Finite Element Methods Lectures

Lecture Notes: Introduction to Finite Element Method Chapter 1.
Introduction
Chapter 1. Introduction
I. Basic Concepts
The finite element method (FEM), or finite element analysis
(FEA), is based on the idea of building a complicated object with
simple blocks, or, dividing a complicated object into small and
manageable pieces. Application of this simple idea can be found
everywhere in everyday life as well as in engineering.
Examples:
• Lego (kids’ play)

• Buildings
• Approximation of the area of a circle:
“ Element” Si
θi
R
1
Area of one triangle: S i = 2 R 2 sin θ i
 2π 
N
1 2
Area of the circle: S N = ∑ Si = R N sin  → π R 2 as N → ∞
i =1 2  N
where N = total number of triangles (elements).
© 1998 Yijun Liu, University of Cincinnati 1

Lecture Notes: Introduction to Finite Element Method Chapter 1. Introduction
Why Finite Element Method?
• Design analysis: hand calculations, experiments, and

computer simulations
• FEM/FEA is the most widely applied computer simulation
method in engineering
• Closely integrated with CAD/CAM applications
• ...
Applications of FEM in Engineering
• Mechanical/Aerospace/Civil/Automobile Engineering
• Structure analysis (static/dynamic, linear/nonlinear)
• Thermal/fluid flows
• Electromagnetics
• Geomechanics
• Biomechanics
• ...
Examples:
...
A Brief History of the FEM
• 1943 ----- Courant (Variational methods)

• 1956 ----- Turner, Clough, Martin and Topp (Stiffness)
• 1960 ----- Clough (“Finite Element”, plane problems)
• 1970s ----- Applications on mainframe computers

• 1980s ----- Microcomputers, pre- and postprocessors
• 1990s ----- Analysis of large structural systems

FEM in Structural Analysis
Procedures:
• Divide structure into pieces (elements with nodes)

• Describe the behavior of the physical quantities on each
element
• Connect (assemble) the elements at the nodes to form an
approximate system of equations for the whole structure
• Solve the system of equations involving unknown
quantities at the nodes (e.g., displacements)
• Calculate desired quantities (e.g., strains and stresses) at
selected elements
Example:

Computer Implementations
• Preprocessing (build FE model, loads and constraints)

• FEA solver (assemble and solve the system of equations)
• Postprocessing (sort and display the results)
Available Commercial FEM Software Packages
• ANSYS (General purpose, PC and workstations)

• SDRC/I-DEAS (Complete CAD/CAM/CAE package)
• NASTRAN (General purpose FEA on mainframes)
• ABAQUS (Nonlinear and dynamic analyses)
• COSMOS (General purpose FEA)
• ALGOR (PC and workstations)
• PATRAN (Pre/Post Processor)
• HyperMesh (Pre/Post Processor)
• Dyna-3D (Crash/impact analysis)
• ...

Objectives of This FEM Course
• Understand the fundamental ideas of the FEM

• Know the behavior and usage of each type of elements
covered in this course
• Be able to prepare a suitable FE model for given problems
• Can interpret and evaluate the quality of the results (know
the physics of the problems)
• Be aware of the limitations of the FEM (don’t misuse the
FEM - a numerical tool)

II. Review of Matrix Algebra
Linear System of Algebraic Equations

a 11 x 1 + a 12 x 2 +...+ a 1n x n = b1
a 21 x1 + a 22 x 2 +...+ a 2 n x n = b2
(1)
.......
a n1 x 1 + a n 2 x 2 +...+ a nn x n = bn
where x1, x2, ..., xn are the unknowns.
In matrix form:
Ax = b (2)
where
a11 a12 ... a1n 
a ... a 2 n 
[ ]
a22
A = aij =  
21
 ... ... ... ... 

a ... ann 
 n1 an2
(3)
 x1  b1 
x  b 
 2  2
x = { xi } =   b = {bi } =  
: :
 xn  bn 
A is called a n×n (square) matrix, and x and b are (column)

vectors of dimension n.

Row and Column Vectors

 w1 
 
v = [ v1 v2 v3 ] w = w2 
w 
 3
Matrix Addition and Subtraction

For two matrices A and B, both of the same size (m×n), the
addition and subtraction are defined by
C=A +B with cij = a ij + bij
D=A−B with d ij = a ij − bij
Scalar Multiplication
λA = λa ij[ ]
Matrix Multiplication
For two matrices A (of size l×m) and B (of size m×n), the
product of AB is defined by
m
C = AB with cij = ∑ a ik bkj
k =1
where i = 1, 2, ..., l; j = 1, 2, ..., n.

Note that, in general, AB ≠ BA , but ( AB )C = A ( BC)
(associative).

Transpose of a Matrix
If A = [aij], then the transpose of A is
A T = a ji [ ]
Notice that ( AB ) T = B T A T .
Symmetric Matrix
A square (n×n) matrix A is called symmetric, if
A = AT or a ij = a ji
Unit (Identity) Matrix

 1 0 ... 0 
 0 1 ... 0 
I= 
... ... ... ...
 0 0 ... 1 
 
Note that AI = A, Ix = x.
Determinant of a Matrix
The determinant of square matrix A is a scalar number
denoted by det A or |A|. For 2×2 and 3×3 matrices, their
determinants are given by
a b 
det   = ad − bc
c d 
and
a11 a12 a13 
det a21 a22 a23  = a11a22a33 + a12a23a31 + a21a32a13
 
a31 a32 a33 
− a13a22a31 − a12a21a33 − a23a32a11
Singular Matrix
A square matrix A is singular if det A = 0, which indicates
problems in the systems (nonunique solutions, degeneracy, etc.)
Matrix Inversion
For a square and nonsingular matrix A (det A ≠ 0), its
inverse A-1 is constructed in such a way that
AA −1 = A −1 A = I
The cofactor matrix C of matrix A is defined by
Cij = ( −1)i + j Mij
where Mij is the determinant of the smaller matrix obtained by

eliminating the ith row and jth column of A.
Thus, the inverse of A can be determined by
1
A −1 = CT
det A
We can show that ( AB ) −1 = B −1A −1 .

Examples:
−1
a b  1  d − b
(1)   =
c d  (ad − bc) − c a 
Checking,
−1
 a b  a b  1  d − b  a b  1 0
=
 c d   c d  (ad − bc) − c a   c d  = 0 1
        
−1
 1 −1 0 
T
3 2 1  3 2 1
(2) − 1 2 − 1 2 2 1 = 2 2 1
1
=
  (4 − 2 − 1)    
 0 − 1 2  1 1 1 1 1 1
Checking,
 1 − 1 0  3 2 1 1 0 0
− 1 2 − 1 2 2 1 = 0 1 0
    
 0 − 1 2  1 1 1 0 0 1
If det A = 0 (i.e., A is singular), then A-1 does not exist!

The solution of the linear system of equations (Eq.(1)) can be
expressed as (assuming the coefficient matrix A is nonsingular)
x = A −1b
Thus, the main task in solving a linear system of equations is to
found the inverse of the coefficient matrix.

Solution Techniques for Linear Systems of Equations

• Gauss elimination methods
• Iterative methods
Positive Definite Matrix

A square (n×n) matrix A is said to be positive definite, if for
any nonzero vector x of dimension n,
x T Ax > 0
Note that positive definite matrices are nonsingular.
Differentiation and Integration of a Matrix

Let
[
A( t ) = a ij ( t ) ]
then the differentiation is defined by
d  da (t ) 
A(t ) =  ij 
dt  dt 
and the integration by
∫ ∫
A(t )dt =  aij (t )dt 
 

Types of Finite Elements
1-D (Line) Element
(Spring, truss, beam, pipe, etc.)
2-D (Plane) Element
(Membrane, plate, shell, etc.)
3-D (Solid) Element
(3-D fields - temperature, displacement, stress, flow velocity)

III. Spring Element

“E verything important is simple. ”
One Spring Element
x
i j
fi ui uj fj
k
Two nodes: i, j
Nodal displacements: ui, uj (in, m, mm)
Nodal forces: fi, fj (lb, Newton)
Spring constant (stiffness): k (lb/in, N/m, N/mm)
Spring force-displacement relationship:
F = k∆ with ∆ = u j − ui
Linear
F Nonlinear
k
k = F / ∆ (> 0) is the force needed to produce a unit stretch.

We only consider linear problems in this introductory

course.
Consider the equilibrium of forces for the spring. At node i,

we have
f i = − F = − k (u j − ui ) = kui − ku j
and at node j,
f j = F = k ( u j − ui ) = − kui + ku j
In matrix form,
 k − k   ui   f i 
− k  = 
 k  u j   f j 
or,
ku = f
where
k = (element) stiffness matrix
u = (element nodal) displacement vector
f = (element nodal) force vector
Note that k is symmetric. Is k singular or nonsingular? That is,
can we solve the equation? If not, why?

Spring System
x
k1 k2
1 2 3
u1, F1 u2, F2 u3, F3
For element 1,
 k1 − k1  u1   f 11 
− k  = 
 1 k1  u2   f 21 
element 2,
 k2 − k 2  u2   f 12 
− k  = 
 2 k 2  u3   f 22 
where f i m is the (internal) force acting on local node i of element

m (i = 1, 2).
Assemble the stiffness matrix for the whole system:
Consider the equilibrium of forces at node 1,
F1 = f 11
at node 2,
F2 = f 21 + f 12
and node 3,
F3 = f 22

That is,
F1 = k1u1 − k1u2
F2 = − k1u1 + ( k1 + k 2 )u2 − k 2 u3
F3 = − k 2 u2 + k 2 u3
In matrix form,
 k1 − k1 0  u1   F1 
− k    
k1 + k 2 − k 2 u2  =  F2 
 1 
 0 − k2 k 2 u3   F3 
or
KU = F
K is the stiffness matrix (structure matrix) for the spring system.
An alternative way of assembling the whole stiffness matrix:

“Enlarging” the stiffness matrices for elements 1 and 2, we
have
 k1 − k1 0  u1   f 11 
− k    
k1 0 u2  =  f 21 
 1 
 0 0 0 u3   0 
0 0 0   u1   0 
0 k    
− k 2  u2  =  f 12 
 2

0 − k 2 k 2  u3   f 22 

Adding the two matrix equations (superposition), we have
 k1 − k1 0  u1   f 11 
− k    
k1 + k 2 − k 2  u2  =  f 21 + f 12 
 1 
 0 − k2 k 2  u3   f 22 
This is the same equation we derived by using the force

equilibrium concept.
Boundary and load conditions:

Assuming, u1 = 0 and F2 = F3 = P
we have
 k1 − k1 0   0   F1 
− k    
k1 + k 2 − k 2  u2  =  P 
 1 
 0 − k2 k 2  u3   P 
which reduces to
 k1 + k 2 − k 2  u2   P 
 −k  = 
 2 k 2  u3   P 
and
F1 = − k1u2
Unknowns are
u2 
U=  and the reaction force F1 (if desired).
u3 

Solving the equations, we obtain the displacements

u2   2 P / k1 
  = 
 3 
u 2 P / k 1 + P / k 2
and the reaction force

F1 = −2 P
Checking the Results

• Deformed shape of the structure
• Balance of the external forces
• Order of magnitudes of the numbers
Notes About the Spring Elements

• Suitable for stiffness analysis
• Not suitable for stress analysis of the spring itself
• Can have spring elements with stiffness in the lateral
direction, spring elements for torsion, etc.

Example 1.1
k1 k2 P k3
x
1 2 3 4
Given: For the spring system shown above,

k1 = 100 N / mm, k 2 = 200 N / mm, k 3 = 100 N / mm
P = 500 N, u 1 = u4 = 0
Find: (a) the global stiffness matrix
(b) displacements of nodes 2 and 3
(c) the reaction forces at nodes 1 and 4
(d) the force in the spring 2
Solution:
(a) The element stiffness matrices are
 100 − 100
k1 =   (N/mm) (1)
− 100 100 
 200 − 200
k2 =   (N/mm) (2)
− 200 200 
 100 − 100
k3 =   (N/mm) (3)
− 100 100 

Applying the superposition concept, we obtain the global stiffness

matrix for the spring system as
u1 u2 u3 u4
 100 − 100 0 0 
− 100 100 + 200 − 200 0 
K= 
 0 − 200 200 + 100 − 100
 0 − 100 100 
 0
or
 100 − 100 0 0 
− 100 300 − 200 0 
K= 
 0 − 200 300 − 100
 0 − 100 100 
 0
which is symmetric and banded.

Equilibrium (FE) equation for the whole system is
 100 − 100 0 0   u1   F1 
− 100 300 − 200 0  u2   0 
   =   (4)
 0 − 200 300 − 100 u3   P 
 0 − 100 100  u4   F4 
 0
(b) Applying the BC ( u1 = u4 = 0 ) in Eq(4), or deleting the 1st and

4th rows and columns, we have

 300 − 200 u2   0 

− 200 300  u  =  P  (5)
  3   
Solving Eq.(5), we obtain
u2   P / 250  2 
 =  =   ( mm) (6)
 3 
u 3 P / 500  3
(c) From the 1st and 4th equations in (4), we get the reaction forces
F1 = −100u 2 = −200 (N)
F4 = −100u 3 = −300 (N )
(d) The FE equation for spring (element) 2 is
 200 − 200  ui   f i 
− 200 200  u  =  f 
  j   j 
Here i = 2, j = 3 for element 2. Thus we can calculate the spring
force as
u2 
F = f j = − f i = [ − 200 200]  
u3 
2 
= [ − 200 200]  
3
= 200 (N)
Check the results!

Example 1.2
4 k4 F1
k1 1
2
4 1 k2 F2 k3
2 3 3 5
x
Problem: For the spring system with arbitrarily numbered nodes

and elements, as shown above, find the global stiffness
matrix.
Solution:
First we construct the following
Element Connectivity Table

Element Node i (1) Node j (2)
1 4 2
2 2 3
3 3 5
4 2 1
which specifies the global node numbers corresponding to the

local node numbers for each element.
Then we can write the element stiffness matrices as follows

u4 u2 u2 u3
 k1 − k1   k2 − k2 
k1 =  k2 = 
− k 1 k1  − k 2 k 2 
u3 u5 u2 u1
 k3 − k3   k4 − k4 
k3 =  k4 = 
− k 3 k 3  − k 4 k 4 
Finally, applying the superposition method, we obtain the global

stiffness matrix as follows
u1 u2 u3 u4 u5
 k4 − k4 0 0 0 
− k k1 + k 2 + k 4 − k2 − k1 0 
 4 
K= 0 − k2 k2 + k3 0 − k3 
 0 − k1 0 k1 0 

 0 0 − k3 0 k 3 
The matrix is symmetric, banded, but singular.

