Introduction to Finite

Element Method
Dr. Rakesh K Kapania
Aerospace and Ocean Engineering Department
Virginia Polytechnic Institute and State University, Blacksburg,
VA

AOE 5024, Vehicle Structures

Summer, 2016

2016
c Rakesh K. Kapania, Mitchell Professor, Aerospace and Ocean Engineering,
Virginia Polytechnic Institute and State University, Blacksburg, VA, 24061-0203.
Introduction to Finite Element Method

I Complex structures under different types of loads

I Solution is often governed by ordinary or partial differential
equations
I Simple domains or simple forcing functions → analytical
techniques → closed form solution
I Complicated problems → approximate solutions

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 2
Approximate methods

Approximate methods
1. Semi-Analytical Methods
2. Numerical Methods
Semi-Analytical Methods
I Series Expansion using Fourier, Bessel and other orthogonal
functions
I Rayleigh-Ritz Method based on a variational principle
I Methods of Weighted Residual (MWR) (Galerkin Method and
the Collocation Method)
I The Green’s Functions approach (Integral Equations
Approach)

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 3
Approximate methods (contd...)

Numerical Methods
I Finite difference method
I Finite element method
I Spectral method
I Boundary element method
Numerical methods (except the finite difference method) are
essentially derived from one or the other semi-analytical methods.
For example
Finite element method → discrete implementation of the
Rayleigh-Ritz Method or the Method of Weighted Residuals
Spectral method → Series expansion method with use of FFT
(coefficients)
Boundary element method → Integral equation approach.
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 4
Approximate methods (contd...)

A key feature of both of these categories is to convert an ordinary

or partial differential equations into a set of linear (or nonlinear
depending on the problem at hand) algebraic equations, i.e.
Ka = Q.

K is called the stiffness matrix.

a is the vector of generalized co-ordinates
Q is a vector of generalized forces.
This set of equations can be solved numerically.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 5
Semi- Analytical Methods

In the semi-analytic methods, the solution u (x, y ) is generally

expressed as
N
u (x, y ) = φo (x, y ) + ∑ an φn (x, y )
n =1

where an are called the generalized co-ordinates and φn (x, y ) are

called the trial functions and are defined over the entire domain Ω.
The function φo (x, y ) is chosen such that it satisfies the
non-homogeneous part of the boundary conditions.
Note that this function is completely known and there is no
undetermined coefficient in front of this function.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 6
Trial Functions

The trial functions i.e. functions φn (x, y ) have global support.

I φn (x, y ) satisfy the homogeneous part of the boundary
conditions and must satisfy certain requirements of
completeness and linear independence (the trial functions
should be linearly independent)
I Selection of the trial functions is non- trivial
I Accuracy of the solution generally increases as the number of
terms N in the expansion are increased
I Semi-analytical methods differ from each other depending
upon the manner in which the coefficients an are determined

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 7
Methods of Weighted Residuals

In the Methods of Weighted Residuals the above approximation

to the solution is substituted in the governing equations and the
resulting error or the residual R (x, y ) is minimized in various ways.
In general, the following condition is used to obtain the values an .
Z
R (x, y )ψj (x, y )d Ω = 0 , j = 1, 2, 3, ...N
Ω

where the ψj (x, y ) are called the weight (hence the name Methods
of Weighted Residuals) or the test functions, Ω, represents the
domain. N is the number of terms used in the original expansion.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 8
Methods of Weighted Residuals (contd...)

The above condition is thus applied N different times and leads to

a set of N algebraic equations in the coefficients, an .

The weight functions may or may not be same as the trial

functions. For example, in the collocation scheme, the weight
functions are the Dirac’s Delta functions and in the Galerkin
Method the weight functions and the trial functions are the same.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 9
Ritz Method (Principle of Stationary Potential
Energy)

The unknown coefficients an ’s are determined by using the

underlying variational principle.
In structures, for example, the coefficients are determined from the
condition that at equilibrium the total potential energy Π of a
structure under given loading must be minimum, i.e. the first
variation of the total potential energy be equal to zero. In this
method the N algebraic equations are obtained by satisfying the
following N conditions

∂Π
= 0; 1≤n≤N
∂an
Note that the total Potential energy is defined as Π = U + V ,
where U is the strain energy and V is the potential of the applied
loads.
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 10
Principle of Virtual Work

The principle of minimum potential energy on which the Ritz

method is based, is restricted to conservative external forces i.e. to
forces that can be derived from a potential. For non conservative
forces, one has to resort to the more general Principle of Virtual
Work which can be stated as: If a structure is in equilibrium and
remains in equilibrium while it is subjected to a virtual distortion,
the external virtual work δWvir done by the external forces acting
on the structure is equal to the internal virtual work δWvir done by
the internal stresses (δWvir = δWint ).

