1 Finite Element Analysis Methods
1 Finite Element Analysis Methods
1.1 Introduction
The finite element method (FEM) rapidly grew as the most useful numerical analysis tool for engineers and
applied mathematicians because of it natural benefits over prior approaches. The main advantages are that it
can be applied to arbitrary shapes in any number of dimensions. The shape can be made of any number of
materials. The material properties can be non‐homogeneous (depend on location) and/or anisotropic (depend
on direction). The way that the shape is supported (also called fixtures or restraints) can be quite general, as
can the applied sources (forces, pressures, heat flux, etc.). The FEM provides a standard process for converting
governing energy principles or governing differential equations in to a system of matrix equations to be solved
for an approximate solution. For linear problems such solutions can be very accurate and quickly obtained.
Having obtained an approximate solution, the FEM provides additional standard procedures for follow up
calculations (post‐processing), such as determining the integral of the solution, or its derivatives at various
points in the shape. The post‐processing also yields impressive color displays, or graphs, of the solution and its
related information. Today, a second post‐processing of the recovered derivatives can yield error estimates
that show where the study needs improvement. Indeed, adaptive procedures allow automatic corrections and
re‐solutions to reach a user specified level of accuracy. However, very accurate and pretty solutions of models
that are based on errors or incorrect assumptions are still wrong.
When the FEM is applied to a specific field of analysis (like stress analysis, thermal analysis, or vibration
analysis) it is often referred to as finite element analysis (FEA). FEA is the most common tool for stress and
structural analysis. Various fields of study are often related. For example, distributions of non‐uniform
temperatures induce non‐obvious loading conditions on solid structural members. Thus, it is common to
conduct a thermal FEA to obtain temperature results that in turn become input data for a stress FEA. FEA can
also receive input data from other tools like motion (kinetics) analysis systems and computation fluid dynamic
(CFD) systems.
1.2 Basic Integral Formulations
The basic concept behind the FEM is to replace any complex shape with the union (or summation) of a large
number of very simple shapes (like triangles) that are combined to correctly model the original part. The
smaller simpler shapes are called finite elements because each one occupies a small but finite sub‐domain of
the original part. They contrast to the infinitesimally small or differential elements used for centuries to derive
differential equations. To give a very simple example of this dividing and summing process, consider
calculating the area of the arbitrary shape shown in Figure 1‐1 (left). If you knew the equations of the
bounding curves you, in theory, could integrate them to obtain the enclosed area. Alternatively, you could
split the area into an enclosed set of triangles (cover the shape with a mesh) and sum the areas of the
individual triangles:
∑ .
Now, you have some choices for the type of triangles. You could pick straight sided (linear) triangles, or
quadratic triangles (with edges that are parabolas), or cubic triangles, etc. The area of a straight‐sided triangle
is a simple algebraic expression. The area of a curved triangle is relatively easy to calculate by numerical
integration, but is computationally more expensive to obtain than that for the linear triangle. The first two
triangle mesh choices are shown in Figure 1‐1 for a large element size. Clearly, the simple straight‐sided
triangular mesh (on the left) approximates the area very closely, but at the same time introduces geometric
errors along the boundary. The boundary geometric error in a linear triangle mesh results from replacing a
Draft 12.0. Copyright 2009. All rights reserved. 7
FEA Concepts: SW Simulation Overview J.E. Akin
boundary curve by a series of straight line segments. That geometric boundary error can be reduced to any
desired level by increasing the number of linear triangles. But that decision increases the number of
calculations and makes you trade off geometric accuracy versus the total number of required area calculations
and summations.
Area is a scalar, so it makes sense to be able to simply sum its parts to determine the total value, as shown
above. Other topics, like kinetic energy or strain energy, can be summed in the same fashion. Indeed, the very
first applications of FEA to structures was based on minimizing the energy stored is a linear elastic material.
The FEM always involves some type of governing integral statement. That integration is also converted to the
sum of the integrals over each element in the mesh. Even if you start with a governing differential equation, it
gets converted to an equivalent integral formulation by one of the methods of weighted residuals (MWR). The
two most common methods, for FEA, are the Galerkin Method and the Method of Least Squares Figure.