Lecture Notes: Introduction to Finite Element Method Chapter 2. Bar and Beam Elements
Chapter 2. Bar and Beam Elements.

Linear Static Analysis
I. Linear Static Analysis
Most structural analysis problems can be treated as linear

static problems, based on the following assumptions
1. Small deformations (loading pattern is not changed due
to the deformed shape)
2. Elastic materials (no plasticity or failures)
3. Static loads (the load is applied to the structure in a slow
or steady fashion)
Linear analysis can provide most of the information about

the behavior of a structure, and can be a good approximation for
many analyses. It is also the bases of nonlinear analysis in most
of the cases.

II. Bar Element
Consider a uniform prismatic bar:
ui uj
fi i j fj
x A,E
L
L length
A cross-sectional area
E elastic modulus
u = u( x ) displacement
ε = ε ( x) strain
σ = σ ( x) stress
Strain-displacement relation:
du
ε= (1)
dx
Stress-strain relation:
σ = Eε (2)

Stiffness Matrix --- Direct Method

Assuming that the displacement u is varying linearly along
the axis of the bar, i.e.,
u( x ) =  1 −
x x
 ui + u j (3)
 L L
we have
u j − ui ∆
ε= = ( ∆ = elongation) (4)
L L
E∆
σ = Eε = (5)
L
We also have
F
σ= (F = force in bar) (6)
A
Thus, (5) and (6) lead to
EA
F= ∆ = k∆ (7)
L
EA
where k = is the stiffness of the bar.
L
The bar is acting like a spring in this case and we conclude
that element stiffness matrix is

 EA EA 
 k − k  L −
k= = L 
− k k  − EA EA 
 
 L L 
or
EA  1 − 1
k= (8)
L − 1 1 
This can be verified by considering the equilibrium of the forces
at the two nodes.
Element equilibrium equation is
EA  1 − 1  ui   f i 
   =  (9)
L − 1 1  u j   f j 
Degree of Freedom (dof)

Number of components of the displacement vector at a
node.
For 1-D bar element: one dof at each node.
Physical Meaning of the Coefficients in k

The jth column of k (here j = 1 or 2) represents the forces
applied to the bar to maintain a deformed shape with unit
displacement at node j and zero displacement at the other node.

Stiffness Matrix --- A Formal Approach

We derive the same stiffness matrix for the bar using a
formal approach which can be applied to many other more
complicated situations.
Define two linear shape functions as follows
N i (ξ ) = 1 − ξ , N j (ξ ) = ξ (10)
where
x
ξ= , 0≤ξ ≤1 (11)
L
From (3) we can write the displacement as
u( x ) = u(ξ ) = N i (ξ )ui + N j (ξ )u j
or
 ui 
u = Ni [ ]
N j   = Nu (12)
u j 
Strain is given by (1) and (12) as
du  d 
ε= =  Nu = Bu (13)
dx  dx 
where B is the element strain-displacement matrix, which is
dξ
B=
d
dx
[
N i (ξ ) N j (ξ ) =] d
dξ
[
N i (ξ ) ]
N j (ξ ) •
dx
i.e., B = [ − 1 / L 1 / L] (14)

Stress can be written as

σ = Eε = EBu (15)
Consider the strain energy stored in the bar
∫ ∫ (u B T EBu)dV
1 1
U= σ T εdV = T
2 2
V V
(16)
1 T 
2 
V
∫
= u  (B EB )dV u
T

where (13) and (15) have been used.

The work done by the two nodal forces is
1 1 1
W= f i ui + f j u j = u T f (17)
2 2 2
For conservative system, we state that
U =W (18)
which gives
1 T 
∫
u  (B EB )dV u = u T f
T 1
2   2
V
We can conclude that
 
∫
 (B EB )dV u = f
 V
T

or
ku = f (19)
where
k=
∫ (B
V
T
EB )dV (20)
is the element stiffness matrix.

Expression (20) is a general result which can be used for
the construction of other types of elements. This expression can
also be derived using other more rigorous approaches, such as
the Principle of Minimum Potential Energy, or the Galerkin’s
Method.
Now, we evaluate (20) for the bar element by using (14)
L
− 1 / L  EA  1 − 1
k=
∫
0

 1 / L
E[ − 1 / L 1 / L] Adx =
 L − 1 1 
which is the same as we derived using the direct method.

Note that from (16) and (20), the strain energy in the
element can be written as
1
U = u T ku (21)
2

Example 2.1
1 2A,E 2 A,E
1 2 P 3 x
L L
Problem: Find the stresses in the two bar assembly which is

loaded with force P, and constrained at the two ends,
as shown in the figure.
Solution: Use two 1-D bar elements.
Element 1,
u1 u2
2 EA  1 − 1
k1 =
L − 1 1 
Element 2,
u2 u3
EA  1 − 1
k2 =
L − 1 1 
Imagine a frictionless pin at node 2, which connects the two

elements. We can assemble the global FE equation as follows,
 2 − 2 0  u1   F1 
EA     
− 2 3 − 1 u2  =  F2 
L  
 0 − 1 1  u3   F3 

Load and boundary conditions (BC) are,

u1 = u 3 = 0, F2 = P
FE equation becomes,
 2 − 2 0   0   F1 
EA     
− 2 3 − 1 u2  =  P 
L  
 0 − 1 1   0   F3 
Deleting the 1st row and column, and the 3rd row and column,
we obtain,
EA
L
[ 3]{u2 } = { P}
Thus,
PL
u2 =
3EA
and
 u1  0
  PL  
u2  = 1
u  3EA 0
 3  
Stress in element 1 is
 u1 
σ 1 = Eε 1 = EB 1u1 = E[ − 1 / L 1 / L]  
u2 
u2 − u1 E  PL
− 0 =
P
=E = 
L L  3EA  3 A

Similarly, stress in element 2 is

u2 
σ 2 = Eε 2 = EB 2 u 2 = E [ − 1 / L 1 / L ] 
u3 
u3 − u2 E  PL  P
=E = 0 −  =−
L L 3EA  3A
which indicates that bar 2 is in compression.
Check the results!
Notes:
• In this case, the calculated stresses in elements 1 and 2
are exact within the linear theory for 1-D bar structures.
It will not help if we further divide element 1 or 2 into
smaller finite elements.
• For tapered bars, averaged values of the cross-sectional
areas should be used for the elements.
• We need to find the displacements first in order to find
the stresses, since we are using the displacement based
FEM.

Example 2.2
∆
1 A,E 2
1 2 P 3 x
L L
Problem: Determine the support reaction forces at the two ends

of the bar shown above, given the following,
P = 6.0 × 10 4 N , E = 2.0 × 10 4 N / mm 2 ,
A = 250 mm2 , L = 150 mm, ∆=1.2 mm
Solution:
We first check to see if or not the contact of the bar with
the wall on the right will occur. To do this, we imagine the wall
on the right is removed and calculate the displacement at the
right end,
PL (6.0 × 104 )(150)
∆0 = = = 18
. mm > ∆ = 12
. mm
EA (2.0 × 10 )(250)
4
Thus, contact occurs.

The global FE equation is found to be,
 1 − 1 0   u1   F1 
EA     
− 1 2 − 1 u2  =  F2 
L  
 0 − 1 1  u3   F3 

The load and boundary conditions are,

F2 = P = 6.0 × 10 4 N
u1 = 0, u3 = ∆ = 12
. mm
FE equation becomes,
 1 − 1 0   0   F1 
EA     
− 1 2 − 1 u2  =  P 
L  
 0 − 1 1   ∆   F3 
The 2nd equation gives,

EA u2 
[ 2 − 1] ∆  = { P}
L  
that is,
[ 2]{u2 } =  P + ∆ 
EA EA
L  L 
Solving this, we obtain
u2 =  + ∆ = 15
1 PL
. mm

2 EA 
and
 u1   0 
   
u2  = 15
.  ( mm)
u  12 
 3  . 

To calculate the support reaction forces, we apply the 1st

and 3rd equations in the global FE equation.
The 1st equation gives,
 u1 
EA   EA
F1 = [ 1 − 1 0] u 2  = ( − u 2 ) = −5.0 × 10 4 N
L u  L
 3
and the 3rd equation gives,
u1 
  EA
[ 0 − 1 1]u2  = ( − u2 + u3 )
EA
F3 =
L u  L
 3
= −10. × 10 4 N
Check the results.!

Distributed Load
i j
x
qL/2 qL/2
i j
Uniformly distributed axial load q (N/mm, N/m, lb/in) can

be converted to two equivalent nodal forces of magnitude qL/2.
We verify this by considering the work done by the load q,
L 1 1
∫ ∫ ∫
1 1 qL
Wq = uqdx = u(ξ )q ( Ldξ ) = u(ξ )dξ
2 2 2
0 0 0
1
 ui 
∫[N ( ) ]
qL
= i ξ N j (ξ )  dξ
2
0
u j 
1
 ui 
∫ [1 − ξ ]dξ 
qL
= ξ 
2 u
 j
0
1  qL qL   ui 
=   
2 2 2  u j 
qL / 2 
=
1
2
ui[ ]
uj 
qL / 2



that is,
1 qL / 2
Wq = u T f q with f q =   (22)
2 qL / 2 
Thus, from the U=W concept for the element, we have
1 T 1 1
u ku = u T f + u T f q (23)
2 2 2
which yields
ku = f + f q (24)
The new nodal force vector is
 f i + qL / 2 
f + fq =   (25)
f
 j + qL / 2 
In an assembly of bars,
1 2 3
qL/2 qL qL/2
1 2 3

Bar Elements in 2-D and 3-D Space

2-D Case
x
y j
Y
ui’ θ
i vi
ui
Local Global
x, y X, Y
ui' , v i' ui , v i
1 dof at node 2 dof’s at node
Note: Lateral displacement vi’ does not contribute to the stretch

of the bar, within the linear theory.
Transformation
u 
ui' = ui cos θ + v i sin θ = [ l m]  i 
v i 
u 
v i' = − ui sin θ + v i cos θ = [ − m l ]  i 
v i 
where l = cosθ , m = sin θ .

In matrix form,
ui'   l m ui 
 ' =   v  (26)
 i
v  − m l  i 
or,
~
u i' = Tu i
where the transformation matrix
~  l m
T=  (27)
− m l 
~ ~
is orthogonal, that is, T −1 = T T .
For the two nodes of the bar element, we have
 ui'   l m 0 0   ui 
 ' 
 v i  − m l 0 0   v i 
 ' =   (28)
u j   0 0 l m u j 
v 'j   0 0 − m l  v j 
or,
~
T 0 
u ' = Tu with T =  ~ (29)
 0 T
The nodal forces are transformed in the same way,
f ' = Tf (30)

Stiffness Matrix in the 2-D Space

In the local coordinate system, we have
EA  1 − 1  ui   f i 
' '
   '  =  '
L − 1 1  u j   f j 
Augmenting this equation, we write
1 0 − 1 0  ui'   f i ' 
 

EA  0 0 0 0  vi'   0 
 '  =  
L − 1 0 1 0 u j   f j' 
0 0 0 0 v 'j   0 

or,
k 'u ' = f '
Using transformations given in (29) and (30), we obtain
k ' Tu = Tf
Multiplying both sides by TT and noticing that TTT = I, we
obtain
T T k ' Tu = f (31)
Thus, the element stiffness matrix k in the global coordinate
system is
k = TT k 'T (32)
which is a 4×4 symmetric matrix.

Explicit form,
ui vi uj vj
 l2 lm − l 2 − lm 
 
EA  lm m2 − lm − m2  (33)
k=
L  − l 2 − lm l 2 lm 
 2 
 − lm − m 2
lm m 
Calculation of the directional cosines l and m:
X j − Xi Yj − Yi
l = cosθ = , m = sin θ = (34)
L L
The structure stiffness matrix is assembled by using the element
stiffness matrices in the usual way as in the 1-D case.
Element Stress
 ui 
 ui'   
1   l m 0 0   vi 
= Eε = EB  '  = E −
1
 
L  0 0 l m u j 
σ
u j   L
v j 
That is,
 ui 
v 
E  i
σ = [ − l − m l m]   (35)
L u j 
v j 

Example 2.3
3
A simple plane truss is made
45o
of two identical bars (with E, A, and
L), and loaded as shown in the 2 P2
figure. Find
1) displacement of node 2; Y 2 P1
1
2) stress in each bar.
45o
Solution: X
1
This simple structure is used
here to demonstrate the assembly
and solution process using the bar element in 2-D space.
In local coordinate systems, we have
EA  1 − 1
k 1' = = k '
L − 1 1 
2
These two matrices cannot be assembled together, because they

are in different coordinate systems. We need to convert them to
global coordinate system OXY.
Element 1:
2
θ = 45o , l = m =
2
Using formula (32) or (33), we obtain the stiffness matrix in the
global system

u1 v1 u2 v2
1 1 − 1 − 1
1 1 − 1 − 1 
EA  
k 1 = T1T k 1' T1 =
2 L − 1 − 1 1 1
− 1 − 1 1 1 

Element 2:
2 2
θ = 135o , l = − , m=
2 2
We have,
u2 v2 u3 v3
 1 −1 −1 1 
 1 − 1
EA − 1 1 
k 2 = T2 k 2 T2 =
T '
2 L − 1 1 1 − 1
 1 −1 −1 1 
 
Assemble the structure FE equation,
u1 v1 u2 v2 u3 v3
1 1 −1 −1 0 0  u1   F1 X 
1 1 −1 −1 0 0   v1   F1Y 
    
EA − 1 − 1 2 0 − 1 1  u2   F2 X 
   = 
2 L − 1 − 1 0 2 1 − 1 v2   F2Y 
0 0 −1 1 1 − 1 u3   F3 X 
    
 0 0 1 − 1 − 1 1   3   3Y 
v F

Load and boundary conditions (BC):

u1 = v1 = u3 = v3 = 0, F2 X = P1 , F2Y = P2
Condensed FE equation,
EA 2 0 u2   P1 
   = 
2 L 0 2 v2   P2 
Solving this, we obtain the displacement of node 2,
u2  L  P1 
 =  
 2
v EA  P2 
Using formula (35), we calculate the stresses in the two bars,
0
 
E 2 L 0 2
σ1 = [ − 1 − 1 1 1] P  = ( P1 + P2 )
L 2 EA  1  2 A
 P2 
 P1 
 
E 2 L  P2  2
σ2 = [1 − 1 − 1 1]  0  = ( P1 − P2 )
L 2 EA   2 A
 0 
Check the results:

Look for the equilibrium conditions, symmetry,
antisymmetry, etc.