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 11
Principle of Virtual Work (contd..)

The term ’virtual displacement’ implies an imaginary infinitesimal

displacement that is superimposed on the deformed configuration
while keeping the external loads unchanged and which is consistent
with the boundary conditions (i.e. it is zero on that part of the
boundary where the essential boundary conditions are specified).

The virtual displacements, which should be piecewise continuous,

to satisfy the displacement compatibility inside the domain, can be
considered as the delta (δ) operator (a variation of the
displacement field) or they may be considered as the test functions
that are used in the method of weighted residuals (e.g. in Galerkin
Method).

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 12
Principle of Virtual Work (contd...)

Approximate methods using the principle of virtual work can be as

easily derived as those using the principle of minimum potential
energy. Also note that the principle of virtual work is both a
sufficient and necessary condition for satisfying the equilibrium of a
structure. Indeed, you can derive the equations of equilibrium and
the associated boundary conditions from the principle of virtual
work.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 13
Principle of Virtual Work (contd...)

For a general three dimensional structure, the expressions for

external and internal virtual work can be written as :
Z Z Z Z Z
(v )
δWvir = Bi δui dVol + Ti δui dS
Vol S1
Z Z Z
δWint = τij δeij dVol
Vol

where δui is the virtual displacement field, Bi is the body force

(v )
field, Ti is the external surface traction acting on S1 part of the
boundary, v is the unit normal to the body. τi j are the nine
internal stress components (i, j = 1,2,3) and δei j are the virtual
strain components.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 14
Principle of Virtual Work (contd...)

Also note that a repeated index indicates a summation over that

index (i.e. ui vi = u1 v1 + u2 v2 + u3 v3 ). Following this convention,
the right hand side in the second equation is a sum of nine
components.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 15
Ritz Method

For conservative forces, the principle of virtual work and the

principle of stationary potential energy lead to the same equation,
namely δ(U + V ) = 0. Here U is the strain energy and V is the
potential of the applied loads.
For a general 3-D structure the Strain Energy is given as
Z Z Z
U= dUdVol
Vol

Where dU is the strain energy density and is given as

Z eij
dU = τij d eij
0

For a linear elastic material, dU = 12 τij eij Also recall that following
the summation convention, the right hand side is a sum of nine
terms.
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 16
Ritz Method (contd...)

The potential V of the applied loads for a general 3-D structure is

given as
 
Z Z Z Z Z
(v )
V = − Bi ui dVol + Ti ui dS 
Vol S1

Note that δV = −δWvir

For a discrete load, Q, with a corresponding displacement, q, V

can be written as:
V = −Qq

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 17
Example

A Cantilever Beam Restrained By a Spring.

The expression for the strain energy for a beam supported by a

spring at the free end (x = L) is given as
2
d 2w
Z L  2
1 1 
U= EI dx + k w |(x =L)
2 0 dx 2 2
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 18
Example (contd..)

The expression for the potential of the applied load V is given as

Z L
V =− p (x )w (x )dx
0

where w (x ) is the transverse deflection of the beam.

The approximate solution is obtained by using the principle of
minimum total potential energy (Π):

δΠ = 0

where Π = U + V .

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 19
Example (contd...)

To obtain the approximate solution, we assume the transverse

deflection as:
N
w (x ) = ∑ Ci φi (x )
i =1

where φi (x ) are the trial functions.

Since there are no prescribed non homogeneous displacements and

slope boundary conditions, we do not need φ0 (x ).

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 20
Example (contd...)

The trial functions must satisfy some requirements:

(i) functions should be linearly independent. This means that the
following equation will hold if and only if all Ci0 ’s are zero.

N
∑ Ci φi (x ) = 0
i =1

(ii) Functions must satisfy the essential boundary conditions.