Figure 1‐1 An area crudely meshed with linear and quadratic triangles
You may think that the geometric boundary error cited for the linear triangles is eliminated by choosing to use
the mesh of curved quadratic triangles (on the right). The parabola segments pass through three points lying
exactly on the boundary curve, but can degenerate to straight lines in the interior. (To speed plotting of small
elements, most systems draw all the parabolas as two straight line segments, as on the right in Figure 1‐1.)
Thus, the boundary shape error is indeed reduced, at the expense of more complicated area calculations, but it
is not eliminated. Some geometric error remains because most engineering curves are circular arcs, splines,
or nurbs (non‐uniform rational B‐splines) and thus are not matched by a parabola. The most common way to
reduce mesh geometric error is to simply use smaller elements, like Figure 1‐2 shows. The default element
choice in SW Simulation is the quadratic element. Other systems offer a wider range of edge polynomial
degree (e.g. cubic), as well as other shapes like quadrilaterals or rectangles. In three‐dimensional solid
applications some systems offer dozens of choices for the edge degree polynomial order, and shapes including
hexahedral, wedges, and tetrahedral elements. Hexahedral elements are generally more accurate, but can be
more challenging to mesh automatically. Tetrahedral elements can match hexahedral element performance
by using more (smaller) elements, and tetrahedral elements are much easier to mesh automatically. SW
Simulation uses only tetrahedral elements for solid studies.
An example of the small two‐dimensional geometric boundary error due to different curved shapes is seen in
Figure 1‐3 where a circular arc and a parabola pass through the same three points. (A new method, called
isogeometric analysis, can essentially eliminate all geometric errors, but it introduces new approximations in
other study stages, such as in the restraint conditions.)
Draft 12.0. Copyright 2009. All rights reserved. 8
FEA Concepts: SW Simulation Overview J.E. Akin
Figure 1‐2 Mesh refinement quickly reduces geometric boundary errors for linear (left) or quadratic elements
Figure 1‐3 Linear or parabolic elements never exactly match circular shapes
1.3 Stages of Analysis and Their Uncertainties
A FEA always involves a number of uncertainties that impact the accuracy or reliability of each stage of a FEA
and its results. The book, Building Better Products with Finite Element Analysis by Adams and Askenazi [1]
gives an outstanding detailed description of most of the real‐world uncertainties associated with solid
mechanics FEA. All engineers conducting stress studies should read it. That book also points out how poor
solid modeling skills can adversely affect the ability to construct meshes for any type of FEA. Here, the most
important FEA uncertainties are highlighted.
The typical stages of a FEA study are listed below:
1. Construct the part(s) in a solid modeler. It is surprisingly easy to accidentally build flawed models with
tiny lines, tiny surfaces or tiny interior voids. The part will look fine, except with extreme zooms, but it
may fail to mesh. Most systems have checking routines that can find and repair such problems before
you move on to a FEA study. Sometimes you may have to export a part, and then import it back with a
new name because imported parts are usually subjected to more time consuming checks than “native”
parts. When multiple parts form an assembly, always mesh and study the individual parts before
studying the assembly. Try to plan ahead and introduce split lines into the part to aid in mating
Draft 12.0. Copyright 2009. All rights reserved. 9
FEA Concepts: SW Simulation Overview J.E. Akin
assemblies and to locate load regions and restraint (or fixture or support) regions. Today, construction
of a part is probably the most reliable stage of any study.
2. Defeature the solid part model for meshing. The solid part may contain features, like a raised logo,
that are not necessary to manufacture the part, or required for an accurate analysis study. They can
be omitted from the solid used in the analysis study. That is a relative easy operation supported by
most solid modelers (such as the “suppress” option in SW) to help make smaller and faster meshes.
However, it has the potential for introducing serious, if not fatal, errors in a following engineering
study. This is a reliable modeling process, but its application requires engineering judgment. For
example, removing small radius interior fillets can greatly reduces the number of elements and
simplifies the mesh generation. But, that creates sharp reentrant corners that can yield false infinite
stresses. Those false high stress regions may cause you to overlook other areas of true high stress
levels. Small holes lead to many small elements (and long run times). They also cause stress
concentrations that raise the local stress levels by a factor of three. The decision to defeature them
depends on where they are located in the part. If they lie in a high stress region you must keep them.
