FINITE ELEMENT ANALYSIS(IM-405)

TOPIC:MODELING TECHNIQUES

COURSE TEACHER:Ms RABIA SIDDIQUI

 MODELLING TECHNIQUES
In this lecture various modelling techniques will be introduced. Some of these
techniques are a must when carrying out finite element analysis to ensure the
reliability and accuracy of the results obtained. With developments in computer
hardware and software, a FEM analysis can now be performed very easily.
Therefore, FEM packages are very often used as a ‘black box’ for many actual
design projects by analysts who may not have a proper background in finite
element analysis. However, improper use of commercial software can lead to
erroneous results, often hidden behind colourful stress plots or other post-
processed results without the knowledge of the analyst. Having described the
theories and procedures of the FEM, readers should have quite a good idea on
what is really going on in a commercial FE software package.
COURSE TEACHER:Ms RABIA SIDDIQUI
 OBJECTIVES:

• The primary objective of this lecture is therefore to throw some additional light into the
black box, so that one can avoid unnecessary mistakes in creating a FEM model when
using a commercial package.

• Another reason for learning some of these modelling techniques is to improve efficiency
in computing the finite element results, as well as the accuracy of the results. An
experienced analyst should be able to obtain accurate results with as little effort in
modelling and computer resources as possible.

COURSE TEACHER:Ms RABIA SIDDIQUI

 BEING A GOOD ANALYST….
To come up with a good finite element model, the following factors need to be considered:
• Computational and manpower resources that limit the scale of the FEM model.
• Requirement on results that defines the purpose and hence the methods of analysis.
• Mechanical characteristics of the geometry of the problem domain that determine the type
of elements to use.
• Boundary conditions.
• Loading and initial conditions.

COURSE TEACHER:Ms RABIA SIDDIQUI

 CPU TIME ESTIMATION
Despite advances in the computer industry, computer resources can still be one of the decisive
factors on how complex a finite element model can be built. The CPU time required for a static
analysis can be roughly estimated using the following simple relation (called the complexity of a
linear algebraic system.

tCPU  n dof

Where,
ndof =it is the number of total degrees of freedom in the finite element equation system,

α=it is a constant in the range of 2.0 to 3.0, depending on the different solvers used in the FEM
package and the structure of the stiffness matrix.

COURSE TEACHER:Ms RABIA SIDDIQUI

 HOW COMPUTATIONAL TIME FOR STATIC ANALYSIS BE REDUCED?

HOW TO REDUCE THE TOTAL NO. OF DEGREE OF FREEDOM FOR A GIVEN SYSTEM?

EQUATION tCPU  ndof
CLEARLY INDICATES THAT A FINER MESH WITH A LARGE
NUMBER OF DEGREES OF FREEDOM (DOFS) RESULTS IN AN EXPONENTIALLY INCREASING
COMPUTATIONAL TIME. THIS IMPLIES THE IMPORTANCE OF REDUCING THE DOFS.
THEREFORE ONE SHOULD;

1. TO CREATE AN FEM MODEL WITH MINIMUM DOFS BY USING ELEMENTS OF AS LOW A

DIMENSION AS POSSIBLE, AND
2. TO USE AS COARSE A MESH AS POSSIBLE, AND USE FINE MESHES ONLY FOR
IMPORTANT/CRITICAL AREAS. THESE HAVE TO BE DONE WITHOUT SACRIFICING ANY
ACCURACY IN THE RESULTS.

COURSE TEACHER:Ms RABIA SIDDIQUI

 PRACTICAL CONSIDERATIONS IN MODELING
“WHERE FINER MESHES SHOULD BE USED”
The examples in Figure below illustrate some of these “danger regions.”
Away from such regions one can use a fairly coarse discretization within
Use a relatively fine (coarse) discretization in constraints imposed by the need of representing the structural geometry,
regions where you expect a high (low) loading and support conditions reasonably well.
gradient of strains and/or stresses.
Regions to watch out for high gradients(fine
meshes) are:

1. Near entrant corners, or sharply curved

edges.
2. In the vicinity of concentrated (point)
loads, concentrated reactions, cracks and
cutouts.
3. In the interior of structures with abrupt
changes in thickness, material properties
or cross sectional areas.

Some situations where a locally refined finite element

COURSE TEACHER:Ms RABIA SIDDIQUI discretization is recommended
 HOW COMPUTATIONAL TIME FOR STATIC
ANALYSIS BE REDUCED?
Important Factors to consider: b
α(depends upon Bandwidth i.e the structure of stiffness matrix)
    
ndof(the total no of degree of freedom)     0 
One of the very important factors that affect α is the bandwidth of the stiffness matrix.
 
