١٣/٠٨/١٤٤١

Finite Element Method

WHY WE
Need for Computational Methods?
• Solutions Using Either Strength of Materials or Theory of
Elasticity Are Normally Accomplished for Regions and
Loadings With Relatively Simple Geometry
• Many Applicaitons Involve Cases with Complex Shape,
Boundary Conditions and Material Behavior
• Therefore a Gap Exists Between What Is Needed in
Applications and What Can Be Solved by Analytical Closed-
form Methods
• This Has Lead to the Development of Several
Numerical/Computational Schemes Including: Finite
Difference, Finite Element and Boundary Element Methods

١
١٣/٠٨/١٤٤١

Finite Element Analysis

The finite element method is a computational scheme to solve field problems in
engineering and science. The technique has very wide application, and has been used on
problems involving stress analysis, fluid mechanics, heat transfer, diffusion, vibrations,
electrical and magnetic fields, etc. The fundamental concept involves dividing the body
under study into a finite number of pieces (subdomains) called elements (see Figure).
Particular assumptions are then made on the variation of the unknown dependent
variable(s) across each element using so-called interpolation or approximation functions.
This approximated variation is quantified in terms of solution values at special element
locations called nodes. Through this discretization process, the method sets up an
algebraic system of equations for unknown nodal values which approximate the
continuous solution. Because element size, shape and approximating scheme can be
varied to suit the problem, the method can accurately simulate solutions to problems of
complex geometry and loading and thus this technique has become a very useful and
practical tool.

Advantages of Finite Element Analysis

- Models Bodies of Complex Shape

- Can Handle General Loading/Boundary Conditions
- Models Bodies Composed of Composite and Multiphase Materials
- Model is Easily Refined for Improved Accuracy by Varying
Element Size and Type (Approximation Scheme)
- Time Dependent and Dynamic Effects Can Be Included
- Can Handle a Variety Nonlinear Effects Including Material
Behavior, Large Deformations, Boundary Conditions, Etc.

٢
١٣/٠٨/١٤٤١

Basic Concept of the Finite Element

Method
Any continuous solution field such as stress, displacement,
temperature, pressure, etc. can be approximated by a
discrete model composed of a set of piecewise continuous
functions defined over a finite number of subdomains.

One-Dimensional Problem Distribution

T T

Approximate Piecewise
Exact Analytical Solution Linear Solution

x x

Two-Dimensional Discretization

u(x,y)
-1

-2

-3
4
3.5
3
3 2.5
2
2.5 1.5
Approximate 2 1
Piecewise Linear 0.5
1.5 0
Representation -0.5
1 -1
y x

٣
١٣/٠٨/١٤٤١

Discretization Concepts
T

E xact T emperature D is tribution, T(x)

Finite Elem ent Discretization

Linear Inte rpolation Model Q uadratic Interpolation M ode l
T1 (Four E lem ents) T (Tw o E lem ents)
1
T2
T2 T2
T3 T3 T3
T4 T4 T3
T4 T5
T5

T T
T1 T1
T2 T2
T3 T3
T4 T5 T4 T5

x x
Piecewise Linear Appr oxim ation Piecewise Q uadratic A pproxim ation
Tem perature Continuous but w ith Tem perature and Temperature Gradients
D iscontinuous Temperature G radients Continuous

Common Types of Elements

Two-Dimensional Elements
One-Dimensional Elements Triangular, Quadrilateral
Line Plates, Shells, 2-D Continua
Rods, Beams, Trusses, Frames

Three-Dimensional Elements
Tetrahedral, Rectangular Prism (Brick)
3-D Continua

٤
١٣/٠٨/١٤٤١

Discretization Examples

Three-Dimensional
One-Dimensional Two-Dimensional Brick Elements
Frame Elements Triangular Elements

Basic Steps in Finite Element Method

Time Independent Problems
- Domain Discretization
- Select Element Type (Shape and Approximation)
- Derive Element Equations (Variational and Energy Methods)
- Assemble Element Equations to Form Global System

[K]{U} = {F}
[K] = Stiffness or Property Matrix
{U} = Nodal Displacement Vector
{F} = Nodal Force Vector

