You are on page 1of 13

CE 72.

12 Finite Element Methods in Engineering


Lecture 07: Galerkin to Finite Element Formulation
Semester – January 2023

Dr. Chaitanya Krishna


Department of Civil and Infrastructure Engineering
School of Engineering and Technology (SET)

3 February 2023 1
Galerkin → FEM 2

• Galerkin method was introduced by Boris Galerkin


in the early 20th century.
• The Galerkin method is a weighted residual
method used to solve partial differential
equations (PDEs).
• The solution of a PDE is approximated by a finite-
dimensional function space (trial function).
• The residual of the PDE is projected onto the trial
function, and the resulting equation is solved for
the unknown coefficients.

• The Galerkin method was later generalized and


developed into the finite element method (FEM).
• In the FEM, the trial function is divided into a set of
elements, each of which is approximated by a lower-
order polynomial.
• This leads to a piecewise approximation of the
solution, which can handle complex geometries and
discontinuous material properties.
3 February 2023
Piecewise Approximation 3

• Piecewise approximation refers to approximating a continuous


function using a series of simple functions defined over non-
overlapping subdomains.
• In the context of FEM, it is used to approximate the solution of
partial differential equations (PDEs).
• The solution of a PDE is divided into small subdomains, known as elements.
• The solution within each element is approximated by a lower-
order polynomial, such as a linear or quadratic function.
• These polynomials are chosen to be continuous across the
element boundaries.
• The piecewise approximation of the solution is used to represent
the solution over the entire domain of the problem.
• The piecewise approximation allows for efficient computation and
improved accuracy for certain types of problems compared to
other numerical techniques.

3 February 2023
FEM 4

Simple modification:
𝜓𝑚 and 𝜙𝑛 Dividing the domain into smaller shaped region ,
we can construct the approximate solution
effectively.
 Divide this 1D domain between a and b into a set of M small finite
segment:
𝒙𝒊 N points in the 1D domain
x

M segments
𝒙𝒊 = 𝒂 𝒙𝑵 = 𝒃

The process of division of domain into elements and nodes is the


fundamental part of FEM
4
3 February 2023
Elemental Function (Linear) 5

Consider a segment (FE)

A B 𝑥𝑖 𝑥𝑖+1
𝑢ො ෢𝑒 = 𝑎0 + 𝑎1 𝑥 𝜙
𝑢

𝑢2 𝜙𝑖 𝜙𝑖+1
𝑢1 1 1

𝑥 𝑥
𝑥𝑖 𝑥𝑖+1 𝑥𝑖 𝑥𝑖+1

𝑢෣
𝑒 (𝑥 ) = 𝑎 + 𝑎 𝑥 = u
𝑖 0 1 𝑖 1
𝑎0 = 𝑢1𝑥𝑥𝑖+1 −𝑢
−𝑥
1 𝑥𝑖
𝑖+1 𝑖 𝑢2 𝑥𝑖 − 𝑢1 𝑥𝑖+1 𝑢2 − 𝑢1
෢𝑒 =
𝑢 + 𝑥
𝑢𝑒෣
(𝑥𝑖+1 ) = 𝑎0 + 𝑎1 𝑥𝑖+1 = u2 𝑥𝑖+1 − 𝑥𝑖 𝑥𝑖+1 − 𝑥𝑖
𝑎1 = 𝑥𝑢2−𝑢1
−𝑥
𝑖+1 𝑖

𝑥𝑖+1 − 𝑥 𝑥 − 𝑥𝑖 ෢𝑒 = 𝜙𝑖 𝑢1 + 𝜙𝑖+1 𝑢2
෢𝑒 =
𝑢 𝑢1 + 𝑢 𝑢
𝑥𝑖+1 − 𝑥𝑖 𝑥𝑖+1 − 𝑥𝑖 2

3 February 2023
Continuous Polynomials 6

0 𝑥 < 𝑥𝑖 −1
𝑥 − 𝑥𝑖−1 If we could consider 𝜓𝑚 = 𝜙𝑛
𝑥𝑖−1 ≤ 𝑥 < 𝑥𝑖 all the integrals can be
𝑥 − 𝑥𝑖−1
𝜙𝑛 = 𝑥𝑖
𝑖+1 − 𝑥 calculated over the length
𝑥𝑖 ≤ 𝑥 < 𝑥𝑖+1
𝑥𝑖+1 − 𝑥𝑖
0 𝑥𝑖+1 ≤ 𝑥

