Introduction To FEM

By: Yishak Tamiru

 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)

2
 Introduction
• It is originated in the aerospace industry as a tool to study stresses in
complex air- frame structure.
• It is numerical technique for obtaining approximate solution to a wide
variety of engineering problem.
• Analytical methods provide accurate solution with applications limited to
simple geometries.
• Experimental methods are used to test prototypes or full scale models.
• And they are costly and may not be feasible in some cases.
• But Numerical methods are most efficient technique for engineering
analysis which can treat complex geometries also.

3
• Historical Background

Open surface water channel used in flow visualization

Direct observation of the flow field

The wake of a cylinder visualized by surface powder

8
• Nowadays

Transient flow over a cylinder using fluent (X velocity)

[email protected] 9
Fig. Ref.: Fig. Ref.:
Golman, A.J., et al., Injury prediction in a side impact crash using human body Pack, K., S.J.J.I.J.o.S. Marcadet, and Structures, Numerical failure analysis of
model simulation. 2014. 64: p. 1-8. three-point bending on martensitic hat assembly using advanced plasticity and
fracture models for complex loading. 2016. 85: p. 144-159. 10
[email protected]
• Among many FEA/FEM is the most versatile & compressive numerical
technique in the hands of engineers today.
• FEM is used to obtain approximate solutions of boundary value
problems in engineering.
• Boundary value problems is a mathematical problems with one or more
dependant variables satisfying different equation within the known
domain of independent variable & satisfying specific conditions on the
boundary of the domain.
• The boundary conditions are the specified values of the field variables
(or related variables such as derivatives) on the boundaries of the field.

12
• Depending on the type of physical problem being analyzed, the field
variables may include physical displacement, temperature, heat flux, fluid
velocity, etc.
• FEM allows for descretizing the intricate geometries in to small fundamental
volumes called finite elements.
• It is then possible to write the governing equations & material properties for
these elements &this will be in terms of unknown values at points called
nodes.
• These equations are then assembled by taking proper care to loads &
constraints which results in a set of equation which are to be solved & the
results describe the behavior of the original complex body being analyzed.
The solution to these equations would provide an exact closed form solution
to the particular problem being studied.

13
 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

20
Advantages of FEM
Increase productivity (reduce uncertainty)
Minimize physical prototyping (optimize performance before
prototyping)
Innovative products in less time & less costs (reduced testing and
redesign costs thereby shortening the product development time)
Applicable to linear and non linear problem
Easily applied to complex, irregular shaped objects composed of several
different materials and having complex boundary conditions
Applicable to steady state, time dependant and eigenvalue problems
One method can solve a wide variety of problems, including problems in
solid mechanics, chemical reaction, electromagnetics, biomechanics, heat
transfer and acoustics.
22
Eg. Stress Analysis

23
Disadvantages of FEM
Experience and judgment are needed in order to construct a good
finite element model
A powerful computer and reliable FEM software are essential.
Input and output data may be large and tedious to prepare and
interpret.
The FEM is applied to an approximation of the mathematical model
of a system(the source of so called inherited error)
Susceptible to user introduced modeling error:
1. poor choice of elements types
2. distorted elements
3. geometry not sufficiently modeled

24
 Application of FEM in Engineering
FEM is not limited to mechanical system alone but to a range of engineering
problems such as
• Mechanical/Aerospace/Civil/Automobile Engineering
• Structure analysis (static/dynamic, linear/nonlinear)
• Stress analysis
• Dynamic analysis
• Deformation studies
• Thermal /Heat flow analysis
• Fluid flow analysis
• Electromagnetic / Magnetic flux studies
• Geo-mechanics
• Biomechanics 25
 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)

26
 BASIC STEPS IN THE FINITE ELEMENT METHOD
The basic steps involved in any finite element analysis consist of the following:
Preprocessing Phase
1. Create and discretize the solution domain into finite elements; that is, subdivide the problem into
nodes and elements.
2. Assume a shape function to represent the physical behavior of an element; that is, an approximate
continuous function is assumed to represent the solution of an element.
3. Develop equations for an element.
4. Assemble the elements to present the entire problem. Construct the global stiffness matrix.
5. Apply boundary conditions, initial conditions, and loading.
Solution Phase
6. Solve a set of linear or nonlinear algebraic equations simultaneously to obtain nodal results, such as
displacement values at different nodes or temperature values at different nodes in a heat transfer
problem.
Post processing Phase
7. Obtain other important information. At this point, you may be interested in values of principal
stresses, heat fluxes, etc.
27
 In the finite element modeling process there are Six Steps

Step 1 - Discretization: The problem domain is discretized into a

collection of simple shapes, or elements.