But defeaturing them is allowed if you know they occur in a low stress region. Such decisions are
complicated because most parts have multiple possible loading conditions (load cases) and a low stress
region for one load case may become a high stress region for another load case.
3. Combine multiple parts into an assembly. Again, this is well automated and reliable from the
geometric point of view and assemblies “look” as expected. However, geometric mating of part
interfaces is very different for defining their physical (displacement, or temperature) mating. The
physical mating choices are often unclear and the engineer may have to make a range of assumptions,
study each, and determine the worst case result. Having to use physical contacts makes the linear
problem require iterative solutions that take a long time to run and might fail to converge.
4. Select the element type. Some FEA systems have a huge number of available element types (with
underlying theoretical restrictions). The SolidWorks system has only the fundamental types of
elements. Namely, truss elements (bars), frame elements (beams), thin shells (or flat plates), thick
shells, and solids. The system selects the element type (beginning in 2009) based on the shape of the
part. The user is allowed to covert a non‐solid element region to a solid element region, and visa
versa. Knowing which class of element will give a more accurate or faster solution requires training in
finite element theory. At times a second element type study is used to help validate a study based on
what is thought to be the best element type.
5. Mesh the part(s) or assembly, remembering that the mesh solid may not be the same as the part solid.
A general rule in FEA is that your computer never has enough speed or memory. Sooner or later you
will find a study that you cannot execute. Often that means you must utilize a crude mesh (or at least
crude in some region) and/or invoke the use of symmetry or anti‐symmetry conditions. Local solution
errors in a study are proportional to the product of the element size and the gradient of the secondary
variables (i.e., gradient of stress or heat flux). Therefore, you exercise mesh control to place small
elements where your engineering judgment estimates high stress (or flux) regions, as well as large
elements in low stress regions. The local solution error also depends on the relative sizes of adjacent
elements. You do not want skinny elements adjacent to big ones. Thus, automatic mesh generators
have options to gradually vary adjacent element sizes from smallest to biggest.
The solid model sent to the mesh generator frequently should have load or restraint (fixture) regions
formed by split lines, even if such splits are not needed for manufacturing the parts. The mesh
typically should have refinements at source or load regions and support regions.
A mesh must look like the part, but that is not sufficient for a correct study. A single layer of elements
filling a part region is almost never enough. If the region is curved, or subjected to bending, you want
Draft 12.0. Copyright 2009. All rights reserved. 10
FEA Concepts: SW Simulation Overview J.E. Akin
at least three layers of quadratic elements, but five is a desirable lower limit. For linear elements you
at least double those numbers.
Most engineers do not have access to the source code of their automatic mesh generator. When the
mesher fails you frequently do not know why it failed or what to do about it. Often you have to re‐try
the mesh generation with very large element sizes in hopes of getting some mesh results that can give
hints as to why other attempts failed. The meshing of assemblies often fails. Usually the mesher runs
out of memory because one or more parts had a very small, often unseen, feature that causes a huge
number of tiny elements to be created. You should always attempt to mesh each individual part to
spot such problems before you attempt to mesh them as a member of an assembly.
Automatic meshing, with mesh controls, is usually simple and fast today. However, it is only as reliable
as the modified part or assembly supplied to it. Distorted elements usually do not develop in
automatic mesh generators, due to empirical rules for avoiding them. However, distorted elements
locations can usually be plotted. If they are in regions of low gradients you can usually accept them.
You should also note that studies involving natural frequencies are influenced most by the distribution
of the mass of the part. Thus, they can still give accurate results with meshes that are much cruder
than those that would be acceptable for stress or thermal studies.
6. Assign a linear material to each part. Modern FEA systems have a material library containing the
“linear” mechanical, thermal, and/or fluid properties of most standardized materials. They also allow
the user to define custom properties. The values in such tables are often misinterpreted to be more
accurate and reliable than they actually are. The reported values are accepted average values taken
from many tests. Rarely are there any data about the distribution of test results, or what standard
deviation was associated with the tests. Most tests yield results that follow a “bell shaped” curve
distribution, or a similar skewed curve. The tests for stainless steel tend to have narrow distributions,
like that on the left in Figure 1‐4, while the results for cast iron have wider distributions. When you
accept a tabulated property value as a single number to be used in the FEA calculation remember it
actually has a probability distribution associated with it. You need to assign a contribution to the total
factor of safety to allow for variations from the tabulated value.