- Incorporate Boundary and Initial Conditions

- Solve Assembled System of Equations for Unknown Nodal
Displacements and Secondary Unknowns of Stress and Strain Values

٥
١٣/٠٨/١٤٤١

Common Sources of Error in FEA

• Domain Approximation
• Element Interpolation/Approximation
• Numerical Integration Errors
(Including Spatial and Time Integration)
• Computer Errors (Round-Off, Etc., )

Measures of Accuracy in FEA

Accuracy
Error = |(Exact Solution)-(FEM Solution)|

Convergence
Limit of Error as:
Number of Elements
or
Approximation Order
Increases
Ideally, Error → 0 as Number of Elements or
Approximation Order → ∞

٦
١٣/٠٨/١٤٤١

Two-Dimensional Discretization Refinement

(Node)
•

(Discretization with 228 Elements)

•
•
(Triangular Element)

(Discretization with 912 Elements)

One Dimensional Examples

Static Case
Bar Element Beam Element
Uniaxial Deformation of Bars Deflection of Elastic Beams
Using Strength of Materials Theory Using Euler-Bernouli Theory

w1 w2
u1 u2
θ2
1 2 θ1 1 2

Differential Equation : Differential Equation :

d d2 d 2w
− (au ) + cu − q = 0 − (b 2 ) = f ( x )
dx dx 2 dx
Boundary Condtions Specification : Boundary Condtions Specification :
du dw d 2w d d 2w
u,a w, ,b 2 , (b 2 )
dx dx dx dx dx

٧
١٣/٠٨/١٤٤١

Two Dimensional Examples

Triangular Element Triangular Element
Scalar-Valued, Two-Dimensional Vector/Tensor-Valued, Two-
Field Problems Dimensional Field Problems
v3
φ3

3 3 u3
v2
φ2
2
u2
2
v1
1

φ1 1 u1
ElasticityField Equationsin Terms of Displacements
Example Differential Equation :
E ∂  ∂u ∂v 
µ∇2 u +  +  + Fx = 0
∂ 2φ ∂ 2φ
+ = f ( x, y ) 2(1 − ν) ∂x  ∂x ∂y 
∂x 2 ∂y 2
E ∂  ∂u ∂v 
µ∇2 v +  +  + Fy = 0
2(1 − ν) ∂y  ∂x ∂y 
Boundary Condtions Specification :
dφ ∂φ ∂φ
φ, = nx + ny BoundaryConditons
dn ∂x ∂y
 ∂u ∂v   ∂u ∂v 
Tx =  C11 + C12 n x + C66  + n y
 ∂x ∂y   ∂y ∂x 
 ∂u ∂v   ∂u ∂v 
Ty = C66  + nx +  C12 + C22 n y
 ∂y ∂x   ∂x ∂y 

Simple Element Equation Example

Direct Stiffness Derivation
u1 u2

F1 F2
1 2
k
Equilibrium at Node 1 ⇒ F1 = ku1 − ku 2
Equilibrium at Node 2 ⇒ F2 = −ku1 + ku 2

or in Matrix Form

 k − k   u1   F1 
− k  = 
 k  u2   F2 
Stiffness Matrix Nodal Force Vector
[ K ]{u} = {F }

٨
١٣/٠٨/١٤٤١

Common Approximation Schemes

One-Dimensional Examples
Polynomial Approximation
Most often polynomials are used to construct approximation
functions for each element. Depending on the order of
approximation, different numbers of element parameters are
needed to construct the appropriate function.

Linear Quadratic Cubic

Special Approximation
For some cases (e.g. infinite elements, crack or other singular
elements) the approximation function is chosen to have special
properties as determined from theoretical considerations

One-Dimensional Bar Element

Approximation : u = ∑ ψ k ( x )uk = [ N ]{d }
k

du d d[N ]
Strain : e = = ∑ ψ k ( x )u k = {d } = [ B]{d }
dx k dx dx
Stress - Strain Law : σ = Ee = E[ B]{d}