For the problem at hand, it means that both φi (x ) = 0 and
φi0 (x ) = 0 at x = 0.
Note that in the Ritz method the force (natural) boundary
conditions are not explicitly satisfied. As the number of terms in
the expansion for the dependent variable is increased, the natural
boundary conditions tend to be satisfied naturally (Hence the
name Natural Boundary Conditions).
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 21
Example (contd...)

(iii) The trial functions must form a complete set. This means
that the solution of the problem can be approximated arbitrarily
close by a linear combination of the given trial functions as the
number of terms in the expansion are increased.

If we are using sine (or cosine) series, we must use

sin(ix )(cosine(ix )) before we use sin(i + 1)x (cosine(i + 1)x ).
Similarly as we use (x )i +1 , we must use (x )i .

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 22
Example (contd...)

Substituting the expansion for w (x ) in the expression for total

potential energy, and using the principle of minimum potential
energy, we get:
N
δΠ
δΠ = ∑ δCi = 0
i =1 δCi

Since the variation in each of the generalized displacements Ci ’s

arbitrary and independent from the variation in the rest of the
generalized displacements Ci ’s, it means that

∂Π
= 0; 1≤i ≤N
∂Ci

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 23
Example (contd...)

This will yield a set of N equations. For the present beam problem,
∂Π
the expression for can be written as:
∂Ci
 2 
∂ ∂2 w
Z L Z L
∂Π ∂ w ∂wL ∂w
= EI 2 2
dx + kw L − p (x ) dx
∂Ci 0 ∂x ∂Ci ∂x ∂Ci 0 ∂Ci
" # " #
Z L N N
=
0
∑ Cj φj00 (x ) φi00 (x )dx + k ∑ Cj φj (L) φi (L)
j =1 j =1
Z L
− p (x )φi (x )dx
0
N Z L

= ∑ 0
EI (x )φi00 (x )φj00 (x )dx + kφi (L)φj (L) Cj
j =1
Z L N
−
0
p (x )φi (x )dx = ∑ Kij Cj − Qi
j =1
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 24
Example (contd...)

Where the stiffness coefficient Kij and the generalized force Qi are
given as:
Z L
Kij = EI (x )φi00 (x )φj00 (x )dx + kφi (L)φj (L)
0
Z L
Qi = p (x )φi (x )dx
0
For the present problem, we can choose
 x  i +1
φi (x ) =
L
where i varies from 1 to N.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 25
Example (contd...)

Note that these trial functions satisfy all the essential boundary
conditions, namely,
φ(0) = 0 and φ 0 (0) = 0.

Also the φ 0 ’s are linearly independent and note that to satisfy the
essential boundary conditions, we are starting our expansion from
 x 2  x 0
(for i = l) and not from , i.e. a constant term as
L L
required by the completeness requirement.
The solution to our problem does not contain any constant or
linear terms in the expansion. The present expansion will thus
satisfy the completeness requirement.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 26
Example (contd...)

Substituting the chosen trial functions in the expressions for the

stiffness matrix and the generalized forces (note (φi (L) = 1), we
obtain:
Z L  i +j −2 
x
Kij = EI (x )(i )(j )(i + 1)(j + 1) dx + k
0 Li +j +2
    
(i )(j )(i + 1)(j + 1) EI EI
= +α
(i + j − 1) L3 L3
Z L     i +1
x x
Qi = p0 dx
0 L L
p0 L
=
(i + 3)
Using the above expressions we can obtain the approximate
solutions using various values of N.
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 27
Example (contd...)

Case (i) (n = 1) The set of linear equations for this case can be
written as:
EI p0 L
3
( 4 + α ) C1 =
L 4
p0 L
C1 =
4(4 + α)EI

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 28
Example (contd...)

Case (ii) (n = 2) The set of linear equations for this case can be
written as:
    1 
EI 4 + α 6 + α C1
= p0 L 41
L3 6 + α 12 + α C2 5
Solution of this set of two equations will give us the values of the
undetermined constants C1 and C2 .

p0 L4
   
C1 36 + α
=
C2 20EI (12 + 4α) −(14 + α)

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 29
Example (contd...)

