Meshing fix

Fan Lei · March 2023

Hi, currently I'm using cadfeko to build a rough surface model, it is a brick with a rough surface on the top, and the source is an equivalent spherical mode source (as shown in the figure).

I have many surfaces that are randomly generated by Matlab and imported to cadfeko by .stl file, and then I use feko to mesh it again, using coarse mesh. When I run feko solver, due to the roughness, some of the models fail due to the mesh error, such as " ERROR 3864: Inconsistent specification of the media in connection with the orientation of the normal vectors at connected mesh elements", " ERROR 120: Segmentation rules have been violated (triangle is too large)". Furthermore, because of the model size, some of the simulations also fail because of " ERROR 33467: Not enough memory available for dynamic allocation".

I have seen some of the solutions online, in which they said you can delete the triangle that causes the error. But in my files, there are two problems. First, I'm not quite sure how feko files store the model mesh. I use the university server to mesh the model, and when I transfer the meshed model from the server to my own laptop, the mesh is not there, so I cannot edit it. The second problem is that when I tried to mesh a small model on my laptop, although the meshed model contains many triangles according to the meshing info from the mesh menu, in the Construction/Meshes/UnknownMeshParts_1 there is only one triangle labeled as '0', so I cannot modify the triangles.

My question would be is there any way to solve the mesh error? And for the memory issue, is there any way to reduce its frequency of happening? Thank you.

Torben Voigt · March 2023

Hi @Fan Lei,

Would it be possible to attach a model? Meshing errors sometime are caused by too complex geometry (or mesh) and too coarse mesh settings: I can try to resolve this in the model.

As for the memory error, that really points to missing memory (RAM). Do you want to use MLFMM or which solver?

Best regards,
Torben

Fan Lei · March 2023

Torben Voigt_20420 said:
Hi @Fan Lei,

Would it be possible to attach a model? Meshing errors sometime are caused by too complex geometry (or mesh) and too coarse mesh settings: I can try to resolve this in the model.

As for the memory error, that really points to missing memory (RAM). Do you want to use MLFMM or which solver?

Best regards,
Torben

Thank you for the reply, I tried to attach the model but it is too large to be uploaded, could you check this shared google drive link？ https://drive.google.com/file/d/1QFhN4q0aLciEPwNJJ9NPDDJnXLCOwXmh/view?usp=sharing

For the memory error, currently, I'm using the MLFMM solve and I'm using an HPC so it shouldn't be a solver settings or hardware issue. Sometimes, the required memory for meshing just becomes extremely and strangely large that it can be processed.

Torben Voigt · March 2023

Fan Lei said:
Thank you for the reply, I tried to attach the model but it is too large to be uploaded, could you check this shared google drive link？ https://drive.google.com/file/d/1QFhN4q0aLciEPwNJJ9NPDDJnXLCOwXmh/view?usp=sharing

For the memory error, currently, I'm using the MLFMM solve and I'm using an HPC so it shouldn't be a solver settings or hardware issue. Sometimes, the required memory for meshing just becomes extremely and strangely large that it can be processed.

Thanks @Fan Lei ,

I was able to download the model. There are 7,142,880 dielectric triangle which means the computation with MLFMM will be HUGE.

Now before I start investigating the mesh fix, what are your computational resources (memory / cores)? Does it make sense?

Using an asymptotic solver like RL-GO or PO sems more reasonable but then again the roughness is not large in terms of wavelengths which again will introduce problems...

Best regards,
Torben

Fan Lei · March 2023

Torben Voigt_20420 said:
Thanks @Fan Lei ,

I was able to download the model. There are 7,142,880 dielectric triangle which means the computation with MLFMM will be HUGE.

Now before I start investigating the mesh fix, what are your computational resources (memory / cores)? Does it make sense?

Using an asymptotic solver like RL-GO or PO sems more reasonable but then again the roughness is not large in terms of wavelengths which again will introduce problems...

Best regards,
Torben

I used a memory node of the HPC server, so it has 48 cores and 768GB memories.

I generate a set of rough surfaces randomly but they all have the same statistical characteristic. Some of the models are fine and I can obtain results using MLFMM solver and the node, so I think there is no issue with it. Some of them fail because of the mesh problem, which is reasonable, but some of them just fail because of memory errors, which is strange since 768GB should be quite enough according to the successful models.

I can still have results from some of my models, so I'm just wondering if this memory issue has a quick solution so I can obtain results with more efficiency? Thank you.