𝑑𝜙1
𝜙1 𝑑𝑥
𝑥1 𝑥2 𝑥3 𝑥4 𝑥𝑁 𝑥1 𝑥2 𝑥3 𝑥𝑁

𝑑𝜙2
𝜙2 𝑑𝑥
𝑥1 𝑥2 𝑥3 𝑥4 𝑥𝑁 𝑥1 𝑥2 𝑥3 𝑥𝑁

𝑑𝜙3
𝜙3 𝑥1 𝑥2 𝑥3 𝑥4 𝑥𝑁 𝑑𝑥 𝑥1 𝑥2 𝑥3 𝑥4 𝑥𝑁

6 3 February 2023
FE Shape Function 7

𝑁1
𝑑𝜙𝑖
𝜙𝑖
𝑥𝑖−1 𝑥𝑖 𝑥𝑖+1 𝑑𝑥 −𝑖−1 𝑖 −𝑖+1 𝑑𝑁1
𝑑𝑥
𝑁2 𝑑𝑁2
𝑑𝜙𝑖+1 𝑑𝑥
𝜙𝑖+1 𝑥𝑖 𝑥𝑖+1 𝑥𝑖+2 𝑖 𝑖+1
𝑑𝑥

1. In any interval → [ 𝑥𝑖 , 𝑥𝑖+1 ] ; each interval is defined by local


functions 𝑁1 𝑎𝑛𝑑 𝑁2 ,
2. 𝑁1 𝑎𝑛𝑑 𝑁2 are called shape functions
3. Local coordinate in each element is 𝑥1𝑒 , 𝑥2𝑒

3 February 2023
Displacements and Weights 8

𝑢1𝑒
𝑢ො 𝑒 = 𝑁1 𝑢1𝑒 + 𝑁2 𝑢2𝑒 𝑁 𝑁2 { 𝑢𝑒 }
2 𝑥1𝑒 𝑥2𝑒
𝑁1
𝑤𝑒= 𝑁1 𝑤1𝑒 + 𝑁2 𝑤2𝑒 𝑤1𝑒 𝑤2𝑒 {𝑁 }
2

Local Coordinate 𝑥 ′ = 𝑥 − 𝑥1𝑒 𝑥′ 𝑥′


𝑁1 = 1 − 𝑒 ; 𝑁2 = 𝑒
Element Length ℎ𝑒 = 𝑥2𝑒 − 𝑥1𝑒 ℎ ℎ
Expression for 𝑁1 and 𝑁2
will be same for every element

𝑑𝑁1 𝑑𝑁1 −1 𝑑𝑁2 𝑑𝑁2 1


= = 𝑒 = =
𝑑𝑥 𝑑𝑥 ′ ℎ 𝑑𝑥 𝑑𝑥 1 ℎ𝑒
Weak Form 9

𝜕2𝑢 𝜕𝑤 𝜕𝑢
𝐺 𝑤, 𝑢 = න 𝑤 𝑥 𝜌 − 𝑏𝑥 𝑑𝑥 + න 𝐸 𝑑𝑥 − 𝑤 𝑡𝑥ҧ ቚ W
Ω 𝜕𝑡 2 Ω 𝜕𝑥 𝜕𝑥 Γ𝑡

𝐺෠ 𝑤,
ෝ 𝑢ො = 𝐺෡𝑖 𝑤, ෢𝜎 𝑤,
ෝ 𝑢ො + 𝐺 ෢𝑓 𝑤,
ෝ 𝑢ො − 𝐺 ෝ 𝑡𝑥ҧ ቚ ‫𝑡ڿ‬
ෝ 𝑢ො − 𝑤
• Terms:
𝑀 ℎ𝑒
𝑁1 ෢𝑒ሷ
𝑢
𝐺෡𝑖 𝑤,
ෝ 𝑢ො = ෍ 𝑤1𝑒 − 𝑤2𝑒 න 𝜌 𝑁 𝑁2 𝑑𝑥 ′ 1
𝑁2 𝑒 1 ෢𝑒ሷ
𝑢
𝑒=1 0 2

𝑀 ℎ𝑒 𝑑𝑁1
෢𝑒
𝑢
෢𝜎 𝑤,
𝐺 ෝ 𝑢ො = ෍ 𝑤1𝑒 − 𝑤2𝑒 න 𝑑𝑥 ′ 𝐸 𝑑𝑁1 𝑑𝑁2
𝑑𝑥 ′ 1
𝑑𝑁2 𝑒 𝑑𝑥 ′ 𝑑𝑥 ′ ෢𝑒
𝑢
𝑒=1 0 2
𝑑𝑥 ′
𝑀 ℎ𝑒
෢ 𝑒 𝑒 𝑁1
𝐺𝑓 𝑤,
ෝ 𝑢ො = ෍ 𝑤1 − 𝑤2 න 𝑏𝑥 𝑑𝑥 ′
𝑁2
𝑒=1 0
Simplified Expression 10

