CE 72.

12 Finite Element Methods in Engineering

Lecture 04-06: Approximate Solutions
Semester – January 2023

Dr. Chaitanya Krishna

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

3 February 2023 1
Weak Formulation 2

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

How to find an approximate solution to this equation?

3 February 2023
Construct an Approximate Solution 3
• Steps:
i. Express the 𝑢 𝑥, 𝑡 in term of Specified Functions x Unknown Parameters

ii. Similarly , N Terms

the arbitrary function 𝑤(𝑥) as Specified Function x Unknown Parameters
N Terms

𝑢 𝑥, 𝑡 ≈ 𝑢ො 𝑥, 𝑡 = ෍ 𝜙𝑛 𝑎𝑛 𝑡 + 𝑢𝑏 (𝑥, 𝑡) A
𝑛=1
𝑁 Any function that
𝑤 𝑥, 𝑡 ≈ 𝑤
ෝ 𝑥, 𝑡 = ෍ 𝜓𝑚 𝑥 𝑤𝑚 can satisfy the
displacement
𝑚=1 boundary function

Assumption: 𝜙𝑛 𝑥 and 𝜓𝑚 𝑥 are Zero at all the Boundary Conditions

𝑢𝑏 𝑥, 𝑡 = 𝜙𝑏 𝑥, 𝑡 𝑑ҧ (𝑡)

3 February 2023
How to Make Approximations 4

• Point collocation:
• Evaluates the differential equation at specific points
• Uses those evaluations to determine the unknown field variable at
those points
• Interpolates field variable between these points
• Sub-domain collocation:
• Similar to point collocation, but collocates over a sub-domain
• Integrates differential equation over sub-domain
• Interpolates field variable between these sub-domains
• Weighted residual method (Galerkin):
• Minimizes the residual of the differential equation
• Involves multiplying the residual by a weighting function
• Integrates over the domain
• All three methods are used to approximate solutions to
differential equations, with different trade-offs on accuracy,
efficiency, and ease of implementation.

3 February 2023
Galerkin Method 5

• Galerkin method is a way to solve the weak form of a

differential equation using FEM
• It involves the following steps:
• Express the differential equation in weak form
• Choose a set of test functions defined over the domain
• Multiply the weak form by a test function and integrate over the
domain
• Resulting equation relates unknown field variable to the test functions and
load function
• Use the set of algebraic equations to determine the values of the
unknown field variable
• Use the values of the field variable to interpolate over the entire
domain
• Merit: Galerkin method is easy to implement,
computationally efficient and provides high-accuracy solution

3 February 2023
Galerkin Solution of Elasticity Equations 6

• Approximate Galerkin solution for elasticity problem can be

obtained by inserting the constructed solution into weak form
equation
PDE ODE
• Substitute A in W W O

𝑁 𝑁

2 ҧ
ⅆ2 𝑎𝑛 −𝑑 𝑑
𝐺෠ 𝑤,
ෝ 𝑢ො = ා 𝑤𝑚 ා ඲ 𝜓𝑚 𝜌 𝜙𝑛 + 𝜙𝑏 2 ⅆ𝑥
ⅆ𝑡 2 ⅆ𝑡
𝛺
𝑚=1 𝑛=1
𝑁
𝑁 O
ⅆ𝜓𝑚 ⅆ𝜙𝑛 ⅆ𝜙𝑏ഥ
+ා 𝑤𝑚 ෎ න 𝑬 [ 𝑎 𝑡 + ҧ ] dx
𝑑(t)
ⅆ𝒙 ⅆ𝒙 𝑛 ⅆ𝑘
𝛺
𝑛=1
𝑚=

𝑁 𝑁
- ෍ 𝑤𝑚 ‫ 𝑥𝑑 𝑥𝑏 𝑛𝜓 𝛺׬‬− ෍ 𝑤𝑚 𝜓𝑚 𝑥 𝑡𝑥ҧ ȁ𝛤𝑡 = 0
𝑚=1 𝑚=1

3 February 2023
ODE→LA 7

• Since the functions 𝜙𝑛 , 𝜓𝑛 and 𝜙𝑏ത are all known functions, the integrals
may be evaluated as

