Professional Documents
Culture Documents
The finite element is a mathematical method for solving ordinary and partial differential
equations. Because it is a numerical method, it has the ability to solve complex problems
that can be represented in differential equation form. As these types of equations occur
naturally in virtually all fields of the physical sciences, the applications of the finite
element method are limitless as regards the solution of practical design problems.
Due to the high cost of computing power of years gone by, FEA has a history of being
used to solve complex and cost critical problems. Classical methods alone usually cannot
provide adequate information to determine the safe working limits of a major civil
engineering construction or an Automobile or an Aircraft. If a tall building, a large
suspension bridge or an Automobile or a Nuclear reactor failed catastrophically, the
economic and social costs would be unacceptably high.
In recent years, FEA has been used almost universally to solve structural engineering
problems. One discipline that has relied heavily on this technology is the Automotive and
Aerospace industry. Due to the need to meet the extreme demands for faster, stronger,
efficient and light weight Automobiles and Aircrafts, manufacturers have to rely on the
technique to stay competitive. But more importantly, due to safety, high manufacturing
costs of components and the high media coverage that the industry is exposed to,
Automotive and Aircraft companies need to ensure that none of their components fail,
that is to cease providing the service that the design intended.
FEA has been used routinely in high volume production and manufacturing industries for
many years, as to get a product design wrong would be detrimental. For example, if a
large manufacturer had to recall one model alone due to a piston design fault, they would
end up having to replace up to 10 million pistons. Similarly, if an oil platform had to shut
down due to one of the major components failing (platform frame, turrets, etc..), the cost
of lost revenue is far greater than the cost of fixing or replacing the components, not to
mention the huge environmental and safety costs that such an incident could incur.
The finite element method is a very important tool for those involved in engineering
design, it is now used routinely to solve problems in the following areas:
Nowadays, even the most simple of products rely on the finite element method for design
evaluation. This is because contemporary design problems usually cannot be solved as
accurately & cheaply using any other method that is currently available. Physical testing
was the norm in years gone by, but now it is simply too expensive and time consuming.
What is FEA?
The Philosophy of FEA can be explained with a small example such as measuring the
perimeter of a circle.
If one needs to evaluate the perimeter of the circle without using the conventional
formula, one of the approach could be to divide the above circle into a number of equal
segments. Join the beginning and end points of these segments by a straight line. Since it
is very easy to measure the length of a straight line, the length of each line multiplied by
the number of lines gives the perimeter of the circle.
For the above exercise let us ask the following questions?
1.
2.
3.
4.
5.
FEA consists of a computer model of a material or design that is loaded and analyzed for
specific results. It is used in new product design, and existing product refinement. A
Design Engineer shall be able to verify a proposed design, which is intended to meet the
customer specifications prior to manufacturing or construction. Things such as,
Modifying the design of an existing product or structure in order to qualify the product
or structure for a new service condition. Can also be accomplished In case of structural
failure, FEA may be used to help determine the design modifications to meet the new
condition.
Let us understand the above five steps one by one sequentially and see what it really
means to an Engineer.
When we use a CAE software either developed in house of commercially available, the
first three steps are called as Pre-Processing phase, the fourth phase is called solution
phase and the fifth phase is called Post-Processing phase. Since FEA involves matrix
operations, it was referred to as Matrix methods for Structural analysis, in the initial days
where it was used only for structural behavior simulations.
Despite the proliferation and power of commercial software packages available, it is
essential to have an understanding of the technique and physical processes involved in
the analysis. Only then can an appropriate & accurate analysis model be selected,
correctly defined and subsequently interpreted.
Before proceeding further to learn more, we shall familiarize our selves with the
following:
What is an element?
Element is an entity , into which a system under study can be divided into. An element
definition can be specified by nodes. The shape (Area, length and Volume) element
depends upon the nodes with which it is made up of.
What are nodes?
Nodes are the corner point of an element. Nodes are independent entities in space. These
are similar to points in geometry. By moving a node in space an element shape can be
changed.
A line element has the shape of a line/curve and hence requires minimum of two nodes
to define it.
An area element has the shape of quadrilateral/triangle and hence requires minimum of
four/three nodes to define it.
Why are we saying minimum number of nodes? Can an element have more number of
nodes than the minimum? If so under what circumstances? The answer is an obvious yes!
As shown in the above pictures, there can be higher order elements which have additional
nodes ( At the mid of the element edge). That is when the element does not have a mid
side node on the edge, it is called a Linear element. If it has an additional node at the mid
side it is called a Quadratic or Second order element.
For Linear elements the edge is defined by a Linear function called shape function whose
degree is one. Hence the name linear elements. For elements having mid side nodes on
the edge Quadratic function called shape function whose degree is two is used. Hence the
name Second order or Quadratic element. Similarly elements can have more nodes on the
edges and hence the edge can be defined using higher degree equations. Such elements
are called higher degree elements.
The higher order elements when over lapped on geometry can represent, complex
shapes very well within few elements. Also the solution accuracy shall be more with the
higher order elements. But with higher order elements the computational effort required
is more.
Design Validation: FEA was being used in the initial days, to evaluate/check the
design against the intended purpose.
In this method attempts were made to simulate a test condition and then study the results
of FEA to assess whether the design meets the requirements or not. If the simulations
reveal that the design might fail, then the Engineer would modify the design to avoid
failure. This method was helping to avoid the design validation through testing of
prototypes. This approach was used by experienced Engineers, who were used to design
the equipment by conventional approaches, but want to avoid testing.
Design Optimization: At a later stage, when the FEA tools have become more
powerful, optimization algorithms have been embedded into the tools and thus
improving the power of FEA. An FEA software coupled with Optimization tools, under
a set of imposed conditions searches for the better design variables which can satisfy the
design requirements.
FEA requirements:
Having learnt what FEA can do for Engineers now let us focus on what we need to utilize
this tool effectively.
Engineer: The Engineer who wishes to use this tool to advance his design cycles need
not be familiar with the theoretical basis of FEA. It is sufficient if he has good back
ground in strength of materials ( For Structural Analysis), Heat Transfer and
Thermodynamics (For Thermal analysis) coupled with good analytical/reasoning/design
skills.
Besides to get best results through FEA simulations the Engineer should have:
Software: The FEA software has to be chosen based upon a number of considerations
like the following:
Hardware: The computer system that is chosen to run CAE should be good
configuration to run the tool. Most often based upon the design requirements one would
choose the software tool and the user should ensure that the minimum configuration as
specified by the software vendor is available.
Based upon specific requirements, one can opt for a higher computer configuration than
the minimum specified the software vendor. Most PC versions of FE packages for
professional use (as opposed to academic) run efficiently on MS-Windows with the latest
service packs and also with an Intel type of chips. The platform used is usually a function
of the cost the user is willing to pay for the hardware it will run on.
Finally six stress distributions are determined via the stress/strain relationships of the
material.
Commercial packages usually go one further and calculate a range of more usuable stress
fields from the six stress components such as the principal stresses and a host of failure
prediction stresses as described by the most common yield criteria (Von Mises/ /Tresca,
etc.). The displacements can be used in conjunction with the element stiffness to
determine the reaction forces and the forces internal to each element (Element Internal
forces).
A point to note is that at least one of the displacements must be known before the rest can
be determined (before the system of equations can be solved). These known
displacements are referred to as boundary conditions and are oftentimes a zero value.
Without these boundary conditions, we would get the familiar singularity or zero-pivot
error message from the solver, indicating that no unique solution was obtainable.
