FINITE ELEMENT

METHOD

Annie Fournier-Gagnoud

Phelma 2016-2017
 FINITE ELEMENT METHOD

● Variational formulation of the PDE

● Approximation of integrals
● Two approaches exist:
 Minimum of the energy of the system
PDE Energy functional exists ??
 Projective approach or Galerkin method
Variational formulation of the PDE
General approach

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 2

 FINITE ELEMENT METHOD

To present the finite element method we chose

the heat transfer equation:

● Finite element technique :

discretization in finite elements

● Galerkin method

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 3

 FINITE ELEMENT METHOD
Discretization of 

● Domain Ω is the union of small volume elements ωe :

tetrahedrons, hexahedrons, prisms (2 D elements :
triangles, quadrilaterals).
● Boundary Г of the domain Ω is the union of small
elements of the boundary γf.

       
 e e  f f

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 4

 FINITE ELEMENT METHOD
Discretization of 

Hexahedron
Triangle

Tetrahedron
Quadrilateral

Prism

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 5

 FINITE ELEMENT METHOD
Discretization of 

Problem on this edge

2 1 2 1

3 4 3

Problem on this edge

Structured mesh Unstructured mesh

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 6

 FINITE ELEMENT METHOD
Discretization of Interpolation
● Elements to interpolate the unknown on the
study domain
● Approximated function T(x,y,z,t)
● At time t the approximated function is
constructed by finite elements.
● Function must be continuous on the domain
● T is interpolated on each element and T is
continuous from one element to the neighboring
element
● Polynomial interpolation

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 7

 FINITE ELEMENT METHOD
Polynomial interpolation

● 2D case with a mesh in triangles

● Linear interpolation of T on the triangle
T ( x, y )  ax  by  c
● The coefficients a, b, c ?

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 8

 FINITE ELEMENT METHOD
Polynomial interpolation
T ( x, y )  ax  by  c
y 2

 T1  a  x1  b  y1  c
T(x,y)

3 T2  a  x 2  b  y 2  c
T  a  x  b  y  c
 3 3 3
1
x  x1 y1 1 a  T1 
x    
y 2 1  b   T2  
 2
 x 3 y 3 1  c  T3 
A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 9
 FINITE ELEMENT METHOD
Polynomial interpolation

 x1 y1 1 a  T1 
x    
y 2 1  b   T2  
 2
 x 3 y 3 1  c  T3 
   x 2  y 3  x 3  y 2    x 3  y1  x1  y 3    x1  y 2  x 2  y1 
 1
 a    y 2  y 3   T1   y 3  y1   T2   y1  y 2   T3 


 1
 b    x 3  x 2   T1   x1  x 3   T2   x 2  x1   T3 
 
c  1   x  y  x  y   T   x  y  x  y   T   x  y  x  y   T 

 
2 3 3 2 1 3 1 1 3 2 1 2 2 1 3

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 10

 FINITE ELEMENT METHOD
Polynomial interpolation
T ( x, y )  ax  by  c
 1
 a    y 2  y 3   T1   y 3  y1   T2   y1  y 2   T3 


 1
 b    x3  x 2   T1   x1  x3   T2   x 2  x1   T3 
 
c  1   x  y  x  y   T   x  y  x  y   T   x  y  x  y   T 
2 3 3 2 1 3 1 1 3 2 1 2 2 1 3

 
1
T ( x, y )    y 2  y 3   T1   y 3  y1   T2   y1  y 2   T3   x

1
   x 3  x 2   T1   x1  x 3   T2   x 2  x1   T3   y

1
   x 2  y 3  x 3  y 2   T1   x 3  y1  x1  y 3   T2   x1  y 2  x 2  y1   T3 

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 11
 FINITE ELEMENT METHOD
Polynomial interpolation
T ( x, y )  ax  by  c
1
T ( x, y )    y 2  y 3   T1   y 3  y1   T2   y1  y 2   T3   x