Step 2 - Develop Element Equations: Developed using the physics of the

problem, and typically Galerkin’s Method or variational principles.

Step 3 - Assembly: The element equations for each element in the FEM
mesh are assembled into a set of global equations that model the
properties of the entire system.

28
 The six steps…
Step 4 - Application of Boundary Conditions: Solution cannot be obtained unless
boundary conditions are applied. They reflect the known values for certain primary
unknowns. Imposing the boundary conditions modifies the global equations.

Step 5 - Solve for Primary Unknowns: The modified global equations are solved for the
primary unknowns (Elemental Displacements) at the nodes.

Step 6 - Calculate Derived Variables (Elemental stress, strain, reaction forces):

Calculated using the nodal values of the primary variables.

And this six steps, categorized in the FEA program to three main stages as follows:

29
 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

30
Computer Implementations
• Preprocessing (build FE model, loads and constraints)

• FEA solver (assemble and solve the system of equations)

• Post-processing (sort and display the results)

31
 Typical elements commonly used in finite Element analysis

• The type of elements to be used for generating the mesh depends upon the actual
geometry and the type of problem being considered. Some typical elements generally
found in most of the FE software are shown below
a) One dimensional

2 node lines (BE2) 3 node quadratic (BE3)

b) Two dimensional
Triangle

3 - Node lines (TR3) 6 -node quadratic (TR6)

32
Quadrilateral

4-Node linear (QU4) 8 - node quadratic (QU8)

c) Three dimensional
Tetrahedral

Linear quadratic

4 - Node linear
Quadratic
quadrilateral
33
35
 Modeling recommendations

1. Maintain compatibility : e.g. Do not attach quadratic quadrilateral elements to linear

quadrilateral

2. Adopt fine mesh in regions of high stress

3. Reduce bandwidth by suitable numbering of nodes.

• Band width of matrix is the number of columns from first to the last non-zero column.

• The element band width = max. Difference between the largest and smallest node
Numbers of the element

The total number of band with = element band width +1

39
e.g.
1 2 3 4 5

Element band width = 5-4 = 1

Total number of band width = 1 + 1 = 2
e.g. 6 10

12
14 21

BW=21 - 10, Total No. BW= 11 + 1 = 12

4. Exploit symmetry of elements wherever possible methods of element refinement as

per point 2 above
P -method: using size of element but of a higher order interpolating polynomial
h - method: using same element type, but of smaller size

40
 Formulation of finite element problems
In general, there are several approaches to formulating finite element problems:

(1) Direct formulation,

(2) The Minimum Total Potential Energy Formulation,

(3) Rayleigh – Ritz Method and

(4) Weighted Residual Formulations

Again, it is important to note that the basic steps involved in any finite element
analysis, regardless of how we general the finite element model, will be tile same
as those listed above.

41
46
47
48
 DIRECT FORMULATION
The following problem illustrates the steps and the
procedure involved in direct formulation.

50
 DIRECT FORMULATION
EXAMPLE 1.0
Consider a bar with a variable cross section supporting a load P, as
shown in Figure 1.0 The bar is fixed at one end and carries the load P at
the other end. Let us designate the width of the bar at the top by w1, at
the bottom by w2, its thickness by t, and its length by L. The bar's
modulus of elasticity will be denoted by E. We are interested in
determining how much the bar will deflect at various points along its
length when it is subjected to the load P. We will neglect the weight of
the bar in the following analysis, assuming that the applied load is
considerably larger than the weight of the bar:
 Figure 1.0 A bar under axial lodaing
Preprocessing Phase
1. Discretize the solution domain into finite elements.
The given bar is modeled using four individual segments,with
each segment having a uniform cross section.The cross-
sectional area of each element is represented by an average
area of the cross sections at the nodes that define the
element.This model is shown in Figure2.
 Figure 2.0 Subdividing the bar in to elements and nodes

2. Assume a solution that approximates the behavior of an

element.
In order to study the behavior of a typical element, let's
consider the deflection of a solid member with a uniform cross
section A that has a length ℓ when subjected to a force F, as
shown in Figure 3.
The average stress rr in the member is given
by σ = F A (1)
The average normal strain e of the member is
defined as the change in length ∆ℓ per unit
original length ℓ of the member:
ε = ∆l l (2)
Over the elastic region, the stress and strain
are related by Hooke's Law, according to the
equation
σ = Eε (3)
where E is the modulus of elasticity of the
material. Combining Eqs. (1), (2), and (1.3)
and simplifying, we have
 AE 
F = 
 l
 ∆l