Example 2.4 (Multipoint Constraint)
y’
x’
P 3
2 2
1 Y
L
3
1 45o
X
For the plane truss shown above,

P = 1000 kN, L = 1 m, E = 210 GPa ,
A = 6.0 × 10 − 4 m2 for elements 1 and 2,
A = 6 2 × 10 − 4 m2 for element 3.
Determine the displacements and reaction forces.
Solution:
We have an inclined roller at node 3, which needs special
attention in the FE solution. We first assemble the global FE
equation for the truss.
Element 1:
θ = 90 o , l = 0, m=1

u1 v1 u2 v2
0 0 0 0
 0 − 1
( 210 × 109 )(6.0 × 10 − 4 ) 0 1  ( N / m)
k1 =
1 0 0 0 0
0 − 1 0 1 

Element 2:
θ = 0o , l = 1, m = 0
u2 v 2 u3 v 3
1 0 − 1 0
 0 0 0
(210 × 10 9 )(6.0 × 10 − 4 )  0  ( N / m)
k2 =
1 − 1 0 1 0
0 0 0 0

Element 3:
1 1
θ = 45o , l = , m=
2 2
u1 v1 u3 v3
 0.5 0.5 − 0.5 − 0.5
 0.5 − 0.5 − 0.5
(210 × 10 9 )(6 2 × 10 − 4 )  0.5 
k3 =
2 − 0.5 − 0.5 0.5 0.5 
− 0.5 − 0.5 0.5 0.5 

( N / m)

The global FE equation is,

 0.5 0.5 0 0 − 0.5 − 0.5  u1   F1 X 
 . 0 − 1 − 0.5 − 0.5  v1   F1Y 
15
    
5
1 0 −1 0  u2   F2 X 
1260 × 10   v  =  F 
 1 0 0   2   2Y 
 15
. 0.5  u3   F3 X 
    
 Sym. 0.5   3   3Y 
v F
Load and boundary conditions (BC):

u1 = v1 = v2 = 0, and v3' = 0,
F2 X = P, F3 x ' = 0.
From the transformation relation and the BC, we have

 2 2  u3  2
v3' = −    = ( − u3 + v3 ) = 0,
 2 2   v3  2
that is,
u3 − v3 = 0
This is a multipoint constraint (MPC).
Similarly, we have a relation for the force at node 3,
 2 2   F3 X  2
F3 x ' =    = ( F3 X + F3Y ) = 0,
 2 2   F3Y  2
that is,
F3 X + F3Y = 0

Applying the load and BC’s in the structure FE equation by

‘deleting’ 1st, 2nd and 4th rows and columns, we have
 1 − 1 0  u2   P 
   
1260 × 105 − 1 15
. 0.5 u3  =  F3 X 
 
 0 0.5 0.5 v3   F3Y 
Further, from the MPC and the force relation at node 3, the
equation becomes,
 1 − 1 0 u2   P 
   
. 0.5u3  =  F3 X 
1260 × 105 − 1 15
 
 0 0.5 0.5u3  − F3 X 
which is
 1 − 1  P 
  
u 
1260 × 105 − 1 2   2  =  F3 X 
  u3 
 0 1  − F 
 3X 
The 3rd equation yields,
F3 X = −1260 × 105 u3
Substituting this into the 2nd equation and rearranging, we have
 1 − 1 u2   P 
1260 × 105    = 
− 1 3  u3   0 
Solving this, we obtain the displacements,

u 2  1 3 P   0.01191 
  = 5  =  ( m)
 3  2520 × 10   
u P 0.003968 
From the global FE equation, we can calculate the reaction
forces,
 F1 X   0 − 0.5 − 0.5 − 500
F   0 − 0.5 − 0.5 u − 500
 1Y
   2
    
 F2Y  = 1260 × 10  0 0  u3  =  0.0  ( kN )
5
0
F  − 1 15  v  − 500
 3X
  . 0.5  3   
 F3Y   0 0.5 0.5   500 
Check the results!
A general multipoint constraint (MPC) can be described as,
∑ Aj u j = 0
j
where Aj’s are constants and uj’s are nodal displacement

components. In the FE software, such as MSC/NASTRAN,
users only need to specify this relation to the software. The
software will take care of the solution.
Penalty Approach for Handling BC’s and MPC’s

3-D Case
y
x j
Y
i
z
X
Z
Local Global
x, y, z X, Y, Z
ui' , vi' , wi' ui , vi , wi
1 dof at node 3 dof’s at node
Element stiffness matrices are calculated in the local

coordinate systems and then transformed into the global
coordinate system (X, Y, Z) where they are assembled.
FEA software packages will do this transformation

automatically.
Input data for bar elements:
• (X, Y, Z) for each node
• E and A for each element

III. Beam Element
Simple Plane Beam Element

y
vi, Fi vj, Fj
i j
E,I x
θi, Mi θj, Mj
L
L length
I moment of inertia of the cross-sectional area
E elastic modulus
v = v( x ) deflection (lateral displacement) of the
neutral axis
dv
θ= rotation about the z-axis
dx
F = F ( x) shear force
M = M ( x) moment about z-axis
Elementary Beam Theory:

d 2v
EI 2 = M ( x ) (36)
dx
My
σ =− (37)
I

Direct Method
Using the results from elementary beam theory to compute
each column of the stiffness matrix.
(Fig. 2.3-1. on Page 21 of Cook’s Book)
Element stiffness equation (local node: i, j or 1, 2):

vi θi vj θj
 12 6 L − 12 6 L   vi   Fi 
 2    
EI  6 L 4 L − 6 L 2 L  θi   M i 
2
(38)
 = 
L 
3
− 12 − 6 L 12 − 6 L  v j   Fj 
 6 L 2 L2 − 6 L 4 L2  θ   M 
  j   j 

Formal Approach
Apply the formula,
L
k=
∫
0
B T EIBdx (39)
To derive this, we introduce the shape functions,

N 1 ( x ) = 1 − 3x 2 / L2 + 2 x 3 / L3
N 2 ( x ) = x − 2 x 2 / L + x 3 / L2
(40)
N 3 ( x ) = 3x / L − 2 x / L
2 2 3 3
N 4 ( x ) = − x 2 / L + x 3 / L2
Then, we can represent the deflection as,
v ( x ) = Nu
 vi 
θ 
 i (41)
= [ N1 ( x) N 2 ( x) N 3 ( x) N 4 ( x )]  
v j 
θ j 
which is a cubic function. Notice that,

N1 + N 3 = 1
N2 + N3 L + N4 = x
which implies that the rigid body motion is represented by the
assumed deformed shape of the beam.

Curvature of the beam is,

d 2v d 2
2
= 2 Nu = Bu (42)
dx dx
where the strain-displacement matrix B is given by,
d2
[
B = 2 N = N 1" ( x ) N 2" ( x ) N 3" ( x ) N 4" ( x )
dx
]
(43)
= − 2 + 3 − + 2
6 12 x 4 6 x 6 12 x 2 6x 
− − + 2
 L 
2 3
L L L L L L L
Strain energy stored in the beam element is
L T
 My  1  My 
∫ ∫∫
1 1
U= σ T εdV = −  −  dAdx
2 2  I  E I 
V 0 A
L L T
1  d 2v   d 2v 
∫ ∫
1 T 1
= M Mdx =   EI  2 dx
2 EI 2  dx 2   dx 
0 0
L
∫
1
= (Bu)T EI (Bu)dx
2
0
1 T  
L
2  ∫
= u  B EIBdx u
T
 0
We conclude that the stiffness matrix for the simple beam

element is
L
k=
∫
0
B T EIBdx

Applying the result in (43) and carrying out the integration, we

arrive at the same stiffness matrix as given in (38).
Combining the axial stiffness (bar element), we obtain the

stiffness matrix of a general 2-D beam element,
ui vi θi uj vj θj
 EA 0 0 −
EA
0 0 
 L L
 12 EI 6 EI 12 EI 6 EI 
 0 0 − 
 L3 L2 L3 L2 
 0 6 EI 4 EI 6 EI 2 EI 
0 − 2
k=  L 2
L L L 
 EA EA 
− L 0 0
L
0 0 
 12 EI 6 EI 12 EI 6 EI 
 0 − 3 − 0 − 2 
 L L2 L3 L 
 0 6 EI 2 EI 6 EI 4 EI 
0 − 2
 L2 L L L 
3-D Beam Element

The element stiffness matrix is formed in the local (2-D)
coordinate system first and then transformed into the global (3-
D) coordinate system to be assembled.
(Fig. 2.3-2. On Page 24)

Example 2.5
Y
P
M
1 2
1 E,I 2 3 X
L L
Given: The beam shown above is clamped at the two ends and
acted upon by the force P and moment M in the mid-
span.
Find: The deflection and rotation at the center node and the
reaction forces and moments at the two ends.
Solution: Element stiffness matrices are,
v1 θ1 v2 θ2
 12 6 L − 12 6 L 
 2 
EI  6 L 4 L − 6 L 2 L 
2
k1 = 3
L − 12 − 6 L 12 − 6 L 
 6 L 2 L2 − 6 L 4 L2 
 
v2 θ2 v3 θ3
 12 6 L − 12 6 L 
 2 
EI  6 L 4 L − 6 L 2 L 
2
k2 = 3
L − 12 − 6 L 12 − 6 L 
 6 L 2 L2 − 6 L 4 L2 
 

Global FE equation is,

v1 θ1 v2 θ2 v3 θ3
 12 6 L − 12 6 L 0 0   v1   F1Y 
 6 L 4 L2 − 6 L 2 L2 0 0  θ1   M 1 
    
EI − 12 − 6 L 24 0 − 12 6 L  v2   F2Y 
 = 
L3  6 L 2 L2 0 8 L2 − 6 L 2 L2  θ 2   M 2 
 0 0 − 12 − 6 L 12 − 6 L  v3   F3Y 
 2    
 0 0 6 L 2 L 2
− 6 L 4 L θ
 3   M 3 
Loads and constraints (BC’s) are,
F2Y = − P , M2 = M ,
v1 = v3 = θ1 = θ 3 = 0
Reduced FE equation,
EI 24 0  v2  − P 
L3  0 8 L2  θ  =  M 
  2   
Solving this we obtain,
v 2  L − PL 
2
 =  
θ
 2 24 EI  3 M 
From global FE equation, we obtain the reaction forces and
moments,

 F1Y   − 12 6 L  2 P + 3 M / L
M   2   
 1  EI − 6 L 2 L  v2  1  PL + M 
 = 3  =  
F
 3Y  L  − 12 − 6 L  θ 2  4 2 P − 3 M / L 
 M 3   6L 2 L2   − PL + M 

Stresses in the beam at the two ends can be calculated using the
formula,
My
σ = σx = −
I
Note that the FE solution is exact according to the simple beam
theory, since no distributed load is present between the nodes.
Recall that,
d 2v
EI 2 = M ( x )
dx
and
dM
= V (V - shear force in the beam)
dx
dV
= q (q - distributed load on the beam)
dx
Thus,
d 4v
EI 4 = q( x )
dx
If q(x)=0, then exact solution for the deflection v is a cubic
function of x, which is what described by our shape functions.

Equivalent Nodal Loads of Distributed Transverse Load
i x L j
qL/2 qL/2
qL2/12
qL2/12
i j
This can be verified by considering the work done by the

distributed load q.
L L
qL qL/2
qL2/12
L L

Example 2.6
y
p
1 E,I 2 x
L
Given: A cantilever beam with distributed lateral load p as

shown above.
Find: The deflection and rotation at the right end, the
reaction force and moment at the left end.
Solution: The work-equivalent nodal loads are shown below,
y
f
m
1 E,I 2 x
L
where
f = pL / 2, m = pL2 / 12
Applying the FE equation, we have

 12 6 L − 12 6 L  v1   F1Y 
 2    
EI  6 L 4 L − 6 L 2 L  θ1   M 1 
2
 = 
L3 − 12 − 6 L 12 − 6 L  v2   F2 Y 
 6 L 2 L2 − 6 L 4 L2  θ   M 
  2   2 
Load and constraints (BC’s) are,
F2 Y = − f , M2 = m
v1 = θ1 = 0
Reduced equation is,
EI  12 − 6 L v2  − f 
L3 − 6 L 4 L2  θ  =  m 
  2   
Solving this, we obtain,
v 2  L − 2 L f + 3 Lm − pL / 8 EI 
2 4
 =  =  (A)
θ
 2 6 EI  − 3 Lf + 6m   − pL3
/ 6 EI 
These nodal values are the same as the exact solution.
Note that the deflection v(x) (for 0 < x< 0) in the beam by the
FEM is, however, different from that by the exact solution. The
exact solution by the simple beam theory is a 4th order
polynomial of x, while the FE solution of v is only a 3rd order
polynomial of x.
If the equivalent moment m is ignored, we have,
v 2  L − 2 L f  − pL / 6 EI 
2 4
 =  =  (B)
θ
 2 6 EI  − 3 Lf   − pL3
/ 4 EI 

The errors in (B) will decrease if more elements are used. The
equivalent moment m is often ignored in the FEM applications.
The FE solutions still converge as more elements are applied.
From the FE equation, we can calculate the reaction force
and moment as,
 F1Y  L3  − 12 6 L  v2   pL / 2 
 =  2  = 
 1
M EI  − 6 L 2 L θ
 2  5 pL2
/ 12 
where the result in (A) is used. This force vector gives the total
effective nodal forces which include the equivalent nodal forces
for the distributed lateral load p given by,
 − pL / 2 
 
− pL / 12 
2
The correct reaction forces can be obtained as follows,

 F1Y   pL / 2   − pL / 2   pL 
 = − = 2 
 1 
M 5 pL2
/ 12   − pL2
/ 12   pL / 2 
Check the results!