1
   x 3  x 2   T1   x1  x 3   T2   x 2  x1   T3   y

1
   x 2  y 3  x 3  y 2   T1   x 3  y1  x1  y 3   T2   x1  y 2  x 2  y1   T3 

1
T ( x, y )    y 2  y 3   x   x3  x 2   y   x 2  y 3  x3  y 2    T1

1
   y 3  y1   x   x1  x 3   y   x 3  y1  x1  y 3    T2

1
   y1  y 2   x   x 2  x1   y   x1  y 2  x 2  y1    T3

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 12
 FINITE ELEMENT METHOD
Polynomial interpolation
T ( x, y )  ax  by  c
1
T ( x, y )    y 2  y 3   x   x 3  x 2   y   x 2  y 3  x 3  y 2    T1

1
   y 3  y1   x   x1  x 3   y   x 3  y1  x1  y 3    T2

1
   y1  y 2   x   x 2  x1   y   x1  y 2  x 2  y1    T3

1
 1 ( x, y )    y 2  y 3   x   x3  x 2   y   x 2  y 3  x3  y 2  

1
 2 ( x, y )    y 3  y1   x   x1  x 3   y   x 3  y1  x1  y 3  

1
 3 ( x, y )    y1  y 2   x   x 2  x1   y   x1  y 2  x 2  y1  

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 13
 FINITE ELEMENT METHOD
Polynomial interpolation

T ( x, y )  ax  by  c
1
 1 ( x, y )    y 2  y 3   x   x3  x 2   y   x 2  y 3  x3  y 2  

1
 2 ( x, y )    y 3  y1   x   x1  x 3   y   x 3  y1  x1  y 3  

1
 3 ( x, y )    y1  y 2   x   x 2  x1   y   x1  y 2  x 2  y1  

   x 2  y 3  x 3  y 2    x 3  y1  x1  y 3    x1  y 2  x 2  y1 

T ( x, y )   1 ( x, y )  T1   2 ( x, y )  T2   3  x, y   T3

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 14

 FINITE ELEMENT METHOD
Polynomial interpolation

T ( x, y )   1 ( x, y )  T1   2 ( x, y )  T2   3  x, y   T3

Temperature on a point of the triangle is the average temperature of the

temperature at the node with coefficients calculated by the polynomials

Properties of these polynomials

T(x1, y1 )  T1 T(x2 , y2 )  T2 T(x3 , y3 )  T3

 x, y  1(x, y)   2(x, y)   3  x, y  1

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 15

 FINITE ELEMENT METHOD
Polynomial interpolation

T ( x, y )   1 ( x, y )  T1   2 ( x, y )  T2   3  x, y   T3

Temperature on a point of the triangle is the average temperature of the

temperature at the node with coefficients calculated by the polynomials

Properties of these polynomials

T(x1, y1 )  T1 T(x2 , y2 )  T2 T(x3 , y3 )  T3

 x, y  1(x, y)   2(x, y)   3  x, y  1

Reference elements : Lagrange polynomials

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 16

 FINITE ELEMENT METHOD
Reference element

● Each kind of element Reference element.

● Reference element :
– a fixed geometry and fixed size.

●
Reference element : polynomials are known.
 It is not necessary to calculate the coefficients to
have the polynomial functions.

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 17

 FINITE ELEMENT METHOD
Reference element

y η
3
+1 3
2

1 2 ξ
1
0 +1
x

Element Reference element

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 18

 FINITE ELEMENT METHOD
Reference element

y
4 3 η

4 1 3

1 -1 1
ξ
2
1 -1 2

Element Reference element

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 19

 FINITE ELEMENT METHOD
Reference element

● Element with N nodes

● Real coordinates (x, y) of the point M
● Coordinates (ξ, η) of the point M in the reference element
N N
x    i  ,   xi y    i  ,   y i
1 1

● Interpolation of the temperature on a point with

coordinates (ξ,η) in the reference element:
N
T    i   ,    Ti
1
A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 20
 FINITE ELEMENT METHOD
Triangle 3 nodes

Reference element – Polynomial functions

Node i  i  , 
1 1   
2 
3 

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 21

 FINITE ELEMENT METHOD
Quadrilateral 4 nodes
Reference element – Polynomial functions

Node i
 i  , 
η
1 1
1     1    1 3
4 4

2 1
1     1   
4 -1 1
ξ

3 1
1     1    1 -1 2
4

4 1
1     1   
4 Reference element

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 22

 FINITE ELEMENT METHOD
Triangle 6 nodes
Reference element – Polynomial functions

Node i  i  , 
1 ( ) * (2  1)

