In this third and final installment of guidelines for getting an initial FEA program contributing to the design department, Ted Fryberger, president and instructor of DeepSoft Inc., Columbia, Md. (www.deepsoft.com), applies lessons from the first two articles.

"There are several methods for getting model geometry into FEA packages," says Fryberger. "One way uses the FEA software's builtin CAD or solidmodeling tools, another uses standalone 2D or 3DCAD software, and a third uses standalone, 3D solid modelers. The choice is largely personal preference although 3D solid modelers are clearly preferred by many including myself". An older approach builds a text file of FEA commands to construct the 2D or 3D geometry. This technique is not available for all FEA programs and requires a strong knowledge of the FEA software's commands. It would also be difficult to model complex 3D geometry with this technique.
Standalone modelers provide the most power and flexibility for creating geometry and parametric modification. Standalone CAD packages may be more familiar to users than FEAsupplied tools. Many just use the modeling functions supplied with FEA software because it avoids having to purchase and learn a second software package. In addition, it eliminates datatranslation errors that crop up between separate CAD and FEA programs.
"Once geometry is in the FEA software, users assign material properties," says Fryberger. For linearelastic structural analysis — the bulk of most analyses — users assign mass density, modulus of elasticity, Poisson's ratio, and shear modulus of elasticity. A thermal analysis normally requires a conductivity coefficient, a thermal expansion coefficient, and specific heat coefficient. Different types of analyses will require different material properties.
As an example, consider an oceanographic instrumentation housing, a device exposed to external hydrostatic pressure. "The problem is to find deflections and resultant stresses as a function of ocean depth," he says. "I built the model with Autodesk Inventor and created the FEA mesh with Algor InCAD. The latter program, a translation utility, reads in 3Dsolid models and generates a surface mesh on the outside. It then generates a solid mesh of elements working from the external surface inward. The utility lets users control external surface and internal solidmesh densities. Users can also specify types of internal elements such as 3Dsolid bricks only, 3Dsolid tetrahedrals only, or a mix of both elements. This housing FEA model uses a mixture.
"The housing shell is machined from 6061T6 aluminum with a transparent Lexan viewport and Oring seal," says Fryberger. "Although the example FEA model has some complexity, it's small enough that I chose not to use symmetry, so the FEA model looks exactly like the product. Inventor creates a parametric model so we don't have to know the exact final dimensions right now, just a good estimate of what they will be. This is an interim FEA analysis. The housing dimensions will change as designers add more detail. The question is: What are the deflections and stresses for this current design? If they are not acceptable, we want to make changes before getting further along in the design cycle.
The enclosure measures 6.03
3.03
2.0 in. with a 0.125in. wall thickness and 0.25in.thick viewport. The maximum design working depth is 200 feetofseawater (fsw) with structural failure allowed at 250 to 300 fsw.
"A double annulus of fixed boundary conditions was applied around the top and bottom of each of the six countersunk screw holes. Boundary conditions (BC) are used to apply nodal constraints, which prevent XYZ translation, or rotation about specific XYZ axes at a specific node, or both. You can constrain translation along or rotation about a single axis, or all three if required. A total of six degrees of freedom (DOF) are available for a single node," says Fryberger. BC's can constrain any combination of them. A node with a fixed BC specifies a node, which is not allowed to translate in XYZ or rotate about the XYZ axes. "BC's have a big impact on the computational results, so it's important to specify them accurately," he adds. BC's are used to simulate the degree of constraint applied to the structure. An analysis frequently includes several FEA runs for different BC scenarios.
Seawater density is 0.037 lbf/in. ^{3 }This results in a gage pressure of 44.58 lbf/in. at a depth of 100 ft. The pressure at 100 fsw is multiplied by a loadcase scale factor of 1, 2, and 3 to simulate the analysis parameters of 100, 200, and 300 fsw in three load cases. This generates three sets of results from essentially a single FEA run, although the solver does have to perform a subset of the computations three times. Interpreting stress analysis results is another topic and beyond the scope of this article.
"Of course, this example could be just one of a dozen or so new users should run through to improve their skills before applying FEA to real problems," says Fryberger. New users should also take FEA training classes before tackling real problems. The academic realm focuses most on structural theory while most FEA training focuses on the user interface of a specific software tool but not on how to interpret a stress analysis. "Software developers assume you know how to do that," he adds.
Preparation for an FEA class that will enhance training would include reviewing a few deflection and stress calculations if necessary, reading the first chapter or two in an FEA book to better understand the overall concept, and most importantly, spending time working with the FEA software. Simulation software is powerful, complex, and sophisticated, and can have a significant learning curve.
"It is not possible to overestimate the importance of analytical engineering experience," says Fryberger. Also, once the model is running don't assume you're done, because you're just getting started. Numerous successful runs are required to tweak the model, develop confidence in it, test the results, and investigate different loading, material, and constraint scenarios. "This is just for linear static analysis. Dynamics and nonlinear analyses are even more complex," he adds.
In the first two articles ...
The first of Fryberger's articles, At square one with FEA, appeared in the Feb 5 issue and covered concepts, preliminary steps, and suggestions for hardware. The second article appeared in the April 1 issue and discussed meshing, loads, and boundary conditions.
FOR FURTHER READING:
 Algor FEA tutorials, www.algor.com
 Nicholas M. Baran, Finite Element Analysis on Microcomputers, McGraw Hill, 1988, 0070336946
 Constantine Spyrakos, Finite Element Modeling in Engineering Practice, West Virginia University Press, 1994, 0964193914
 Constantine Spyrakos and John Raftoyiannis, Linear and Nonlinear Finite Element Analysis in Engineering Practice, Algor Inc., Publishing Div., 1997, 0965280624