𝑀𝑚𝑛 = න 𝜓𝑚 𝜌 𝜙𝑛 𝑑𝑥 Mass Matrix

𝛺 Specific Load Vector
𝑑𝜓𝑚 𝑑𝜙𝑛
𝐾𝑚𝑛 =න 𝐸 𝑑𝑥 Stiffness Matrix
𝑑𝑥 𝑑𝑥
𝛺

ⅆ2 𝑑ҧ ⅆ𝜓𝑚 𝑑𝜙𝑏−
𝑓𝑚 𝑡 = න 𝜓𝑚 𝑏𝑥 𝑑𝑥 + 𝜓𝑚 𝑡𝑥ҧ ቚ − න 𝜓𝑚 𝜌𝜙𝑏ഥ ⅆ𝑥 . 2 − න 𝐸 ⅆ𝑥 𝑑ҧ 𝑡
𝛤𝑡 ⅆ𝑡 𝑥 ⅆ𝑥
𝛺 𝛺 𝛺

Since the parameters 𝑤𝑚 are arbitrary, the expression each one must be zero.
This leads to the set of equations
𝑁
ⅆ 2 𝑎𝑛
෍ 𝑀𝑚𝑛 2 + 𝐾𝑚𝑛 𝑎𝑛 𝑡 = 𝑓𝑚 𝑡 , m= 1,2,…..N
𝑛=1 ⅆ𝑡

3 February 2023
Static LA Equation 8

• Problems without inertia effects are called static.

• The equations are now purely algebraic
σ𝑁
𝑛=1 𝐾𝑚𝑛 𝑎𝑛 = 𝑓𝑚 , N= 1,2,…..N

Where the load matrix reduces to

ⅆ𝜓𝑚 𝑑𝜙𝑏−
𝑓𝑚 = න 𝜓𝑚 𝑏𝑥 𝑑𝑥 + 𝜓𝑚 𝑡𝑥ҧ ቚ − න 𝐸 ⅆ𝑥 𝑑ҧ
𝛤𝑡 𝑑𝑥 ⅆ𝑥
𝛺 𝛺
Expressing the problem in matrix form
𝐾𝑎 = 𝑓
Where,
K11 K12 …. K1N f1 We know a
K21 K22 …. K2N f2
K = and f =
: : : 𝑁
KN1 KN2 …. KNN fN 𝑑𝜙𝑛
𝜎𝑥 = 𝐸 ෍ .𝑎
𝑑𝑥 𝑛
𝑛=1
The formal solution is given by
𝑎 = 𝐾 −1 𝑓
3 February 2023
Example Question 9

𝐸𝐴𝐵 = 1000 10 , 0< x < 5

𝑡𝑥 = 25 𝑏𝑥 =
A x C B 0 , 5< x < 10
0 5 10
10
➢ Find Displacement and the stresses on each location on the bar
Boundary Conditions are given as : 𝑡𝑥ҧ 10 = 25; 𝑢 𝑥 = 0

𝜕𝜎𝑥 𝜕2 𝑥 𝜕𝜎𝑥
Exact Solution: + 𝑏𝑥 = 𝜌 = −𝑏𝑥 ( integrate )
𝜕𝑥 𝜕𝑡𝑥 𝜕𝑥

𝜎𝑥 = −10 < x − 5 > +25 𝑥 − 5;𝑥 ≤ 5

< 𝑥−5>= ቊ
0 ;𝑥 > 5
𝜕𝑢
E = −10 < x − 5 > +25 (integrate again )
𝜕𝑥
−10 < x − 5 2 > +25x + 𝑐
1000 u 𝑥 =
2
Because u 𝑥 = 0 ⇒ , c = 125 , Thus ,
1 2+
u 𝑥 = −5 < 𝑥 − 5 > 25𝑥 + 125
1000
3 February 2023
Weak form of the problem
• weak formulation for the problem
0

