you are here:   / News & Insights / Engineering Advantage Blog / The Importance of Validating Your FEA Process – Part 2

Engineering Advantage

The Importance of Validating Your FEA Process – Part 2

Validation of Your FEA Process | FEA Consultant
October 9, 2015 By: Nick Veikos

In a previous post on Validation of FEA processes, I discussed the importance of validation and described a couple of general techniques for validating simulation solutions, namely physical testing and improvement of model fidelity.

After posting, I realized I forgot to mention another useful and important method for Validation: parameter sensitivity studies. If changing the value of a model parameter, say bolt pretension, has a big impact on the results, and this parameter cannot be controlled well in practice, validation may be very difficult to achieve. In this case, a more sophisticated, stochastic, approach will be required, where validation must be considered in an “average” sense.

A perfect example of this comes to mind, again from my early days in engineering. We were experiencing resonance failures of a crimped locking ring on a gas turbine shaft. As a junior engineer, fresh out of school and FEA training, I couldn’t be trusted with anything too complicated, but my manager thought I could handle something simple, like a locking ring, which was essentially a thin cylinder with four equally spaced dents around the circumference, serving to prevent the nut from loosening.

I built my model, with geometry reflecting the drawing dimensions, and the analysis showed that the natural frequencies were outside the anticipated excitation range. So, after a more senior engineer verified that the rookie had not messed up the analysis, things got more interesting. It turned out the process for crimping was not standardized in any way, consisting of a technician whacking the locking ring with a hammer and a blunt tool to form the crimps. The crimp geometry depended on who was making them, and what kind of mood they were in that day. It seemed that no two were alike. After doing further analysis, using measured geometry, it turned out that the crimping details had a big effect on the natural frequencies. Mystery solved! A standardized process was put in place for the crimping, and the problem went away.

Had we performed a sensitivity study on the lock ring geometry up-front, it would have been immediately evident that the crimp geometry variations could create problems for us. At the time, however, sensitivity studies were not common; meshes had to be created by hand, and CPU time was dear. There are no excuses like that today – with parametrically associative geometry and mesh, along with 16 core computers, there is little reason not to perform these validation studies as part of every project.

It will take careful planning to put together a solid Validation process in an organization, and each one will be slightly different based on product, type of simulation, resource availability, and other factors. The relative weighting of testing, model fidelity, and sensitivity studies will also be company, or even project specific. Keeping all this in mind, some practical aspects of a Validation plan include:

  • Make sure the simulation model used for the validation is verified and is not introducing its own numerical errors into the picture.
  • When comparing with test data, ensure that the testing was carried out under the same conditions as the analysis, that the test results are valid, and that the test is representative of the conditions experienced in service.
  • Verify that the finite element numerical assumptions are consistent with the true response. For example, typical shell element assumptions of plane sections remaining plane will not produce acceptable results in shear-dominated problems where warping is significant. When in doubt, build a model with more fidelity and fewer assumptions.
  • Try to ensure that you are capturing all the physics. For example, trying to validate a dynamic event using a static assumption, which neglects inertia and damping forces, will never turn out well.
  • Understand the material behavior and ensure that it is being properly addressed in the simulation. Particular care should be given to elastomers, polymers, and composites. In addition to complex behavior, which is sometimes sensitive to manufacturing processes, these properties tend to have significant scatter.
  • Verify that constraints and loading are modeled properly – these are the most likely sources of error. For example, keeping the direction of a load fixed in space, rather than allowing it to follow the deformation can yield significantly different results (Figure 1 above). Applying a dynamic load statically is another common error, along with neglecting to consider order of load application in path-dependent problems.
  • Check to see if the model omits important features. In an effort to simplify the analysis, features such as fillets, holes, or ribs are ignored. In other cases, only a small portion of the true physical part or assembly is modeled, with boundary conditions imposed at the cut-off. Sometimes these simplifications are too aggressive, resulting in unrealistic outcomes.
  • Understand the sensitivity of parameters on the simulation response, especially those parameters which cannot be well controlled in manufacturing, testing or service. One example is weld bead size; six different welders will likely create six different weld geometries. If the system response is sensitive to this quantity, it may be challenging to validate.

In my opinion, the above list will form the basis for a good Validation program. But there is always room for improvement. Can anyone in help fill in what I have missed?