Learning Forum

[Wanqi Zheng]

Hi, I'm trying to run simulation for acoustic plane wave scattering from a solid sphere scatterer. This problem has an analytical solution that I can compare the solid stresses and the acoustic pressure at the cross-section of the sphere. 

I built the model in coupled harmonic analysis system, modelled as a water cube with a sphere in the middle and surrounded with a PML layer. 

I did a test run with 15 EPW mesh last week, and the stresses and acoustic pressure matched closely with the analytical solution. However, when I ran the exact model with the same settings again today, the results were very different and the difference to the analytical solution is more than 20% (vs. 2% that I got last week). 

I am very confused because this should give the same result. I even redo it in the new workbench file, created a brand new model and set up (with the same parameters in the previous file). And again, the result in the new file is different from the previous two attempts. So now I have three set of data for the same model. 

Did any one encounter similar issues before? Can anyone please give me some suggestions on how to cope with this?

 

Thank you very much.

[dlooman]

Sometimes a coupled model is just naturally ill-conditioned.  Are you using distributed processing?  That could lead to different solutions if the model is ill-conditioned.  

[Wanqi Zheng]

 

Hi thanks for your reply.
Yes I am using distributed processing. I made the model as one quarter of the cube and applied symmetry BC at the cutting planes. I applied an incident plane wave coming from the top. Do you think the symmetry conditions can be a problem? Do you think I’ll be better off with other analysis systems, or do I need to somehow refine the conditions in my model? I found the acoustic pressure in the water region is quite consistent, but the stresses in the sphere can vary a lot between runs.

[Wanqi Zheng]

 

 

[ErKo]

 

 

 

 

 

Hello

Please wait for others to chime in and help (it is hard to impossible to know why it is different)


All the best

Erko

 

 

 

 

 

[dlooman]

Shared memory parallel is worth a try.  There is a history of ill-conditioning in acoustic-structural harmonic analyses, so that's probably the issue.