You are on page 1of 9

Building better products with FEA analysis

Pg.17

Process. Most engineering problems are solved using the following four steps.

1. Establish a clearly defined goal.


2. Compile and qualify the inputs
3. Solve the problem with the most appropriate means
4. Verify and document the results.

Pg. 18

What is the goal of the analysis? An excellent illustration of this decision involves developing
an analysis for overall displacements versus localized stress data. If stresses are not deemed
important, many details can be left off the model and your options for idealizations to improve
efficiency are expanded.

Predictive engineering versus failure verification. At the conception stage. Letting the results of
the analysis drive the choice of materials, features, wall thickness, and so forth is called
predictive engineering.

Assumptions and approximations must be minimized. Prior to undertaking an analysis,


evaluate whether failure is consistent across several parts or is isolated to one sample. An
isolated instance might suggest thatan expected loading scenario was responsible.

If analysis is to be used early in the design process, start on simplified geometry. Complex
solutions can be mode easily approximated as simple ones at this stage. Refine your analyses
as the part definition and behavior become more defined.

Pg. 19

Trend analysis versus absolute data. A trend analysis will not necessarily yield actual
performance data, but rather will show the effect of geometric changes and the sensitivity to
parameters such as material properties and applied load. i.e. a linear trend analysis shows that
decreasing the nominal wall thickness reduces cost by 10% with less than a 1% increase in
stress and deformation. Does this change make sense?

What input is required for the solution and what levels of uncertainty does it introduce? In
addition to filling the blanks on a data form or in a closed form equation, care must be taken to
qualify the data being used.

Pg. 20

What is the most efficient means to solve the problem? Even when FEA as a requirement has
been identified, try to explore simplifications which can be solved manually to provide a
starting point for the selection of inputs, and to qualify the general accuracy of the results.

Introduction to assumptive approach. If you make enough assumptions, you can analyse
anything

Assumptions regarding material uniformity, assembly variability, user inconsistency, and


general unpredictability need to be weighted, qualified, and documented.
Pg. 21

The key to success in light of these seemingly uncontrollable circumstances is a scientific


qualification of the uncertainties involved.

Pg. 22

Common misconceptions about FEA.

Meshing is everything. Accuracy local to a poorly shaped element will be affected by that
element.

Local mesh refinement tools are critical to ensuring a good mesh with gradual transitions
between densities. Choosing the right type of mesh or element for the problem is equally
important to well-shaped elements. A linear tetrahedral mesh on a thin-walled plastic part is
probably not going to provide reliable results, regardless of the uniformity of the elements.

While a good mesh can be obtained with patience in most pre-processors, well-conceived
boundary conditions and representative material properties will never be automated because
the engineer`s judgment must direct and qualify the necessary assumptions.

FEA Replaces testing. Only through correlation of test models and actual prototypes can the
methods and assumptions used in FEA be qualified. After qualification, it may be decided that
the analysis results for similar studies in the future are reliable and some development mental
testing may be eliminated

It is the best to say that FEA augments testing and vice-versa. Analytical results can suggest
strain gage placement and orientation. However, when applied correctly, it can be said that a
solid predictive engineering program can reduce testing in the design stage as the confidence
in simulation results grows.

Test models and additional research are almost always required to determine loads and
constraints. The same can be said about material properties. Many problems must be
bracketed (run at extremes of their inputs) to fully understand the sensitivity of the results to
these inputs.

With the simplicity of use comes the removal of the options required for accurate analyses

Check your egos at the door!!

Pg. 57

Results interpretation. The goals set forth at the beginning of the study. These should tell you
where to look and what you look for. In most cases, you will be looking for some evidence of
failure or assurance that failure is unlikely.

Pg. 58

Typical failure modes

Yielding (ductile materials). Yielding near stress concentrations is not considered a failure if it
produces localized strains, which merely redistribute the stress, whereupon yielding, ceases.

Insufficient stiffness Moving parts may have undesirable resonant frequencies if they are too
flexible.
Bukcling, Fatigue, Creep.

Classic Failure theories. It is extremely important to understand the load path and the material
behaviour for the particular test condition regardless of the means of analysis.

Pg. 59

Ductile failure theory. A load that produces yielding sets up residual stresses that extend the
elastic range under future loads in the same direction but decrease the elastic range under
future loads in the opposite direction. This is called the Bauschinger effect. Ductile failure is
characterized by slow crack or void propagation after significant plastic deformation.

Maximum normal stress theory. 1 and 3 equals the failure strength of the material.

Maximum shear stress theory (Tresca criterion). Yielding begins when the maximum shear
stress becomes equal to one-half the yielding strength. Failure in tension of ductile materials
occurs on one of the 45o maximum shear planes. This theory only predicts yield failure, hence
it is only good for ductile materials.

Distortion energy (Von Misses-Hencky) theory. Failure by yielding will occur whenever the von
Mises, or effective stress (`), equals the yield strength of the material. This stress quantity is
derived using a strain energy hypothesis and is given by the following equation.

The beauty of the above equation is that it represents the entire stress state, no matter how
complex it is.