2  * (2  1)
6 5
3  * (2  1)
4 4 4

5 4
6 4   1   
A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 23
 FINITE ELEMENT METHOD
Galerkin method
● Static heat transfer equation on Ω
div(k  gradT )  Q
● Find the function T(x, y, z)
● T on a vector space of functions (Sobolev space)
● βi the basis functions of the space.
● Scalar product of two functions f and g by the
following relation:
 f .g   f  g d


A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 24

 FINITE ELEMENT METHOD
Galerkin method

● For a function T(x,y,z)

  −k ∇
∇⋅  T  −Q=

●
is not zero , depends on T
● Only one vector is orthogonal to all βi: vector 0

T =0 ⇔ ∀ i ∭  i⋅ d =0



A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 25

 FINITE ELEMENT METHOD
Galerkin method

● Number of scalar products = number of elements

of the basis
● Which basis functions will we take? i   i
● Polynomials defined for the polynomial
interpolation of the elements.
 Continuity between the elements

∀i ∭ ⃗ ( −k ∇
βi ∇⋅ ⃗ T ) d Ω−∭ β Q d Ω=0
i
Ω Ω

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 26

 FINITE ELEMENT METHOD
Galerkin method
∀i ∭ ⃗ ( −k ∇
βi ∇⋅ ⃗ T ) d Ω−∭ β Q d Ω=0
i
Ω Ω

i   i
∀i ∭ ⃗ ( −k ∇
αi ∇⋅ ⃗ T ) d Ω−∭ α Q d Ω=0
i
Ω Ω

● Boundary conditions don’t appear

 Dirichlet : OK
 Neumann : this formulation must be modify

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 27

 FINITE ELEMENT METHOD
Galerkin method: Weak Formulation

∀i ∭ ⃗ ( −k ∇
αi ∇⋅ ⃗ T ) d Ω−∭ α Q d Ω=0
i
Ω Ω

∭ ⃗ (−α k ∇
∇⋅ ⃗ T ) d Ω=∭ α ∇⋅
⃗ ( −k ∇
⃗ T ) d Ω−∭ k ∇
⃗ α ⋅∇
⃗ T dΩ
i i i
Ω Ω Ω

∭ ⃗ (−k ∇
α i ∇⋅ ⃗ T ) d Ω=∭ ∇⋅
⃗ (−α k ∇
⃗ T ) d Ω+∭ k ∇
⃗ α ⋅∇⃗ T dΩ
i i
Ω Ω Ω

∭ ⃗ (−α k ∇
∇⋅ ⃗ T ) d Ω=−∬ α k ∇
⃗ T⋅⃗n d Γ
i i
Ω Γ

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 28

 FINITE ELEMENT METHOD
Galerkin method: Weak Formulation

∀i ∭ ⃗ ( −k ∇
αi ∇⋅ ⃗ T ) d Ω−∭ α Q d Ω=0
i
Ω Ω

∭ ⃗ (−k ∇
α i ∇⋅ ⃗ T ) d Ω=∭ ∇⋅
⃗ (−α k ∇
⃗ T ) d Ω+∭ k ∇
⃗ α ⋅∇⃗ T dΩ
i i
Ω Ω Ω

∭ ⃗ (−α k ∇
∇⋅ ⃗ T ) d Ω=−∬ α k ∇
⃗ T⋅⃗n d Γ
i i
Ω Γ

Weak formulation
⃗ T⋅⃗n d Γ+∭ k ∇
∀ i −∬ αi k ∇ ⃗ α ⋅∇⃗ T d Ω−∭ α Q d Ω=0
i i
Γ Ω Ω