Case (iii) N = 3 The set of three linear equations is:

1
    
4+α 6+α 8+α C1  4
EI   
1

6+α 12 + α 18 + α  C2 = p0 L 5
L3
8+α 18 + α 1445 +α C3 1
   
6

Using Gauss elimination, the solution is

(8α + 123)p0 L4
C1 =
240EI (α + 3)
(13α + 72)p0 L4
C2 = −
240EI (α + 3)
p0 L4
C3 =
48EI

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 30
Example (contd...)

Case (iv) N = 4
The set of four equations is:
 1
   
4+α 6+α 8+α 10 + α  C1   4 
1
   
EI  6 + α 12 + α 18 +α 24 + α 

C2
 
5

3
 144
 = p0 L 1
L  8 + α 18 + α 5 +α 40 + α  
 C3 
 
 6 

400 1
10 + α 24 + α 40 + α 7 +α C4
   
7

Solution of these four equations using Gauss elimination yields:

(3α + 20)p0 L4
(7α + 20)p0 L4 C2 = −
C1 = 80EI (α + 3)
240EI (α + 3)
p0 L4
C3 = 0 C4 =
120EI
It can be shown that this is also the exact solution to this problem.
As a result, we do not have to proceed any further.
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 31
Example (contd...)

I For Complex problem exact solution is not known

I We have to decide how many terms in the expansion will yield
an accurate solution
I We stop when we see that addition of more terms does not
make any significant change in the desired response quantity
(generally stresses)
I For a given number of terms, the accuracy of the stresses will
be less than that for the displacements
I Differentiation of an approximation is almost always less
accurate than the function

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 32
Penalty Approach

Solution for various values of the spring constant can be easily

determined by using appropriate value of α.

As a special case, consider the case of a simply-supported end at

x = L. This case is achieved by letting the value of the spring
constant to be infinite (α → infinity). The limiting value of the
solution is:
p0 L4 7  x 2
 
3  x 3 1  x 5
w (x ) = − +
EI 240 L 80 L 120 L

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 33
Penalty Approach (contd...)

Note that in obtaining this solution we did not satisfy the essential
boundary condition at x = L explicitly. This suggests a way around
satisfying the essential boundary conditions explicitly, by placing a
spring with very large stiffness at the point where the displacement
is zero. This approach is called the penalty approach.

Instead of looking for trial functions that are zero at x = 0, and,

x = L and have slope = 0 at x = 0, we only satisfied the essential
boundary conditions at x = 0 exactly and will satisfy the essential
boundary condition at x = L only approximately.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 34
Limitations of the Ritz Method

In the above approach, difficulty is that the selection of trial

functions is very time consuming and for complicated domains with
complex set of restraints, it may well be impossible to come up
with a set of trial functions that will satisfy all the essential
boundary conditions.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 35
Finite Element Method

In finite element method, in which the domain is divided

(“discretized”) into a large number of smaller domains ( called
finite elements) and simple polynomials that are defined only
locally (i.e. only over a given finite element) are used as trial
functions. Using either a weighted residual method or Ritz
method, a set of algebraic equations are derived for every finite
element and are then assembled (in a procedure called
“assemblage”) together to form a global set of equations.
Another advantage the finite element method has over the global
methods is that the matrix K is banded (sparse). This is due to
the fact that a given trial function is defined only locally.
It can easily solve linear and nonlinear problems and static and
dynamic problems.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 36
Finite Element Method (contd...)

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 37
Finite Element Method (contd...)

Consider the solution of the ordinary differential equation

governing the axial response of a bar with axial rigidity EA(x ), and
subjected to an axial force p (x ):
 
d du
EA(x ) = −p (x )
dx dx
Let the boundary conditions be : u (0) = 0 and u (L) = 1. An
approximate solution of this equation can be found by satisfying
the stationary condition for the total potential potential energy Π,
which for an axial bar is given as:
 2
1 L
Z L
du
Z
Π= EA(x ) dx − p (x )u (x )dx
2 0 dx 0

In the finite element method, we assume u (x ) to be :

N
u (x ) = ∑ ai φi (x )
i =1
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 38
Finite Element Method (contd...)

In the finite element method, the generalized coefficients are the

values of the dependent variables (the axial displacement u in the
present case) at the nodes (i.e ai = ui ). This is an important
difference between the finite element method and the Ritz method.