Example 2.7
Y
P
1 E,I 2
2 X
1 3 k
L L 4
Given: P = 50 kN, k = 200 kN/m, L = 3 m,

E = 210 GPa, I = 2×10-4 m4.
Find: Deflections, rotations and reaction forces.
Solution:
The beam has a roller (or hinge) support at node 2 and a
spring support at node 3. We use two beam elements and one
spring element to solve this problem.
The spring stiffness matrix is given by,
v3 v4
 k − k
ks = 
− k k 
Adding this stiffness matrix to the global FE equation (see

Example 2.5), we have

v1 θ1 v2 θ2 v3 θ3 v4
12 6 L − 12 6 L 0 0 0  v1   F1Y 
 4 L2 − 6 L 2 L2 0 0 0 θ1   M 1 
    
 24 0 − 12 6L 0 v2   F2Y 
EI      
8 L 2
− 6L 2L 2
0 θ2  =  M 2 
L 
3
   
 12 + k ' − 6 L − k 'v3   F3Y 
 
 4 L2 0 θ3   M 3 
 Symmetry k ' v4   F4Y 
in which
L3
k '= k
EI
is used to simply the notation.
We now apply the boundary conditions,
v1 = θ1 = v2 = v4 = 0,
M 2 = M 3 = 0, F3Y = − P
‘Deleting’ the first three and seventh equations (rows and
columns), we have the following reduced equation,
 8 L2 − 6 L 2 L2  θ 2   0 
EI     
− 6 L 12 + k ' − 6 L   v 3  = − P 
L3  2
 2 L − 6 L 4 L2  θ3   0 
Solving this equation, we obtain the deflection and rotations at

node 2 and node 3,

θ 2  3
  PL 2
 
v3  = − 7 L 
θ  EI (12 + 7 k ' )  
 3 9
The influence of the spring k is easily seen from this result.
Plugging in the given numbers, we can calculate
θ 2  − 0.002492 rad 
   
 3 
v = − 0 .01744 m 
θ   − 0.007475 rad 
 3  
From the global FE equation, we obtain the nodal reaction
forces as,
 F1Y   − 69.78 kN 
 M  − 69.78 kN ⋅ m
 1  
 = 
F
 2Y   116 .2 kN 
 F4Y   3.488 kN 
Checking the results: Draw free body diagram of the beam
69.78 kN 50 kN
1 2 3
69.78 kN⋅m
116.2 kN 3.488 kN

Lecture Notes: Introduction to Finite Element Method Chapter 3. Two-Dimensional Problems
Chapter 3. Two-Dimensional Problems
I. Review of the Basic Theory
In general, the stresses and strains in a structure consist of

six components:
σ x , σ y , σ z , τ xy , τ yz , τ zx for stresses,
and
ε x , ε y , ε z , γ xy , γ yz , γ zx for strains.
σy
τ xy
τ yz
σx
τ zx
y
σz
x
z
Under contain conditions, the state of stresses and strains

can be simplified. A general 3-D structure analysis can,
therefore, be reduced to a 2-D analysis.

Plane (2-D) Problems

• Plane stress:
σ z = τ yz = τ zx = 0 (ε z ≠ 0) (1)
A thin planar structure with constant thickness and

loading within the plane of the structure (xy-plane).
y y
p
x z
• Plane strain:
ε z = γ yz = γ zx = 0 (σ z ≠ 0) (2)
A long structure with a uniform cross section and

transverse loading along its length (z-direction).
y y
p
x z

Stress-Strain-Temperature (Constitutive) Relations

For elastic and isotropic materials, we have,
 εx   1/ E −ν / E 0  σ x   ε x 0 
      
 ε y  = − ν / E 1/ E 0  σ y  +  ε y 0  (3)

γ   0 1 / G  τ xy  γ xy 0 
 xy   0
or,
ε = E −1σ + ε0
where ε 0 is the initial strain, E the Young’s modulus, ν the
Poisson’s ratio and G the shear modulus. Note that,
E
G= (4)
2(1 + ν )
which means that there are only two independent materials
constants for homogeneous and isotropic materials.
We can also express stresses in terms of strains by solving
the above equation,
σ x  1 ν 0   ε x   ε x 0 
 
σ =
E ν 1   ε  −  ε 
 y 
0
  y   y 0 
(5)
τ  1 − ν
2
 xy   0 0 (1 − ν ) / 2 γ xy  γ xy 0 
or,
σ = Eε + σ 0
where σ 0 = −Eε 0 is the initial stress.

The above relations are valid for plane stress case. For
plane strain case, we need to replace the material constants in
the above equations in the following fashion,
E
E→
1− ν 2
ν
ν→ (6)
1− ν
G→G
For example, the stress is related to strain by
σ x  1 − ν ν 0   ε x   ε x 0 
  E  ν     
σ
 y = 1−ν ε − ε
  y   y 0 
0
(1 + ν )(1 − 2ν ) 
τ 
 xy   0 0 (1 − 2ν ) / 2  γ xy  γ xy 0 
in the plane strain case.
Initial strains due to temperature change (thermal loading)

is given by,
 ε x 0  α∆T 
   
ε
 y0   = α∆T  (7)
γ   0 
 xy 0   
where α is the coefficient of thermal expansion, ∆T the change
of temperature. Note that if the structure is free to deform under
thermal loading, there will be no (elastic) stresses in the
structure.

Strain and Displacement Relations

For small strains and small rotations, we have,
∂u ∂v ∂u ∂v
εx = , ε y = , γ xy = +
∂x ∂y ∂y ∂x
In matrix form,
 ε x  ∂ / ∂x 0 
    u 
εy  =  0 ∂ / ∂y   , or ε = Du

(8)
γ  ∂ / ∂y ∂ / ∂x  v 
 xy   
From this relation, we know that the strains (and thus
stresses) are one order lower than the displacements, if the
displacements are represented by polynomials.
Equilibrium Equations
In elasticity theory, the stresses in the structure must satisfy
the following equilibrium equations,
∂σ x ∂τ xy
+ + fx = 0
∂x ∂y
(9)
∂τ xy ∂σ y
+ + fy = 0
∂x ∂y
where fx and fy are body forces (such as gravity forces) per unit
volume. In FEM, these equilibrium conditions are satisfied in
an approximate sense.

Boundary Conditions
ty
p
y tx
St
Su
x
The boundary S of the body can be divided into two parts,

Su and St. The boundary conditions (BC’s) are described as,
u = u, v = v, on S u
(10)
tx = tx , ty = ty , on S t
in which tx and ty are traction forces (stresses on the boundary)

and the barred quantities are those with known values.
In FEM, all types of loads (distributed surface loads, body
forces, concentrated forces and moments, etc.) are converted to
point forces acting at the nodes.
Exact Elasticity Solution

The exact solution (displacements, strains and stresses) of a
given problem must satisfy the equilibrium equations (9), the
given boundary conditions (10) and compatibility conditions
(structures should deform in a continuous manner, no cracks and
overlaps in the obtained displacement fields).

Example 3.1
A plate is supported and loaded with distributed force p as
shown in the figure. The material constants are E and ν.
The exact solution for this simple problem can be found

easily as follows,
Displacement:
p p
u= x, v = −ν y
E E
Strain:
p p
εx = , ε y = −ν , γ xy = 0
E E
Stress:
σ x = p, σ y = 0, τ xy = 0
Exact (or analytical) solutions for simple problems are

numbered (suppose there is a hole in the plate!). That is why we
need FEM!

II. Finite Elements for 2-D Problems
A General Formula for the Stiffness Matrix

Displacements (u, v) in a plane element are interpolated
from nodal displacements (ui, vi) using shape functions Ni as
follows,
 u1 
v 
L  
1
u   N 1 0 N2 0
 = u  or u = Nd (11)
v   0 N1 0 N2 L  2 
v
 2
 M 
where N is the shape function matrix, u the displacement vector

and d the nodal displacement vector. Here we have assumed
that u depends on the nodal values of u only, and v on nodal
values of v only.
From strain-displacement relation (Eq.(8)), the strain vector
is,
ε = Du = DNd, or ε = Bd (12)
where B = DN is the strain-displacement matrix.

Consider the strain energy stored in an element,
∫ ∫ (σ ε x + σ y ε y + τ xy γ xy )dV
1 1
U= σ T ε dV = x
2 2
V V
∫ (Eε ) ∫
1 1 T
= ε dV = ε Eε dV
T
2 2
V V
∫
1
= d T B T EB dV d
2
V
1
= d T kd
2
From this, we obtain the general formula for the element
stiffness matrix,
k=
∫
V
B T EB dV (13)
Note that unlike the 1-D cases, E here is a matrix which is given
by the stress-strain relation (e.g., Eq.(5) for plane stress).
The stiffness matrix k defined by (13) is symmetric since E
is symmetric. Also note that given the material property, the
behavior of k depends on the B matrix only, which in turn on
the shape functions. Thus, the quality of finite elements in
representing the behavior of a structure is entirely determined by
the choice of shape functions.
Most commonly employed 2-D elements are linear or
quadratic triangles and quadrilaterals.

Constant Strain Triangle (CST or T3)

This is the simplest 2-D element, which is also called
linear triangular element.
v3
3
(x3, y3) u3
y
v
v2
v1 u 2
(x, y) u2
1 (x2, y2)
u1
(x1, y1)
x
Linear Triangular Element
For this element, we have three nodes at the vertices of the

triangle, which are numbered around the element in the
counterclockwise direction. Each node has two degrees of
freedom (can move in the x and y directions). The
displacements u and v are assumed to be linear functions within
the element, that is,
u = b1 + b2 x + b3 y , v = b4 + b5 x + b6 y (14)
where bi (i = 1, 2, ..., 6) are constants. From these, the strains
are found to be,
ε x = b2 , ε y = b6 , γ xy = b3 + b5 (15)
which are constant throughout the element. Thus, we have the

name “constant strain triangle” (CST).

Displacements given by (14) should satisfy the following

six equations,
u1 = b1 + b2 x1 + b3 y1
u2 = b1 + b2 x 2 + b3 y 2
M
v 3 = b4 + b5 x3 + b6 y3
Solving these equations, we can find the coefficients b1, b2, ...,
and b6 in terms of nodal displacements and coordinates.
Substituting these coefficients into (14) and rearranging the
terms, we obtain,
 u1 
v 
 1
u N1 0 N2 0 N3 0  u2 
 =   (16)
v   0 N1 0 N2 0 N 3  v 2 
 u3 
 
v3 
where the shape functions (linear functions in x and y) are
1
N1 = {( x2 y3 − x3 y 2 ) + ( y2 − y3 ) x + ( x3 − x 2 ) y}
2A
1
N2 = {( x3 y1 − x1 y3 ) + ( y3 − y1 ) x + ( x1 − x3 ) y} (17)
2A
1
N3 = {( x1 y2 − x 2 y1 ) + ( y1 − y2 ) x + ( x2 − x1 ) y}
2A
and

1 x1 y1 
A = det 1 x 2 y2 
1
(18)
2  
1 x3 y 3 
is the area of the triangle (Prove this!).

Using the strain-displacement relation (8), results (16) and
(17), we have,
 u1 
v 
 εx   y 23 0 y31 0 y12 0  1 
  1  u 
ε
 y = Bd = 0 x32 0 x13 0 x 21   2  (19)
2A   v2
γ   x32 y 23 x13 y 31 x 21 y12   
 xy   u3 
 
v3 
where xij = xi - xj and yij = yi - yj (i, j = 1, 2, 3). Again, we see
constant strains within the element. From stress-strain relation
(Eq.(5), for example), we see that stresses obtained using the
CST element are also constant.
Applying formula (13), we obtain the element stiffness
matrix for the CST element,
k=
∫
V
B T EB dV = tA( B T EB ) (20)
in which t is the thickness of the element. Notice that k for CST

is a 6 by 6 symmetric matrix. The matrix multiplication in (20)
can be carried out by a computer program.

Both the expressions of the shape functions in (17) and

their derivations are lengthy and offer little insight into the
behavior of the element.
η=0
ξ=0 η=b
3
ξ=a η=1
ξ=1 (a, b)
2
1
The Natural Coordinates
We introduce the natural coordinates (ξ , η ) on the

triangle, then the shape functions can be represented simply by,
N1 = ξ , N 2 = η, N 3 = 1 − ξ − η (21)
Notice that,
N1 + N 2 + N 3 = 1 (22)
which ensures that the rigid body translation is represented by
the chosen shape functions. Also, as in the 1-D case,
1, at node i;
Ni =  (23)
0, at the other nodes
and varies linearly within the element. The plot for shape
function N1 is shown in the following figure. N2 and N3 have
similar features.

ξ=0
3
N1
ξ=1
1
2
1
Shape Function N1 for CST
We have two coordinate systems for the element: the global

coordinates (x, y) and the natural coordinates (ξ , η ) . The
relation between the two is given by
x = N1x1 + N 2 x 2 + N 3 x3
(24)
y = N1 y1 + N 2 y 2 + N 3 y3
or,
x = x13ξ + x 23η + x 3
(25)
y = y13ξ + y 23η + y 3
where xij = xi - xj and yij = yi - yj (i, j = 1, 2, 3) as defined earlier.
Displacement u or v on the element can be viewed as
functions of (x, y) or (ξ , η ) . Using the chain rule for derivatives,
we have,
∂ u  ∂ x ∂ y  ∂ u  ∂ u 
 ∂ ξ   ∂ ξ ∂ ξ   ∂ x   ∂ x 
∂ u  = ∂ x ∂ y   ∂ u 
= J 
∂u
(26)
      
∂ η  ∂ η ∂ η  ∂ y  ∂ y 
where J is called the Jacobian matrix of the transformation.