S-N data can be used reliably for design only if the test conditions for generating S-N
data match the service conditions for the component. Unless the material fatigue
properties and the effects of machining are well known, the stresses predicted by FEA
probably cannot be used to predict fatigue life. However, the FEA stress data can be used
to redesign fatigues that have known failure problems. The fatigue life cannot be
predicted from stress unless the material's fatigue characteristics are known.
Geometric VS FE Entities:
When we are building a geometry we have various entities like points, lines, curves,
areas, surfaces, volumes and solids. Incase of FEA we have only two entities called nodes
and elements. We use geometric entities as aids to build the FEA entities. But what we
really need for the simulations is only the FEA entities.
Plane elements: These elements are used to simulate the in-plane behavior of the
component whether plane stress or plane strain. These elements also have the same shape
as shell elements.
Special Elements:
Surface elements: These elements are used to represent the surface effects such as
surface tension, pressure, insulation and convection. These elements are over laid on the
free surface of other elements like shell or solid elements as the case may be. Hence the
shape can be a line (In case of Plane elements) and a quadrilateral/triangle incase of a
shell or solid element.
Super element: The super element can be considered to be a heavy nucleus formed
by combining a number of similar elements ( Line or Shell or Solid). Hence the number
of nodes for a super element is arbitrary ( But a minimum of two are required). Like wise
the shape is also arbitrary. The super elements takes the combined shape of all the
elements from which it is formed and retains the nodes of all the constituent element on
the periphery.
Depending upon the software, there are certain rules as to how these elements need to be
crated.It is important to note that most of the FEA software vendors have specific
limitations on the use of axi-symmetric elements, some define the axis of symmetry to be
the global x-axis, where others define it as the global y-axis, such a limitation usually has
a further limitation that no nodes can have a negative coordinate value.
Some FE packates have a special class of axi-symmetric element that allows non axisymmetric loads and boundary conditions. For these elements, each load is defined as a
series of harmonic functions written in the form of a Fourier series. These special type of
axi-symmetric element are usually referred to as harmonic elements.
Rigid Elements: Rigid elements are generally used to model rigid parts of a
structures without having to assign the computationally expensive usual elastic elements.
They are useful in dynamic analysis to account for distributed mass and inertia. They are
also regularly used in kinematic type simulations away from any areas of interest in the
model.
Contact Elements: These are special case of spring elements. That is a contact
element is nothing but a compression only spring element with some additional features.
These elements are used to simulate the contact between the different surfaces in a
system. Based upon the situation, the element can take various shapes. These elements
are used to define a contact between two surfaces. On the creation of the elements, a set
of slave nodes and master nodes are defined. Contact occurs if either group of nodes
attempts to penetrate the other. Friction properties can usually be applied to the element.
As with gap elements, they are usually NOT recommended for use with higher order
elements. Element connecting between two nodes is called a GAP element.
Gap elements are used to model point-to-point contact conditions. The elements have a
large stiffness when active and a small (but non-zero) stiffness when in an inactive state.
The element becomes active when it comes within a specified proximity of a specified
object in the analysis.. The active state can usually accommodate both compression and
tension. An initial opening and pre-load force and a friction coefficient can usually be
specified. A nonlinear analysis is required when using gap elements for the open/closed
gap behavior to exist in the model. Gap elements can be used in linear static solutions but
will only function as springs with the Compression Stiffness in the axial direction and the
Transverse Stiffness in the transverse direction.
There are also other forms of elements called as surface to surface contact or surface to
ground elements. The shapes of these elements are as shown below. The surface type o
contact elements give more accurate results. But they are computationally more
expensive.
The Apex of the surface to contact element shown below becomes the ground, if
these elements are used as surface to ground elements
at each node
which is used best to represent the behavior of a system is called the degree of freedom.
Each node of an element is only capable of accepting a load/conditions pertaining to its
mobility/Characteristic. In general the Translation degrees of freedom for a structure in a
particular global direction is represented by U with a suffix indicating the axis. Similarly,
in general, the rotation degrees of freedom for a structure in a particular global direction
is represented by R/Rot with a suffix indicating the axis
For example if a node has only four degrees of freedom say Ux and Uy (Translations in
Global X and Y directions) and Rx and Ry (Rotations in Global X and Y directions), the
FE model can not be used to study the behavior of system subjected to load acting in
Global Z direction and a moment acting along the Z direction.
Degree of freedom for shell elements: The shell elements have usually six
degrees of freedom. However, there is a variant of shell element called membrane
element, which can not capture bending. This membrane element has only three
translation degrees of freedom (Ux, Uy and Uz along the Global X, Y and Z directions
respectively.
Degrees of freedom for Solid Element: The solid element has three degrees
of freedom which are translations along the three global axes (Ux, Uy and Uz).
Degrees of freedom for surface elements: The surface effect elements are
always over laid on the edges/surface of the respective elements and hence share he
nodes of those elements. Hence the nodes of these elements do not have any special
degrees of freedom and hence inherit the degrees of freedom of the nodes of the
respective elements.
Degrees of freedom for Spring Element: The spring element can have any
where between 3 to 6 degrees of freedom based upon the element configuration.
Since Super
elements are formed by combining the regular elements, they inherit the physical
properties of the constituent elements from which the Super element is formed. Hence no
separate physical properties are required for these elements.
Pre-Processing:
The preprocessor stage in general FE packages involves the following:
Specifying the title: This is giving name for the problem. This is optional but very useful,
especially if a number of design iterations are to be completed on the same base model.
Setting the type of analysis: To simulate the conditions properly, correct type of analysis
method/solver/tool need to be used, e.g. structural, fluid, thermal or electromagnetic, etc.
(sometimes this can only be done by selecting a particular element type).
Creating the model: The model is drawn in 1D, 2D or 3D space in the appropriate units
(M, mm, in, etc..). The model may be created in the pre-processor, or it can be imported
from another CAD drafting package via a neutral file format (IGES, STEP, ACIS,
Parasolid, DXF, etc.).
Defining the element type: This may be 1D, 2D or 3D, and specific to the analysis type
being carried out (you need thermal elements to do thermal analyses).
Mesh Generation: As discussed previously, Mesh generation is the process of dividing
the analysis continuum into a number of discrete parts or finite elements. The finer the
mesh, the better the result, but the longer the analysis time. Therefore, a compromise
between accuracy and solution speed is usually made. The mesh may be created manually
or generated automatically as the ones shown below in the pictures. In the manually
created mesh, you will notice that the elements are smaller at the joint. This is known as
mesh refinement, and it enables the stresses to be captured at the geometric discontinuity
(the junction).
Manual meshing is a long and tedious process for models with any degree of geometric
complication, but with useful tools emerging in pre-processors, the task is becoming
easier. Automatic mesh generators are very useful and popular. The mesh is created
automatically by a mesh engine, the only requirement is to define the mesh density along
the edges of the model. Automatic meshing has limitations as regards mesh quality and
solution accuracy. The currently available Automatic brick element (Hexa) meshers are
limited in function as they can mesh only few regular shapes like cubes etc., but are
steadily improving. The present popular method of mesh generation is by simply
selecting the mesh command on the preprocessor list through Graphical User Interface
(GUI).
Often the creation of mesh (Finite Element generation) can consume significant amount
of time in the total cycle time required for a CAE simulation. It is very difficult to create
all the elements by manually picking the nodes. Even to do this, the nodes should be
present. To aid the process of meshing certain kinds of algorithms are built into each
software. Each software has certain capabilities and hence certain kinds of meshing
operations can be done with ease based upon the software.
Usually meshing capabilities of most of the CAE tools are poor. Hence Engineers
generally employ various kinds of third party tools for pre-processing ( Mesh and
Boundary/Field condition simulations) which are capable of writing the FE data
(Elements and Nodes) into the format readable by respective CAE solvers.