(4)

Note that Eq. (4) is similar to the equation for

a linear spring, F = kx. Therefore, a centrally
 loaded member of uniform cross section may
be modeled as a spring with an equivalent
stiffness of
k eq =
AE
l
(5)

Figure 3.0 A solid member of uniform crossection subjected to a force F

the bar is represented by a model consisting of
four elastic springs (elements) in series, and
the elastic behavior of an element is modeled
by an equivalent linear spring according to the
equation
 f = keq (Ui+1 +
_ Ui) = Aavg E
l
(U i +1 − U i ) =
( Ai +1 + Ai )
2l
E (U i +1 − U 1 )

(6)
where the equivalent element stiffness is given
by
Keq = ( Ai +1 + Ai )
2l
E (7)

Ai and Ai+l are the cross-sectional areas of the member at

nodes i and i+1, respectively, and ℓ is the length of the
element.
The free body diagram of nodes, which shows the
forces acting on nodes 1 through 5 of this model, is
depicted in Figure 4. Static equilibrium requires that the
sum of the forces acting on each node be zero. This
requirement creates the following five equations:
node 1: R1 -- kl(U2 - Ui) = 0 R1

node2:kl(U2-Ul)-k2(u3-u2)=0 (8)
Node 1

K1(U2-U1)

node 3: k2(u3-u2)-k3(u4-u3) = 0 Node 2

K2(U3-U2)

node 4: k3(u4 - u3) - k4(u5 - u4) = 0 Node 3

node 5: k4(us-u4) – P = O
K3(U4-U3)

Node 4

K4(U5-U4)

Node 5

Figure 4.0 Free body diagram of the nodes P

Rearranging the equilibrium equations given by Eq. (8)
by separating the reaction force R1 and the applied
external force P from the internal forces, we have:
kl U1 - klU2 = - R1 .................................. (a)
klU2 - kl Ul - k2 u 3 + k2 u2 = 0 ....................(b)
(9)
k2 u 3 - k2 u 2 -k3 u 4 + k3 u 3 = 0 .........................(c)
k3 u 4 - k3 u 3 - k4 u 5 +- k4 u 4 = 0 .......................(d)
k4 u s- k4 u 4 = P..............................................(e)

Presenting the equilibrium equations of Eq. (9) in a

matrix form, we have:

 k1 − k1 0 0 0  u1  − R1 
 − k1 k1 + k 2 − k2 0 0  u2   0 
     
 0 − k2 k 2 + k3 − k3 0



u3  =

 0

 (10)
  
 0 0 − k3 k3 + k 4 − k 4 u4   0 
   

 0 0 0 − k4 k4  
 u5 
  P
 

It is also important to distinguish between the reaction
forces and the applied loads in the load matrix.
Therefore, the matrix relation of Eq. (10) can be written
as:

− R1   k1 − k1 0 0 0  u1   0
 0   − k1 k1 + k 2 − k2 0 0  u2   0 
     
 0  =  0 − k2 k 2 + k3 − k3 0  u3  −  0

(11)
 0   
 0 0 − k 3 k 3 + k 4 − k 4  u4   0
     
 0   0 0 0 − k4 k 4   u 5   P 

We can readily show that under additional nodal loads and other fixed boundary
conditions, the relationship given by Eq. (11) can be put into the general form
{R} = [K] {u} – {F} (12)
 which stands for:
{reaction matrix} = [stiffness matrix] {displacement matrix} - {load matrix}

we find that because the bar is fixed at the top, the

displacement of node 1 is zero. Thus, the first row of the
system of equations given by Eq. (10) should read u1 = 0.
Thus, application of the boundary condition leads to the
following matrix equation:
 1 0 0 0 0  u1  0
 − k1 k1 + k 2 − k2 0 0  u2  0
    
 0 − k2 k 2 + k3 − k3 0  u3  =  0  (13)
   u4  0
 0 0 − k 3 k 3 + k 4 − k 4
   
 0 0 0 − k4 k 4   u 5   P 
The solution of the above matrix yields the nodal
displacement values. In the next section, we will develop the
general elemental stiffness matrix and discuss the
construction of the global stiffness matrix by inspection.
 Develop equations for an element.
• Because each of the elements in Example 1.1 has
two nodes, and with each node we have
associated a displacement, we need to create two
equations for each element.
• These equations must involve nodal displacements
and the element's stiffness. Consider the internally
transmitted forces fi and fi+1 and the end
displacements ui and ui+1 of an element, which are
shown in Figure 5.
Static equilibrium conditions require that the sum
of fi and fi+1 be zero, so that fi and fi+1 are given
in the positive y-direction. Thus, we write the
transmitted forces at nodes i and i+1 ac-
cording to the following equations:
fi = k eq (u i − u i +1 )
f i +i = k eq (u i +1 − u i )
(14)
Equation (1.14) can be expressed in a matrix
form by

 fi 
  = 
f
 k eq − k eq   ui 
   (15)
 i+1   eq eq   ui+1 
− k k
Figure 5.0 Internally transmitted forces through an
arbitrary element.