From (25), we calculate,

x y13  1  y 23 − y13 
J =  13  , J −1 = (27)
 x 23 y 23  2 A − x 23 x13 
where det J = x13 y 23 − x 23 y13 = 2 A has been used (A is the area of

the triangular element. Prove this!).
From (26), (27), (16) and (21) we have,
∂ u  ∂ u 
 ∂ x  1  y 23 − y13   ∂ ξ 
∂ u  = − x  
  2 A  23 x13   ∂ u 
∂ y  ∂ η  (28)
1  y 23 − y13   u1 − u3 
=  
2 A − x 23 x13  u2 − u3 
Similarly,
∂ v 
 ∂ x  1  y 23 − y13  v1 − v 3 
∂ v  = − x   (29)
  2 A  23 x13 v 2 − v3 
∂ y 
Using the results in (28) and (29), and the relations

ε = Du = DNd = Bd, we obtain the strain-displacement matrix,
 y 23 0 y 31 0 y12 0
1 
B= 0 x 32 0 x13 0 x 21  (30)
2A  
 x32 y 23 x13 y 31 x 21 y12 
which is the same as we derived earlier in (19).

Applications of the CST Element:

• Use in areas where the strain gradient is small.
• Use in mesh transition areas (fine mesh to coarse mesh).
• Avoid using CST in stress concentration or other crucial
areas in the structure, such as edges of holes and corners.
• Recommended for quick and preliminary FE analysis of
2-D problems.

Linear Strain Triangle (LST or T6)

This element is also called quadratic triangular element.
v3
u3
3
v6 v5
y
u6 6 5 u5
v2
v1
u4 2 u2
1 4
u1 v4
x
Quadratic Triangular Element
There are six nodes on this element: three corner nodes and
three midside nodes. Each node has two degrees of freedom
(DOF) as before. The displacements (u, v) are assumed to be
quadratic functions of (x, y),
u = b1 + b2 x + b3 y + b4 x 2 + b5 xy + b6 y 2
(31)
v = b7 + b8 x + b9 y + b10 x + b11 xy + b12 y 2 2
where bi (i = 1, 2, ..., 12) are constants. From these, the strains

are found to be,
ε x = b2 + 2b4 x + b5 y
ε y = b9 + b11 x + 2b12 y (32)
γ xy = ( b3 + b8 ) + ( b5 + 2b10 ) x + ( 2b6 + b11 ) y
which are linear functions. Thus, we have the “linear strain

triangle” (LST), which provides better results than the CST.
In the natural coordinate system we defined earlier, the six

shape functions for the LST element are,
N 1 = ξ ( 2ξ − 1)
N 2 = η( 2η − 1)
N 3 = ζ ( 2ζ − 1)
(33)
N 4 = 4ξ η
N 5 = 4ηζ
N 6 = 4ζ ξ
in which ζ = 1 − ξ − η . Each of these six shape functions
represents a quadratic form on the element as shown in the
figure.
ξ=0
3
ξ=1/2
6 5
ξ=1 N1
1
4 2
1
Shape Function N1 for LST
Displacements can be written as,

6 6
u = ∑ N i ui , v = ∑ N i vi (34)
i =1 i =1
The element stiffness matrix is still given by

k = ∫ B T EB dV , but here BTEB is quadratic in x and y. In
V
general, the integral has to be computed numerically.
Linear Quadrilateral Element (Q4)

η v3
v4 u3
η =1 u4
3 ξ
4
v2
v1 2 u2
1
y u1
η = −1
ξ = −1 ξ =1
x
Linear Quadrilateral Element
There are four nodes at the corners of the quadrilateral

shape. In the natural coordinate system (ξ , η ) , the four shape
functions are,
1 1
N 1 = (1 − ξ )(1 − η ), N 2 = (1 + ξ )(1 − η )
4 4 (35)
1 1
N 3 = (1 + ξ )(1 + η ), N 4 = (1 − ξ )(1 + η )
4 4
4
Note that ∑ N i = 1 at any point inside the element, as expected.
i=1
The displacement field is given by

4 4
u = ∑ N i ui , v = ∑ N i vi (36)
i =1 i =1
which are bilinear functions over the element.

Quadratic Quadrilateral Element (Q8)

This is the most widely used element for 2-D problems due
to its high accuracy in analysis and flexibility in modeling.
η
η =1 7
3 ξ
4
6
8
5 2
1
y
η = −1
ξ = −1 ξ =1
x
Quadratic Quadrilateral Element
There are eight nodes for this element, four corners nodes
and four midside nodes. In the natural coordinate system (ξ , η ) ,
the eight shape functions are,
1
N 1 = (1 − ξ )(η − 1)(ξ + η + 1)
4
1
N 2 = (1 + ξ )(η − 1)(η − ξ + 1)
4 (37)
1
N 3 = (1 + ξ )(1 + η )(ξ + η − 1)
4
1
N 4 = (ξ − 1)(η + 1)(ξ − η + 1)
4

1
N 5 = (1 − η )(1 − ξ 2 )
2
1
N 6 = (1 + ξ )(1 − η 2 )
2
1
N 7 = (1 + η )(1 − ξ 2 )
2
1
N 8 = (1 − ξ )(1 − η 2 )
2
8
Again, we have ∑ N i = 1 at any point inside the element.
i=1
The displacement field is given by

8 8
u = ∑ N i ui , v = ∑ N i vi (38)
i =1 i =1
which are quadratic functions over the element. Strains and

stresses over a quadratic quadrilateral element are linear
functions, which are better representations.
Notes:
• Q4 and T3 are usually used together in a mesh with
linear elements.
• Q8 and T6 are usually applied in a mesh composed of
quadratic elements.
• Quadratic elements are preferred for stress analysis,
because of their high accuracy and the flexibility in
modeling complex geometry, such as curved boundaries.

Example 3.2
A square plate with a hole at the center and under pressure
in one direction.
y
A
x
B
The dimension of the plate is 10 in. x 10 in., thickness is

0.1 in. and radius of the hole is 1 in. Assume E = 10x106 psi, v
= 0.3 and p = 100 psi. Find the maximum stress in the plate.
FE Analysis:
From the knowledge of stress concentrations, we should
expect the maximum stresses occur at points A and B on the
edge of the hole. Value of this stress should be around 3p (=
300 psi) which is the exact solution for an infinitely large plate
with a hole.

We use the ANSYS FEA software to do the modeling

(meshing) and analysis, using quadratic triangular (T6 or LST),
linear quadrilateral (Q4) and quadratic quadrilateral (Q8)
elements. Linear triangles (CST or T3) is NOT available in
ANSYS.
The stress calculations are listed in the following table,
along with the number of elements and DOF used, for
comparison.
Table. FEA Stress Results
Elem. Type No. Elem. DOF Max. σ (psi)
T6 966 4056 310.1
Q4 493 1082 286.0
Q8 493 3150 327.1
... ... ... ...
Q8 2727 16,826 322.3
Discussions:
• Check the deformed shape of the plate
• Check convergence (use a finer mesh, if possible)
• Less elements (~ 100) should be enough to achieve the
same accuracy with a better or “smarter” mesh
• We’ll redo this example in next chapter employing the
symmetry conditions.

FEA Mesh (Q8, 493 elements)
FEA Stress Plot (Q8, 493 elements)

Transformation of Loads
Concentrated load (point forces), surface traction (pressure
loads) and body force (weight) are the main types of loads
applied to a structure. Both traction and body forces need to be
converted to nodal forces in the FEA, since they cannot be
applied to the FE model directly. The conversions of these
loads are based on the same idea (the equivalent-work concept)
which we have used for the cases of bar and beam elements.
qB
q fB
qA
fA
s
B B
A L A
Traction on a Q4 element
Suppose, for example, we have a linearly varying traction q

on a Q4 element edge, as shown in the figure. The traction is
normal to the boundary. Using the local (tangential) coordinate
s, we can write the work done by the traction q as,
L
Wq = t ∫ un ( s )q( s )ds
0
where t is the thickness, L the side length and un the component

of displacement normal to the edge AB.

For the Q4 element (linear displacement field), we have

un ( s ) = (1 − s / L )unA + ( s / L )unB
The traction q(s), which is also linear, is given in a similar way,
q( s ) = (1 − s / L )q A + ( s / L )qB
Thus, we have,
 L
1 − s / L    q A 
Wq = t ∫  [ unA unB ]   [1 − s / L s / L]q  ds
0  s / L   B
L
 (1 − s / L ) 2 ( s / L )(1 − s / L ) q A 
= [ unA unB ]t ∫   ds q 
0 ( s / L )(1 − s / L )
2
( s / L)   B
tL 2 1 q A 
= [ unA unB ] 
6 1 2 q B 
and the equivalent nodal force vector is,

 f A  tL 2 1 q A 
 =   
 f B  6 1 2 q B 
Note, for constant q, we have,
 f A  qtL 1
 = 
 fB  2 1
For quadratic elements (either triangular or quadrilateral),
the traction is converted to forces at three nodes along the edge,
instead of two nodes.
Traction tangent to the boundary, as well as body forces,
are converted to nodal forces in a similar way.
Stress Calculation
The stress in an element is determined by the following
relation,
σ x   εx 
   
σ
 y = E  ε y  = EBd (39)
τ  γ 
 xy   xy 
where B is the strain-nodal displacement matrix and d is the
nodal displacement vector which is known for each element
once the global FE equation has been solved.
Stresses can be evaluated at any point inside the element
(such as the center) or at the nodes. Contour plots are usually
used in FEA software packages (during post-process) for users
to visually inspect the stress results.
The von Mises Stress:

The von Mises stress is the effective or equivalent stress for
2-D and 3-D stress analysis. For a ductile material, the stress
level is considered to be safe, if
σ e ≤ σY
where σ e is the von Mises stress and σ Y the yield stress of the
material. This is a generalization of the 1-D (experimental)
result to 2-D and 3-D situations.

The von Mises stress is defined by

1
σe = (σ 1 − σ 2 ) 2 + (σ 2 − σ 3 ) 2 + (σ 3 − σ 1 ) 2 (40)
2
in which σ 1 , σ 2 and σ 3 are the three principle stresses at the
considered point in a structure.
For 2-D problems, the two principle stresses in the plane
are determined by
σx +σ y σx −σ y 
2
σ1 = +   + τ xy
P 2
2  2 
(41)
σx +σ y σ −σ y 
2
σ2 = −  x  + τ xy
P 2
2  2 
Thus, we can also express the von Mises stress in terms of

the stress components in the xy coordinate system. For plane
stress conditions, we have,
σ e = (σ x + σ y ) 2 − 3(σ x σ y − τ xy2 ) (42)
Averaged Stresses:
Stresses are usually averaged at nodes in FEA software
packages to provide more accurate stress values. This option
should be turned off at nodes between two materials or other
geometry discontinuity locations where stress discontinuity does
exist.

Discussions
1) Know the behaviors of each type of elements:

T3 and Q4: linear displacement, constant strain and stress;
T6 and Q8: quadratic displacement, linear strain and stress.
2) Choose the right type of elements for a given problem:

When in doubt, use higher order elements or a finer mesh.
3) Avoid elements with large aspect ratios and corner angles:

Aspect ratio = Lmax / Lmin
where Lmax and Lmin are the largest and smallest characteristic
lengths of an element, respectively.
Elements with Bad Shapes
Elements with Nice Shapes

4) Connect the elements properly:

Don’t leave unintended gaps or free elements in FE models.
A C
B D
Improper connections (gaps along AB and CD)
Readings:
Sections 3.1-3.5 and 3.8-3.12 of Cook’s book.

Lecture Notes: Introduction to Finite Element Method Chapter 4. FE Modeling and Solution Techniques
Chapter 4. Finite Element Modeling and

Solution Techniques
I. Symmetry
A structure possesses symmetry if its components are

arranged in a periodic or reflective manner.
Types of Symmetry:
• Reflective (mirror, bilateral) symmetry
• Rotational (cyclic) symmetry
• Axisymmetry
• Translational symmetry
• ...
Examples:

Applications of the symmetry properties:

• Reducing the size of the problems (save CPU time, disk
space, postprocessing effort, etc.)
• Simplifying the modeling task
• Checking the FEA results
• ...
Symmetry of a structure should be fully exploited and

retained in the FE model to ensure the efficiency and quality of
FE solutions.
Examples:
Cautions:
In vibration and buckling analyses, symmetry concepts, in
general, should not be used in FE solutions (works fine in
modeling), since symmetric structures often have antisymmetric
vibration or buckling modes.

II. Substructures (Superelements)
Substructuring is a process of analyzing a large structure as

a collection of (natural) components. The FE models for these
components are called substructures or superelements (SE).
Physical Meaning:
A finite element model of a portion of structure.
Mathematical Meaning:
Boundary matrices which are load and stiffness matrices
reduced (condensed) from the interior points to the exterior or
boundary points.

Advantages of Using Substructures/Superelements:

• Large problems (which will otherwise exceed your
computer capabilities)
• Less CPU time per run once the superelements have
been processed (i.e., matrices have been saved)
• Components may be modeled by different groups
• Partial redesign requires only partial reanalysis (reduced
cost)
• Efficient for problems with local nonlinearities (such as
confined plastic deformations) which can be placed in
one superelement (residual structure)
• Exact for static stress analysis
Disadvantages:
• Increased overhead for file management
• Matrix condensation for dynamic problems introduce
new approximations
• ...