Apart from the above the pre-processing stage also includes specification of material
properties, Physical properties and applications of loads and Boundary conditions.
Meshing Requirements:
Meshing Requirements for 1D: 1D element can be manually created by
picking of nodes or using the meshing capabilities in a CAE tool they can be generated
overlaid on lines.
Most of the software require specification of a third node to represent the local XZ cross
section of the element. The length direction of the element represents the local XDirection. If the third nodes is not defined for these elements, a default direction based
upon the software specification is assumed. Except for the definition of the third node,
there are no practical problems that can be encountered while creation of these elements.
Meshing requirements for 2D: 2D (Plane) element creation can be done either
by manual picking or using the meshing capabilities available in the CAE software. In all
most all the cases, the latter method is preferred.
Care should be taken to get good quality mesh. The number of triangles should be kept as
minimum as possible. Availability of clean areas generated in CAD software can reduce
the effort required for mesh generation.
Meshing Algorithms: Based upon the CAE software, the meshing algorithms can
vary to great extent. All CAE softwares may or may not support all the algorithms
Mesh density: User should take precaution to generate more number of elements
(Fine mesh) at regions of interest ( Where accuracy in results is required). As matter of
precaution Engineer can not go on putting more number of elements at all zones as it
amounts to more solution time and also more disk space interms of disk space required
for results storage.
Hence more number of elements needs to be generated only at locations/regions required.
Also meshing approach used also can cater to improve the solution accuracy, which are
as follows. Usually the first of these approaches is usually followed to get good results.
P-Adaptivity: This approach involves in adding mid side nodes to the edges of
elements to increase the solution accuracy. This is equivalent to increasing the
Polynomial order of the equation representing the edge of the element. Hence the name
P-Adaptivity. Computationally this kind of approach consumes lot of solver time.
Usually when one wants to follow this approach there are capabilities available in certain
CAE software which determine the degree of polynomial required to get good results.
Usually Polynomial order is limited to 9. These kinds of solvers are usually used by inexperienced Engineers.
R-Adoptivity: This is called relocation adoptivity. Under this approach the Engineer
generates he mesh that he thinks is he best for analysis requirements. Later based upon
set of algorithms, from the results of analysis the software relocates the nodes to adjust
the mesh density. This way the mesh shall be coarsened/refined at locations wherever it is
required.
Mesh Generation:
In order to carry out a finite element analysis, the model we are using must be divided
into a number of small pieces known as finite elements. Since the model is divided
into a number of discrete parts, FEA can be described as a discretization technique. In
simple terms, a mathematical net or "mesh" is required to carry out a finite element
analysis.
If the system under investigation is 1D in nature, we may use line elements to
represent our geometry and to carry out our analysis. If the problem can be described
in 2 dimensions, then a 2D mesh is required. Correspondingly, if the problem is
complex and a 3D representation of the continuum is required, then we use a 3D
mesh.
Mesh Density : The art of using FEM lies in choosing the correct mesh density
required to solve a problem. If the mesh is too coarse, then the element will not allow a
correct solution to be obtained. Alternatively, if the mesh is too fine, the cost of analysis
in computing time can be out of proportion to the results obtained. In order to define a
relevant mesh, some idea of the parameter distributions (stress, temperature, pressure,
etc.) within the component is required. If the answer is known, then a good mesh can be
defined. A fine mesh is required where there are high parameter gradients and strain and
a course mesh is sufficient in areas that have result contours of reasonably constant slope.
Few points to be noted from the above quarter model of a plate with hole are:
For the coarse mesh the number of elements are very less
The mesh does not capture the round hole accurately
It would have been sufficient if we would have just increased the number of
elements around the hole than increasing the number of elements all around like
in the Fine mesh
Since, linear element was used we were not able to represent the hole correctly
in the Coarse mesh
If we would have used, higher order elements in the Coarse mesh model, we
would have captured, the hole geometry properly
A Good Mesh : A good or appropriate mesh is one that enables accurate resolution
of the underlying physical phenomena, yet is coarse enough to allow a fast solution time.
Linear elements require a finer mesh than parabolic (quadratic) ones, which in turn
require a finer mesh than cubic elements. It follows that any user of a finite element
package must have some knowledge of underlying physical phenomena and the manner
in which it behaves. We must be able to identify regions of high gradients (rapidly
changing fields of stress, temperature, pressure, etc.).
Element Distortion
Every element is defined in terms of the basic shape of a parent element. Due to the
geometry of a component, elements may become distorted in an effort to force a mesh
within the boundary of the model. When elements are distorted from their parent
shape they become less accurate. As the distortion is increased, the solution error
becomes greater. Therefore, the user should attempt to keep the elements as near to
the basic parent element shape as possible when creating a mesh.
Mesh requirements
The Finite Element Method (FEM) has certain requirements on a mesh:
The mesh must be valid, (no holes, self-intersections, or faces joined at two or
more edges).
The mesh must conform to the boundary of the domain.
The density of the mesh must be controllable, to allow trade-off between accuracy
and solution time.
The grid density will vary depending on local accuracy requirements, but any
variations must be smooth to reduce or eliminate numerical diffusion/refraction
effects.
There are some requirements on the shape of elements. In general, the elements
should as equiangular as possible in equilateral triangles & regular tetrahedron.
Highly distorted elements (long, thin triangles, squashed tetrahedron) can lead to
numerical stability problems caused by round-off errors. This requirement is
modified for boundary layers, where highly stretched elements are desired and
facilitated in the FEM formulation. The min-max-angle property is still required
in this case.
Element Limitations
The meaning of Linear Static: Linear analyses deal with problems in which the
structural response is linear. Therefore, if the applied forces are doubled, then the
displacements and internal stresses also double. Problems that fall outside this domain are
usually classified as non-linear.
Static or steady state analyses are those where the solution is independent of time. Inertial
forces are either ignored or neglected and so there is no requirement to calculate actual
time derivatives. Problems that require inertial terms to be evaluated are usually
classified as dynamic and/or transient analyses.
Linear static analyses are usually sufficient for situations where loads are known and the
instance at which peak stress occurs is obvious. When performing a linear static stress
analysis, the analyst applies static loads (forces, pressures or prescribed displacements) to
the model.
Assumptions of Linear Static Analyses: As with all types of analyses, linear static ones
are based on a set of assumptions. The main assumptions are listed here:
Structural deformations are proportional to the loads applied. This infers that
the loading pattern does not changed due to the deformed shape and no
geometric stiffening occurs due to the application of the load.
All materials behave in a linear elastic fashion. Hence, the material deforms
along the straight line portion of the stress-strain curve (no plasticity or
failures occur). Highly localized stress concentrations are usually permitted as
long as gross yielding does not take place.
Loads are all static. This means that the loads are applied to the structure in a
slow or steady fashion and in a way that makes them time independent (are
assumed to be constant for an infinite period of time).
Limitations of Linear Static Analyses: There is a point when linear static analyses are
not sufficient to represent the real behavior of the system being modeled. As mentioned
previously, all physical phenomena are non-linear and dynamic to some degree, some are
negligibly non-linear/transient and some are grossly non-linear and time dependent. The
decision to go NL and/or Transient should be based on careful consideration of the
physics of the problem at hand.
Some suggested considerations are:
Let us assume that one end is fixed ( At node 1 as shown in figure above). Then the
matrix of unknowns shall be as follows:
For system involving large number of elements with more degrees of freedom, this can
result in substantially huge matrices. Hence it becomes almost impossible to calculate the
solution manually for any system.
Since the solution methodology involves large number of matrix operation FEA was
called as Matrix methods for structural analysis by Structural Engineers. As time
progressed the methodology was well developed to take care of non-structural systems as
well.