Draft 12.0. Copyright 2009. All rights reserved. 11
FEA Concepts: SW Simulation Overview J.E. Akin
results. The axial stress in a bar is equal to the elastic modulus times the strain, . Thus, if E is
only known to three or four significant figures then the reported stress result should have no more
significant figures. (It is true that the computer uses many digits to obtain the most accurate answer,
but you should not accept the displayed numbers blindly.)
Material data are usually more reliable than the loading values (considered next), but less accurate
that the model or mesh geometries.
7. Select regions of the part(s) to be loaded and assign load levels and load types to each region. In
mathematical terminology, load or flux conditions on a boundary region are called Neumann boundary
conditions, or non‐essential conditions. The geometric regions can be points (in theory), lines,
surfaces, or volumes. If they are not existing features of the part, then you should apply split lines to
the part to create them before activating the mesh generator. Point forces, or heat sources, are
common in undergraduate studies, but in a FEA they cause false infinite stresses, or heat flux. If you
include them do not be mislead by the high local values. Refining the mesh does not help much since
the smallest element still reports near infinite values. In reality, point loads are better modeled as a
total force, or pressure, acting over a small area formed by prior split lines.
Saint Venant’s Principle states that two different, but statically equivalent, force systems acting on a
small portion of the surface of a body produce the same stress distributions at distantness large in
comparison with the linear dimensions of the portion where the forces act. In undergraduate statics
and dynamics courses engineers are taught to think in terms of point forces and couples. Solid
elements do not accept pure couples as loads, but statically equivalent pressures can be applied to
solids and yield the correct stresses. Indeed, a couple at a point is almost impossible to create, so the
distribution of pressures is probably more like the true situation.
The magnitudes of applied loads are often guesses, or specified by a governing design standard. For
example, consider a wind load. A building standard may quote a pressure to be applied for a given
wind speed. But, how well do you know the wind speed that might actually be exerted on the
structure? Again, there probably is some type of “bell curve” around the expected average speed.
You need to assign a contribution to the total factor of safety to allow for variations in the uncertainty
of the load value or actual spatial distribution of applied loads.
Loading data are usually less accurate than the material data, but much more accurate that the
restraint or supporting conditions considered next.
8. Determine (or more likely assume) how the model interacts with the surroundings not included in your
model. These are the restraint (support, or fixture) regions. In mathematical terminology, these are
called the essential boundary conditions, or Dirichlet boundary conditions. You cannot afford to model
everything interacting with a part. For many decades engineers have developed simplified concepts to
approximate surroundings adjacent to a model to simplify hand calculations. They include roller
supports, smooth pins, cantilevered (encastre, or fixed) supports, straight cable attachments, etc.
Those concepts are often carried over to FEA approaches and can over simplify the true support
nature and lead to very large errors in the results.
The choice of restraints (fixations, supports) for a model is surprisingly difficult and is often the least
reliable decision made by the engineer. Small changes in the supports can cause large changes in the
results. It is wise to try to investigate a number of likely or possible support conditions in different
studies. When in doubt, try to include more of the surrounding support material and apply assumed
support conditions to those portions at a greater distance from critical part features.
You need to assign a contribution to the total factor of safety to allow for variations in the uncertainty
of how or where the actual support conditions occur.
Draft 12.0. Copyright 2009. All rights reserved. 12
FEA Concepts: SW Simulation Overview J.E. Akin
9. Solve the linear system of equations, or the eigenvalue problem. With today’s numerical algorithms
the solution of the algebraic system or eigen‐system is usually quite reliable. It is possible to cause ill‐
conditioned systems (large condition number) with meshes having large elements adjacent to small
ones, but that is unlikely to happen with automatic mesh generators.
10. Check the results. Are the reactions at the supports equal and opposite to the sources you thought
that you applied? Are the results consistent with the assumed linear behavior? The engineering
definition of a problem with large displacements is one where the maximum displacement is more
than half the smallest geometric thickness of the part. The internal definition is a displacement field
that significantly changes the volume of an element. That implies the element geometric shape
noticeably changed from the starting shape, and that the shape needs to be updated in a series of
much smaller shape changes. Are the displacements big enough to require re‐solution with large
displacement iterations turned on? Have you validated the results with an analytic approximation, or
different type of finite element? Engineering judgments are required.