Pg. 60

If 1 and 2 are similar in sign and magnitude, the maximum normal stress theory reasonably
predicts behaviour. The maximum shear stress theory would be conservative but acceptable
from a design stand-point. Meanwhile, the best match with experimental data is provided by
the distortion energy theory.

Brittle Failure Theory

Maximum normal stress. Failure occurs when the ultimate strength, not yield, is reached.

Pg. 61

Coulomb-Mohr theory. Fracture occurs when:

Both 3 and Suc are always negative(compression).


This theory is more frequently applicable to brittle materials because they are stronger in
compression. When compression is dominant (c >> t ), the Mohr criterion is the most reliable
predictor.

Modified Mohr theory. Applicable when 1 is in tension and 2 is in compression.

Pg. 62

Other failure theories

Buckling. Typical conditions where buckling is a concern include a slender (Euler) column under
axial loading, a thin-walled cylinder under external pressure, a thin plate under edge pressure,
and a deep, thin, cantilevered beam under a transverse end load applied at the top surface.

Pg. 67

Fatigue. The corresponding endurance or fatigue limit (Se) is defined as the maximum cyclic
stress which a part can sustain for an infinite number of cycles. Note that for nonferrous
metals and alloys, the strength of the material never stabilizes but keeps decreasing with time.
Hence, these materials do not have an endurance limit.

For ferrous alloys with an ultimate strength below 200ksi, Se is approximately 100ksi. Because
nonferrous metals and alloys lack an endurance limit, a fatigue strength (Sf) is usually reported
for 50(107) cycles of reversed stress. This strength is often as low as Sut for some aluminium
alloys.

Here, ka is a surface factor, kb is a size factor, kc is a load factor, kd is a temperature factor,


and ke is an all encompassing, other miscellaneous effects factor.

The first is the desired number of cycles the part must withstand, which will dictate the value
used for its material strength. The second is the loading history of the part, which will provide
values for the mean and amplitude stress states that the part will experience. Third is the
parameter which always relates the strength and stress in a part-the desired safety factor in its
design. Several methods are available to relate cyclic loading data to fatigue life.
Pg. 70

Some of the more significant factors that affect the endurance life of a part:

Stress concentrators. High local stresses decrease fatigue life.


Surface roughness. Smooth surfaces are more crack resistant because roughness
creates stress concentrations.
Surface conditioning. Hardening process tend to increase fatigue strength while
planting and corrosion protection tend to diminish fatigue strength.
Environment. A corrosive environment greatly reduces fatigue strength. A combination
of corrosive attack and cyclic stresses is called corrosion fatigue.

Creep. Plastic deformation under a sustained load. The amount of creep experienced is a
function of time, temperature and load.

For most metals until operating temperature reaches 35 to 70 percent of respective melting
point.

Pg. 72

Dynamic analysis.

Strictly speaking, such analysis should be referred to as vibration and time response analyses,
because large displacement, completely rigid body motion is not the realm of FEA.

Modal analysis. The building block of all dynamic analyses is the modal analysis, which reports
the natural frequencies and corresponding principal mode shapes.

Pg. 73

The more mass (inertia) that the beam has, the harder it is for the beam to change directions
when fluctuating and consequently, the slower the motion.

Pg. 76

Most mechanical structures are underdamped. In fact, their damping ratio is usually well
below 10%. For a damping ratio of 10%, the difference between wn and wd of about 0.5%.
Hence, for a modal analysis, because the increased complexity of the solution has virtually no
effect on its numerical value, damping is generally not taken into account.

Pg. 79

The number of eigenvector solutions is equal to the number of dynamic degrees of freedom of
the system. An additional, trivial solution occurs when this vector is zero, indicating rigid body
motion of the system.
Pg. 80

Hence, sometimes the goal of an analysis is simply to help design the natural frequencies of a
system away from its known operating frequencies and to note which modes, if any, are
excited as the system ramps up to its operating state.

Frequency Response Analysis. When the excitation does not change with time, the solution is a
steady state response at the operating frequency of interest. This is known as frequency
response analysis. The relevant results of this analysis are typically displacements, velocities,
and accelerations of the system, which can be used to calculate forces and stresses in the
structure.

Pg. 82

Note that for lightly damped systems, as the system approaches resonance, the amplitude
ratio becomes divergent. Because excessive response amplitudes are not desired, you must
either verify the operating frequencies remains away from the resonant frequency of the
system, or add damping to the system. The latter is usually accomplished by means of an
external viscoelastic mount device, also note that as the driving frequency goes to zero, the
amplitude ratio becomes 1 (static solution), yet as it goes to infinity, the ratio becomes 0. This
last observation is an important, albeit surprising, fact: regardless of the driving function`s
amplitude (F0), if the system is shaken fast enough, its response will be zero.

Pg. 83

Note that as the frequency of the input spans the range from zero to infinity, the system
response will go from being completely in phase with the input (0 lag) to being completely out
of phase(180 lag). Always passing through 90 at resonance. Also note that for lightly damped
systems, the response will be (approximately) always either in phase or out of phase with the
input, with a quick phase shift at resonance.