Loads:
Some type of load is usually applied to the analysis model. The loading may be in the
form of a point load, a pressure or a displacement in a stress (displacement) analysis, a
temperature or a heat flux in a thermal analysis and a fluid pressure or velocity in a fluid
analysis. The loads may be applied to a point, an edge, a surface or a even a complete
body. The loads should be in the same units as the model geometry and material
properties specified. In the cases of modal (vibration) and buckling analyses, a load does
not have to be specified for the analysis to run (Only for Linear analysis).
The following is the description about some of the loads that can be applied:
Pressure Load: Pressure load is used to simulate the distributed loads that act on a
structure/component. Pressure loads always act normal to surface. As discussed
previously, shell normals play a very important role in deciphering the direction of load
application. Also each Hexahedron element has 6 faces. Hence care should be taken
while specifying the correct face of Hexahedron element to simulate the correct load.
Boundary conditions:
If you apply a load to the model, then in order to stop it accelerating infinitely through the
computer's virtual ether (mathematically known as a zero pivot), at least one constraint or
boundary condition must be applied. Structural boundary conditions are usually in the
form of zero displacements, thermal BCs are usually specified temperatures, fluid BCs
are usually specified pressures. A boundary condition may be specified to act in all
directions (x,y,z), or in certain directions only. They can be placed on nodes, points, areas
or on lines. BC's on lines can be in the form of symmetric or anti-symmetric type
boundary conditions, one allowing in plane rotations and out of plane translations, the
other allowing in plane translations and out of plane rotations for a given line. The
application of correct boundary conditions are a critical to the accurate solution of the
design problem. At least one BC has to be applied to every model.
These boundary conditions serve as reference point while solving for the unknowns.
Hence care should be taken in a model to ensure that all DOF are constrained ( Not
necessarily at the same place) so that a proper reference is established for solution. If this
is not taken care, the solution might diverge or we might get junk results.
We can either specify a restraint (specify a zero displacement in a direction) or also
specify a finite value of known displacement to simulate things like pre-stressing.
Solvers:
The FE solver can be logically divided into three main parts, the pre-solver, the
mathematical-engine and the post-solver. The pre-solver reads in the model created by
the pre-processor and formulates the mathematical representation of the model. All
parameters defined in the pre-processing stage are used to do this, so if you left
something out, chances are the pre-solver will complain and aborts the solution process.
If the model is correct the solver proceeds to form the element-stiffness matrix for the
problem and calls the mathematical-engine which calculates the primary unknown result
(displacement, temperatures, etc.). The results are returned to the solver and the postsolver is used to calculate derived results (strains, stresses, heat fluxes, velocities, etc.) for
each node within the component or continuum. All these results are written to a file
which may be read by the post-processor.
Frontal Solver : The advantages of a Frontal solver is that it requires only a very
small RAM in the computer. Even very large problems can be solved on a computer
having very limited RAM as low as 16 MB.
The element numbering sequence is very important for this solver. The maximum RAM
required for an analysis depends upon the maximum wave front length. At any point of
time only few equation remain in the RAM ( Core memory). Frontal solvers were widely
used in the initial days since the computing power available was less. If the total number
of degrees of freedom in a problem is n the number of multiplications involved are the
order of n3 .
The disadvantage of the Frontal solver is that it requires huge disk space for storing of
matrices during solution. Also the process of solving includes generation of upper or
lower triangle matrix and subsequent solution by Gaussian elimination method. This way
each row in the stiffness matrix is inter linked to other row for row operations.
But in case if the behavior of a system is to be studied under multiple loading conditions,
the solution time shall be almost similar to that required for a single solution. This is due
to the fact the stiffness and the unknown matrices (Left hand side) remain same for all
kinds of loading and only the right hand side ( Force matrix) shall be changing each time.
Hence different force matrices can be solved simultaneously for the same left hand side
to get quick solution in case of multiple loads.
Iterative solver: If the computer RAM is abundant this is the most ideal solver. The
run time using this solver is not effected either node numbering scheme or element
numbering scheme adopted in the problem.
The disadvantage of this solver is it requires very high RAM. This solver keeps all the
equation s in the RAM at any point of time. This does not involve any row operations for
the stiffness matrix. Hence all the rows have independent existence. Hence incase if
multiple processors are available the rows can be passed onto them for solution. This
makes this solver IDEAL for super computing.
For multiple load cases, he solution needs to be run multiple times for the same set of
boundary conditions. . If the total number of degrees of freedom in a problem is n the
number of multiplications involved are the order of k*n2 . Here usually k is less than
n/4. That is why iterative solvers are minimum 4 times faster than the Frontal solvers
for the same problem.
But convergence problems might arise if the elements are of bad quality or the problem
has too much variations in the stiffness values across various zones.
Sparse Direct Matrix Solver : Disk space wise and run time wise this solver lies
in between the Frontal and Iterative solvers. Node numbering scheme is important for
optimum run time using this solver.
The RAM requirements are also moderate for this solver since only either upper half or
lower half of the matrix needs to be stored in the RAM. Multiple load cases can be solved
with ease like incase of frontal solver.
Post Processing:
As discussed previously the process of reviewing the results with the help of CAE tool is
called post processing. In this stage, the results of the analysis are read and interpreted.
They can be presented in the form of a table, a contour plot, deformed shape of the
component or the mode shapes and natural frequencies if frequency analysis is involved.
Other results are available for fluids, thermal and electrical analysis types. Most postprocessors provide an animation service, which produces an animation and brings your
model to life.
If the results are available only in the form of numbers at each nodal location in the
component or system, it becomes very difficult for the engineer to interpret the same and
take some decisions. Contour plots are usually the most effective way of viewing results
for structural type problems. Hence the usual trend is to show the results also in the form
of colored contours. Each color represents a range of values for the region enclosed by it.
This approach gives a good insight into the state of the system on a Macro level. If one
wants the exact value at a particular location, he can refer to the actual numbers. This
makes the process of judgment very easy and user friendly. Slices can be made through
3D models to facilitate the viewing of internal stress patterns.
All post-processors now include the calculation of stress and strains in any of the x, y or z
directions, or indeed in a direction at an angle to the coordinate axes. The principal
stresses and strains may also be plotted, or if required the yield stresses and strains
according to the main theories of failure (von mises, St. Venant, Tresca etc.). Other
information such as the strain energy, plastic strain and creep strain may be obtained for
certain types of analyses.
shall be the average value of result obtained by extrapolating the Gauss point results for
all the elements. That is why these nodal results are some times referred to as Average
Nodal Results.
If mesh density is not managed properly, the element stresses and the average nodal value
can be quiet different. User should ensure good mesh density to ensure minimal variance
between the elemental and nodal results.
Plot paths: Plot paths are another way of deciphering the FEA results. Most of the
CAE post processing tools provide this capability. This is a good means of studying how
a result displacement/stress/strain varies along a direction of interest.
Sources Of Error:
There could be many reasons for obtaining poor results. Few of the contributing factors
could be as follows:
Problems With Units :In many cases, problems with units are easy to detect. To ensure
that the units employed are appropriate
Check the FE model for the correctness of results before the solution
Review the deformed model shape to ensure it is roughly what you expected.
Incorrect Loading :There are two main ways of loading the model incorrectly, namely
errors with units and oversights in the application of loads. The first is common with the
event of pushbutton analysis packages, this type of error is usually easy to detect.
Loading oversights are much more subtle and often impossible to detect. For example, if
a bending moment is required and is formed by applying a force at the end of the moment
arm, then a direct force is created as well as the bending moment. This is a common
mistake in torsion problems.