∫ σδedV = P u
Ω
i i + Pj u j + ∫ fδudV ⇒
Ω
L  Pi  L
∫ A[ B] E[ B]dx{d } = {δd}   + {δd}T ∫ A[ N ]T fdx ⇒
T T T
{δd }
 Pj 
0 0
L L
∫ A[ B] E[ B]dx{d} = {P} + ∫ A[ N ] fdx
T T
0 0

L
[ K ] = ∫ A[ B]T E[ B]dx = Stiffness Matrix
0

 Pi  L
[ K ]{d } = {F } {F } =   + ∫ A[ N ]T fdx = Loading Vector
 Pj 
0

 ui 
{d } =   = Nodal Displacement Vector
u j 

٩
١٣/٠٨/١٤٤١

One-Dimensional Bar Element

Axial Deformation of an Elastic Bar

f(x) = Distributed Loading A = Cross-sectional Area

E = Elastic Modulus
Typical Bar Element
dui ui Ω uj du j
Pi = − AE Pj = − AE
dx (i ) L (j ) dx
(Two Degrees of Freedom)

Virtual Strain Energy = Virtual Work Done by Surface and Body Forces

∫ V
σ ij δeij dV = ∫ Ti n δui dS + ∫ Fi δui dV
St V

For One-Dimensional Case

∫ σδedV = P u
Ω
i i + Pj u j + ∫ fδudV
Ω

Linear Approximation Scheme

ui uj
x (local coordinate system)
(1) L (2)
u(x)
Approximate Elastic Displacement
u = a1
u = a1 + a 2 x ⇒ 1
u2 = a1 + a2 L
x
u − u1  x  x (1) (2)
⇒ u = u1 + 2 x = 1 − u1 +  u2
L  L  L
= ψ1 ( x )u1 + ψ 2 ( x )u2 ψ1(x) ψ2(x)
 u   x x   u1 
u = [ψ1 ψ 2 ] 1  = 1 −    = [ N ]{d } 1
u2   L L  u2 
[ N ] =Approximation Function Matrix
{d } = Nodal Displacement Vector
x
(1) (2)
ψk(x) – Lagrange Interpolation Functions

١٠
١٣/٠٨/١٤٤١

Element Equation
Linear Approximation Scheme, Constant Properties
 1
L −  1 1 
L AE  1 − 1
[ K ] = ∫ A[ B] E[ B]dx = AE[ B] [ B]∫ dx = AE  L −
T T
L=
0 0 1  L L  L − 1 1 
 
 
 L 
 x
 P1  L  P1  L
−   P  Af L 1
{F } =   + ∫ A[ N ] fdx =   + Af o ∫  L dx =  1  + o  
T

P
 2 0 P
 2 0 x  P2  2 1
 
 L 
u 
{d } =  1  = Nodal Displacement Vector
u2 

AE  − 1 1   u1   P1  Af o L 1
[ K ]{d } = {F } ⇒  1 − 1 u  =  P  + 2 1
L   2   2  

Quadratic Approximation Scheme

u1 u2 u3
x
(1) (2) (3)
L
Approximate Elastic Displacement u(x)
u1 = a1
L L2
u = a1 + a2 x + a3 x 2
⇒ u2 = a1 + a2 + a3
2 4 x
u3 = a1 + a2 L + a3 L2 (1) (2) (3)
u = ψ1 ( x )u1 + ψ 2 ( x )u2 + ψ 3 ( x )u3
ψ2(x)
 u1  ψ3(x)
  ψ1(x)
u = [ψ1 ψ 2 ψ 3 ]u2  = [ N ]{d}
u3  1
Element Equation
x
 7 − 8 1   u1   F1  (1) (2) (3)
AE     
− 8 16 − 8 u2  =  F2 
3L  
   