     
Bandwidth is define as” the diagonalized non-zero entries of the stiffness matrix.”    0 
 ndof
 
Mathematically,  sy.   
 
Bandwidth (b)= (D+1) ⋅ f    
 
Where, 
D= the maximum largest difference in the node numbers occurring for all elements

of the assemblage,

f= the number of dof at each node

A smaller bandwidth leads to a smaller value of α, and hence a faster computation. The bandwidth of a global matrix depend upon
node numbering scheme and no. of degree of freedom per node. If we can minimize the bandwidth, storage requirement as well as
solution time can be minimized. Since no.of degree of freedoms per node is generally fixed for any type of given system therefore
,bandwidth can be minimized by using proper node numbering scheme.
COURSE TEACHER:Ms RABIA SIDDIQUI
 AS AN EXAMPLE, CONSIDER A THREE-BAY FRAME WITH
RIGID JOINTS, 20 STORIES HIGH, SHOWN IN FIGURE

Assuming that there are 3 dof per node, there are 252
unknowns in the final equations (including the dof corresponding
to the fixed nodes), and if the entire stiffness matrix is stored in the
computer, it will require 252x252 = 63,504 locations.
Bandwidth=(D+1)*f
D=20,f=3
Therefore, Bandwidth=(20+1)*3=63
Storage requirement of non-zero entries=63*240=15,120 locations
COURSE TEACHER:Ms RABIA SIDDIQUI
 Alternate Solution:(Place the node numbering column wise)

D=4, f=3
Bandwidth=(4+1)*3=15
Total no. of locations for non-zero entries=15*240=3600 locations

COURSE TEACHER:Ms RABIA SIDDIQUI

COURSE TEACHER:Ms RABIA SIDDIQUI
 AUTOMATIC MESH GENERATION:

Mesh generation is the process of dividing a physical domain into smaller subdomains (called elements) to
facilitate an approximate solution of the governing ordinary or partial differential equation. For this, one-
dimensional domains (straight or curved lines) are subdivided into smaller line segments, two-dimensional
domains (planes or surfaces) are subdivided into triangle or quadrilateral shapes, and three-dimensional domains
(volumes) are subdivided into tetrahedron and hexahedron shapes. If the physical domain is simple and the
number of elements used is small, mesh generation can be done manually.

However, most practical problems, such as those encountered in aerospace, automobile, and construction
industries have complex geometries that require the use of thousands and sometimes millions of elements. In
such cases, the manual process of mesh generation is impossible and we have to use automatic mesh
generation schemes based on the use of a CAD or solid modeling package.

COURSE TEACHER:Ms RABIA SIDDIQUI

 Automatic mesh generation involves the subdivision of a given domain, which may be in the form of a
curve, surface, or solid (described by a CAD or solid modeling package)into a set of nodes (or
vertices) and elements (subdomains) to represent the domain as closely as possible subject to the
specified element shape and size restrictions. Many automatic mesh generation schemes use a
“bottom-up” approach in that nodes (or vertices or corners of the domain) are meshed first, followed
by curves (boundaries), then surfaces, and finally solids. Thus, for a given geometric domain of the
problem, nodes are first placed at the corner points of the domain, and then nodes are distributed
along the geometric curves that define the boundaries. Next, the boundary nodes are used to develop
nodes in the surface(s), and finally the nodes on the various surfaces are used to develop nodes
within the given volume (or domain). The nodes or mesh points are used to define line elements if the
domain is one-dimensional, triangular, or quadrilateral elements if the domain is two-dimensional, and
tetrahedral or hexahedral elements if the domain is three-dimensional.

Mesh Generation Schemes:

The most common methods used in the development of automatic mesh generators

are the tesselation and octree methods.

COURSE TEACHER:Ms RABIA SIDDIQUI
TESSELATION METHOD:
In the tesselation method, the user gives a
collection of node points and also an
arbitrary starting node. The method then
creates the first simplex element using the
neighboring nodes. Then a subsequent or
neighboring element is generated by
selecting the node point that gives the
least distorted element shape. The
procedure is continued until all the
elements are generated.

COURSE TEACHER:Ms RABIA SIDDIQUI

Octree methods:
The octree methods belong to a class of mesh generation schemes known as tree structure methods,
which are extensively used in solid modeling and computer graphics display methods. In the octree
method, the object is first considered enclosed in a three dimensional cube. If the object does not
completely (uniformly) cover the cube, the cube is subdivided into eight equal parts. In the two-
dimensional analog of the octree method, known as the quadtree method, the object is first
considered enclosed in a square region. If the object does not completely cover the square, the
square is subdivided into four equal quadrants. If any one of the resulting quadrants is full (completely
occupied by the object) or empty (not occupied by the object), then it is not subdivided further. On the
other hand, if any one of the resulting quadrants is partially full (partially occupied by the object), it is
subdivided into four quadrants. This procedure of subdividing partially full quadrants is continued until
all the resulting regions are either full or empty, or until some predetermined level of resolution is
achieved. At the final stage, the partially full quadrants are assumed to be either full or empty
arbitrarily based on a prespecified criterion.