11. Post‐process the solution for secondary variables. For structural studies you generally wish to
document the deflections and stresses. For thermal studies you display the temperatures and heat
flux vectors. With natural frequency models you show (or animate) a few mode shapes. You can
control the number of contours employed, as well as their maximum and minimum ranges. The latter
is important if you want to compare two designs on a single page. Limit the number of digits shown on
the contour scale to be consistent with the material modulus (or conductivity, etc.). Contour plots
often do not reproduce well in a report, but graphs generally do, so learn to include graphs in you
documentation.
12. Determine (or more likely assume) what failure criterion applies to your study. This stage involves
assumptions about how a structural material might fail. There are a number of theories. Most are
based on stress values or distortional energy levels, but a few depend on strain values. If you know
that one has been accepted for your selected material then use that one (with a contribution to the
overall factor of safety). Otherwise, you should evaluate more than one theory and see which is the
worst case. Also keep in mind that loading or support uncertainties can lead to a range of stress levels,
and variations in material properties affect the strength and unexpected failures can occur if those
types of distributions happen to intersect, as sketched in Figure 1‐5.
13. Optionally, post‐process the secondary variables to measure the theoretical error in the study, and
adaptively correct the solution. This converges to an accurate solution to the problem input, but
perhaps not to the problem to be solved. Accurate garbage is still garbage.
14. Document, report, and file the study. The part shape, mesh, and results should be reported in image
form. Assumptions on which the study was based should be clearly stated, and hopefully confirmed.
The documentation should contain an independent validation calculation, or two, from an analytical
approximation or a FEA based on a totally different element type. Try to address the relative
uncertainties of the main analysis stages, as summarized in Figure 1‐1‐6.
Technical communication and documentation is always important. In America, engineers are
supposed to retain their calculations for at least seven years. Will your report be clear and helpful if
you have to defend it years later? Paper hardcopies are the most reliable for long term storage. (Can
you read the electronic media you used five years ago?)
Draft 12.0. Copyright 2009. All rights reserved. 13
FEA Concepts: SW Simulation Overview J.E. Akin
Figure 1‐5 Distributions of loads/restraints and material strengths can cause failure
Figure 1‐1‐6 Relative uncertainty of major modeling stages
You usually assume that the materials are linear. If not (creeping, hyperelastic, inelastic, plastic, viscoelastic,
etc.), define the appropriate material data and the nonlinear equations to be solved. Then the matrix system
becomes non‐linear. Your original results check may lead you to conclude that the problem is actually an
iterative one due to large displacements, or the need to insert physical contact interfaces.
1.4 Part Geometric Analysis and Meshing Failures
Before attempting meshing your part, for a finite element analysis, you should check your solid model for
potentially fatal geometric flaws that may not be noticed except at greatly magnified views. Within
SolidWorks this is called a Geometric Analysis. To utilise that feature, a geometric analysis check the
Angel_Connector part will be outlined:
1. With the part open, go to ToolsÆ Check will open the Check Entity panel.
Draft 12.0. Copyright 2009. All rights reserved. 14
FEA Concepts: SW Simulation Overview J.E. Akin
2. In that panel check the boxes for most entities, select Check.
3. Highlight each item in the Result List. As you scroll down the Result list the short edge location on the
part is illustrated by a yellow arrow. Either the feature needs to be eliminated (best), or the mesh will
need to be fine there (ok, usually).
4. To consider a potential mesh refinement you should determine the size of the small feature. Use
ToolsÆMeasure to open up the Measure panel. Select the XYZ option, click on a edge of the feature
to see its length displayed.
Draft 12.0. Copyright 2009. All rights reserved. 15
FEA Concepts: SW Simulation Overview J.E. Akin
5. Attempt to create a mesh: MeshÆCreate Mesh and select a default element size. As expected, that
process fails and a failure diagnostic message appears:
6. Right click on Mesh to open the Failure Diagnostics panel. Scroll down the lists of faces or edges that
caused the mesh failure. In this case, there is a highly distorted surface that formed with the fillets.
Sometimes this type of surface can be removed by suppressing the fillets, or by building them in a
different order. Sometimes the surface can split by inserting split lines to make more manageable
regions. Fixing the surface is better that having to try to control the mesh in such regions.