The overall damping of a system is usually the most difficult parameter to obtain. The only
accurate means to do so is experimentally, often from the free vibration decay caused by a
spike input such as a hit with a rubber hammer while making use of Eq. 2.83. Selected
representative damping ratios are provided in table 2.3.
Pg. 84

Yet, Eq. 2.95 turns out to be a good approximation as long as every mode up to a frequency at
least two to three times the highest operating frequency of the system is used.

Depending on the FEA package you use, you might be able to select the method for carrying
out your analysis. If the model is small, or subject to only a few excitation frequencies, the
direct method should be used. This method should also be used if a higher level of accuracy is
deemed necessary. Yet, for a large models or models subject to many excitation frequencies,
the modal method will prove to be the better choice.

Pg. 85

Note that eq. 2.97 is only an approximation if the modal content is incomplete. For this
approximation to be valid, the frequency content of the transient load must be evaluates to
determine a cut-off frequency. This will be the frequency above which no modes are
noticeably excited.

If the model is small, or subject to only a few time steps, or subject to high excitation
frequencies, the direct method should be used. This method should alse be used if a higher
level of accuracy is deemed necessary. Yet, fo large models or models subject to many
excitation time steps, the modal method is the preferred choice.

When this function is applied to a structure, there will be a settling time period before
reaching steady state response. This settling is governed by the transient part of Eq. 2.91.
Hence, if you are interested in the effects observed at the start of steady state operation, a
transient response analysis should be performed.

Keep in mind though, that if transient solution is of no interest, frequency response analysis is
a much more appropriate and efficient solution.

Pg. 86

The excitation in a transient analysis can be a load, acceleration, or displacement of a given


duration with a specified amplitude.

These studies can often be used to correlate damping values and model approximations prior
to a transient response analysis
Pg. 87

At the beginning of the FEA process:

What are the fundamental principles that will govern the analysis? How
can they be qualified and quantified?
When setting up your FEA model:

What kind of stress state will be caused by the geometry and boundary
conditions of the model?
Will there be bending? Torsion?
Can the geometry be simplified or broken into a sum of simpler, more
manageable shapes?
Attempt a prediction of the resulting stress state contour or even some of its numerical
magnitudes. This will prove invaluable when evaluating results.

Next, verify that the boundary conditions make sense. Always draw yourself a free body
diagram and evaluate the equilibrium state.

How confident are you that unexpected and/or undesired reactions are
not being introduced by your representation of the constraints?
Are the loads themselves adding or subtracting stiffness to the structure
by their representation?
Yet, if your calculation is an order of magnitude off, a closer inspection is warranted.

Ch. 3 FEA Capabilities and Limitations

Pg. 90

Regardless of the analysis method used, the accuracy of the solution is always dependent on
the validity of all assumptions.

Will I get the right answer to the question I ask? Better yet: Will I be able to ask the right
questions or enough questions to fully understand the answer?

However, once you learn to qualify the assumptions and unknowns, or the questions you
cannot ask, extracting useful design information out of FEA becomes easy.

Pg. 92

The effect may be small but nonetheless significant. Minor variations in topology may alter
strain gage readings by a sufficient magnitude to call into question the FEA results.

Pg. 93

At a minimum, fillets, chamfers, and other small features may be neglected in the FEM.
Pg. 97

Consequently, for general deformed structures, the minimum energy error can only be as
small as the similarity of the desired shape between any two nodes and the ability of an
element or edge to approximate that shape. Consequently, as the distance between two
nodes decreases, the validity of a linear shape approximation increases and the minimum error
decreases to a point where there is little change in error from making the distance smaller.

Convergence. This is mostly commonly accomplished by increasing or refining the mesh


density, either locally to areas of rapid changes in curvature, or to the entire model.

H-elements versus P-elements Codes which utilize p-element technology typically provide an
internal capability for evaluating convergence and automatically increasing the polynomial
order where required.

Pg. 102

Mesh. The quality of a mesh is best characterized by the convergence of the problem. The
global displacements should converge to a stable value and any other results of interest should
converge locally.

Although a good looking mesh is not necessarily the best mesh, a bad looking mesh almost
always indicates a problem. A good looking mesh should have well shaped elements.
Equilateral triangles and squares are ideal. Transitions between densities should be smooth
and gradual without skinny, distorted elements. When you determine that a mesh is ready to
solve, you are assuming that it will accurately represent the stiffness, or other property of
interest, of the intended structure.

Pg. 104

Geometry Concerns. This is often called stress stiffening ( not to be confused with strain
hardening, a change in material properties). While it is usually straightforward to estimate a
nonlinearity of a system due to material properties, it is difficult, if not impossible, to
determine if geometric stiffening will be affecting the system.

To account for this effect, a nonlinear solution with large displacement option must be run. If
the difference between the linear solution and the nonlinear run is significant in the area of
concern, the nonlinear results should be used.

Pg. 105

Boundary Conditions. Loading must be constant in magnitude, orientation, and distribution.

If a pair of surfaces is not in contact in the unloaded state but come in contact as a load is
applied, a reaction force is introduced at the contact region, thereby changing the load
distribution in the system.

Pg. 106

Static Assumption.

You might also like