EN0175 09 / 07 / 06

1.2 Introduction to FEM

Finite element method is a very powerful tool for numerical analysis in solid and structural
mechanics as well as in many other engineering disciplines. Here we present an introduction to the
basic concepts of FEM in the simplest context of a one dimensional elasticity of a bar.

In lecture 1, we have derived the governing equation for 1D elasticity of a bar as,

∂σ ∂ 2u
x + f =ρ 2
∂x ∂t
L
σ = Eε

∂u
ε=
P ∂x
We can consider two simple problems, one with traction BC and one with displacement BC.

P1: P2:

u (0 ) = 0 u (0 ) = 0

ρg ρg

σ (L ) = P u (L ) = 0

⎧ E u '' + ρ g = 0 ⎧ E u '' + ρ g = 0
⎨ ⎨
⎩u (0) = 0, E u (L ) = P ⎩u (0) = u (L ) = 0
'

1
EN0175 09 / 07 / 06

General solution for E u + ρ g = 0 :

''

0
ρg
u=− x + C1 x + C2
2

2E

i) Solution for specified boundary conditions of P1:

ρg P ρ gL
u=− x2 + x+ x
2E E E
σ = Eu ' = P + ρ g (L − x )
Actually, we can simply get this solution by considering the equilibrium below

σ (x )

ρ g (L − x )

ii) Solution for specified boundary conditions of P2:

ρg ρg ρg
u=− x2 + Lx = x (L − x )
2E 2E 2E
ρg
σ = Eu ' = (L − 2 x )
2

σ >0

σ <0

How to use FEM to solve the same problems? (Discretization)

2
EN0175 09 / 07 / 06

elements
nodes

1) Nodal displacement

u1 , u2 , L , un or u j ( j = 1, 2, L , n )

u (x j ) = u j

2) Use interpolation to represent u ( x ) in terms of u j

n
u ( x ) = ∑ u k wk ( x ) , where wk ( x ) is a function based on element type.
k =1

For linear element:

wk ( x )

1
0 k L

⎧1, j = k
wk (x j ) = ⎨
⎩0, j ≠ k

u (x )

Exact
FEM

0 L x
(More elements lead to more accuracy)

3
EN0175 09 / 07 / 06

Original problem: to solve u ( x ) (infinite DOE)

FEM problem: to solve u j ( j = 1, 2, L , n )

E u '' + f = 0 (strong form)

∫ (E u )
+ f w(x ) dx = 0 (weak form)
L
''
0

The discrete formula of the weak form is

∫ (E u )
+ f w j ( x ) dx = 0 , ( j = 1, 2, L, n)
L
''
0

⎧ x − xk −1
⎪ , k ∈ [xk −1 , xk ]
⎪ xk − xk −1
⎪ x −x
wk ( x ) = ⎨ k +1 , k ∈ [xk , xk +1 ]
⎪ xk +1 − xk
⎪
⎪0, otherwise
⎩

( n algebraic equations to be solved for n unknowns u j )

n
u ( x ) = ∑ u k wk ( x ) , let us calculate the equation in parts.
k =1

E u '' w j ( x ) dx = Eu ' w j ( x ) 0L − ∫ Eu ' w 'j (x ) dx

L L
∫ 0 0
n
= − ∫ E ∑ wk' ( x )w 'j ( x ) dx u k
L

0
k =1
n
= −∑ K jk u k
k =1

Ewk' ( x )w 'j ( x )dx which is called Stiffness Matrix. (It turns out this is still
L
where K jk = ∫ 0

correct even if Young’s modulus varies along the length of the bar, i.e. E = E (x ) . Can u

explains why?).

f w j ( x ) dx = F j (nodal force)
L
∫ 0

Therefore, we have
n

∑K
k =1
jk uk = Fj

which can be written in matrix form as

4
EN0175 09 / 07 / 06

KU = F
⎡ k11 k12 L k1n ⎤ ⎡ u1 ⎤ ⎡ F1 ⎤
⎢k k 22 L k2n ⎥⎥ ⎢u2 ⎥⎥ ⎢F ⎥
where K = ⎢ , U= ⎢ , F=⎢ ⎥
21 2

⎢L L L L⎥ ⎢L⎥ ⎢L ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣k n1 k n 2 L k nn ⎦ ⎣un ⎦ ⎣ Fn ⎦

Remarks:

1) K jk = K kj (symmetric matrix)

2) K jk = 0 , if j − k >1

0
[K ] =
0
K is a sparse matrix. (there is very efficient numerical solver for such a property)

What if u ( L) = u L ≠ 0 ?

0
1
2
3

n
n+1 u = uL

n
u ( x ) = ∑ uk wk ( x ) + u L wk +1 ( x )
k =1

5
EN0175 09 / 07 / 06

E u '' w j (x ) dx = Eu ' w j ( x ) 0L − ∫ Eu ' w 'j ( x ) dx

L L
∫ 0 0
n
= −∑ K jk u k − u L ∫ E wn' +1 (x )w 'j ( x )dx
L

0
k =1

( j = 1,2,..., n ). If we define the nodal forces as

F j = ∫ f w j (x ) dx − u L ∫ E wn' +1 ( x )w 'j ( x )dx

L L

0 0

n
The same form ∑K
k =1
jk u k = F j is obtained.

What if we have traction BC σ = P @ x = L ?

In this case, the nodal displacement at the boundary node x = L is unknown. We need to solve

u j for j = 1,2,..., n, n + 1 .

Consider
n +1
u ( x ) = ∑ u k wk ( x )
k =1

E u '' w j ( x ) dx = Eu ' w j ( x ) 0L − ∫ Eu ' w 'j (x ) dx

L L
∫ 0 0
n +1
= Pw j ( x n +1 ) − ∑ ∫ Ewk' ( x )w 'j ( x ) dx u k
L

0
k =1
n +1
= Pw j ( x n +1 ) − ∑ K jk u k
k =1

Therefore:
n +1

∑K u k = Pw j ( xn +1 ) + ∫ f w j ( x ) dx = F j
L
jk
0
k =1

∂ 2u ∂ 2u
What if we are dealing with dynamic problem, i.e. E u + f = ρ u
&& ( E 2 + f = ρ 2 ) ? ''

∂x ∂t

0 = ∫ (E u '' + f − ρ u&&)w j ( x ) dx
L

n +1
E u '' w j ( x ) dx = P (t )w j ( x n +1 ) − ∑ ∫ Ewk' ( x )w 'j ( x ) dx u k (t )
L L
∫ 0
k =1
0

f w j ( x ) dx remains the same as in static case. In the dynamic case, we will have
L
The term ∫ 0

6
EN0175 09 / 07 / 06

the additional term:

n +1
ρ u&& w j (x ) dx = −∑ ρwk (x )w j (x ) dx u&&k (t )
L L
−∫ ∫
0 0
k =1
n +1
= −∑ M jk u&&k (t )
k =1

= ∫ ρwk (x )w j ( x ) dx is called the Mass Matrix. (The density should be kept

L
where M jk
0

inside the integral if the density varies along the length of the bar, i.e. ρ = ρ (x ) .)

Therefore we can write:

n +1 n +1

∑ M jk u&&k (t ) + ∑ K jk uk = P(t )w j (xn+1 ) + ∫ f w j (x )dx = F j (t )

L

0
k =1 k =1

In matrix form

&& + KU = F
MU
Remarks:

1) M jk = M kj .

2) M jk = 0 , if j − k >1

0
[M ] =
0
Similar to the stiffness matrix, M is also a symmetric and sparse matrix.