How to manually increase mesh size in high frequency

Yun_20312

I am using FEKO 2019 version and what I do is analyzing RCS of certain aircraft with full wave solver like MLFMM, in 8 GHz frequency.

the frequency is fixed since it is part of research what I'm doing now.

In previous research I used PO for high frequency RCS analysis but found there was a problem in that solver, so I'm using MLFMM now.

but the thing is with MFLMM or MoM, under 8GHz frequency, it takes too long time to complete analyzing RCS and I need even more computing resources.

to handle this problem I tried to increase size of mesh, but I have this error: ERROR 120: Segmentation rules have been violated (triangle is too large)

I can't find solution except lowering frequency to around 1/4 of original, 2 GHz, or less.

but I need to continue using 8 GHz frequency.

I want to ask: Would there be any way to increase mesh size without error for fast speed analysis?

Torben Voigt

Hi Yun,

As you probably know, the simulation time can be reduced by using more parallel CPU cures. I would not recommend increasing the mesh further than lambda/7, this also often means longer runtime with MLFMM as convergence can be degraded.

Feko 2019 is quite outdated. The MLFMM solver in particular has been significantly improved since then, in terms of memory requirements and runtime. You should therefore have fewer problems with Feko 2023.1.2.

Basically, with MLFMM, a new complete simulation is required for each incident plane wave, which is not the case with classical MoM. Therefore, MoM with HOBF (Higher Order Basis Functions) or MoM with CBFM (Characterized Basis Functions Method) can usually be recommended for RCS with many plane waves. At 8 GHz, however, I think that the memory requirement will still be very large.

If I were you, I would compare MLFMM with LE-PO at a frequency (where the runtime is still tolerable) to see how the results differ. I know of many examples where LE-PO and MLFMM only show very small differences in the results.

I would set the mesh like this:

I hope this helps you.

Best regards,
Torben

Yun_20312

Torben Voigt_20420 said:

Hi Yun,

As you probably know, the simulation time can be reduced by using more parallel CPU cures. I would not recommend increasing the mesh further than lambda/7, this also often means longer runtime with MLFMM as convergence can be degraded.

Feko 2019 is quite outdated. The MLFMM solver in particular has been significantly improved since then, in terms of memory requirements and runtime. You should therefore have fewer problems with Feko 2023.1.2.

Basically, with MLFMM, a new complete simulation is required for each incident plane wave, which is not the case with classical MoM. Therefore, MoM with HOBF (Higher Order Basis Functions) or MoM with CBFM (Characterized Basis Functions Method) can usually be recommended for RCS with many plane waves. At 8 GHz, however, I think that the memory requirement will still be very large.

If I were you, I would compare MLFMM with LE-PO at a frequency (where the runtime is still tolerable) to see how the results differ. I know of many examples where LE-PO and MLFMM only show very small differences in the results.

I would set the mesh like this:

I hope this helps you.

Best regards,
Torben

thank you for your help. I have this computer: AMD Threadripper 2950x 16-core process and 128GB ram, I exactly made same input as the schreen shot, but unfortunately the required ram is not sufficient. I think I need to lower the frequency, or use and test LE-PO too.