Strain & Stress Postprocessing for unstructured mesh

[Osama Mahmoud]

Hello all,

I have been facing issues with postprocessing strains and stresses for unstructured mesh. so I wanted more clarification on how to do so.

I created an example of a 2D cantilever column subjected to sinusoidal displacement at the top free end. The results for displacements match between a structured and unstructured mesh, but the difference in strains and stresses between the 2 meshes is high (especially stress_xx). A checkerboard pattern is occurring in stress_xx for the unstructured mesh. I would have expected the exact same results, so is there a reason for that or am I defining the postprocessors wrong? I have attached my input files for both cases

Note: I rotated the 2D example I created and the most significant errors seem to shift from stress_xx to stress_yy if that can help in figuring out the reason for that

Thanks,

Osama Abdelaleim

[Chandrakanth Bolisetti]

Hi Osama,

 

Have you tried refining the unstructured mesh in the locations where the checkered pattern occurs and then comparing the values?

 

--

Best regards,

Chandu

--
https://mooseframework.org/mastodon
https://github.com/idaholab/mastodon
---
You received this message because you are subscribed to the Google Groups "mastodon-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mastodon-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mastodon-users/CAFoX9dQTjQVEe%2BT02%3D3JQXPq3vi4uEk6duG7V%3DKbD9DX0COO0g%40mail.gmail.com.

[Osama Mahmoud]

Hi Chandu,

Yes, I have tried refining the mesh and comparing the values and the problem still persists. Again the displacements at a node at one of these elements matches to 10^-5 of the displacement of a node at the same location in the structured mesh. I would think if it is a mesh refinement issue, I would have gotten a different displacement at this node. I have also done some trials with different orders for the stress in the aux kernel as well as finite displacements vs small displacements and this problem still persists in all cases. Here is a diagram of the stress history at the same location between both meshes.  

Kind regards,

Osama Abdelaleim

To view this discussion on the web visit https://groups.google.com/d/msgid/mastodon-users/E4EACC17-50E9-4A36-9F7C-CD6370340189%40inl.gov.

[Swetha Veeraraghavan]

Hi Osama and Chandu, 

I did some basic checks - like volumetric locking correction, mesh refinement etc and the checkerboard pattern still persists. Volumetric locking is the usual suspect for checkerboard pattern, but that din't help here. Maybe Ben or Al would have a better idea of what's happening here. 

Regards,

Swetha

To view this discussion on the web visit https://groups.google.com/d/msgid/mastodon-users/CAFoX9dSzYebCar8adagXeRPCgM9opi8oojKSx8nytFoHXESysQ%40mail.gmail.com.

[Chandrakanth Bolisetti]

Hi Osama,

 

Ben and Swetha looked at this and here’s what they have to say:

 

I think those are just artifacts of the way stresses are averaged to output a single value per element. I tried switching the stress AuxVariables from CONSTANT/MONOMIAL to FIRST/MONOMIAL, and things look much better.

 

Yes, it does look better with FIRST/MONOMIAL (attached image - unstructured mesh in the left) with the aberrations an order less than the maximum stress. Given that stress_yy and the displacements look fine, it may just be stress averaging artifacts. 

 

Please let us know if you have more issues.

 

--

Best regards,

Chandu

 

From: mastodon-users <mastodo...@googlegroups.com> on behalf of Swetha Veeraraghavan <swetha.vee...@gmail.com>
Date: Tuesday, September 15, 2020 at 7:44 AM
To: mastodon-users <mastodo...@googlegroups.com>
Subject: Re: [EXTERNAL] [mastodon-users] Strain & Stress Postprocessing for unstructured mesh

 

Hi Osama and Chandu, 

 

I did some basic checks - like volumetric locking correction, mesh refinement etc and the checkerboard pattern still persists. Volumetric locking is the usual suspect for checkerboard pattern, but that din't help here. Maybe Ben or Al would have a better idea of what's happening here. 

 

Regards,

Swetha

 

On Mon, 14 Sep 2020 at 22:59, 'Osama Mahmoud' via mastodon-users <mastodo...@googlegroups.com> wrote:

Hi Chandu,

 

Yes, I have tried refining the mesh and comparing the values and the problem still persists. Again the displacements at a node at one of these elements matches to 10^-5 of the displacement of a node at the same location in the structured mesh. I would think if it is a mesh refinement issue, I would have gotten a different displacement at this node. I have also done some trials with different orders for the stress in the aux kernel as well as finite displacements vs small displacements and this problem still persists in all cases. Here is a diagram of the stress history at the same location between both meshes.  

 

 

 

Kind regards,

Osama Abdelaleim

 

To view this discussion on the web visit https://groups.google.com/d/msgid/mastodon-users/CAHhC%3De0525E6WA%3DP%2B%2BNb%3DbCARKZp9g7D6m3s3Au0fg3NFf0NNQ%40mail.gmail.com.

[Osama Mahmoud]

Hi Chandu,

Thanks. I realize that it is a postprocessing issue as the displacement field matches well (10^-5 difference), but still the differences between the stresses in structured and unstructured meshes is high even when using higher order approximation for stresses. I still can't see why the difference in the stresses is more than one order of magnitude higher if the displacements match well. I am not sure if it is a mapping issue in the postprocessor, but I guess I will stick to structured meshes then. I have attached the results for the stress history for the same location in the 2 meshes using linear approximation for stress.

Regards,

Osama Abdelaleim

To view this discussion on the web visit https://groups.google.com/d/msgid/mastodon-users/89CFB3F5-D524-45D6-8629-63A27392141A%40inl.gov.

[Chandrakanth Bolisetti]

Hi Osama,

 

My apologies – I thought the displacements weren’t matching either. Stresses and strains are very local responses and can be quite sensitive to the mesh you use. Comparison of stresses makes more sense if it is averaged across a cross-section, or a certain area, etc. As seen in the plots from the previous email, the unstructured mesh still has a slight checkerboard pattern, which is likely a result of the element shapes, etc., and the stresses are relatively not uniform. I am not surprised that the stresses at one single point are quite different. If you measure the averaged stress or strain in a part of the beam, they should be close when the mesh is fine enough.

 

--

Best regards,

Chandu

 

From: Osama Mahmoud <osama_...@aucegypt.edu>
Date: Wednesday, September 16, 2020 at 10:50 AM
To: Chandu Bolisetti <Chandrakant...@inl.gov>
Cc: mastodon-users <mastodo...@googlegroups.com>
Subject: Re: [EXTERNAL] [mastodon-users] Strain & Stress Postprocessing for unstructured mesh

 

Hi Chandu,

 

Thanks. I realize that it is a postprocessing issue as the displacement field matches well (10^-5 difference), but still the differences between the stresses in structured and unstructured meshes is high even when using higher order approximation for stresses. I still can't see why the difference in the stresses is more than one order of magnitude higher if the displacements match well. I am not sure if it is a mapping issue in the postprocessor, but I guess I will stick to structured meshes then. I have attached the results for the stress history for the same location in the 2 meshes using linear approximation for stress.

 

 

Regards,

Osama Abdelaleim