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 29

 FINITE ELEMENT METHOD
Galerkin method: Weak Formulation

⃗ T⋅⃗n d Γ+∭ k ∇
∀ i −∬ αi k ∇ ⃗ α ⋅∇⃗ T d Ω−∭ α Q d Ω=0
i i
Γ Ω Ω

⃗ T⋅⃗n =h ( T−T ) +ε σ ( T 4 −T 4 )
−k ∇ r SB r

∀i ∬ i ( r ) ∬ i SB (
α h T −T d Γ+ α ε σ T
4
−T )r d Γ+∭ k ∇⃗ αi⋅∇⃗ T d Ω
4

Γ Γ Ω

−∭ αi Q d Ω=0
Ω

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 30

 FINITE ELEMENT METHOD
Galerkin method: Weak Formulation

∀i ∭ ⃗ α ⋅∇
k∇ ⃗ T d Ω+∬ α h (T −T ) d Γ+∬ α ε σ ( T 4 −T 4 ) d Γ
i i r i SB r
Ω Γ Γ

−∭ αi Q d Ω=0
Ω

• Generally this method is well adapted to Neumann conditions

• In this formulation the geometry doesn’t create difficulty for the
boundary condition.
• Complex geometry can be modeled

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 31

 FINITE ELEMENT METHOD
Finite Element Formulation

∭ ⃗ ⃗ ∬ i ( r ) ∬ i SB ( 4 4
∀i k ∇ α ⋅
i ∇ T d Ω+ α h T−T d Γ+ α ε σ T −T r )d Γ
Ω Γ Γ

−∭ αi Q d Ω=0
Ω

       
N
T = ∑  j   ,  ,  ⋅T j
 e e  f f j=1

N N
∂T ∂j 
=∑   ,  ,  ⋅T j ∇ T =∑ 
∇  j   ,  ,  ⋅T j
∂ x j=1 ∂ x j =1

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 32

 FINITE ELEMENT METHOD
Finite Element Formulation

     
N
 
N
T =∑ α j ( ξ , η, ζ )⋅T j ⃗
∇ T =∑ ⃗
∇ α j ( ξ , η , ζ )⋅T j
 e e  f f j=1 j=1

N N
∀i ∑e ∭
ω
⃗ α⋅
k∇ i
e
∑
j=1
⃗α T
∇ j j ( ) d Ω+∑ ∬ αi h
f γf (∑ )
j=1
αj T j d Γ
N
+ ∑ ∬ αi ε σ SB T 3
f γf (∑ )
j=1
α j T j d Γ=∑ ∬ α i hT r d Γ
f γf

+ ∑ ∬ αi ε σ SB T 4r d Γ+ ∑ ∭ αi Q d Ω
f γf e ωe

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 33

 FINITE ELEMENT METHOD
Finite Element Formulation
N N
⃗ ⃗
∀i ∑e ∑
j=1 ( ω
∭ k ∇ α ⋅
i
e
∇ α j d Ω ⋅T j + ∑ ∑
) f j=1
(∬ α hα d Γ )⋅T
γf
i j j

N
3
+∑ ∑ ∬ i SB α j d Γ ⋅T j=∑ ∬ α i hT r d Γ
f j=1
( α ε σ
γf
T
) f γf

+ ∑ ∬ α i ε σ SB T 4r d Γ+ ∑ ∭ αi Q d Ω
f γf e ωe

 System of non linear equations

 Coefficients of the matrix are integrals : numerical method: Gauss method
for quadrilateral
 We obtain a sparse matrix

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 34

 FINITE ELEMENT METHOD
Finite Element Formulation
 Two numberings:
 The number of the nodes on the mesh global node number,
 The number of the node on the element local node number
 Two create the system local node number j must be
transformed into the global node number
 In practice sub matrix of the system are calculated
and assembly of the matrix is done after
 The integration is done on reference element, a
change of variable must be done: where J is the
Jacobian matrix and the coordinate on the reference
element.
A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 35
 FINITE ELEMENT METHOD
Finite Element Formulation

 Integration : Change of variable must be

done :
 x x x 
    
 
d  dxdydz  det  J  ddd J  y y y 
    
 z z z 
 
    

