Accuracy of FEA – a needless complication


I had to write about an interesting discussion I had yesterday with a CAE peer – a discussion on estimating the accuracy of analysis results. Finite Element theory has traditionally preached stress convergence. Instructors and technical staff tell the user to work towards making sure the mesh is fine enough at high stress areas so that the stress values do not change appreciably over successive runs.

To supplement this procedure, COSMOS has gotten ever so easier to use as well. Any trained designer/engineer can pick up meshing tips pretty fast. COSMOS makes it so straight forward, that the user only has to know two options in meshing – creating a mesh, and applying mesh controls.

Amongst all this commotion on accuracy, the user ends up overlooking a fundamental objective in performing analysis – GAINING DESIGN INSIGHT!!  More often than not, the objective of analysis is to understand where the high stresses are in a model, and try to mitigate their effects. The question here is:

Is it that critical to know exactly what the values of high stresses are at that location?

More often than not, you would answer no to this question. Granted that there is going to be a need every now and then to work towards the exact values!! But in most cases, one can bank upon design experience to estimate the necessary additional strength at a high stress location.

The locations of the high stress are not going to change, the values might!! But once the locations are identified, then in most cases, it is just a matter of applying design principles to strengthen these areas. The exact converged value of stress becomes a mute number that only quantifies the already known weak locations.

I see major advantages to considering design insight:

    1. Saves time involved in working towards stress convergence, and worrying about divergence issues.
    2. Saves discussions on what are the most accurate stress results
    3. Gets one away from modeling FEA-focused features (like fillets etc.) to avoid singularities
    4. Gets one thinking about the design than analysis

Using analysis to understand the model may be the thing of the future. Use it to your advantage – if you know a design can fail if the stresses seem to localize at a weak spot, try to focus time and energy on directly strengthening that location.

Hey!! If a dartboard can give you points even if you miss bulls-eye, then why not analysis!!

One comment on “Accuracy of FEA – a needless complication

  1. I agree with you about the physical insights one should have about the problem. I have come across many fresh engineers with knowledge of operating FE software reporting stress values to 4th decimal ‘accuracy’. As a designer I go for the nearest whole number, although I am a theoretical man. In FE formulation there are many types of errors, some of them being: geometrical approxiamtion, polynomial function chosen, interpolant used and even the very mathematical hypothesis used to develop such elements. I am for test/experimental values that are reported with 6th decimal also. FE is only part of the whole design process. For any robust product one should definitely test. The data obtained from test can be used to improve the FE model(or any type of mathematical model) that represents the product. All the talk of stress peaks and singularity will be settled with a test. As the old adage goes: One test will prove /disprove thousand theories. So engineering should go in a balaced manner- analysis and test/experiment. Of course on expects a good fundamental knowledge from the engineer/scientist.

Comments are closed.