Another common oversight is to apply a static steady state load value where there will be
much higher loads experienced due to impluse. For example, if large boulders are loaded
onto a truck trailer, they probably will not be lowered in gently. Therefore, the impulse
loads experienced by the trailer chassis will far exceed the static load of the boulders,
especially when the trailer is almost full.
Element Polynomial Order : For the same mesh density (same number of elements),
parabolic elements provide better results because they represent the a models curved
boundaries more accurately, as well as producing better mathematical formulations
However, parabolic elements require much greater computational resources than linear
elements (usually takes about double the time to solve a parabolic element mesh than it
does to solve a linear element mesh of the same problem, with the same mesh density
being employed).
Also, Solution accuracy depends primarily on how well the assumed variation of stress
within an element fits the actual variation of stress within an element fits the actual
variation in a structure. Linear pyramid (tetrahedral) elements often provide inaccurate
results, especially for bending type problems. However, accuracy can be improved
dramatically by choosing parabolic elements instead.
Well shaped linear brick (hexahedral) elements can provide accurate results. The
elements perform at their best as regular prisms (cubes). Automatic hexa (brick) meshers
meshing complicated geometry, however will introduce distorted or transitional elements
in unexpected zones, thus considerably affecting the accuracy of the results. Parabolic
pyramid elements provide results at least as accurate as linear brick elements and are less
sensitive to distortion to poor orientation in the model.
Both 20 noded brick and 10 noded pyramid elements provide good stress results for
reasonable meshes with a comparable number of nodes, while 8 noded bricks and 4
noded pyramids require many more elements for solids with curved boundaries to
achieve the same geometrical and stress accuracy.
against using a very dense mesh, is that the analysis will take a considerable time longer,
due to the extra nodes that are required to be solved for.
An alternative is to refine the mesh at the local details. This requires specialist knowledge
on how the results will vary over the model domain.
Computed values such as stress and strain, are evaluated at locations on the element
known as Gauss points. These points are always well inside element boundaries. Values
at other positions are interpolated or extrapolated. If this is done across a boundary
between two elements, then it should be reasonably accurate, but extrapolating to the
edge of an element on the edge of a structure, where the stresses will probably be at the
highest and of most interest, can lead to significant errors in rapidly changing stress fields
if the mesh density or the element order is too low.
The best way of ensuring that your mesh is sufficiently refined is to plot the results with
node and element averaging switched off. A problem will materialize in the form of a
discontinuity (or step) of the solution parameter between elements (the contours will be
stepped across the element boundary rather than smooth as they should be).
The analyst should be aware of derivative type solutions. In a displacement analysis, the
stress is obtained by differentiating the displacement results. Therefore, if the element is
parabolic, then a linear variation of stress will be obtained. It is important that this is kept
in mind when defining the mesh density and grade. Remember that derivatives always
require more elements to converge. Heat flux in thermal analyses are also derivative
based results.
Formulation Error :Formulation error results from the use of elements that do not
precisely describe the behavior of the physical problem. Elements which are used to
model physical problems that they are not suited, are sometimes referred to as illconditioned or mathematically unsuitable elements for the problem in hand.
For example a particular finite element might be formulated on the assumption that
displacements vary in a linear manner over the domain. Such an element will produce no
formulation error when it is used to model linearly varying physical problem (linear
varying displacement field in this example), but would create a significant formulation
error if it used to represent a parabolicly or cubicly or logarithmically varying
displacement field. Formulation error can be reduced by selecting the proper element
type and mesh.
round off errors. Therefore, this problem mainly concerns the FE developer. The user can
also contribute to the numerical accuracy. In an analysis session, it is usual procedure for
the pre-processor to create an input file for the actual program that solves the problem.
Some of these keyword files are generated automatically by the software, and is hidden
from the user. In the case of high level analyses, most keyword files are user generated,
and it is possible to send the data to the solver with a reduced number of decimal places,
without even realizing it. Hence the solver solves a problem that does not represent the
geometry or loading conditions of the physical model, or the intentions of the user.
Unless
otherwise specified the applied forced loads or the displacement boundary conditions act
along the direction parallel to the global axes.
At times we might have to apply a net force acting in a direction which is at an angle to
any of the global axes. Or at times we need to apply a force acting radially out/in wards.
Under such circustanges we either need to resolve the loads into components along the
global axes and apply them.
Alternately most of the CAE software provide an option of creating a local co-Ordinate
system, into which the nodes at which the loads are to be applied can be moved. Then the
loads can be applied in the local direction (Parallel to the local axes).
Most of the
material soften upon heating. Hence the material properties change to some extent. Hence
in simulations involving temperature effects the material properties need to be input as a
property of the system. Based upon the temperature assigned at a node, the software picks
up the relevant material property.
Many structural applications of the finite element method contain long slender regions,
thin zones and complex chunky portions. Long slender regions are best represented with
beam elements, thin zones can usually be modeled using shell elements, and chunky
portions are best represented as three-dimensional elements. In order that each region is
represented by an appropriate element type, some sort of scheme is required to form a
link between the meshes of dissimilar dimension.
Transitioning schemes fall into two main categories. The first category are transitions
from one element type to another that are of the same dimension (e.g both are shell or
solid elements). The second type of transition is where element of different dimension are
joined (e.g. a shell to solid transition).
Transition Elements : Transition elements, which are also known as interface elements,
have been developed for both similar dimension and mixed dimension transitions.
Interface elements for similar dimension transitions can come in two main types, those
that have a one to one relationship and those that have a one to many relationship. The
most common one is the one to one, it is used for coupling elements that have edges with
different numbers of nodes. This type has been used mainly to couple elements with
different orders of interpolation. An example of a one-to-one transition element is shown
below:
It should be noted that the element on the right has only two nodes on one side, this is to
facility the proper coupling of eight noded quadrilaterals to four noded quadrilaterals on
the left. The formulation of the transition element will ensure that kinematically
equivalent nodal forces are applied to the elements on either side of it. These types of
transition elements can be cumbersome to use as they need to be oriented in a specific
manner and so usually require that the mesh is created manually.
Mixed dimensioned transition elements are used in regions where there is a mismatch
between element dimensions. The image shown here indicates a transition from where
the shell and solid are stitched together. The advantage of this type of transition is that
small 3D local details can be evaluated in structures that are otherwise thin-walled in
nature. Notice how there is a mismatch between the element densities, there are four
elements across the width in the shell side while there are seven elements on the solid
side of the transition.
Rigid Links :Rigid linking elements, otherwise known as kinematic elements, are
another way of connecting different element types within the same finite element model.
Rigid Link elements enforce kinematic relationships between the displacements at the
interface nodes.
Rigid links elements ususally come in two forms, 2D and 3D. Planar rigid links are
intended to connect plane stress, plane strain, beam, and axisymmetric solid elements.
The rigid linking formulation connects two nodes and produces constrained
displacements that satisfy kinematic relationships. 2D rigid link elements usually have
the requirement that the model must lie in the global XY plane.
The 3D variant of the rigid link element can be arbitrarily oriented in 3D space. It is
designed to connect three dimensional elements such as beams, shells, and solids. Again
the rigid linking connects two nodes and produces constrained displacements that satisfy
kinematic relationships, but it is more general than the planar type.
There is usually one master node and there can be one or more slave nodes in each
kinematic element. When the master node moves in space, the rigid linking mechanism
enforces the slave nodes to move with it, that is they maintain a constant relative distance
and orientation to the master node. As the slave nodes maintain constant proximity, the
nodes associated with the rigid element cannot deform relative to each other. This
restricts them from undergoing load based deformations and so a stress disturbance is
observed in the region where the link was applied. Therefore, kinematic elements should
really only be used in areas of a model where the stress quantities are not required.