 1 − 8 7  u3   F3 

١١
١٣/٠٨/١٤٤١

Lagrange Interpolation Functions

Using Natural or Normalized Coordinates
1 , i = j
ψ i (ξ j ) = 
0 , i ≠ j
ξ 1
ψ1 = (1 − ξ)
2
(1) −1 ≤ ξ ≤ 1 (2) 1
ψ 2 = (1 + ξ)
2
1
ψ1 = − ξ(1 − ξ)
ξ 2
ψ 2 = (1 − ξ)(1 + ξ)
(1) (2) (3)
1
ψ3 = ξ(1 + ξ)
2
9 1 1
ψ1 = − (1 − ξ)( + ξ)( − ξ)
16 3 3
27 1
ξ ψ2 = (1 − ξ)(1 + ξ)( − ξ)
16 3
27 1
(1) (2) (3) (4) ψ3 = (1 − ξ)(1 + ξ)( + ξ)
16 3
9 1 1
ψ 4 = − ( + ξ)( − ξ)(1 + ξ)
16 3 3

Simple Example
P

A1,E1,L1 A2,E2,L2
Take Zero Distributed Loading
1 2
f =0
(1) (2) (3)
Global Equation Element 1 Global Equation Element 2
 1 − 1 0 U 1   P1 
(1)
0 0 0  U 1   0 
A1 E1      (1)  A2 E 2     
− 1 1 0 U 2  =  P2  0 1 − 1 U 2  =  P1( 2 ) 
L1   L2  
 0 0 0 U 3   0  0 − 1 1  U 3   P2 
( 2 )

Assembled Global System Equation

 A1 E1 A1 E1 
 L − 0 
 1 L1  U1   P1
(1)
  P1 
− A1 E1 A1 E1 A2 E 2 A2 E 2     (1) (2)   
+ − =
 2  2
U P + P  =  P2 
 L1 L1 L2 L2    
1


(2)  P 
AE A2 E 2  U 3   P2   3
 0 − 2 2 
 L2 L2 

١٢
١٣/٠٨/١٤٤١

Simple Example Continued

P

A1,E1,L1 A2,E2,L2
1 2
(1) (2) (3)
Reduced Global System Equation
Boundary Conditions  A1 E1 A1 E1 
 L − 0 
U1 = 0  1 L1   0   P1 
(1)

− A1 E1 A1 E1 A2 E 2 A2 E 2     
P2( 2 ) = P + − U 2  =  0 
 L1 L1 L2 L2    
 
+ P1( 2) = 0 A2 E 2  U 3   P 
(1)
P2 A2 E 2
 0 − 
 L2 L2 

 A1 E1 A2 E 2 A2 E 2 
 L + L −
L2  U 2   0  For Uniform AE  2 − 1 U 2   0 
 1  =  = 
A2 E 2  U 3   P  L − 1 1  U 3   P 
2

 − A2 E 2 Properties A, E , L
 L2 L2 
PL 2 PL
Solving ⇒ U 2 = , U3 = , P1(1) = − P
AE AE

One-Dimensional Beam Element

Deflection of an Elastic Beam
f(x) = Distributed Loading
x

I = Section Moment of Inertia

E = Elastic Modulus
Typical Beam Element d  d 2w   d 2w 
Q1 =  EI 2  , Q2 =  EI 2 
w1 w2 dx  dx 1  dx 1
θ1 θ2
Ω d  d 2w   d 2w 
Q3 = −  EI 2  , Q4 = − EI 2 
(1) L (2)
M2 dx  dx  2  dx  2
M1
V1 V2 dw dw
u1 = w1 , u2 = θ1 = − , u3 = w2 , u4 = θ 2 = −
(Four Degrees of Freedom) dx 1 dx 2

Virtual Strain Energy = Virtual Work Done by Surface and Body Forces

∫ σδedV = Q u
Ω
1 1 + Q2 u2 + Q3u3 + Q4 w4 + ∫ fδwdV ⇒
Ω
L L
EI ∫ [ B] [ B]dx{d } = Q1u1 + Q2 u2 + Q3u3 + Q4 w4 + ∫ f [ N ]T dV
T
0 0

١٣
١٣/٠٨/١٤٤١

FEA Beam Problem

f Uniform EI

a b
1 2
(1) (2) (3)
Element 1
 6 / a3 − 3 / a2 − 6 / a3 − 3 / a2 0 0 U1   6  Q1(1) 
   − a   (1) 
− 3 / a
2
2/ a 3 / a2 1/ a 0 0 U 2    Q2 
− 6 / a 3 3 / a2 6 / a3 3 / a2 0 0 U 3  fa  6  Q3(1) 
2 EI     = −   +  (1) 
− 3 / a
2
1/ a 3 / a2 2/a 0 0 U 4  12  a  Q4 
 0 0 0 0 0 0 U 5  0  0 
      