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

𝜕𝑤 𝜕𝑢
𝐺 𝑤, 𝑢 = ‫׬‬Ω 𝑤 𝑥 −𝑏𝑥 𝑑𝑥 + ‫׬‬Ω 𝐸 𝑑𝑥 − 𝑤𝑡𝑥ҧ ȁΓ𝑡
𝜕𝑥 𝜕𝑥
10 10
𝜕𝑤 𝜕𝑤
න .1000 𝑑𝑥 − න 𝑤 𝑥 . 𝑏𝑥 𝑑𝑥 − 𝑤 10 . 25 = 0
𝜕𝑥 𝜕𝑥
10
0
10
0
10
0
𝜕𝑤 𝜕𝑤
1000 න . 𝑑𝑥 − න 𝑤 𝑥 .10 . 𝑑𝑥 − න 𝑤 𝑥 . 0𝑑𝑥 − 𝑤 10 . 25 = 0
𝜕𝑥 𝜕𝑥
0 0 5

10
𝜕𝑤 𝜕𝑢
1000 . . 𝑑𝑥 − න 𝑤 𝑥 .10 . 𝑑𝑥 − 𝑤 10 . 25 = 0
𝜕𝑥 𝜕𝑥
0
LA Form
𝑁
𝑥2 𝑥3 𝑥𝑛 𝑥 𝑛
𝑢ො 𝑥 = 𝑎1 𝑥 + 𝑎2 2 + 𝑎3 3 + … … . . 𝑎𝑛 𝑛 = ෍ 𝑎𝑛
𝐿 𝐿 𝐿 10
𝑛=1
N Terms
𝑁
𝑥 𝑥2 𝑥3 𝑥𝑛 𝑥 𝑚𝑤
𝑤
ෝ 𝑥 = 𝑤1 + 𝑤2 2 + 𝑤3 3 + … … . . 𝑤𝑛 𝑛 = ෍ 𝑚
𝐿 𝐿 𝐿 𝐿 10
𝑚=1
Weak Formulation can be written as N
: Terms
𝑁 10 5
𝑥 𝑛−1 𝑥 𝑚−1 𝑥 𝑚 10 𝑑𝑥
෍ න 10𝑚𝑛 𝑑𝑥 𝑎𝑛 = න + 25
10 10 10
𝑛=1 0 0

10
𝑥 𝑛−1 10𝑚𝑛. 1 ∗ 10 10𝑚𝑛
𝐾𝑚𝑛 = න 10𝑚𝑛 𝑑𝑥 = =
10 𝑚+𝑛−1 𝑚+𝑛−1
0

𝐾𝑚𝑛 . 𝑎𝑛 = 𝑓𝑚
5
𝑥 𝑚 10 𝑑𝑥
100 𝑚+1 +
𝑓𝑚 = න + 25 = 1/2 25
10 m+1
0
Displacements and Stresses
• After Finding Values of 𝑎𝑛

𝑁 𝑁
𝑥 𝑛
𝑢ො = ෍ 𝑎𝑛 𝜙𝑛 = ෍ 𝑎𝑛
𝐿
𝑛=1 𝑛=1

𝑁
𝜕𝑢ො 𝑥 𝑛−1
𝜎ො = 𝐸 = ෍ 𝑎𝑛 . 𝑛 𝑛
𝜕𝑥 𝐿
𝑛=1
Results 13

N. Terms 𝑎1 𝑎𝟐 𝑎𝟑 𝑎𝟒 𝑎𝟓

N=1 0.3750 0 0 0 0
N=2 0.6250 -0.2500 0 0 0
N=3 0.7012 -0.7107 0.3125 0 0
N=4 0.7812 -0.7187 0.3125 0 0
N=5 0.7344 -0.2500 -1.0937 1.6406 -0.6562

Observations :
1. When N=1 ,the error seems to be
quite high
2. Numerically stable
3. Displacements are converging
4. Ends have same displacement , This is
usually seen in 1-D Problems

3 February 2023