𝑀𝑒 𝑢෠ሷ + 𝐾𝑒 𝑢ො = 𝑓 𝑒

ℎ𝑒
𝑒 𝑒
𝑁 𝑀 𝑀12
𝑀𝑒 = න 1 𝜌𝑒 𝑁1 𝑁2 𝑑𝑥 ′ = 11
𝑒 𝑒
𝑁2 𝑀21 𝑀22
0

ℎ𝑒 𝑑𝑁1
𝑒 𝑒
𝑑𝑥 ′ 𝐸 𝑑𝑁1 𝑑𝑁2 𝐾11 𝐾12
𝐾𝑒 = න 𝑑𝑥 ′ =
𝑑𝑁2 𝑒 𝑑𝑥 𝑑𝑥
𝑒
𝐾21 𝑒
𝐾22
0
𝑑𝑥 ′
ℎ𝑒
𝑒
𝑁 𝑓
𝑓 𝑒 = න 1 b𝑥 d𝑥 ′ = 1𝑒
𝑁2 𝑓2
0
For a Linear Shape Function 11

𝑀𝑒 𝐾𝑒 𝑓𝑒
𝑒
𝑀11 𝑒
𝑀12 ෢1ሷ
𝑢 𝑒
𝐾11 𝑒
𝐾12 𝑢
ෞ1 𝑓1𝑒
𝑒 𝑒 + 𝑒 𝑒 =
𝑀21 𝑀22 ෢
𝑢2ሷ 𝐾21 𝐾22 𝑢
ෞ2 𝑓2𝑒

• For the linear polynomial , we assumed earlier:

𝑒
1 𝑒 2 1 𝐸 1 −1 1 1
𝑀𝑒 = 𝜌𝑒 ℎ ; 𝐾𝑒 = 𝑒 ; 𝑓 = 𝑏𝑥 ℎ𝑒
𝑒
6 1 2 ℎ −1 1 2 1
Assembling 12
1 2 3 M Global Nodes
Local Nodes Element Nos …
1 2 3 4 N
𝑢
ෞ1 𝑢
ෞ2 𝑢
ෞ3 𝑢
ෞ4 𝑢ෞ𝑁 1 2 3 4 ….
@𝑥1𝑒 →1 1 2 3 4
@𝑥2𝑒 →2 2 3 4 5

For a static problem , if the inertial form is ignored: 𝐾𝑢 = 𝑓


1 2 3 4 ------------------------------------N

1 1 𝑢
ෞ1 𝑓1
𝐾11 𝐾12 𝑓2
𝑢
ෞ2
1 1 + 𝐾2 2
𝐾21 𝐾22 11 𝐾12 𝑢
ෞ3 𝑓3
2 2 3 3 _ _
𝐾21 𝐾22 + 𝐾11 𝐾12
3 3 _ _
𝐾21 𝐾22 =
_ _
_ _
_ _
_ _
𝑢ෞ𝑁 𝑓𝑁
Boundary Condition 13
Take the effect of boundary to the other side

N
3
1

5
𝑢
ෞ𝑎 0 𝑢
ෞ𝑎

-
1 0 0 0 0 0 0 0 𝑢෤ 1 𝑓1 𝐾11 𝑢
1 1 1
𝐾11 𝐾12 0 0 0 0 0 0 𝑢෤ 2 𝑓2 𝐾12 𝑢
2 1 0 1 2 2 𝑢෤ 3 𝑓3 0
𝐾21 𝐾22 + 𝐾11 𝐾12 0 0 0 0 _ _ 0
0 𝑓 + 𝑡 ҧ
3 3 3
_ _ 𝑗 𝑥 0
0 2
𝐾21 2
𝐾22 +𝐾11 𝐾12 0 0 0 = −
_ 𝑓𝑗 0
- 0 _ _ 0
3 3 4
0 0 𝐾21 𝐾22 +𝐾11 0 0 0 _ _ 0
- 0 _ _ _
N 0 0 0 0 0 0 0 0 𝑢෤ 𝑁 𝑓𝑁 _

Boundary Conditions = Essential Boundary + Natural Boundary


𝑢ഥ𝑖 𝑥𝑎 = 𝑢ෞ𝑎 𝑎𝑛𝑑 𝑓𝑗 = 𝑓𝑗 + 𝑡𝑥ҧ

Solve:
𝐾𝑢 = 𝑓 ⇒ 𝑢 = 𝐾 −1 𝑓

You might also like