III. Equation Solving
Direct Methods (Gauss Elimination):

• Solution time proportional to NB2 (N is the dimension of
the matrix, B the bandwidth)
• Suitable for small to medium problems, or slender
structures (small bandwidth)
• Easy to handle multiple load cases
Iterative Methods:
• Solution time is unknown beforehand
• Reduced storage requirement
• Suitable for large problems, or bulky structures (large
bandwidth, converge faster)
• Need solving again for different load cases

Gauss Elimination - Example:

 8 − 2 0   x1   2 
 − 2 4 − 3  x  = − 1 or Ax = b .
  2   
 0 − 3 3   x 3   3 
Forward Elimination:
(1)  8 − 2 0 2
Form ( 2)  − 2 4 − 3 − 1 ;
 
(3)  0 − 3 3 3 
(1) + 4 x (2) ⇒ (2):

(1) 8 − 2 0 2
(2) 0 14 − 12 − 2 ;
 
(3) 0 − 3 3 3 
14
(2) + (3) ⇒ (3):
3
(1) 8 − 2 0 2
(2) 0 14 − 12 − 2 ;
 
(3) 0 0 2 12 
Back Substitution:
x 3 = 12 / 2 = 6 1.5
 
x 2 = ( −2 + 12 x 3 ) / 14 = 5 or x =  5 .
x1 = (2 + 2 x 2 ) / 8 = 1.5 6
 

Iterative Method - Example:

The Gauss-Seidel Method
Ax = b (A is symmetric)
N
or ∑ aij x j = bi , i = 1, 2, ..., N .
j =1
Start with an estimate x ( 0 ) and then iterate using the following:

1  i −1 N

∑ ∑ a ij x j
( k +1) ( k +1)
= − −
(k )
xi  b i a ij x j ,
a ii  j =1 j = i +1 
for i = 1, 2, ..., N .
In vector form,
−1
[
x ( k +1) = A D b − A L x ( k +1) − A L x ( k ) ,
T
]
where
A D = 〈 a ii 〉 is the diagonal matrix of A,
A L is the lower triangular matrix of A,
such that A = A D + A L + A L .
T
Iterations continue until solution x converges, i.e.

x ( k +1 ) − x ( k )
≤ε,
x (k )
where ε is the tolerance for convergence control.

IV. Nature of Finite Element Solutions

• FE Model – A mathematical model of the real structure,
based on many approximations.
• Real Structure -- Infinite number of nodes (physical
points or particles), thus infinite number of DOF’s.
• FE Model – finite number of nodes, thus finite number
of DOF’s.
ð Displacement field is controlled (or constrained) by the

values at a limited number of nodes.
Recall that on an element :

4
u = ∑ N α uα
α =1
Stiffening Effect:
• FE Model is stiffer than the real structure.
• In general, displacement results are smaller in
magnitudes than the exact values.

Hence, FEM solution of displacement provides a lower

bound of the exact solution.
∆ (Displacement)
Exact Solution
FEM Solutions
No. of DOF’s
The FEM solution approaches the exact solution from

below.
This is true for displacement based FEA only!

V. Numerical Error
Error ≠ Mistakes in FEM (modeling or solution).
Types of Error:
• Modeling Error (beam, plate … theories)
• Discretization Error (finite, piecewise …)
• Numerical Error ( in solving FE equations)
Example (numerical error):
u1 u2
P
1 k1 2 k2 x
FE Equations:
 k1 − k 1   u1   P 
− k  = 
 1 k 1 + k 2  u 2   0 
and Det K = k 1 k 2 .
The system will be singular if k2 is small compared with k1.

P
u2 = u1 −
u2
k1
k1
u2 = u1
k1 + k 2
k2 << k1 (two lines close):

ð System ill-conditioned.
P/k1 u1
P
u2 = u1 −
u2 k1
k1
u2 = u1
k1 + k 2
k2 >> k1 (two line apart):
ð System well conditioned.
P/k1 u1
• Large difference in stiffness of different parts in FE

model may cause ill-conditioning in FE equations.
Hence giving results with large errors.
• Ill-conditioned system of equations can lead to large
changes in solution with small changes in input
(right hand side vector).

VI. Convergence of FE Solutions

As the mesh in an FE model is “refined” repeatedly, the FE
solution will converge to the exact solution of the mathematical
model of the problem (the model based on bar, beam, plane
stress/strain, plate, shell, or 3-D elasticity theories or
assumptions).
Types of Refinement:
h-refinement: reduce the size of the element (“h” refers to the
typical size of the elements);
p-refinement: Increase the order of the polynomials on an
element (linear to quadratic, etc.; “h” refers to
the highest order in a polynomial);
r-refinement: re-arrange the nodes in the mesh;
hp-refinement: Combination of the h- and p-refinements
(better results!).
Examples:
…

VII. Adaptivity (h-, p-, and hp-Methods)

• Future of FE applications
• Automatic refinement of FE meshes until converged
results are obtained
• User’s responsibility reduced: only need to generate a
good initial mesh
Error Indicators:
Define,
σ --- element by element stress field (discontinuous),
σ*--- averaged or smooth stress (continuous),
σE = σ - σ* --- the error stress field.
Compute strain energy,

M
1 T −1
U = ∑U i , Ui = ∫ 2 s E s dV ;
i =1 V i
M
1 *T −1 *
U = ∑U i* ,
*
U* =
i ∫ 2 s E s dV ;
i =1 V i
M
1 T −1
U E = ∑U E i , U Ei = ∫ 2 s E E s E dV ;
i =1 V i
where M is the total number of elements, Vi is the volume of the

element i.

One error indicator --- the relative energy error:

1/ 2
 UE 
η=  . (0 ≤ η ≤ 1)
U + U E 
The indicator η is computed after each FE solution. Refinement

of the FE model continues until, say
η ≤ 0.05.
=> converged FE solution.
Examples:
…

Lecture Notes: Introduction to Finite Element Method Chapter 5. Plate and Shell Elements
Chapter 5. Plate and Shell Elements
I. Plate Theory
• Flat plate
• Lateral loading
• Bending behavior dominates
Note the following similarity:

1-D straight beam model ó 2-D flat plate model
Applications:
• Shear walls
• Floor panels
• Shelves
• …

Forces and Moments Acting on the Plate:
∆y
y
∆x My
q(x,y)
Qy Mxy
t Mx
Qx
x Mid surface
Mxy
Stresses:
τyz y
σy
τxy
τxz τxy
σx
x

Relations Between Forces and Stresses

Bending moments (per unit length):
t/2
M x = ∫− t / 2 σ x zdz , ( N ⋅ m / m) (1)
t/2
M y = ∫− t / 2 σ y zdz , ( N ⋅ m / m) (2)
Twisting moment (per unit length):

t/2
M xy = ∫− t / 2 τ xy zdz , ( N ⋅ m / m) (3)
Shear Forces (per unit length):

t/2
Q x = ∫− t / 2 τ xz dz , ( N / m) (4)
t/2
Q y = ∫− t / 2 τ yz dz , ( N / m) (5)
Maximum bending stresses:

6M 6M y
(σ x ) max =± 2x, (σ y ) max = ± . (6)
t t2
• Maximum stress is always at z = ± t / 2
• No bending stresses at midsurface (similar to the beam
model)

Thin Plate Theory ( Kirchhoff Plate Theory)
Assumptions (similar to those in the beam theory):

A straight line along the normal to the mid surface remains
straight and normal to the deflected mid surface after loading,
that is, these is no transverse shear deformation:
γ xz = γ yz = 0 .
Displacement:
∂w
z
∂x
w
x
w = w( x, y ), ( deflection)
∂w
u = −z , (7)
∂x
∂w
v = −z .
∂y

Strains:
∂ 2w
ε x = −z 2 ,
∂x
∂ 2w
ε y = −z 2 , (8)
∂y
∂ 2w
γ xy = −2 z .
∂x ∂y
Note that there is no stretch of the mid surface due to the
deflection (bending) of the plate.
Stresses (plane stress state):

σ x  1 ν 0  ε x 
  E ν 1  ε  ,
σ
 y =
 y 
0

τ  1 − ν
2
 xy   0 0 (1 − ν ) / 2 γ xy 
or,
 ∂2w 
 2 
σ x   1 ν 0   ∂x2 
 
σ = −
E 
ν  ∂ w  .
 y z 2 
1 0
  ∂y 2 
(9)
τ  1 − ν
 xy   0 0 (1 − ν )  ∂ 2 w 
 
 ∂x∂y 
Main variable: deflection w = w( x, y ) .

Governing Equation:
D∇ 4 w = q ( x , y ) , (10)
where
∂4 ∂4 ∂4
∇ ≡ ( 4 + 2 2 2 + 4 ),
4
∂x ∂x ∂y ∂y
Et 3
D= (the bending rigidity of the plate),
12(1 − ν )
2
q = lateral distributed load (force/area).
Compare the 1-D equation for straight beam:

d 4w
EI 4
= q( x ) .
dx
Note: Equation (10) represents the equilibrium condition

in the z-direction. To see this, refer to the previous figure
showing all the forces on a plate element. Summing the forces
in the z-direction, we have,
Q x ∆y + Q y ∆x + q∆x∆y = 0,
which yields,
∂Q x ∂Q y
+ + q( x, y ) = 0 .
∂x ∂y
Substituting the following relations into the above equation, we
obtain Eq. (10).

Shear forces and bending moments:

∂M x ∂M xy ∂M xy ∂M y
Qx = + , Qy = + ,
∂x ∂y ∂x ∂y
 ∂2w ∂2w   ∂2w ∂2w 

M x = D 2 + ν 2  , M y = D 2 + ν 2  .
 ∂x ∂y   ∂y ∂x 
The fourth-order partial differential equation, given in (10)

and in terms of the deflection w(x,y), needs to be solved under
certain given boundary conditions.
Boundary Conditions:
∂w
Clamped: w = 0, = 0; (11)
∂n
Simply supported: w = 0, M n = 0; (12)
Free: Q n = 0, M n = 0; (13)
where n is the normal direction of the boundary. Note that the
given values in the boundary conditions shown above can be
non-zero values as well.
s n
boundary

Examples:
A square plate with four edges clamped or hinged, and
under a uniform load q or a concentrated force P at the center C.
y
C L
L
x Given: E, t, and ν = 0.3
For this simple geometry, Eq. (10) with boundary condition

(11) or (12) can be solved analytically. The maximum
deflections are given in the following table for the different
cases.
Deflection at the Center (wc)

Clamped Simply supported
Under uniform load q 0.00126 qL4/D 0.00406 qL4/D
Under concentrated force P 0.00560 PL2/D 0.0116 PL2/D
in which: D= Et3/(12(1-v2)).
These values can be used to verify the FEA solutions.

Thick Plate Theory (Mindlin Plate Theory)

If the thickness t of a plate is not “thin”, e.g., t / L ≥ 1 / 10
(L = a characteristic dimension of the plate), then the thick plate
theory by Mindlin should be applied. This theory accounts for
the angle changes within a cross section, that is,
γ xz ≠ 0, γ yz ≠ 0 .
This means that a line which is normal to the mid surface before
the deformation will not be so after the deformation.
∂w
z θy  ≠ − 
 ∂x 
w
∂w x
∂x
New independent variables:

θ x and θ y : rotation angles of a line, which is normal to the
mid surface before the deformation, about x- and y-axis,
respectively.

New relations:
u = zθ y , v = − zθ x ; (14)
∂θ y
εx = z ,
∂x
∂θ
ε y = −z x ,
∂y
∂θ ∂θ
γ xy = z ( y − x ), (15)
∂y ∂x
∂w
γ xz = +θ y ,
∂x
∂w
γ yz = −θ x.
∂y
Note that if we imposed the conditions (or assumptions)
that
∂w ∂w
γ xz = + θ y = 0, γ yz = − θ x = 0,
∂x ∂y
then we can recover the relations applied in the thin plate
theory.
Main variables: w( x, y ),θ x ( x, y ) and θ y ( x, y ) .
The governing equations and boundary conditions can be

established for thick plate based on the above assumptions.

II. Plate Elements

Kirchhoff Plate Elements:
4-Node Quadrilateral Element
z y
Mid surface 4
3
x
1 2
∂w  ∂w  t
w1 ,   ,    ∂w   ∂w 
w2 ,   ,  
 ∂x  1  ∂y  1  ∂x  2  ∂y  2
∂w ∂w
DOF at each node: w, , .
∂y ∂y
On each element, the deflection w(x,y) is represented by
 4
∂w ∂w 
w( x, y ) = ∑  N i wi + N xi ( ) i + N yi ( ) i  ,
i =1  ∂x ∂y 
where Ni, Nxi and Nyi are shape functions. This is an
incompatible element! The stiffness matrix is still of the form
k = ∫ B T EBdV ,
V
where B is the strain-displacement matrix, and E the stress-

strain matrix.

Mindlin Plate Elements:
4-Node Quadrilateral 8-Node Quadrilateral
z y z y
4 4 7 3
3
8 6
x x
1 2 1 2
t t 5
DOF at each node: w, θx and θy.

On each element:
n
w( x, y ) = ∑ N i wi ,
i =1
n
θ x ( x, y ) = ∑ N iθ xi ,
i =1
n
θ y ( x, y ) = ∑ N iθ yi .
i =1
• Three independent fields.

• Deflection w(x,y) is linear for Q4, and quadratic for Q8.

Discrete Kirchhoff Element:

Triangular plate element (not available in ANSYS).
Start with a 6-node triangular element,
z y 3
4 6
1 2
t 5 x
∂w ∂w
DOF at corner nodes: w, , ,θ x ,θ y ;
∂x ∂y
DOF at mid side nodes: θ x ,θ y .
Total DOF = 21.

Then, impose conditions γ xz = γ yz = 0 , etc., at selected
nodes to reduce the DOF (using relations in (15)). Obtain:
z y 3
1 2
x
∂w   ∂w 
At each node: w,θ x  = ,θ y  = .
 ∂x   ∂y 
Total DOF = 9 (DKT Element).
• Incompatible w(x,y); convergence is faster (w is cubic
along each edge) and it is efficient.

Test Problem:
P
y
C
L
L
x L/t = 10, ν = 0.3
ANSYS 4-node quadrilateral plate element.
ANSYS Result for wc

Mesh wc (× PL2/D)
2×2 0.00593
4×4 0.00598
8×8 0.00574
16×16 0.00565
: :
Exact Solution 0.00560
Question: Converges from “above”? Contradiction to what

we learnt about the nature of the FEA solution?
Reason: This is an incompatible element ( See comments
on p. 177).

III. Shells and Shell Elements
Shells – Thin structures witch span over curved surfaces.
Example:
• Sea shell, egg shell (the wonder of the nature);
• Containers, pipes, tanks;
• Car bodies;
• Roofs, buildings (the Superdome), etc.
Forces in shells:
Membrane forces + Bending Moments
(cf. plates: bending only)

Example: A Cylindrical Container.
p p
internal forces:
p
p
membrane stresses
dominate
Shell Theory:
• Thin shell theory
• Thick shell theory
Shell theories are the most complicated ones to formulate
and analyze in mechanics (Russian’s contributions).
• Engineering ≠ Craftsmanship
• Demand strong analytical skill

Shell Elements:
plane stress element plate bending element
flat shell element
cf.: bar + simple beam element => general beam element.
DOF at each node:
w
v
u θx
θy
Q4 or Q8 shell element.