In the Ritz method, the generalized co-ordinates ai ’s, in general,

do not have any physical significance.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 39
Finite Element Method (contd...)

By taking the first variation of the potential energy with respect to

the nodal displacement values, we obtain:
N
∂Π
δΠ = ∑ ui = 0
i =1 ∂ui

Since the variation of each of the generalized coordinates (nodal

displacements ui ’s) is arbitrary and independent from the variation
of all the other generalized co-ordinates, we get a set of N
equations:
∂Π
= 0, i = 1, 2...N
∂ui

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 40
Finite Element Method (contd...)

The left hand side in the above equation can also be written as:
Z L     Z L
∂Π ∂u ∂ ∂u ∂
= EA(x ) dx − p (x ) (u (x ))dx
∂ui 0 ∂x ∂ui ∂x ) 0 ∂ui

" ! #
Z L N
=
0
EA(x ) ∑ uj φj0 (x ) φi0 (x ) − p (x )φi (x ) dx
j =1
N Z L
 Z L
= ∑ 0
EA(x )φi0 (x )φj0 (x )dx uj −
0
p (x )φi (x )dx
j =1

where a prime indicates the derivative with respect to x.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 41
Finite Element Method (contd...)

For a linear problem, the resulting set of equations will be linear

and can be written as: Ka = Q. Here,

K → the stiffness matrix

a → the vector of nodal displacements
Q → the nodal load vector.

The elements of the stiffness matrix K, and the load vector Q are
given as:
Z L
Kij = EA(x )φi0 (x )φj0 (x )dx
0
Z L
Qi = p (x )φi (x )dx
0

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 42
Finite Element Method (contd...)

The stiffness coefficient Kij have a physical meaning: it is the force

required at node i due to a unit displacement at node j provided all
other displacements are zero. Note that the stiffness matrix is
symmetric but sparse. For example, the stiffness coefficients K13 ,
K14 , ..... = 0. To illustrate this , let us examine the stiffness
co-efficient K13 . Z
K13 = φ10 (x )φ30 (x )dx

Product is zero because the trial function φ1 (x ) is zero on that

part of the domain on which the trial function φ3 (x ) is non-zero
and vice-versa. Similarly we can see that for the second-order
ordinary differential equations, the stiffness matrix will have a
bandwidth of 3.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 43
Finite Element Method (contd...)

For example, in the third row, the three nonzero coefficients will be
: K32 , K33 , and K34 , and in the i th row, the three non-zero
coefficients will be Ki,i −1 , Ki,i , and Ki,i +1 . Let Pi be an axial force
acting at node i.

In that case, the ith generalized force will be modified as

Q i = Pi + Qi

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 44
Finite Element Method (contd...)

This can be easily seen from the fact that if a point force Pk is
acting at the k th node, the expression for the total potential energy
will be modified to contain an additional term, −Pk uk . When the
variation of this term is taken, it leads to the additional force term
Pi in the expression for the generalized force Q i .
If for the given problem, we divide the domain (0, L) using four
elements (or five nodes), the resulting set of equations will be as
follows:
    
K11 K12 0 0 0 
 u1 = 0 
 
 Q 1 

 K21 K22 K23 0 0 
 u2





 Q 2



 
 0 K32 K33 K34 0 
  u3 = Q 3
 0 0 K43 K44 K45   u4
 
Q

 4 
 
 
 
  
0 0 0 K54 K55 u5 = 1
   
Q5

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 45
Finite Element Method (contd...)

Note that in the above equation, there are five unknowns

including: the two generalized forces Q 1 and Q 5 and the three
generalized (nodal) displacements u2 , u3 , and u4 .

Also note that the 5 by 5 global stiffness matrix K in the above set
of linear equations is singular. This is due to the fact that the axial
bar is not restrained from moving freely in the space.

If we determine the eigenvalues of this global stiffness matrix, one

of the eigenvalue will be found to be zero and all the elements in
the corresponding eigenvector will be unity (i.e. a rigid body
motion in the x direction).

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 46
Finite Element Method (contd...)

For numerical results to converge, it is very important that the

global stiffness matrix be singular and contain appropriate rigid
body modes.

Lack of this singularity will imply that a rigid body motion (in our
case, all ui = 1) will result in a nonzero force vector, i.e. a free
body motion will result in a non-zero axial strain which is of course
not true.

