Meshing Approaches

[isponmo]

Dear All,

This is a rather basic question, but it still confuses me a lot: what is the best meshing approach? Meshing everything in detail, with fillets, etc.? simplifying the geometry and removing small features? Using solid elements for everything? Using shells for thin parts?

I guess the answer depends on the purpose of the analysis. The thing is that for the same kind of analysis done by different people, I have seen rather different approaches.

To put an example, imagine that you have to analyse a piece of equipment (for example, an electronic box) both to assess the dynamic characteristics of the system and to calculate the stresses. The client wants to know if the equipment will survive certain dynamic loading conditions.

In that case, I would say that the best approach would be to mesh everything in detail, with refined mesh in the sensitive areas, and keeping features such as fillets to properly predict the stress levels in stress concentration areas. However, having a look at reports from different analyses made by different people with identical purposes, I have often seen rather simplified geometries, with all the small features suppressed. I think that such mesh would work for, say, estimating the natural frequencies and so on, but how effective is it to assess stresses?

I know it is basic question, but so far I did not find a precise answer... What is your opinion about this topic?

Thank you very much in advance!

Kind regards,

I.

 

[rickfischer51]

The other guys could be doing it wrong, or they could know from experience what is important and what isn't. If they looked at stresses in corners for a couple dozen problems and never saw a problem, and knew there was never a problem with cracked corners in the field, they might have stopped worrying about fillets. Or, they are just lazy, ignorant of basic engineering design and stress analysis, or copying what the other guys did, or maybe their boss told them to do it that way. You should go through the process both ways yourself at least once to know what's important.

Another possibility is that the object being designed is designed to a code, and that the simplified analyses were used to get reactions that were used in the code. For instance, when I do a weld analysis, I do not model the weld. I just take reactions at the weld joint, put them into a spreadsheet, calculate my weld stress and apply an allowable per the code.

Generally, for stiffness related issues you can suppress the small features. For stress related issues, you need to include them. In the case of your dynamically loaded box, you could model it with shells, suppress small features,and optimize the box to deflection issues (remember, modal and deflections are stiffness related). Then remodel with solid elements and look at stresses. Look at clues in the shell models for where the stresses might be high. Often, once deflections are in line, stresses are not a problem (this could be why some of the analyses you've looked at are simplified). If your box is made from sheet metal, a shell model might be completely adequate.

Rick Fischer
Principal Engineer
Argonne National Laboratory

 

[isponmo]

Dear Rick,

Thank you very much for your answer. It is clearer now

Just another question: considering the computational power we have nowadays, is it really worth spending time creating two models, one for stiffness with shells and another with solids for stress? Or is it better to go directly for the solid model? (I am talking about relatively simple models, not models with lots of non-linearities, etc.) Excluding computational savings, is there any advantage in not using solid models?

Again, thank you very much.

Kind regards,

I.

 

[rstupplebeen]

It really just depends on what you are doing. I do a lot of optimization running similar models many times and I will sacrifice on accuracy to get the job done by using axi-symmetry, shells or a coarse mesh. Then do a sensitivity study on the optimized design.

Another way to answer the question is to predict the required preprocessing time and weigh that vs the time savings or accuracy improvements.

I hope this helps even though there are no clear cut answers.

Rob Stupplebeen

http://www.optimaldevice.com

 

[rickfischer51]

Depending on the code and the element choices and your model, consider:

1. The thickness of a shell is a parameter. Very quick to analyze the the effect of a thickness change.
2. Shells usually can be easily post-processed for various stress components (membrane, bending, etc) where solids usually need stress linearization. What does your design code require?
3. Solids may need several elements through the thickness to give accurate stress distribution through the thickness. Again an issue if a code requires stresses averaged though the thickness and the stress distribution is not linear.
4. Layered shells can be used to analyze layered material, like fiber reinforced composite.
5. Fast is always better. When your model doesn't work and you have a deadline and you are trying to debug it and the help line is not returning your calls, you want fast. In my experience, added computational power never is used to increase speed. It is used to increase model complexity to get "a better solution."

Just my 2 cents. Not every model is right for shells, and some are not good candidates for solids. You beed to use experience and judgement.

Rick Fischer
Principal Engineer
Argonne National Laboratory

 

[isponmo]

Dear Rob and Rick,

Thank you very much for your comments. I understand them and find them very valuable.

Kind regards,

I.