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 36

 FINITE ELEMENT METHOD
Finite Element Formulation

 Calculation of grad :  i

  i   i  i   i   i 
       
x
   x  x  x  x
 i   i 
 y          
  i  , , , , , , , ,
  x x x y y y z z z
 z 

Reverse matrix J

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 37

 FINITE ELEMENT METHOD
Finite Element Formulation

 Calculation of reverse matrix J

[ ]
∂ ∂ ∂
∂x ∂y ∂z
−1
J =
∂ ∂ ∂  i
∂x ∂y ∂z
∂ ∂ ∂
∂x ∂y ∂z

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 38

 FINITE ELEMENT METHOD
Finite Element Formulation

21 22 23 24 25
13 14 15 16

16 17 18 19 20
9 10 11 12 4 3

11 12 13 14 15
5 6 7 8
1 2

6 7 8 9 10
1 2 3 4

Local numbering
1 2 3 4 5 of nodes

Global numbering of nodes

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 39

 FINITE ELEMENT METHOD
Finite Element Formulation
21
13
22
14
23
15
24
16
25 Equation at the node 13
16 17 18 19 20
9 10 11 12 N  
15
   

e6,7,10,11 j1 
k 13   jd Tj 


11 12 13 14
5 6 7 8 e

6
1
7
2
8
3
9
4
10   
e6,7,10,11 
13Q d

1 2 3 4 5

Global numbering of nodes

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 40

 FINITE ELEMENT
METHOD 21 22 23 24 25
13 14 15 16

16 17 18 19 20
9 10 11 12

Equation at the node 2 11

5
12
6
13
7
14
8
15

N 6 7 8 9 10

∑ ∑ ∭ ⃗ α ⋅∇
⃗ α d Ω ⋅T
e=1,2 j=1 ( ω
k∇
e
2 j
) j
1
1

2
2

3
3

4
4

5
N
+∑ ∑ (∬
f =1,2 j=1γ f
)
α2 h α j d Γ ⋅T j
N

+∑ ∑(∬ α 2 ε σ SB T α j d Γ )⋅T j = ∑ ∬ α2 h T r d Γ
3

γ
f =1,2 j=1 f =1,2 γ f =1,2 f

+ ∑ ∬ α2 ε σ SB T 4r d Γ+ ∑ ∭ α2 Q d Ω
f γ f =1,2 e=1,2 ωe

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 41

 FINITE ELEMENT METHOD
Time Dependant phenomena

● Unsteady heat transfer equation:

T
 CP    (k  T )  Q  0
t
● For the time : same scheme as in finite
difference method : implicit method

T
i  
 i  CP
t
     (k T)    Q  0

i

i

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 42

 FINITE ELEMENT METHOD
Time Dependant phenomena

 T  Tn  Tn1 N N
   T = ∑  j   ,  ,  ⋅T j 
∇ T =∑ 
∇  j   ,  ,  ⋅T j
 t  n t j=1 j =1

T n−T n−1 ⃗ α ⋅( ∇
⃗ T) dΩ
∭ Ω
αi ρC P
Δt
d Ω+∭ k ∇
Ω
i n

4 4
+∬ α i h (T n −T r ) d Γ+∬ αi ε σ SB ( T n −T r ) d Γ−∭ αi Q n d Ω=0
Γ Γ Ω

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 43

43
 FINITE ELEMENT METHOD
Time Dependant phenomena

∭ αi ρ C P T n d Ω+ Δ t⋅∭ k ∇ ⃗ α ⋅( ∇
⃗ T) dΩ
i n
Ω Ω

+ Δt⋅∬ αi h T n d Γ+Δ t⋅∬ αi ε σ SB T 4n d Γ=∭ αi ρ C P T n−1 d Ω

Γ Γ Ω

+ Δ t⋅∬ αi h T r d Γ+Δ t⋅∬ αi ε σ SB T 4r d Γ+ Δ t⋅∭ αi Q n d Ω

Γ Γ Ω

Stable scheme

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 44

 FINITE ELEMENT METHOD
Conclusion

● Complicated method compared to the FDM or FVM

● To construct the system of equations we have to
developed and used different numerical techniques:
 mesh generation,
 interpolations,
 numerical integrations,
 construction of elementary matrix,
 assembly the matrix,
 numerical method to solve linear system with sparse matrix.

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 45

 FINITE ELEMENT METHOD
Conclusion

● Important advantages :
 Well adapted to take in account Neumann
boundary condition,
 Well adapted to complex geometry,
 Well adapted to multi domain geometry,
 Adapted to different physical problems,
 General software : Flux-expert, Comsol

A. Fournier-Gagnoud FINITE ELEMENT METHOD (16-17) 46