Curved shell elements:
θz
i w
v
i u θx
θy
• Based on shell theories;

• Most general shell elements (flat shell and plate
elements are subsets);
• Complicated in formulation.

Test Cases:
L/2
q L/2
F
A
R A
80o R
F
Roof Pinched Cylinder
F2
F R F
b
A
A F L
F1
F
Pinched Hemisphere Twisted Strip (90o)
ð Check the Table, on page 188 of Cook’s book, for

values of the displacement ∆A under the various loading
conditions.
Difficulties in Application:
• Non uniform thickness (turbo blades, vessels with
stiffeners, thin layered structures, etc.);
ð Should turn to 3-D theory and apply solid elements.

Lecture Notes: Introduction to Finite Element Method Chapter 6. Solid Elements
Chapter 6. Solid Elements for 3-D Problems
I. 3-D Elasticity Theory
Stress State:
y
F
x
z
y,v
σy
τ yx
τ yz τ xy
τ zy
σx
τ zx τ xz
σz x, u
z, w

σx 
σ 
 y 
σz 
ó = {σ }=   , or [σ ] (1)
 τ xy
ij

 τ yz 
 
 τ zx 
Strains:
 εx 
 εy 
ε 
å = {ε }=  z
γ xy  , or [ε ]
ij (2)
γ 
 yz 
 γ zx 
Stress-strain relation:
1 − v v v 0 0 0 
σx   v 1− v v 0 0 0  εx 
σ    
 y   v v 1− v 0 0 0  ε y 
 0 1 − 2v
σz


=
E
 0 0 0 0   ε z 
 τ xy + −  γ xy 
(1 v )(1 2 v ) 2
  1 − 2v
 τ yz   0 0 0 0 0  γ yz 
   2  
 τ zx   0 1 − 2v  γ zx 
0 0 0 0
 2 
or ó = Eå (3)

Displacement:
 u( x, y , z )   u1 
   
u =  v ( x, y , z )  =  u2  ( 4)
 w( x, y , z )   u 
   3
Strain-Displacement Relation:
∂u ∂v ∂w
εx = , εy = , εz = ,
∂x ∂y ∂z
∂v ∂u ∂w ∂v ∂u ∂w
γ xy = + , γ yz = + , γ xz = + (5)
∂x ∂y ∂y ∂z ∂z ∂x
or
1  ∂ui ∂u j 
ε ij =  +  , (i, j = 1, 2, 3)
2  ∂x j ∂xi 
or simply,
ε ij =
1
2
(ui , j + u j ,i ) ( tensor notation)

Equilibrium Equations:
∂σ x ∂τ xy ∂τ xz
+ + + fx = 0 ,
∂x ∂y ∂z
∂τ yx ∂σ y ∂τ yz
+ + + fy = 0 , ( 6)
∂x ∂y ∂z
∂τ zx ∂τ zy ∂σ z
+ + + fz = 0 ,
∂x ∂y ∂z
or
σ ij , j + f i = 0
Boundary Conditions (BC’s):

ui = ui , on Γu ( specified displacement )
ti = ti , on Γσ ( specified traction ) ( 7)
( traction ti = σ ij n j )
p
n
Γσ
Γ ( = Γu + Γσ )
Γu
Stress Analysis:
Solving equations in (6) under the BC’s in (7).

II. Finite Element Formulation

Displacement Field:
N
u= ∑ N i ui
i =1
N
v= ∑ N i vi (8)
i =1
N
w= ∑ N i wi
i =1
Nodal values
In matrix form:
 u1 
 v1 
u   N1 0 0 N2 0 0 L  w1 
   
v  = 0 N1 0 0 N2 0 L  u2  (9)
 ( 3×1)  L
w 
0 0 N1 0 0 N2 ( 3×3 N )  v2 
w2 
 M ( 3N ×1)
or u=Nd
Using relations (5) and (8), we can derive the strain vector
ε =B d
(6×1) (6×3N)×(3N×1)

Stiffness Matrix:
k = ∫ B T E B dv (10)
v
(3×N) (3N×6)×(6×6)×(6×3N)
Numerical quadratures are often needed to evaluate the

above integration.
Rigid-body motions for 3-D bodies (6 components):

3 translations, 3 rotations.
These rigid-body motions (singularity of the system of
equations) must be removed from the FEA model to ensure the
quality of the analysis.

III Typical 3-D Solid Elements

Tetrahedron:
linear (4 nodes) quadratic (10 nodes)
Hexahedron (brick):
Penta:
Avoid using the linear (4-node) tetrahedron element in 3-D

stress analysis (Inaccurate! But it is OK for dynamic analysis).

Element Formulation:
 Linear Hexahedron Element
6
5
y 8 7 2
1
4 3 mapping (x↔ξ)
x (-1≤ ξ,η,ζ ≤ 1)
z
η
(-1,1,-1) 4 3 (1,1,-1)
(-1,1,1) 8 7 (1,1,1)
o ξ
(-1,-1,-1) 1 2 (1,-1,-1)
(-1,-1,1) 5 6 (1,-1,1)
ζ
Displacement field in the element:

8 8 8
u = ∑ N i ui , v = ∑ N i vi , w = ∑ N i wi (11)
i =1 i =1 1i =1

Shape functions:
1
N 1 (ξ ,η , ζ ) = (1 − ξ ) (1 − η ) (1 − ζ ) ,
8
1
N 2 (ξ ,η , ζ ) = (1 + ξ ) (1 − η ) (1 − ζ ) ,
8
1
N 3 (ξ ,η , ζ ) = (1 + ξ ) (1 + η ) (1 − ζ ) , (12)
8
M M
1
N 8 (ξ ,η , ζ ) = (1 − ξ ) (1 + η ) (1 + ζ ) .
8
Note that we have the following relations for the shape
functions:
N i ( ξ j ,η j , ζ j ) = δ ij , i, j = 1,2,L, 8.
8
∑ N i ( ξ ,η ,ζ ) = 1.
i =1
Coordinate Transformation (Mapping):

8 8 8
x = ∑ N i xi , y = ∑ N i yi , z = ∑ N i zi . (13)
i =1 i =1 i =1
The same shape functions are used as for the displacement

field.
⇒ Isoparametric element.

Jacobian Matrix:
 ∂u   ∂x ∂y ∂z   ∂u 
 ∂ξ   ∂ξ ∂ξ ∂ξ   ∂x 
     ∂u 
 ∂u   ∂x ∂y ∂z 
 =   (14)
 ∂η   ∂η ∂η ∂η   ∂y 
 ∂u   ∂x ∂y ∂z   ∂u 
 ∂ζ     ∂z 
 ∂ζ ∂ζ ∂ζ 
≡ J Jacobian matrix
 ∂u   ∂u 
 ∂x   ∂ξ 
 ∂u   
−1  ∂u   ∂u 8
∂N i 
⇒  =J   ,  = ∑ ∂ξ i u , etc. 
 ∂y   ∂η   ∂ξ i =1 
 ∂u   ∂u 
 ∂z   ∂ζ 
and
 ∂v   ∂v 
 ∂x   ∂ξ 
 ∂v   
−1  ∂v 
 =J   , (15)
 ∂y   ∂η 
 ∂v   ∂v 
 ∂z   ∂ζ 
also for w.

 ∂u 
 ∂x 
 ∂v 
ε x   
ε   ∂ y 
  
y ∂w 
 ε z   ∂z 
å =   =  ∂x ∂u  = L use (15) = B d
γ xy   + 
γ yz   ∂x ∂y 
   ∂w ∂v 
γ zx   + 
∂y ∂z
 
∂u
 +  ∂w
 ∂z ∂x 
where d is the nodal displacement vector,

i.e.,
å = Bd (16)
(6×1) (6×24)×(24×1)

Strain energy,
1 T 1
U=
2V∫ ó å dV =
2V∫ ( Eå ) T å dV
1
= ∫ å T E å dV
2V
1 T T 
= d  ∫ B E B dV  d (17)
2 V 
Element stiffness matrix,

k = ∫ B T E B dV (18)
V
(24×24) (24×6)×(6×6)×(6×24)
In ξηζ coordinates:
dV = (det J ) dξ dη dζ (19)
1 1 1
⇒ k = ∫ ∫ ∫ B T E B (det J ) dξ dη dζ ( 20)
−1 −1 −1
( Numerical integration)
• 3-D elements usually do not use rotational DOFs.

Loads:
Distributed loads ⇒ Nodal forces
pA/3 pA/12
p
Area =A Nodal forces for 20-node

Hexahedron
Stresses:
ó = Eå = EB d
Principal stresses:
σ 1 ,σ 2 ,σ 3 .
von Mises stress:

1
σ e = σ VM = (σ 1 − σ 2 ) 2 + (σ 2 − σ 3 ) 2 + (σ 3 − σ 1 ) 2 .
2
Stresses are evaluated at selected points (including nodes)

on each element. Averaging (around a node, for example) may
be employed to smooth the field.
Examples: …

Solids of Revolution (Axisymmetric Solids):
Baseball bat shaft
Apply cylindrical coordinates:

( x, y, z) ⇒ (r, θ, z)
z,w
θ
r, u
z, w
σz
θ
r, u
τ rz
r σθ σr

Displacement field:
u = u ( r , z ) , w = w( r , z ) (No v − circumferential component )
Strains:
∂u u ∂w
εr = , εθ = , εz = ,
∂r r ∂z
∂w ∂u
γ rz = + , (γ rθ = γ zθ = 0) ( 21)
∂r ∂z
u
r (r+u)dθ
dθ
rdθ
Stresses:
σ r  1 − v v v 0  εr 
σ   v 1− v v 0  ε 
 θ E    θ
  = v 1− v   ( 22)
 v 0 
σ z  (1 + v ) (1 − 2v )  0 1 − 2v  ε z 
τ rz   0 0 γ rz 
2 

Axisymmetric Elements:
2 η
2 2
r, u 3
r, u
3 ξ
3 4 1
1 1
3-node element (ring) 4-node element (ring)
∫
k = B T E B rdr dθ dz
V
( 23)
or
2π 1 1
k=
∫∫∫
0 −1 −1
B T E B r (det J ) dξ dη dθ
1 1
= 2π
∫∫
−1 −1
B T E B r (det J ) dξ dη ( 24)

Applications:
• Rotating Flywheel:
z
ω angular velocity (rad/s)
Body forces:
fr = ρ rω 2 ( equivalent radial centrifugal/ inertial force)
fz = − ρ g ( gravitational force)

• Cylinder Subject to Internal Pressure:
p
r0
q = ( p ) 2π r0
• Press Fit:
ri
r0
ri +δ
ring ( Sleeve) shaft
at r = ri :
uo − ui = δ
⇒ MPC
“i” “o”

• Belleville (Conical) Spring:
p z
This is a geometrically nonlinear (large deformation)

problem and iteration method (incremental approach) needs to
be employed.

Lecture Notes: Introduction to Finite Element Method Chapter 7. Structural Vibration and Dynamics
Chapter 7. Structural Vibration and Dynamics
• Natural frequencies and modes F(t)

• Frequency response (F(t)=Fo sinωt)
• Transient response (F(t) arbitrary)
I. Basic Equations
A. Single DOF System
k m - mass
f=f(t) k - stiffness
m 
c 
c - damping
 f (t ) - force
ku
c u& m f(t)
x, u
From Newton’s law of motion (ma = F), we have

mu&& = f(t)−k u −cu& ,
i.e.
mu&&+cu& +ku = f(t) , (1)
where u is the displacement, u& = du / dt and u&& = d 2u / dt2 .

Free Vibration: f(t) = 0 and no damping (c = 0)

Eq. (1) becomes
mu&&+k u =0 . (2)
(meaning: inertia force + stiffness force = 0)
Assume:
u(t) = U sin (ω t) ,
where ω is the frequency of oscillation, U the amplitude.

Eq. (2) yields
−Uù 2 m sin( ù t)+kU sin( ù t)= 0
i.e.,
[−ω 2
]
m+k U = 0.
For nontrivial solutions for U, we must have

[−ω 2
]
m+k = 0,
which yields
k
ω = . (3)
m
This is the circular natural frequency of the single DOF

system (rad/s). The cyclic frequency (1/s = Hz) is
ω
f= , (4)
2π

u
u = U s in w t
U
t
U
T=1/f
Undamped Free Vibration
With non-zero damping c, where

0 < c < c c = 2 mω = 2 k m (cc = critical damping) (5)
we have the damped natural frequency:
ωd = ω 1 − ξ 2 , (6)
c
where ξ = (damping ratio).
cc
For structural damping: 0 ≤ ξ < 0.15 (usually 1~5%)

ωd ≈ ω . (7)
Thus, we can ignore damping in normal mode analysis.

u
Damped Free Vibration

B. Multiple DOF System

Equation of Motion
Equation of motion for the whole structure is
&& + C u& + Ku = f ( t ) ,
Mu (8)
in which: u  nodal displacement vector,
M  mass matrix,
C  damping matrix,
K  stiffness matrix,
f  forcing vector.
Physical meaning of Eq. (8):
Inertia forces + Damping forces + Elastic forces
= Applied forces
Mass Matrices
Lumped mass matrix (1-D bar element):
ρAL 1 ρ,A,L 2 m = ρAL
m1 = 2
2 2
u1 u2
Element mass matrix is found to be
 ρAL 
 0 
m= 2
ρAL 
 0 
1442424 3
diagonal matrix

In general, we have the consistent mass matrix given by
m= ∫V
ρ N T NdV (9)
where N is the same shape function matrix as used for the

displacement field.
This is obtained by considering the kinetic energy:
1 T 1
Κ = u& m u& (cf. mv 2 )
2 2
1 1
= ∫ ρ u& 2 dV = ∫ ρ (u& ) u& dV
T
2 V 2 V
1
= ∫ ρ (N u& ) (N u& )dV
T
2 V
1
= u& T ∫ ρ N T N dV u&
2 1V 42 43
m
Bar Element (linear shape function):
1 − ξ 
m = ∫ ρ  [1 − ξ ξ ]ALdξ
V
 ξ 
1 / 3 1 / 6 u&&1 (10)
= ρAL  
1 / 6 1 / 3 u&&2

Element mass matrices:

⇒ local coordinates ⇒ to global coordinates
⇒ assembly of the global structure mass matrix M.
Simple Beam Element:

v1 v2
θ1 ρ, A, L θ2
m = ∫ ρNT NdV
V
 156 22 L 54 − 13L  v&&1

 − 3L2  θ&&1
ρAL  22 L 4 L2 13L

= (11)
420  54 13L 156 − 22 L v&&2
 
− 13L − 3L
2
− 22 L 4 L2  θ&&2
Units in dynamic analysis (make sure they are consistent):

Choice I Choice II
t (time) s s
L (length) m mm
m (mass) kg Mg
a (accel.) m/s2 mm/s2
f (force) N N
ρ (density) kg/m3 Mg/mm3

II. Free Vibration

Study of the dynamic characteristics of a structure:
• natural frequencies
• normal modes (shapes)
Let f(t) = 0 and C = 0 (ignore damping) in the dynamic

equation (8) and obtain
M&u& + Ku = 0 (12)
Assume that displacements vary harmonically with time, that
is,
u ( t ) = u sin( ω t ),
u& ( t ) = ω u cos( ω t ),
&& ( t ) = − ω 2 u sin( ω t ),
u
where u is the vector of nodal displacement amplitudes.