Assemble the elements to present the entire

problem.
Applying the elemental description given by Eq. (15) to
all elements and assembling them (putting them
together) will lead to the formation of the global
stiffness matrix. The stiffness matrix for element (1) is
  k1 − k1 
[k ](1)
=  
− k k
 1 1 
and it position in the global fness matrix is given by

 k1 − k1 0 0 0
 − k1
 k1 0 0 0 
[k ](1G ) = 0 0 0 0 0
 
 0 0 0 0 0
 0 0 0 0 0 

The nodal displacement matrix is shown alongside the

position of element 1 in the global stiffness matrix to
aid us to observe the contribution of a node to its
neighboring elements. Similarly, for elements (2), (3),
and (4), we have
 k2 − k2 
[k ]( 2) = − k k 2 
and its position in the global matrix
 2
  k3 − k3 
[k ]
( 3)
=   and its position in the global matrix
−
 3 k k 3 

 0 0 0 0 0

 0 0 0 0 0 
[k ](3G ) = 0 0 k3 − k3 0
 
 0 0 − k3 k3 0
 0 0 0 0 0
  k4 − k4 
[k ]( 4) =   and its position in the global matrix
− k 4 k 4 

 0 0 0 0 0 

 0 0 0 0 0 
[k ](4G ) = 0 0 0 0 0 
 
 0 0 0 k 4 − k 4
 0 0 0 − k 4 k 4

The final global stiffness matrix is obtained by assembling, or adding, together each
element's position in the global stiffness matrix:

[k ](G ) = [k ](1G ) + [k ](2G ) + [k ](3G ) + [k ](4G )

 k1 − k1 0 0 0 
 − k1
 k1 + k 2 − k2 0 0 
[k ](G ) = 0 − k2 k 2 + k3 − k3 0 (16)
 
 0 0 − k3 k3 + k 4 − k 4
 0 0 0 − k4 k 4 
The gobal stiffness matrix obtained using elemental description is identical to the global
matrix we obtained earlier from the analysis of the FBD of the nodes, as given by the left
hand side of Eq. (10).

Apply boundary conditions and load.

The bar is fixed at the top, which leads to the
boundary condition u1 = 0..the external load P is
applied at node 5. Applying these conditions
results in the following set of linear equations.
 1 0 0 0 0   u1   0 
 − k1 k1 + k 2 − k 2 0 0   u2   0 
    
 0 − k 2 k 2 + k3 − k3 0  u3  =  0  (17)
   u4   0 
 0 0 − k 3 k 3 + k 4 − k 4
   
 0 0 0 − k4 k 4   u5   P 

Solve a system of algebraic equations

simultaneously.
In order to obtain numerical values of the nodal
displacements, let us assume that E = 10.4 x 106 lb/in2
(aluminum), w1 = 2 in, w2 = 1 in, t = 0.125 in, L = 10
in, and P = 1000 lb.
TABLE 1 Properties of the elements in Example 1.0
Element Nodes Average Length(in) Modulus Element's
cross- of stiffness
sectional elasticity coefficient
area (in2) (lb/in2) (lb/in)

1 1 2 0.234375 2.5 10.4 x 106 975 x 103

2 2 3 0.203125 2.5 10.4 x 106 845 x 103
3 3 4 0.171875 2.5 10.4 x 106 715 x 103
4 4 5 0.140625 2.5 10.4 x 106 585 x 103

The variation of the cross-sectional area of the bar in the y-

direction can be expressed by:
  w2 − w1    (1 − 2) 
A( y ) =  w1+   y t =  2 + y (0.125) = 0.25 − 0.0125 y (18)
  L    10 
Using Eq. (18), we can compute the cross-sectional areas at each
node:
A1 = 0.25 in2
A2 = 0.25 - 0.0125(2.5) = 0.21875 in2
A3 = 0.25 - 0.0125(5.0) = 0.1875 in2
A4 = 0.25 - 0.0125(7.5) = 0.15625 in2
A5 = 0.125 in2
Next, the equivalent stiffness coefficient for each element is
computed from the equations
 ( Ai +1 + Ai )
keq = E
2