COURSE TEACHER:Ms RABIA SIDDIQUI

COURSE TEACHER:Ms RABIA SIDDIQUI
COURSE TEACHER:Ms RABIA SIDDIQUI
 PRACTICAL CONSIDERATIONS IN MODELING
Aspect Ratio and Element Shape
The aspect ratio is define as the ratio of the longest dimension to the shortest dimension of
a quadrilateral element.
In general, as the aspect ratio increases, the inaccuracy of the finite element solution
increases.
In general, elements that yield the best results are compact and regular in shape will:(1)
aspect ratios near one; and (2) corner angles of quadrilaterals near 90°.

COURSE TEACHER:Ms RABIA SIDDIQUI

COURSE TEACHER:Ms RABIA SIDDIQUI
COURSE TEACHER:Ms RABIA SIDDIQUI
COURSE TEACHER:Ms RABIA SIDDIQUI
 Practical Considerations in Modeling

Use of Symmetry
1-The use of symmetry will often
expedite the modeling of a
problem.
2-Symmetry allows us to consider
a reduced problem instead of
the actual problem.
3-This will allow us to use a finer
discretization of element with
less computational cost.
COURSE TEACHER:Ms RABIA SIDDIQUI
 PRACTICAL CONSIDERATIONS IN MODELING
Mesh Refinement: Process of modifying a mesh in order to improve the accuracy of
solution is called mesh refinement.
Mesh revision or refinement is used to increase the accuracy of the results produced by a
finite element model.
First, start with a basic mesh using the fewest, reasonable number of elements, and obtain
a benchmark result. Then refine the model by doing actions like increasing the mesh
density and compare the results.
This process continues until the results converge to some user-defined criteria.

COURSE TEACHER:Ms RABIA SIDDIQUI

COURSE TEACHER:Ms RABIA SIDDIQUI
 The mesh should be refined
until convergence is achieved
(i.e., the results change very
little from the previous
refinement). Some software
automates the refinement
process (adaptive meshing).

COURSE TEACHER:Ms RABIA SIDDIQUI

 PRACTICAL CONSIDERATIONS IN MODELING
CONVERGENCE OF SOLUTION
WHEN THE MESH SIZE IS REDUCED - THAT IS THE NUMBER OF ELEMENTS IS
INCREASED - WE ARE ENSURED OF MONOTONIC CONVERGENCE OF THE
SOLUTION WHEN COMPATIBLE AND COMPLETE DISPLACEMENT FUNCTIONS ARE
USED.

COURSE TEACHER:Ms RABIA SIDDIQUI

 Errors in FEA:

COURSE TEACHER:Ms RABIA SIDDIQUI

Errors in FEM:
Error is define as” the difference between exact and approximated value”.

There are 3 main sources of errors in FEA:

• A. Modeling Error (due to simplifying assumptions in mathematical model such as when
using beam elements we assume that the cross-sections stay planar and do not change
shape
• B. Discretization Error (i.e., due to piecewise approximation which can be minimized by
using higher order shape functions or smaller elements)
• C. Numerical Error (due to limited number of significant digits maintained by computer)

COURSE TEACHER:Ms RABIA SIDDIQUI

 Modeling Error: To do a proper FE
analysis, the analyst must understand
how the structure is likely to behave
and how elements are able to behave.
E.g., if the analyst knows the
displacement varies linearly, 4-node
quad. elements will work, but if they
vary quadratically, 8-nodequad
elements must be used.

COURSE TEACHER:Ms RABIA SIDDIQUI

 If a mesh is repeatedly refined, will the results
converge to a solution?
• Yes, if the elements used pass the “patch” test.

Use a patch test or single element test to determine how

an element works under different circumstances. • Study
different states of stress and strain.

In a patch test, the FE model must have:

A simple arrangement of elements with one internal node.

Supports sufficient to stop rigid body motion.

Work equivalent loads consistent with a constant state of stress

(and strain)

• To pass the test, the results must exactly represent the correct
constant stress (and strain), within numerical error.
COURSE TEACHER:Ms RABIA SIDDIQUI
 THE TYPES OF NON LINEARITY IN FEA:
• Geometric Non-Linearity
• Material Non-Linearity
• Boundary Conditions Non-Linearity
An important point to be considered is that, for any non linearity, in the
linear regime the stiffness matrix in FEA tends to be
constant (obvious). Once the non linearity sets in, the stiffness matrix
keeps on changing with each increment. Once the stiffness matrix keeps
on changing, analysis is going to get complex and time consuming.