Eq. (12) yields,
[K − ω 2
]
M u = 0 (13)
This is a generalized eigenvalue problem (EVP).
Solutions?

Trivial solution: u = 0 for any values of ω (not interesting).

Nontrivial solutions: u ≠ 0 only if
K − ω2 M = 0 (14)
This is an n-th order polynomial of ω2, from which we can

find n solutions (roots) or eigenvalues ωi.
• ωi (i = 1, 2, …, n) are the natural frequencies (or
characteristic frequencies) of the structure.
• ω1 (the smallest one) is called the fundamental frequency.
• For each ωi , Eq. (13) gives one solution (or eigen) vector
[K − ω i
2
]
M ui = 0 .
u i (i=1,2,…,n) are the normal modes (or natural

modes, mode shapes, etc.).
Properties of Normal Modes

u iT K u j = 0,
u iT M u j = 0 , for i ≠ j, (15)
if ωi ≠ ω j . That is, modes are orthogonal (or independent) to

each other with respect to K and M matrices.

Normalize the modes:
u iT M u i = 1,
u iT K u i = ω i2 . (16)
Note:
• Magnitudes of displacements (modes) or stresses in normal
mode analysis have no physical meaning.
• For normal mode analysis, no support of the structure is
necessary.
ωi = 0 ⇔ there are rigid body motions of the whole or a
part of the structure.
⇒ apply this to check the FEA model (check for
mechanism or free elements in the models).
• Lower modes are more accurate than higher modes in the
FE calculations (less spatial variations in the lower modes
⇒ fewer elements/wave length are needed).

Example: y
v2
ρ, A, EI θ2
1 2 x
L
v2  0
[K − ω M   =  ,
2
]
θ2  0
EI  12 − 6L ρAL  156 − 22L
K= 3  2 
, M=  2 
.
L − 6L 4L  420 − 22L 4L 
EVP: 12 −156λ − 6L + 22Lλ

= 0,
− 6L + 22Lλ 4L − 4L λ
2 2
in which λ = ω ρ AL / 420 EI .
2 4
Solving the EVP, we obtain,

1
 EI  2  v 2   1 
ω 1 = 3.533  ,   = 1.38 ,
4 
 ρ AL  θ 2 1  L 
#3 #2
1
 EI  2  v 2   1 
#1 ω 2 = 34 .81 
4 
,   = 7.62 .
 ρAL  θ 2  2  L 
Exact solutions:
1 1
 EI  2
 EI  2
ω1 = 3.516  ,
4 
ω2 = 22.03  .
4 
 ρAL   ρAL 
We can see that mode 1 is calculated much more accurately
than mode 2, with one beam element.

III. Damping
Two commonly used models for viscous damping.
A. Proportional Damping (Rayleigh Damping)

C = αM + β K (17)
where the constants α & β are found from
αω 1 β αω 2 β
ξ1 = + , ξ2 = + ,
2 2ω1 2 2ω 2
with ω1 , ω 2 , ξ1 & ξ 2 (damping ratio) being selected.
Damping ratio
B. Modal Damping
Incorporate the viscous damping in modal equations.

IV. Modal Equations

• Use the normal modes (modal matrix) to transform the
coupled system of dynamic equations to uncoupled
system of equations.
We have
[K − ω i M ui = 0 ,
2
] i = 1,2,..., n (18)
where the normal mode u i satisfies:
 u iT K u j = 0,
 T for i ≠ j,
ui M u j = 0,
and
 u iT M u i = 1 ,
 T for i = 1, 2, …, n.
 i
u K u i = ω 2
i ,
Form the modal matrix:

Ö (n×n ) = [u 1 u 2 L u n ] (19)
Can verify that
ω12 0 L 0 
 
0 ω 2
M
ÖT KÖ = Ù =  2  (Spectralmatrix),
M O 0
 2 (20)
 0 L 0 ω n
ÖT MÖ = I.

Transformation for the displacement vector,
u = z1 u 1 + z 2 u 2 + L + z n u n = Φ z , (21)
where
 z1 ( t ) 
 z (t ) 
 
z= 2 
 M 
 z n ( t ) 
are called principal coordinates.

Substitute (21) into the dynamic equation:
M Φ &z& + C Φ z& + K Φ z = f ( t ).
Pre-multiply by ΦT, and apply (20):
&z& + C φ z& + Ω z = p ( t ), (22)
where C φ = α I + β Ω (proportional damping),
p = Φ T
f (t) .
Using Modal Damping
 2 ξ 1ω 1 0 L 0 
 0 2 ξ 2ω 
=  
2
Cφ
 M O M  . (23)
 
 0 L 2 ξ nω n 

Equation (22) becomes,

&z&i + 2 ξ i ω i z& i + ω i2 z i = p i ( t ), i = 1,2,…,n. (24)
Equations in (22) or (24) are called modal equations.

These are uncoupled, second-order differential equations,
which are much easier to solve than the original dynamic
equation (coupled system).
To recover u from z, apply transformation (21) again, once
z is obtained from (24).
Notes:
• Only the first few modes may be needed in constructing
the modal matrix Φ (i.e., Φ could be an n×m rectangular
matrix with m<n). Thus, significant reduction in the
size of the system can be achieved.
• Modal equations are best suited for problems in which
higher modes are not important (i.e., structural
vibrations, but not shock loading).

V. Frequency Response Analysis

(Harmonic Response Analysis)
&& + Cu& + Ku = F
Mu 1sin
23ωt
Harmonicloading
(25)
Modal method: Apply the modal equations,
&z&i + 2ξ iω i z&i + ω i2 zi = pi sin ω t , i=1,2,…,m. (26)

These are 1-D equations. Solutions are
pi ω i2
zi ( t ) = sin( ω t − θ i ), (27)
(1 − η ) + ( 2ξ iηi )
2 2 2
zi
where
 2ξ iη i
 θ = arctan , phase angle
1 − ηi
i 2

 ηi = ω ω i ,
 ci ci
ω/ωi
ξ
 i c = = , damping ratio
 c 2 m ω i
Recover u from (21).

Direct Method: Solve Eq. (25) directly, that is, calculate
iω t
the inverse. With u = u e (complex notation), Eq. (25)
becomes
[K + iω C − ω 2 M u = F . ]
This equation is expensive to solve and matrix is ill-
conditioned if ω is close to any ωi.

VI. Transient Response Analysis

(Dynamic Response/Time-History Analysis)
• Structure response to arbitrary, time-dependent loading.
f(t)
u(t)
Compute responses by integrating through time:
u1
u n u n+1
u2
t0 t1 t2 t n t n+1 t

Equation of motion at instance t n , n = 0, 1, 2, 3, ⋅⋅⋅:
&& n + Cu& n + Ku n = f n .
Mu
Time increment: ∆t=tn+1-tn, n=0, 1, 2, 3, ⋅⋅⋅.

There are two categories of methods for transient analysis.
A. Direct Methods (Direct Integration Methods)
• Central Difference Method
Approximate using finite difference:
1
u& = ( u n + 1 − u n − 1 ),
2 ∆ t
n
1
u
&& = ( u n +1 − 2 u n + u n −1 )
(∆ t)2
n
Dynamic equation becomes,

 1   1 
M ( u + − 2 u + u − )  + C  ( u n +1 − u n −1 ) + Ku n = fn ,
 ( ∆t )  2 ∆t
n 1 n n 1

2

which yields,
Au n +1 = F(t )
where
 1 1
 A = M+ C,
 (∆ t )2
2∆t
    
 F ( t ) = f n −  K − 2 2 M  u n −  1 2 M − 1 C  u n −1.
  (∆ t )   (∆ t ) 2∆t 

un+1 is calculated from un & un-1, and solution is

marching from t 0 , t1, L t n , t n + 1, L , until convergent.
This method is unstable if ∆t is too large.

• Newmark Method:
Use approximations:
( ∆t ) 2
u n +1 ≈ u n + ∆tu& n + [(1 − 2 β )u&& n + 2 βu&& n +1 ], → ( u&& n +1 = L)
2
u& n +1 ≈ u& n + ∆t [(1 − γ ) u && n +1 ],
&& n + γu
where β & γ are chosen constants. These lead to
Au n +1 = F (t)
where
γ 1
A = K + C + M ,
β∆t β (∆ t)2
F ( t ) = f ( f n + 1 , γ , β , ∆ t , C , M , u n , u& n , u
&& n ).
This method is unconditionally stable if

1
2 β ≥ γ ≥ .
2
1 1
e . g ., γ = , β =
2 4
which gives the constant average acceleration method.

Direct methods can be expensive! (the need to
compute A-1, often repeatedly for each time step).

B. Modal Method
First, do the transformation of the dynamic equations using
the modal matrix before the time marching:
m
u = ∑ u i zi (t ) =Φ z ,
i =1 i = 1,2,⋅⋅⋅, m.
&z&i + 2 ξ i ω i z& i + ω i z i = p i ( t ),
Then, solve the uncoupled equations using an integration

method. Can use, e.g., 10%, of the total modes (m= n/10).
• Uncoupled system,
• Fewer equations,
• No inverse of matrices,
• More efficient for large problems.
Comparisons of the Methods

Direct Methods Modal Method
• Small model • Large model
• More accurate (with small ∆t) • Higher modes ignored
• Single loading • Multiple loading
• Shock loading • Periodic loading
• … • …

Cautions in Dynamic Analysis

• Symmetry: It should not be used in the dynamic analysis
(normal modes, etc.) because symmetric structures can
have antisymmetric modes.
• Mechanism, rigid body motion means ω = 0. Can use
this to check FEA models to see if they are properly
connected and/or supported.
• Input for FEA: loading F(t) or F(ω) can be very
complicated in real applications and often needs to be
filtered first before used as input for FEA.
Examples
Impact, drop test, etc.

Finite Element Methods Lectures

Uploaded by

Copyright:

Available Formats

Finite Element Methods Lectures

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Finite Element Methods Lectures

Uploaded by

Copyright:

Available Formats

Lecture Notes: Introduction to Finite Element Method Chapter 1.

• Lego (kids’ play)

© 1998 Yijun Liu, University of Cincinnati 1

Why Finite Element Method?

• Design analysis: hand calculations, experiments, and

Applications of FEM in Engineering

A Brief History of the FEM

• 1943 ----- Courant (Variational methods)

• 1970s ----- Applications on mainframe computers

© 1998 Yijun Liu, University of Cincinnati 3

FEM in Structural Analysis

• Divide structure into pieces (elements with nodes)

© 1998 Yijun Liu, University of Cincinnati 4

• Preprocessing (build FE model, loads and constraints)

Available Commercial FEM Software Packages

• ANSYS (General purpose, PC and workstations)

© 1998 Yijun Liu, University of Cincinnati 5

Objectives of This FEM Course

• Understand the fundamental ideas of the FEM

© 1998 Yijun Liu, University of Cincinnati 6

II. Review of Matrix Algebra

Linear System of Algebraic Equations

where x1, x2, ..., xn are the unknowns.

 ... ... ... ... 

A is called a n×n (square) matrix, and x and b are (column)

© 1998 Yijun Liu, University of Cincinnati 7

Row and Column Vectors

Matrix Addition and Subtraction

where i = 1, 2, ..., l; j = 1, 2, ..., n.

© 1998 Yijun Liu, University of Cincinnati 8

Unit (Identity) Matrix

where Mij is the determinant of the smaller matrix obtained by

© 1998 Yijun Liu, University of Cincinnati 10

If det A = 0 (i.e., A is singular), then A-1 does not exist!

© 1998 Yijun Liu, University of Cincinnati 11

Solution Techniques for Linear Systems of Equations

Positive Definite Matrix

Differentiation and Integration of a Matrix

© 1998 Yijun Liu, University of Cincinnati 12

Types of Finite Elements

1-D (Line) Element

(Spring, truss, beam, pipe, etc.)

2-D (Plane) Element

(Membrane, plate, shell, etc.)

3-D (Solid) Element

(3-D fields - temperature, displacement, stress, flow velocity)

III. Spring Element

One Spring Element

k = F / ∆ (> 0) is the force needed to produce a unit stretch.

© 1998 Yijun Liu, University of Cincinnati 14

We only consider linear problems in this introductory

Consider the equilibrium of forces for the spring. At node i,

© 1998 Yijun Liu, University of Cincinnati 15

where f i m is the (internal) force acting on local node i of element

© 1998 Yijun Liu, University of Cincinnati 16

An alternative way of assembling the whole stiffness matrix:

© 1998 Yijun Liu, University of Cincinnati 17

Adding the two matrix equations (superposition), we have

This is the same equation we derived by using the force

Boundary and load conditions: