 (x,y) y N4 (x,y) x

0 N 2 (x, y) y N 2 (x, y) x
B

u2

N1 (x,y) x 0 N1 (x,y) y

We will denote the columns of the B-matrix as

y3 y y y2 y1 y 0 0 0 0 y y4 1 B= x x2 x1 x x x4 x3 x 0 0 0 0 4ab x x2 y y4 x1 x y3 y x x4 y y2 x3 x y1 y

B u1

Notice that the strains (and hence the stresses) are NOT constant within an element

N 1 (x,y) 0 x N 1 (x,y) ; and so on... 0 = ; B v1 = y N (x,y) 1 N 1 (x,y) y x

The stiffness matrix corresponding to this element is

k = e B D B dV
T V
u1 v1 u2

which has the following form

v2 u3 v3 u4 v4

k11 k 21 k 31 k k = 41 k 51 k 61 k 71 k 81

k12 k 22 k 32 k 42 k 52 k 62 k 72 k 82

k13 k 23 k 33 k 43 k 53 k 63 k 73 k 83

k14 k 24 k 34 k 44 k 54 k 64 k 74 k 84

k15 k 25 k 35 k 45 k 55 k 65 k 75 k 85

k16 k 26 k 36 k 46 k 56 k 66 k 76 k 86

k17 k 27 k 37 k 47 k 57 k 67 k 77 k 87

k18 k 28 k 38 k 48 k 58 k 68 k 78 k 88

u1 v1 u2 v2 u3 v3 u4 v4

Notice that these formulae are quite general (apply to all kinds of finite elements, CST, quadrilateral, etc) since we have not used any specific shape functions for their derivation.

The individual entries of the stiffness matrix may be computed as follows

k11 = e Bu1 D Bu1 dV; k12 = e Bu1 D Bv1 dV; k13 = e Bu1 D Bu2 dV,...
T T T V V V

k 21 = e Bv1 D Bu1 dV; k21 = e Bv1 D Bv1 dV;.....

T T V V

Example
y

1000 lb 300 psi 3 2 in Thickness (t) = 0.5 in E= 30106 psi =0.25

Realize that this is a plane stress problem and therefore we need to use

D=

1 0 3.2 0.8 0 E 1 0 = 0.8 3.2 0 10 7 psi 2 1 1 0 1.2 0 0 0 2

1 3 in

2
x

Write down the shape functions

1 (x x2 )( y y 4 ) = ( x 3)( y 2) 4 ab 6 1 N2 = (x x1 )( y y3 ) = x ( y 2) 4 ab 6 1 (x x4 )( y y 2 ) = xy N3 = 4 ab 6 1 (x x3 )( y y1 ) = ( x 3) y N4 = 4 ab 6 N1 =
x 0 3 3 0 y 0 0 2 2

(a) Compute the unknown nodal displacements. (b) Compute the stresses in the two elements. This is exactly the same problem that we solved in last class, except now we have to use a single 4-noded element

Compute only the relevant columns of the B matrix

We have 4 nodes with 2 dofs per node=8dofs. However, 5 of these are fixed. The nonzero displacements are
u2 u3 v3

Hence we need to solve

u2 u3 v3

N 2 ( 2 y ) x 6 B u2 = 0 = 0 N 2 x y 6 N 3 y x 6 = 0 = 0 N 3 x y 6

k11 k 21 k 31

k12 k 22 k 32

k13 u 2 0 k 23 u 3 = 0 k 33 v3 f 3 y

B u3

Need to compute only the relevant terms in the stiffness matrix

k11 = e Bu2 D Bu2 dV; k12 = e Bu2 D Bu3 dV; k13 = e Bu2 D Bv3 dV
T T T V V V

k 21 = e Bu3 D Bu2 dV; k22 = e Bu3 D Bu3 dV; k13 = e Bu3 D Bv3 dV
T T T V V V

B v3

k31 = e Bv3 D Bu2 dV; k 22 = e Bv3 D Bu3 dV; k13 = e Bv3 D Bv3 dV
T T T V V V

0 0 N 3 x = = y 6 N 3 y x 6

k11 = e Bu2 D Bu2 dV

T V

How do we compute f3y

= 0.5

x=0 y=0

2 y 8 7 5 2 (0.1067 10 0.533 10 )( 6 ) + 3.33 10 x dxdy

f 3 y = 1000 + f S 3 y

N 3 edge = N 3
4 3

y =2

= 0.656 107

f S3 y = t

x =0

N3

along edge 3 4 3

( 300 ) dx x dx 3
4

x xy = = 6 y =2 3

Similarly compute the other terms

= ( 0 .5)( 300 ) 3 = 150 2 = 225 lb

x =0

f 3 y = 1000 + f S 3 y = 1225 lb

How about a 9-noded rectangle?

Corner nodes 2 b 6 b 3 y 5 a a 9 7 1 8 4 x
x(a + x) y(b + y) x(a x) y(b + y) N1 = 2 2 N 2 = 2a 2 2b 2 2a 2b x(a x) y(b y ) x(a + x) y(b y) N 3 = N4 = 2 2a 2 2b 2 2b 2 2a

Midside nodes
2 2 a 2 x 2 y(b + y) x (a x ) b y N5 = 2 2 N 6 = 2a 2 b 2 a 2b 2 2 a 2 x 2 y(b y) x( a + x) b y N7 = N8 = 2 2 2 2b 2 2a b a

Center node
a 2 x2 b2 y 2 N9 = 2 2 a b

Question: Can you generate the shape functions of a 16-noded rectangle? Note: These elements, whose shape functions are generated by multiplying the shape functions of 1D elements, are said to belong to the Lagrange family