To obtain a unique solution for a given problem we remove this

singularity by using the essential boundary conditions. In the
present problem, this will be accomplished by substituting u1 = 0
and u5 = 1.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 47
Finite Element Method (contd...)

The three displacements can be found by solving the following

reduced set of equations:
      
K22 K23 0  u2   Q2   0 
 K32 K33 K34  u3 = Q3 − 0 u
 5
0 K43 K44 u4 Q4 K45
    

The first vector on right hand side can be obtained by performing

the required integrals in the expressions for generalized forces and
the second vector contains products of some stiffness coefficients
with the generalized displacement u5 = 1.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 48
Finite Element Method (contd...)

Once the three unknown displacements are determined from the

above equations, we can obtain the unknown generalized forces Q 1
and Q 5 by substituting the values of all the nodal displacements in
the full stiffness matrix. Note that Q 1 represents the reaction at
the node 1 and the Q 5 represents the force required to produce
u5 = 1.

We can easily change the boundary conditions in the finite element

method. For example, suppose that we have applied a force P at
the free edge (x = L). In that case, the displacement at the free
edge (u5 if we divide the domain in four elements) will be unknown
but the force Q 5 = Q5 + P5 will be known.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 49
Finite Element Method (contd...)

As another example, suppose we have a spring of stiffness k at the

free end (x = L). This is equivalent to applying a force P5 = −ku5
at the free end. Since this term contains the unknown u5 we can
take it to the left hand side in the fifth (last) equation. This will
change the stiffness coefficient K55 to K55 + k.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 50
Example

∂4 w
EI = p (p is uniform)
∂x 4

w (0) = w 0 (0) = 0
EIw 00 |x =L = M0
EI (w 00 ) 0 = 0
Z L  
1 00 2 ∂w 
I (w ) = EI (w ) − pw dx − M0
0 2 ∂x x =L
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 51
Example (contd...)

N
Assume w (x ) = ∑ Ci φi (x ) (N-parameter Ritz method)
j =1

φi must satisfy essential boundary conditions. i.e. φ(x ) = 0 at

x = 0; φ 0 (0) = 0
Furthermore, the φi ’s should be linearly independent
i.e.
a1 φ1 + a2 φ2 + · · · + an φn = 0

if
a1 = a2 = · · · = an = 0

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 52
Example (contd...)

i.e. if we chose φ1 = x 2 , we cannot choose another φj = ax 2 ,

where a is a constant.

Let φj (x ) = x j +1 , j = 1, 2, 3, · · · , N

since x j +1 = 0 at x = 0

∂x j +1
= (j + 1)x j = 0
∂x
at
x = 0

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 53
Example (contd...)

Furthermore,
x j +1 is linearly independent of x j +2 , x j +3 , · · ·
w = C1 x 2 + C2 x 3 + C3 x 4 + C4 x 5 + · · · + CN x N + 1

if N = 2, we stop at x 3
if N = 3, we stop at x 4
.. ..
. .
N
w = ∑ Cj x j + 1
j =1

N
w 00 = ∑ Cj (j + 1)(j )x j −1
j =1

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 54
Example (contd...)

 !2 !
N N
1 L
Z
I =
2 0
EI ∑ Cj φj00 − p ∑ Cj φj  dx
j =1 j =1
"  #
N 
− M0 ∑ j 
C j φ 0
j =1 x =L

= I [C1 , C2 , C3 , · · · , CN ]

∂I ∂I ∂I
= 0 = 0··· = 0
∂C1 ∂C2 ∂CN

which are a set of linear equations.

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 55
Example (contd...)

" #
Z L N Z l
∂I EI
∂Ci
=
0
2·
2 ∑ Cj φj00 φi 00 dx −
0
pφi dx − M0 φi0 |x =L = 0
j =1

for i = 1, 2, · · · , N

N Z L
 Z l
∑ 0
EI φi00 φj00 dx Cj =
0
pφi dx + M0 φi0 |x =L
j =1

for i=1,2,· · · ,N

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 56
Example (contd...)

Let
Z L
Kij = EI φi00 φj00 dx = B (φi , φj )
0
Z L
fi = pφi dx + M0 φi0 |x =L = l (φi )
0
N
∑ Kij Cj = fi
j =1
for i=1,2,3,· · · ,N
 
K11 K12 · · · K1N    
 K21 K22 · · · C1 f1
K2N 

 
 
 

 C2  f2
  
  

=
 
  .. ..
 .. .. .. ..   .   . 
 . . . . 
 
 
 

CN fN
   
KN1 KN2 · · · K1N
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 57
Example (contd...)

Note Kij = Kji

[K ]{C } = {f }
where [K] is the stiffness matrix, { C } is the generalized
displacement vector, and { f } is the generalized load vector.
Also it can be shown that [K] is positive definite. So,

{ C } = [ K ] −1 { f }

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 58
Example (contd...)

Let N=2: w = C1 x 2 + C2 x 3
i.e. φ1 = x 2 , φ2 = x 3 , φ100 = 2, φ200 = 6x
Z L Z L
K11 = EI φ100 φ100 dx K12 = K21 = EI φ100 φ200 dx
0 0
Z L Z L
= EI · 2 · 2 dx = EI · 2 · 6x dx
0 0
= 4EIL = 6EIL2

Z L Z L
K22 = EI φ200 φ200 dx = EI · 6 · 6 dx
0 0
= 12EIL3

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 59
Example (contd...)

Z L Z L
f1 = pφ1 dx + M0 (φ10 )|x =L f2 = pφ2 dx + M0 (φ20 )|x =L
0 0
Z L Z L
= px 2 dx + M0 (2x )|x =L = px 3 dx + M0 (3x 2 )|x =L
0 0
pL3 pL4
= + 2M0 L = + 3M0 L2
3 4
( )
pL3
4L 6L2
  
C1 3 + 2M0 L
EI = pL4
6L2 12L3 C2 2
4 + 3M0 L

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 60
Example (contd...)

 −1 ( pL3
)
4L 6L2
  
C1 1 3 + 2M0 L
= pL4
C2 EI 6L2 12L3 + 3M0 L2
4

( pL3
)
12L3 −6L2

1 1 3 + 2M0 L
= · pL4
EI 12L4 −6L2 4L + 3M0 L2
4

4pL5 + 24M0 L4 − 46 pL6 − 18M0 L4

 
1
=
12EIL4 −2pL5 − 12M0 L3 + pL4 · L + 12M0 L3

5 6 + 6M0 L4
 
1 2 pL
=
12EIL4 −pL5
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 61
Example (contd...)

w (x ) = C1 x 2 + C2 x 3
5pL2 + 12M0 −pL
   
2
= x + x3
24EI 12EI

Now , for N = 3

w (x ) = C1 x 2 + C2 x 3 + C3 x 4

A better approximation
Objective : To determine C1 , C2 , C3

φ1 = x 2; φ10 = 2x; φ1 = 2
φ2 = 3
x ; φ20 = 3x ; 2
φ2 = 6x
φ3 = 4
x ; φ30 = 4x 3 ; φ3 = 12x 2
Rakesh
c K. Kapania AOE 5024, Vehicle Structures 62
Example (contd...)

K13 = K31 K23 = K32

Z L Z L
= EI φ100 φ300 dx = EI φ200 φ300 dx
0 0
Z L Z L
= EI · 2 · 12x 2 dx = EI · 6x · 12x 2 dx
0 0

K13 = 8EIL3 K23 = 18EIL4

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 63
Example (contd...)

Z L
Z L
f3 = pφ3 dx + M0 (φ10 )|x =L
K22 = EI φ300 φ300 dx 0
0
Z L
p · x 4 dx + M0 · 4x 3 x =L

Z L
=
= EI · (12x 2 )2 dx 0
0

144EIL5
K22 = pL5
5 f3 = + 4M0 L3
5

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 64
Example (contd...)

 
PL3
8l 2
  
4 6L  C1  
 3 + 2M0 L 
EIL  6L 12L2 18L 3 C2 = PL4 2
4 + 3M0 L
144 4
8L2 18L3 5 L C3 PL5
+ 4M0 L3 
  
 
5

SOLVING GIVES:

pL2 + 2M0
C1 =
4EI
−pL
C2 =
6EI
p
C3 =
24EI

Rakesh
c K. Kapania AOE 5024, Vehicle Structures 65