Commercial CAE software have implemented an improved version of rigid links that
allows warping of the cross-section. They also allow linear contractions between nodes
that comprise the coupling element, this means that the stress disturbances are reduced
considerably.
the very high stiffness would be able to transfer all the energy to the adjacent nodes.
CAE tools provide for definition of such rigid elements.
Sub-Modeling:
At times we are interested in a studying the behavior of a large model without losing
accuracy of the models at some critical locations. But if the local regions need good
results the mesh there needs to be very fine. Achieving the required kind of mesh
transition might result in large number of elements where in the computer resources
might not be sufficient to solve the problem at one go.
Hence under such situations a multi pronged approach called sub-modeling is employed.
Sub-modeling is based on St. Venant's principle, which states that if an actual distribution
of forces is replaced by a statically equivalent system, the distribution of stress and strain
is altered only near the regions of load application. Therefore, if the boundaries of the sub
model are far enough from the stress concentration, reasonably accurate results can be
calculated in the sub model.
Sub modeling allows the user to study a localized part of a model with a refined mesh
based on the solution from an initial, global model with a coarse mesh. The sub model is
analyzed as an independent simulation. The only link between it and the global analysis
is the transfer of the values of variables to the relevant boundary nodes of the sub model.
Sub model boundaries do not have match element boundaries in the global model, the
evaluation of the nodal values in the sub model can be determined by interpolating
between results in the coarse model. Most of the CAE software are capable of doing this
automatically.
Under this approach, first a coarse model is made and analyzed for behavior. Then a
piece of the model containing the region of interest is chosen. The boundary of the model
are chosen such that the results remain more or less same irrespective of the mesh
density.
The FE mesh for such a model is then refined to the level of accuracy required and then
the boundary conditions are applied at the cut boundaries. Since the number of nodes at
the cut boundary are more compared to those in the initial model, the results in the
original model are transferred to the boundaries of the cut model by interpolation. Loads
any present in the region are gain applied and then the solution is re-run to get accurate
results at regions of interest. The approach can be repeated for multiple regions of
interest.
Sub models can usually be used with all types of analysis procedures, including nonlinear
and dynamic analysis models.
This fast analysis technique is also known as the cut-boundary displacement method or
the specified boundary displacement method. The cut boundary is the boundary of the
sub model which represents a cut through the coarse model. Analysis results calculated
on the cut boundary of the coarse model are specified as boundary conditions for the sub
model.
rubber
are quiet different from those exhibited by materials like steel. The behavior of rubber
like mateial can be best simulated y using material models such as Mooney rivilin or
Neo-Hookean models.
To aid such simulations one needs to input the uni-axial compression or tension test
results for the material along with the material properties, for the material.
Planar symmetry : This is the most common type of symmetry found in finite element
models. Reflective symmetry is a condition where the same pattern is seen to be mirrored
in a plane. The image shown below indicates that the model is doubly symmetric, one
plane in the horizontal direction and another in the vertical direction. As this is a 2D
model the symmetric planes are comprised of lines. Planar symmetry can occur in 3D
also, the plane of symmetry would be defined by a surface.
It is important to apply a symmetric constraint to all new edges and surfaces that are
created due to taking advantage of the symmetry property. In this case, the vertical
constraint is to prevent any horizontal movement, while the horizantal constraint would
be to prevent any vertical movement. Had this been a 2D mesh in 3D space, further
conditions would have to have been applied, i.e. constrain the model from rotation about
the planar axes.
Axisymmetry or Rotational Symmetry:If a shape can be defined by rotating a crosssection about a line, then it is said to be axi-symmetric. If the loads and boundary
conditions are also axi-symmetric in nature, then an axi-symmetric analysis may be
carried out.
Axi-symmetric elements are 2D planar in nature, and are used to model a revolved 3D
part in 2D space. Each element deforms as if it were a solid ring rotated about the axis of
revolution. Axi-symmetric elements are available most finite element packages and in a
range of element shapes and types. No special boundary conditions have to be applied to
these elements to achieve the symmetry condition.
Cyclic Symmetry : Cyclic symmetry is the geometric repetition in the form of cyclic
sectors. The structure is composed of a series of identical sectors that are arranged
circumferentially to form a ring. The image shown below is of a wheel model. The
cyclic portion is cut out, leaving the remainder to illustrate that it comprises eight
cyclic portions.
In order that the cyclic constraint be achieved, the new surfaces that comprise the
cyclic boundary must be constrained in the circumferential direction. This is achieved
by assigning the surfaces a cylindrical coordinate system and subsequently
constraining the circumferential freedom, which is usually dof 2 (the theta direction).
Some FE packages have this transformation already built in to their functionality, you
simply specify the lines or surfaces that are to be applied the cyclic constraint, it
carries out the transformation automatically.
In cases were only minor details disrupt a structure's symmetry, you can oftentimes
ignore them or treat them as being symmetric, in order to gain the benefits of using a
smaller analysis model. The gain in model simplification usually outweighs the cost in
reduced accuracy when un-symmetric features are deliberately ignored. Such modeling
practices should be carried out carefully.
Non-Linear Analysis:
Non-Linear Analyses : In order to explain non-linearity in stress analyses, lets examine
the nature of linear solutions. Many assumptions are made in linear analyses, the two
primary ones being the stress/strain relationship and the deformation behavior. The stress
is assumed to be directly proportional to strain and the structure deformations are
proportional to the loads. A fishing rod is an example of a non-linear structure made of
linear material. A stress analysis problem is linear only if all conditions of proportionality
hold. If any one of them is violated, then we have a Non-Linear problem.
Most real life structures, especially plastics, are non-linear, perhaps both in structure and
in material. Most plastic materials have a non-linear stress strain relationship. The nonlinearity arising from the nature of material is called 'Material Non-linearity'.
Furthermore, thin walled plastic structures exhibit a non-linear load-deflection
relationship, which could arise even if the material were linear (fishing rod). This kind is
called geometric non-linearity.
All non- linearities are solved by applying the load slowly (dividing it into a number of
small loads increments). The model is assumed to behave linearly for each load
increment, and the change in model shape is calculated at each increment. Stresses are
updated from increment to increment, until the full applied load is reached.
In a nonlinear analysis, initial conditions at the start of each increment is the state of the
model at the end of the previous one. This dependency provides a convenient method for
following complex loading histories, such as a manufacturing process. At each increment,
the solver iterates for equilibrium using a numerical technique such as the NewtonRaphson method. Due to the iterative nature of the calculations, non-linear FEA is
computationally expensive, but reflects the real life conditions more accurately than
linear analyses. The big challenge is to provide a convergent solution at minimum cost
(the minimum number of increments).
While simulating the structural behavior there could be various kinds of non- linearities
which are as follows:
Geometric Non-Linearity: At times the deformation of a component in a direction can be
more than three times its dimensions in a particular direction. Such problems fall under
the category of geometric non-linearity.
Such problems instead of solving in a single go, have to solved in multiple phases each
time updating the shape of the component from the results of the previous step. Here
instead of applying the load in one single step, shall be applied in multiple steps
incrementally so that deformed shape at the end of each step shall be input geometry for
the next step.
Material non-linearity: At times the stress at a location in FE model shall be more than
the yield. Under such a situation, the stress strain behavior of the system shall be no
longer linear and the system has to follow the stress strain curve of the material. Under
such situation also, the solution instead of running in a single go is run in multiple steps.
At the end of each step, the solver checks for the plastic strains while trying to follow the
material stress strain curve. Here also the stiffness matrix at the end of a step shall be
input for the next step.
