Discussion Forum

Hi Allison,

a few observations:

• the 3D mesh is much coarser than the 2D mesh. Are you still sampling the field at 6 elements per wavelength?

• there is no PML in the 3D model, this may explain deviations in the measured amplitudes due to reflections.

Cheers Edgar

Hi Edgar,

Thank you so much for looking over my models! I really appreciate it!

You are correct that I have no PML in the 3D model. I removed it to make sure the PML was not the cause for the early wave arrival times in the air domain. I have since updated my geometry to add the PML back into the 3D model so that the 2D and 3D geometries will match.

I will look closer into the meshing. The 6 elements per wavelength that I used for the 2D model mesh seemed to work really well and give accurate results. I tried making the mesh for the 2D model more coarse, and found similar acoustic pressure trends to the 3D model, so it looks like the coarse mesh may be the cause of my problems. Do you have any tips for designing a 3D mesh that is fine enough to give accurate results, but won't weigh down the run time of the model too much?

Best, Allison

Hi Allison,

yes I know the pain. I am doing a lot of acoustics and there is no way around sufficient meshing. You can try to make the air space as small as possible and often I am using linear discretization instead of the default quadratic. That saves a lot of computational costs. Often the results are accurate enough. In case you can use frequency domain you could try the boundary element method, which avoids having to mesh the air domain, but needs 4th order discretization on the boundaries.

Cheers Edgar

One more idea: exploit any symmetries. I only took a quick look at your file but it gave me the impression of having planes of symmetry. Alas that approach will only work if you are not working your way towards a system with no symmetries - and I am guessing that is where you are headed.

Jeff

Hi Edgar,

That is really helpful advice. I appreciate it. I am a student who just started learning Comsol, so I am not familiar with linear discretization as opposed to quadratic. I will look into that further as any chance of lowering computation costs while maintaining accuracy will be best. I am hoping to stay in the time domain as I want to limit how many plate edge reflections are included in my solution, so I am not sure the frequency domain will work for this particular simulation, but I will keep it in mind for future simulations!

• Allison

Hi Jeff,

I appreciate your advice as well. I am working with a symmetric model to ensure everything is working correctly in the simulation, then will be moving away from a system with symmetries. Unfortunately it doesn't look like I'll be able to exploit any symmetries as I will be moving the source and sensors to no longer by symmetric in any direction. For troubleshooting, I will try and exploit some of my symmetries to make sure my 3D model has a fine enough mesh to give accurate results.

• Allison

Hi Allison,

you have a good strategy. From the simple to the complex and continuous verification. That is how it should be done.

Good luck Edgar