COURSE TEACHER:Ms RABIA SIDDIQUI

• In simpler terms, material non linearity is the transfer of material
behavior from Elastic to Plastic (Where Elastic corresponds to a linear
behavior and Plastic corresponds to a non linear behavior). Even for a
plastic material, initial portions of the stress strain graph tend to be linear
and then become non linear post the Proportional limit (like in the figure
below).

COURSE TEACHER:Ms RABIA SIDDIQUI

Boundary conditions non linearity happens when the Boundary conditions defined in the
Simulation change during an analysis. For example, consider a deflecting Cantilever beam.

Now, consider an object in the path of the deflection

Between the two above figures, the Boundary Conditions has changed. The Cantilever
beam can no longer deflect as it’s deflection is prevented by the object. This is what is
meant by Boundary Condition Non Linearity.

COURSE TEACHER:Ms RABIA SIDDIQUI

Consider a flat panel with about 4 stiffeners. Now on application of compressive load, the
initially flat panel takes up a bow shape (circular). This is an example of Geometric non
linearity.

Once Geometric non linearity onsets, parts in addition to having axial stiffness (In plane
stiffness axial to the element), also have an out of plane stiffness. Geometric non linearity is
the outcome of large rotations or Buckling.
COURSE TEACHER:Ms RABIA SIDDIQUI
COURSE TEACHER:Ms RABIA SIDDIQUI
COURSE TEACHER:Ms RABIA SIDDIQUI
 CONSTANT STRAIN TRIANGLE: Let us consider a Linear (First order element) of
Triangular configuration. It will have 3 nodes which are the three vertex of the triangle.

Consider the displacement equation assigned for this element - u = a + bx, now the strain which is
the derivative of displacement = b . Now, this is what we mean by constant strain. That is, the
following happens,

The original element will either magnify with constant

strain at all places or zoom out with constant strain at all
places.

COURSE TEACHER:Ms RABIA SIDDIQUI

 LINEAR STRAIN TRIANGLE

Now, consider a quadratically formulated triangle. (6 nodes)

Consider the displacement equation assigned for this element u = a + bx + cx2, now the strain
which is the derivative of displacement = b + 2cx . It has linear variation. Now, this is what we
mean by Linear strain variation. That is, the following happens,

Consider the displacement equation assigned for this element u = a + bx + cx2, now the
strain which is the derivative of displacement = b + 2cx . It has linear variation. Now, this
is what we mean by Linear strain variation. That is, the following happens,

The element is free to undergo linear variation unlike

the case of Constant Strain element where all
places are bounded to have equal responses.
It is not preferred to use constant strain elements for
areas of high stress concentrations, place of high
strain gradients. For general analysis, CST is not
COURSE TEACHER:Ms RABIA SIDDIQUI
preferred.
 WHAT IS SHEAR LOCKING?

The above is a Fully Integrated element. The circles are the locations of the integration points. There
are four Integration points with 2 integration point in each direction. Two integration points in each
direction is an representation of Linear formulation. Hence, this is a fully integrated linear element.
In the element, assume red arbitrary horizontal and vertical lines. Red lines are the iso lines. The
intersection of the red horizontal and the vertical lines is the Integration points location. As visible, the
angle between the red horizontal and the vertical line = 90 degrees. And the points a and b are
equidistant from “o” where o is the midpoint of the top line.

COURSE TEACHER:Ms RABIA SIDDIQUI

 2-The top horizontal black line has elongated indicating
1-On bending, the tension and bottom horizontal black line has decreased in
element is supposed to length indicating compression. All the horizontal lines ( red
/ expected to take up or black have curved). Points a and b have shifted but are
the following shape. still equidistant from “o”. The horizontal red line and the
vertical red line have curved , thus the angle at the
integration point is still 90 degrees. This is the expected
behavior, but the actual behavior is,

3-In the actual behavior, the horizontal line refuses to curve. Due to this, the angle at the
integration points is less than 90 degrees. Since the angle has changed Sigma 12 will also
change. But this is wrong as there is no shear in bending.
This kind of behavior where the edges are unable to curve and leading to a change in angle at the
position of the integration point is Shear Locking.
This is observed only in Linear elements , as strain captured in a linear element is constant strain.
Edges of a Linear element cannot curve.Hence, in FEA terms, shear locking causes the element
to behave more stiff than what it is in actual case. Only linear elements are affected.
COURSE TEACHER:Ms RABIA SIDDIQUI