 0 0 0 0 0 0 U 6   0   0 

Element 2
0 0 0 0 0 0  U1   0 
0 0 0 0 0 0  U 2   0 

0 0 6 / b3 − 3 / b2 − 6 / b3 − 3 / b 2  U 3  Q1( 2) 
2 EI    =  
0 0 − 3 / b2 2/b 3 / b2 1 / b  U 4  Q2( 2) 
0 0 − 6 / b3 3 / b2 6 / b3 3 / b 2  U 5  Q3( 2) 
    
0 0 − 3 / b2 1/ b 3 / b2 2 / b  U 6  Q4( 2) 

FEA Beam Problem

1 2
(1) (2) (3)

Global Assembled System

6 / a 3 − 3 / a2 − 6 / a3 − 3 / a2 0 0  U1   6   Q1(1) 
   − a   
 ⋅ 3 / a2 (1)
2/ a 1/ a 0 0  U 2     Q2 
 ⋅ ⋅ 6 / a 3 + 6 / b3 3 / a 2 − 3 / b 2 − 6 / a3 − 3 / a 2  U 3  fa  6  Q3(1) + Q1( 2) 
2 EI     = −   +  (1) ( 2) 
 ⋅ ⋅ ⋅ 2/a +2/b 3 / a2 1 / a  U 4  12  a  Q4 + Q2 
 ⋅ ⋅ ⋅ ⋅ 6 / a3 3 / a 2  U 5   0   Q3( 2) 
      
 ⋅ ⋅ ⋅ ⋅ ⋅ 2 / a  U 6   0   Q4( 2) 

Boundary Conditions Matching Conditions

U1 = w1(1) = 0 , U 2 = θ1(1) = 0 , Q3( 2) = Q4( 2) = 0 Q3(1) + Q1( 2) = 0 , Q4(1) + Q2( 2) = 0

Reduced System
6 / a 3 + 6 / b 3 3 / a 2 − 3 / b 2 − 6 / a3 − 3 / a 3  U 1  6  0
      
⋅ 2/ a+ 2/b 3 / a2 1 / a  U 2  fa a  0
2 EI    = −  +  
 ⋅ ⋅ 6 / a3 3 / a 2  U 3  12 0  0
   0  0
 ⋅ ⋅ ⋅ 2 / a  U 4 

Solve System for Primary Unknowns U1 ,U2 ,U3 ,U4

Nodal Forces Q1 and Q2 Can Then Be Determined

١٤
١٣/٠٨/١٤٤١

Special Features of Beam FEA

Analytical Solution Gives Analytical Solution Gives

Cubic Deflection Curve Quartic Deflection Curve

FEA Using Hermit Cubic Interpolation

Will Yield Results That Match Exactly
With Cubic Analytical Solutions

Truss Element
Generalization of Bar Element With Arbitrary Orientation

k=AE/L

s = sin θ , c = cos θ

١٥
١٣/٠٨/١٤٤١

Frame Element
Generalization of Bar and Beam Element with Arbitrary Orientation
w1 w2
θ1 θ2
u1 Ω u2
P1 P2
(1) L (2)
M2
M1
V1 V2
 AE AE 
 L 0 0 − 0 0 
L
 12 EI 6 EI 12 EI 6EI   u   P 
 0 0 −  1 1
 L3 L2 L3 L2   w  Q 
 0 6EI 4 EI 6 EI 2 EI     1 
1
0 − 2
 L2 L L L   θ1  = Q2 
 AE AE    
− 0 0 0 0   u2   P2 
 L L  w  Q 
12 EI 6 EI 12 EI 6EI   2   3 
 0 − − 0 
− 2  θ 2  Q4 
 L3 L2 L3 L 
 0 6EI 2 EI
0
6 EI
− 2
4 EI 
 L2 L L L 

Element Equation Can Then Be Rotated to Accommodate Arbitrary Orientation

١٦