Contact non-Linearity: Contact problems are non-linear in nature because unlike in the
regular problems where in the stress increases linearly with the increase in load value, the
stress value in a contact problem may not increase or if it increases, it may not be
proportional to load. This is due to the fact the load bearing area (Contact Area) increases
with the increase in load.
Like other non-linear analysis here also the load is applied in steps to capture the gradual
engagement of the contact regions.
Thermal Analysis:
Thermal analysis is used to determine the temperature distribution, heat accumulation or
dissipation, and other related thermal quantities in an object. The nodal degrees of
freedom (primary unknown data) are the temperatures. The primary heat transfer
mechanisms are conduction, convection and radiation. In addition, less dominant
phenomena such as change of phase (melting or freezing) and internal heat generation
can occur.
Conduction: Conduction describes the temperature within the solid body, but does not
account for how heat will flow to and from the component. In order to carry out analyses
using a conduction model alone, temperatures must be described as part of the boundary
condition description. Heat flows (otherwise known as heat flux) are at times specified
along boundaries in addition to temperature as boundary conditions.
Convection: The convection heat transfer mechanism is due to the temperature gradient
between a fluid and a solid. This mechanism is complex as a boundary layer usually
exists within the fluid adjacent to the solid boundary. The heat flux is a function of the
temperature difference, T, and a heat transfer coefficient, h.
The heat transfer coefficient is dependent on many factors such as fluid pressure,
velocity, density, specific heat (ratio of specific heats if the fluid is compressible),
viscosity and conductivity. It is also dependent on surface properties such as roughness
and geometry.
Steady state Thermal analysis: In a steady state thermal analysis we can estimate the
temperature at various location in the FE model which could be under the individual or
combined influence of Conduction, Convection and Radiation.
Here it is assumed that the conditions of the system do not vary with time. In this kind of
analysis one can specify the temperature at know points as boundary conditions. Also the
heat transfer rate at known surfaces in the form of convection heat transfer co-efficients
along with the ambient temperatures can be specified.
The outputs that can be obtained through this analysis are temperatures values at various
nodes and the thermal flux values at each node. By careful formulation of the problem,
the heat transfer rate through various zones or the same through different modes of heat
transfer (Conduction/Convection/Radiation) can be estimates.
Transient thermal analysis: In a transient thermal analysis, the behavior of the system
with respect to time ( Under the influence of time varying conditions can be studied).
The results that can be obtained in a Transient thermal analysis are similar to those
obtained in a Steady state thermal analysis. But the results can be obtained over a range
of time.
Some of the CAE software has the ability to perform the coupled field analysis
automatically where as some do not have that capability. However in both cases the
Engineer can run one simulation, obtain out put results and apply them as inputs for the
other analysis.
Hence to perform a stress analysis under thermal loading, initially a thermal analysis is
run and the temperature results obtained are transferred to structural model as thermal
loads.
Along with the thermal loads ( In the form of temperature) other mechanical loads also
can be applied for structural model to study the behavior under combined loading.
Modal Analysis:
Vibration usually becomes a concern when the amplitudes grow large enough to cause
either excessive stress, or if it disturbs the people in, on or near the surroundings. As far
as most structures are concerned, vibration will disturb the people around the structure
long before stress becomes an issue. There are many items of equipment (balances,
microscopes, cameras, transmission equipment etc.) that are very sensitive to vibration.
Modal analyses are important in machines where there is likely to be cyclic out of
balance forces, such as in rotating machinery (engines, electric and pneumatic motors,
generators, industrial equipment, etc.) and fluid flow applications (due to alternating
vortex shedding). The chief aim of any vibration analysis is to ensure that the system is
not subject to a dangerous resonant condition during the range of operation.
The equation for equilibrium for a dynamic system without any external forces is given
by:
[M]*Double differential [X] + [K]*[X] = 0 where,
[X] is the displacement matrix
[K] is the stiffness matrix
and [M] is the mass matirx
problem in matrix terminology is called as Eigen value problem. Hence modal analysis is
also to referred to as Eigen value analysis.
The square root of Eigen values is nothing but the natural frequency of the system. A
coarse mesh is sufficient for a modal/dynamic analysis compared to a stress analysis.
Consistent Vs lumped mass matrices: The mass matrix that is used for the
Modal analysis can be in two forms.
In one of the forms, the total mass is just lumped at all the nodes of the element
proportionately based upon the distance of the node from the C.G of the element. Under
such conditions, the mass matrix shall be a diagonal matrix and it becomes very easy to
compute the inverse of the mass matrix and also matrix multiplication wise.
Where as in the other form, the mass is distributed in tune with the polynmial used to
define the edge of the element. Mass matrix obtained using such an approach is not a
diagonal matrix and hence the effort required to inverse the mass matrix or multiplication
of inverted mass matrix with stiffness matrix requires considerable effort.
But the results obtained by consistent mass matrix are more accurate and
computationally more time consuming.
Different kinds of solvers for Modal analysis : There are various kinds of
solvers used for Modal analysis. Of them Lancozos method and Sub-Space iteration
method are most popular.
Of these two Lancozos method is more robust and can generate Eigen values even for
meshed with bad quality.
Reduced methods:
At times an experienced engineer shall be sure that a information about mode shapes, in a
particular mode or direction. Hence to decrease the computational time, even before the
analysis can be performed to evaluate the natural frequencies, Engineer can instruct the
software not to calculate the natural frequencies that shall have mode shapes in a
particular direction.
This way the computational time can be reduced considerable without studying for the
mode shapes and natural frequencies which are of not importance to the Engineer.
If Super elements are used in a dynamic analysis, the CAE software shall be ignoring
some modes and mode shapes automatically and this results in a reduced modal analysis.
Also at times if the engineer can not decide upon the un-wanted modes, the CAE
software based upon certain logic can eliminate some modes and mode shapes.
Random Vibration analysis: This analysis is similar to the shock analysis. But
the loads applied are not realistic. Probabilistic loads arrived at methodically shall be
applied and the system response is studied for the same.
The output for such an analysis shall be the applied Time Vs the RMS response of the
system in terms of displacement/Strain/Stress.
Buckling analysis:
Buckling is a critical state of stress and deformation, at which a slight disturbance causes
a gross additional deformation, or perhaps a total structural failure of the part. Buckling
failures do not depend on the strength of the material, but are a function of the component
dimensions and modulus of elasticity. Therefore, materials with a high strength will
buckle just as quickly as low strength ones.
If a structure has one or more dimensions that are small relative to the others (slender or
thin-walled), and is subject to compressive loads, then a buckling analysis may be
necessary. From an FE analysis point of view, a buckling analysis is used to find the
lowest multiplication factor for the load that will make a structure buckle. The result of
such an analysis is a number of buckling load factors (BLF). The first BLF (the lowest
factor) is always the one of interest. If it is less than unity, then buckling will occur due to
the load being applied to the structure.
Buckling analysis aims at determining, Eulers critical loads required to buckles a column
or structure. There are two methods of performing a Buckling analysis. Of them one is a
Eigen value analysis which calculates the buckling modes through linear techniques. The
analysis is also used to find the shape of the buckled structure. The other method
requires a Geometric non-Linear analysis to be run to determine the buckling load. In the
non-linear method it is tough to calculate the Buckling loads of higher mode since the
mesh distorts almost after the first Buckling load is evaluated ( At times the mesh may
get distorted even before the first Buckling load can be calculated).
In buckling mode of failure, soon after the column fails, the stiffness characteristics of the
structure changes and hence more or additional load is required to cause further failure by
buckling. Hence there can be multiple Buckling load factors for a structure.