(0.21875 + 0.25) (10.4 x 10 6 )

k1 = = 975 x 10 3 b
2 (2.5) in

(0.1875 + 0.21875) (10.4 x 10 6 )

k2 = = 845 x 10 3 b
2 (2.5) in

(0.15625 + 0.1875) (10.4 x 10 6 )

k3 = = 715 x 10 3 b
2 (2.5) in

(0.125 + 0.15625) (10.4 x 10 6 )

k4 = = 585 x 10 3 b
2 (2.5) in

and the elemental matrices are

  k − k1  3  975 − 975
[ k ] (1) =  1  = 10 − 975 975
 − k 1 k   

 k − k2  3  845 − 845
[ k ] (2 ) =  2  = 10 − 845 845
 − k 2 k 2   

 k3 − k3  3  715 − 715
[ k ] (3 ) =   = 10 − 715 715
 − k .3 k 3   

 k4 − k4  3  585 − 585
[k ] (4 )
=   = 10 − 585 585
 − k 4 k 4   

Assembling the elemental matrices leads to the generation of the global stiffness matrix:

 975 − 975 0 0 0 
 − 975 975 + 845 − 845 0 0 
 
[k ](G ) 3
= 10  0 − 845 845 + 715 − 715 0 
 
 0 0 − 715 715 + 585 − 585
 0 0 0 − 585 585 
Applying the boundary condition u1 = 0 and the load P = 1000 lb, we get

 1 0 0 0 0   u1 
  0 
 − 975 1820
 − 845 0 0   u2   
3  0 
10  0 − 845 1560 − 715 0   u3  =  
  0
 0 0 − 715 1300 − 585  u4   3 
   10 
 0 0 0 − 585 585   u 5 

Because in the second row, the -975 coefficient gets multiplied by u1 = 0, we need only to
solve the following 4 x 4 matrix:
 1820 − 845 0 0   u2   0 
 
− 845 1560 − 715 0 

 u 3   0 
10 3   = 
 0 − 715 1300 − 585  u4   0 
 
 0 0 − 585 585   u 5   10 3 

Therfoer the displacement solution is: u1 = 0, u2 = 0.001026 in, u3 = 0.002210 in, u4 = 0.003608
in, and u5 = 0.005317 in.
Postprocessing Phase, Obtain other information.
For the given example 1.0, we may be interested in obtaining other information, such as the
average normal stresses in each element. These values can be determined from the equation
Aavg E
f (u − u ) (ui +1 − ui ) (ui +1 − ui )
σ= = k eq i +1 i =  =E (19)
Aavg Aavg Aavg 
Since the displacements of different nodes are known, Eq. (1.19) could have been obtained
directly from the relationship between the stresses and strains,
(ui +1 − ui ) (20)
σ = Eε = E

Employing Eq. (20) , we compute the average normal stress for each element is

σ (1)
=E
(u 2 − u1 )
=
(10.4 x 10 ) (0.001026 − 0) = 4268 b
6

 2.5 in 2
 (u 3 − u 2 )
σ (2 ) = E = 4925 b
 in 2
(u 4 − u 3 )
σ (3 ) = E = 5816 b
 in 2
(u 5 − u 4 )
σ (4 ) = E = 7109 b
 in 2

f=p
f=p

y
P p

P
Figure 6.0 The internal forces
In the above figure, we note that for the given problem,
regardless of where we cut a section through the bar,
the internal force at the section is equal to 1000 lb. So,
f 1000
σ (1) = = = 4267 b 2
Aav 0.234375 in

f 1000
σ (2 ) = = = 4923 b 2
Aav 0.203125 in

f 1000
σ (3 ) = = = 5818 b 2
Aav 0.171875 in

f 1000
σ (4 ) = = = 7111 b 2
Aav 0.140625 in
Ignoring the errors we get from rounding off our answers, we find that these
results are identical to the element stresses computed from the displacement
information. This comparison tells us that our displacement calculations are
good for this problem.
Reaction Forces
For the above problem, the reaction force may be computed in a
number of ways. First, referring to Figure 4, we note that the statics
equilibrium at node 1 requires
R1 = k1 (u2 - ul) = 975 x 103(0.001026 - 0) = 1000 lb
The statics equilibrium for the entire bar also requires that
R1 =P = 1000lb
As you may recall, we can also compute the reaction forces from the
general reaction equation
{R} = [K]{u} - (F}
{reaction malrlx} = [stiffness matrix] {displacemenl matrix} -
{load matrix}
below
below

uU
Temperature effect
Temperature effect
 Thank you

[email protected] 88