7. First, try to get some type of mesh output by specifying a small element size along the edges of the
distorted region MeshÆApply Mesh Control. Specify a local element size that will assure that one or
two elements will fit along the smallest edge.
Draft 12.0. Copyright 2009. All rights reserved. 16
FEA Concepts: SW Simulation Overview J.E. Akin
Surprisingly, this executed. But it yielded a distorted mesh in the region of the small edge. Ideally, the surface
triangles (one face of the tetrahedron) would be isosceles. That gives an element “aspect ratio” (say the ratio
of the long side divided by the short one) of unity. Here the triangles are curved. A few are also badly
distorted and not desirable for analysis if they are in an expected high stress region.
8. One measure of the quality of an element is its aspect ratio. Think of that as the ratio of the diameter
of the enclosing sphere to the diameter of the enclosed sphere. Alternately, use the ratio of the
longest element edge length to its shortest. An ideal aspect ratio should be near unity. Check the
mesh quality by looking at a plot of the aspect ratio of the elements. Select Mesh ÆCreate Mesh Plot
Æ Aspect Ratio. That shows an aspect ratio of more than ten, which is on the high side (five is a good
upper bound goal).
9. Try to improve this mesh by removing the bad surface, or subdividing it into controllable regions. At
the narrow region, insert a split line that avoids very small intersection angles with both curves.
Draft 12.0. Copyright 2009. All rights reserved. 17
FEA Conceptss: SW Simulaation Overview
w J.E. Akin
The ssmall slender partition will need very sm mall elements, but the largger partition can have largger ones.
Especially if you u
use the transittion control rratio to give fiive or more ggrowth layers at an enlargeement
ratio of about 1.2 instead of th
he default value of 1.5. Usse Mesh Æ Apply Mesh Co ontrol to speccify
elemment sides of 00.02 and 0.05
5, respectivelyy in the Mesh h Control pan nel. They givee a much bettter mesh
in this region.
Another partt, the Five_Ho
ole_Link, show ws a similar m
mesh distortio on that gives very bad elem ment aspect rratios
and might haave caused thhe mesh generation to fail.. It is wise to carry out a geometry analysis at variou us stages
of your solid construction. This part will be revieweed in a similarr manner, andd you will disccover multiple
problem regiions.
1. ToolssÆ Check will open the Ch
heck Entity p
panel. In that panel check tthe boxes for most entitiess, select
Checck.
Draft 12.0. Copyright 2009. Alll rights reservved. 1
18
FEA Concepts: SW Simulation Overview J.E. Akin
2. Try to generate a mesh for the part, and examine each of the regions in the Result List, via
MeshÆCreate Mesh. A mesh is created.
Note that the aspect ratios are bad in the first short edge region. For an ideal element shape, in addition to a
small aspect ratio, you also want the corner angles of the element to be equal. While that is not likely to occur
you should avoid corner angles less than 15 degrees or the element accuracy decreases due to too much
distortion (it develops a non‐positive geometric Jacobian). To avoid that condition, the mesh generator checks
the geometric Jacobian value at a number of points (under your control) in each element. Ideally, all the
values would be the same constant (e.g. all its edges are straight), or they should not vary greatly, and they
should never reach a fatal negative value (since that can cause a negative element volume).
The other regions in the check list for this part are where two curves become tangent. A common cause of
failure in mesh generation is to have two solid regions or two joining surfaces meet at a near zero angle. That
often happens in practice and often requires intervention to be able to create a mesh for analysis. If a
tangency condition is really required in the part, then you must force smaller element sizes there via the Mesh
Control option.
Draft 12.0. Copyright 2009. All rights reserved. 19
FEA Conceptss: SW Simulaation Overview
w J.E. Akin
If the part can be modified
d to avoid thee tangency, th
hen meshing becomes mu uch easier. Th
hat is illustratted
below wheree it was feasib
ble to avoid a tangency req quirement in this application.
Parts and asssemblies often fail to meshh by bad or unseen model features. Yoou need to leaarn how check your
part geometrry as it growss in complexitty. Try meshing each part at each majo
or change in complexity.
Draft 12.0. Copyright 2009. Alll rights reservved. 2
20