Linear Buckling : This method is very conservative. This is similar to Eigen value
analysis to estimate the natural frequencies. In this method a load is applied at a certain
location and the Buckling analysis gives the load factors as solution.
That is the applied load multiplied by the Buckling load factors obtained gives the Eulers
critical load required to buckle a column or structure. A negative factors indicate that the
direction of the load should be reversed. If more than one load is to be applied to evaluate
the Buckling load factors then they should be applied in the same proportion as they
would be in reality so as to get the accurate factor.Like in case of dynamic analysis there
shall be multiple Eigen values and associated Eigen vectors for the case of Buckling also.
The second, third and so on Buckling factors represent the higher buckling modes and
their behavior while failing under buckling mode.
An important note is that the eigenvalue method does not take into account of any initial
imperfections in the structure and so the results rarely correspond with practical tests.
Eigenvalue solutions usually over estimate the buckling load and give no information
about the post-buckling state of the structure. Sudden buckling simply does not occur in
the real world.
An interesting variation arises in the case of automotive applications. In the case of front
end collision, the hood is expected to crumple (buckle) in order to absorb the energy of
collision, as well as to save the passenger compartment. In such cases, we are not
designing against, but for buckling.
Initial stage: Before planning for a simulation, try to assess what contribution the
analysis will provide to the overall design objective. Ensure that the scope of the analysis
is sufficiently detailed in any specifications, as it will avoid problems at a later stage.
Details such as the source of geometry (file type), loading data, outputs required and time
frame should be carefully considered.
Before deciding on how detailed a FEA model will have to be in order to give adequate
resolution of results, gloss over the theories relating to the design problem at hand. Many
problems have closed form solutions, or closed form solutions can provide very close
approximations. Hand calculations and simple computer tools will provide indispensable
information.
It is worthwhile forming very simple parameterized analyses. By varying the parameters
(thickness, number of supports, etc.), it will become obvious what parameters are
dominant in regard to structural response. Isolating the leading parameters early in the
design cycle will facilitate an efficient and fruitful outcome. This process is known as
design optimization.
It will be worth spending some time over relevant documentation on the physical
phenomena being studied.
displacement
elements).
If the stress varies in two directions only, then a plane stress analysis is
suitable. If strain is constant through the thickness, then plane strain
analysis is appropriate.
If there are any rotating parts, then a frequency analysis may be required
to ensure that they do not fail prematurely due to whirl fatigue.
If there are cyclic forces, then a full dynamic analysis may be required. If
the number of loading cycles is greater than 20,000 you may need to
consider fatigue.
If individual loads vary, then some sort of time integration scheme will
need to be adopted.
If there are long and slender members or thin sheets that are subject to
compressive loads, then a buckling analysis is required. If the fundamental
buckling load factor is less than unity while the yield stress of the material
has been exceeded, then a geometrically and material non-linear analysis
will be required. It is very important to know in advance the type of
expected buckling (global or local, bifurcation or snap-through).
If there are complex components that are not of immediate interest, then
they may be condensed out as a super element. Their effect will be
included in the overall structural response, but will reduce analysis times
considerably. Sub-structuring is also useful where there are repetitions of
geometry in the model. The repeated geometry can be meshed and substructured out. This super element can then be copied as required, hence
forming
the
bulk
of
the
structure
using
super
elements.
Large gradients in stress levels will require a high mesh density to capture
the behavior appropriately.
Model Preparation :Planning the creation of an analysis model saves time as it can help
you avoid making mistakes that may take long time to correct, or even prevent the total
re-modeling of the problem. The following precautionary measures can be adopted:
If using shell quadrilateral elements, try to ensure that the elements are all
oriented in the same global direction. As stresses can only be shown on
one side of the shell at a time, if some shells have surface normals that are
pointing inwards and the rest point outwards, the stress patterns what we
see through post-processor may not be correct.
Remember that distributed loads are not evenly divided among the nodes
of higher order (quadratic or higher) element edges or faces. Midside
nodes carry most of the load while the chord (corner) nodes carry much
less or even negative loads. Therefore do not apply distributed loads to
higher order element nodes manually. If applying them via the preprocessor GUI, check to ensure that constant loading is not being applied
to all nodes. The unequal loads applied at the nodes is known as
kinematically equivalent loading.
A free edge plot can be used to identify what element edges form an
external boundary. It can help identify locations where internal parts of the
structure are not joined. Holes or discontinuities in the mesh can arise
where different parts of the model were mesh separately. Node merging or
equivalencing can be used to correct such problems, all coincident nodes
are effectively replaced by one single node.
A plot of loads and constraints on the model will provide a check against
errors that were made at the time of user input. This check will save you
running the analysis for the wrong loading scenario.
Results Interpretation:
Before accepting the results of a simulation, it is worthwhile carrying out a number of
simple checks to ensure that your solution is appropriate and representative. FEA
provides an approximation to the problem being considered and so any results should be
treated with due caution. It is important to establish best practices for checking FEA
models in an organized manner, and should include the contribution and response of team
wherever possible.
The Simulation Report :Most solvers have some means of outputting annotations that
indicate the state and progress of the analysis into a file. This fils is commonly filled with
numerous warning messages, most of them irrelevant, but some may be important. It is
worth browsing through this file with a text editor and make a note of all relevant
warning/error messages. Some of the common messages that can be see in these files:
The Shear Modulus has not been specified, a value of (0.39 * Young's
Modulus) is assumed.
The state of the System Matrix :Some solvers yield details of the system matrix. The
pivot ratio is common output, large values indicate that accuracy is poor. This may be
due to insufficient or incorrect boundary conditions being applied or due to the lack of
element continuity. If the ratio of maximum to minimum element stiffness is given, its
ideal value is unity. a very high value indicates that material properties may be applied
incorrectly or that element constants such as thickness or beam properties are defined
incorrectly (probably in the wrong units).
Logical Checks: At times the analysis Engineer might not be knowing the exact value to
be expected from the analysis ( Quantitative results). But he can always check the results
qualitatively in the following ways:
The Ballpark Check: For the verification of results, it is best to attempt to estimate the
results with a simpler solution, preferably before an analysis result is available. Your
approximate solution should be in the same order of magnitude as the FEA solution.
Simple approximations may be values obtained from analytic theory, empirical equations
or other sources such as tables.
Node Averaging :During an analysis, the stresses within an element are determined at
points interior to the element known as Gauss points. The values at the nodes are
determined by extrapolating the Gauss point stresses to the boundary. Therefore, there
can be multiple values of stress for the same node. In order that a smooth and continuous
plot is obtained, post-processors average the stresses at each node.
Most post-processors generate contour plots with node averaging switched on as default.
It is worthwhile switching off node averaging, as the un-averaged plot can give good
indication of the discretisation error in the model. When averaging is switched off, the
contour plot may look jagged in places, highlighting that the stresses change considerably
across inter-element boundaries. This indicates that the true stresses were not captured at
that location due to the low level of discretisation (coarse mesh). In this case, it is
worthwhile refining the mesh in the affected regions and re-running the analysis.
Some post-processors have facilities to estimate the discretisation error in a solution. The
estimation is usually made on some sort of algorithm centered around the relative energy
error within each element. Some software goes further and carries out adaptive solutions,
by refining the mesh in one of two ways, increasing the number of elements (hrefinement) or increasing the polynomial order of the elements in question (prefinement).
Linear Vs Non-Linear: Based on the results of the analysis, it may become more
apparent that a linear solution is not representative of the problem being solved. Common
indications that a linear analysis is insufficient are:
If any gaps over close, causing inter penetration, then a contact analysis is
required.
If the behavior is inconsistent with assumptions, it is possible that nonlinear behavior is occurring.