you are here:   / News & Insights / Engineering Advantage Blog / Practical FEA Simulations

Engineering Advantage

Practical FEA Simulations

Practical FEA Simulations | FEA Consulting
May 30, 2017 By: Peter Barrett

In an ideal world, one would have access to all the input data needed to perform all finite element analyses.  Clean CAD models would be provided that include detailed fillets in high stress areas, while simplified geometry is provided in regions that are free of stress gradients. Nonlinear temperature and time dependent material properties would be available to cover all loading conditions needed for the simulation in nice neat tables easily cut and paste into the appropriate FE material models. Loading would be well defined with details on the magnitude, sequence and duration. Accurate bolt loads and sequential tightening schemes would also be nicely packaged in a table. In my 35 years of FEA consulting, I have very rarely been provided with all the input data needed for my simulation, and the data I do get is usually in the form of screen shots, web links, etc. where units conversions, interpolations, extrapolations, followed by engineering judgement are always required to transform the data into a usable form.

In a perfect world, one would always verify one’s FE model with physical testing for difficult to determine input such as friction, creep, etc. Unfortunately, practical FE projects are rarely provided the luxury of this kind of input and test data, due to money, and especially time (creep tests, in particular, can take years to perform).  It is in these situations where the experienced analyst using the limited input data they are dealt, along with their experience, calculates valuable and accurate results, meeting customer timing and budget constraints - using some well-established tricks of the trade. This post provides ten rules to follow for performing practical, meaningful FEA simulations while avoiding the GIGO “Garbage in, Garbage out” syndrome.

1. When defining material properties, vast amounts of material test data are often readily available. There is no reason to perform stress-strain tests for steel, for example. MATWEB and the Mil Handbook (MIL-HDBK-5H) are excellent sources of material data. Often, contacting the material supplier or visiting their website provides access to detailed analysis-ready material properties as well.

2. Parametric studies can often be used to bound variable input, such as friction coefficients, to determine their effect on the behavior.  In many cases, a conservative zero friction assumption can create the highest stress state.

3. It is good practice to assume conservative estimates for most of the input if you don’t have the time to perform a statistical analysis. You don’t design a chair for an average weight person, you design it for the heaviest person that will ever sit in it.

4. Engineering experience provides valuable information on when small features can be ignored, acceptable mesh density, etc. For example, I have performed enough analyses of notches to have a good feel that accurate stresses in a fillet region requires a max element span in the range of 10 to 20 degrees for use in design evaluations.  We don’t have to run a mesh density study on each and every model we analyze. 

5. It is always good practice to take advantage of symmetry when you can. If the geometry is 90% axisymmetric, the loads are axisymmetric and the materials are axisymmetric, an axisymmetric analysis is all that is needed. If the customer is not satisfied, then run comparable linear-elastic 3d vs. 2d analyses to validate the axisymmetric assumption. Limit the refined mesh, nonlinear contact and material nonlinear response to the 2-d model.

6. Start with a static analysis in all simulations, even though, technically, all loads are a function of time.  In general, if the excitation frequency is less than 1/5 of the structure’s lowest natural frequency, a static analysis may be acceptable. There are also some special cases where static analyses can be used as a reasonable approximation, even when significant dynamic loads are present. Figure 1 illustrates dynamic amplification scaling factors based on the ratio of the structure’s fundamental period vs. the duration of the dynamic pulse that can be used in performing equivalent static analyses as a first order approximation to the dynamic response.

7. Connections can be modeled a number of different ways. Contact elements can be an efficient way of transferring load between connecting parts even when they are welded together, since they do not require a flow through mesh. I always recommend starting with linear models that define the connections as either bonded or no-separation contact during the debugging stage, before frictional contact is implemented, if needed.

8. When performing dynamic analysis such as harmonic response simulations, damping is often the most difficult input to obtain. Sensitivity studies can be performed to demonstrate the impact of damping levels on the solution response. For base isolation materials, the use of a single combined spring-damper element at each support can provide local damping while eliminating the local elastomer mode shapes that do not impact the peak metal stresses, and just add to the computational effort.

9. Often, using empirical safety factors is most the efficient means of accounting for those inputs with the most scatter.

10. Although it is always the goal to solve the “real” problem as closely as possible, often more lessons are learned by the analyst that runs a series of simplified models with different input assumptions than from a single “robust” analysis. 

These are just a few examples of practical tips utilized in FEA simulations.  I would love to hear others’ practical engineering assumptions used in their simulations!

Figure 1-  Reference - Clough and Penzien Dynamics of Structures - Displacement Response Amplification Factor for Shock Loading