Solid Mesh Convergence Error

[luca.albertazzi]

Hi,

in the result environment I see there is the possibility to select the Solid mesh convergence error.

If the max error was, for examples 0,03 this would mean that in that point I'll have a 3% of error?

thx

luca

[Andrew.Sartorelli]

Hi @luca.albertazzi,

For Solids

A stress error of 0.012 means a Normal Stress convergence within 0.2%

A stress error of 0.030 means a Normal Stress convergence within 2%

For Shells

A stress error of 0.015 means a Von Mises convergence within 2%

For solids it may be fairly time intensive to get a Normal Stress convergence of 0.2%, so looking for 2% convergence may be better. These numbers should also be applicable to contact problems. I always recommend customers perform a mesh sensitivity/convergence study on their models, this value should quickly give you an idea if you need to go further in those studies.

Regards,

Andrew

______________________________________________________________

If my post answers your question, please click the "Accept as Solution" button. This helps everyone find answers more quickly!

[luca.albertazzi]

Hi Andrew, thx for the answer. Yes, I'm interested in a mesh convergence study. Anyway I don't understand where these numbers come from. Why: - stress error 0.012 implies Normal Stress convergence within 0.2%; - stress error 0.030 implies Normal Stress convergence within 2%? It doesn't seem linearly proportional. I was expected that 0.03 implies 3%. No? thx luca

[luca.albertazzi]

Hi Andrew, thx for the answer.

Yes, I'm interested in a mesh convergence study.

Anyway I don't understand where these numbers come from.

Why:

- stress error 0.012 implies Normal Stress convergence within 0.2%;

- stress error 0.030 implies Normal Stress convergence within 2%?

It doesn't seem linearly proportional. I was expected that 0.03 implies 3%. No?

thx

luca

p.s.: I re-post the answer because it has lost the original format.

[Andrew.Sartorelli]

Hi Luca,

I agree it can be a bit confusing. The stress error is calculating from the following equation:

I'll believe that the correlation of value to convergence was determined by testing a variety of models, but I'll have to dig through my notes on the topic when it came up last.

Regards,

Andrew

______________________________________________________________

If my post answers your question, please click the "Accept as Solution" button. This helps everyone find answers more quickly!

[luca.albertazzi]

Ok now It's clear the reason of the non-linearity.

But if it's like this, how can I know which is the convergence (in %) associated to the stress error that I see in the results?

thx

luca

[luca.albertazzi]

Hi Andrew,

have you some news for me? It would be really appreciated.

thx

Luca

[domas.valiukas]

Hi Andrew,

if I get a solid mesh convergence error of 0,134 MPa? To which percent error it implies? I didn't understand the formula You wrote on Your post.

Could You explain it a little bit?

Thanks.

[John_Holtz · Accepted solution]

Hi Luca,

Personally, I think there is no correlation between the "solid mesh convergence error" and any percent error, but I do not have the theoretical background to understand what the solid mesh convergence error is doing with all of the calculations. I have attached the complete calculation method in case someone has the background to explain it.

In my opinion, the "error" can only be used to indicate that the stress is continuous from one element to the next (and therefore the results have presumably converged) or that the stress is not continuous (and the results have not converged). I wrote "presumably converged" because there are always cases where the error is 0 at a node, yet the results can be inaccurate. (Nodes along a symmetry plane of a full model should have a convergence error of 0 because the results are symmetric, but this does not guarantee that the results are accurate.)

By the way: the mesh convergence error should not have any units, assuming that the attached paper is correct. The results are a stress divided by a stress, so they should be unit-less. This issue has been reported to development.

[luca.albertazzi]

Hi John,

thanks for the explanation, I have appreciated it.

I hope this document will be added as part of the Manual because it is not so easy without a guideline.

Thx again

Luca