Menu

Discussion Forum

New Discussion
Filter

Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Acoustic-Structure Interaction Time Dependent Computational Cost Reduction Advice

Posted Mar 11, 2024, 2:14 p.m. EDT Structural & Acoustics, Acoustics & Vibrations, Mesh Version 6.0 4 Replies

Allison King
Send Private Message Flag post as spam

Please login with a confirmed email address before reporting spam

Hello! I am looking to model a force excited plate that radiates sound into an air domain where the sound is measured by microphones. I have modeled this using a 3D geometry, time dependent acoustic-structure interaction model. I have measured frequencies up to ~20kHz in the air domain, which requires a fine mesh and fine time stepping to resolve in the model. I have followed the Knowledge Base 1118 recommendations for setting up my mesh and solver settings for the frequencies observed, and I am not able to run the simulation on my computer as I run out of physical RAM and virtual memory. I have attached the model for your reference.

I am looking to reduce the run time and size of this simulation and am seeking assistance/advice on some of the following proposed methods. I have already trimmed the air domain size as much as possible and switched my Acoustic Pressure physics to be a linear discretization to save computational costs. I will not have any symmetry to exploit either, so I think the model needs to stay as 3D as I want to include the vibration reflections at the plate's edge.

  1. Any recommendations for reducting computation cost to keep this a time dependent study?
  2. I have looked into using the Exterior Field Calculation in the air domain, but am a little unsure of how to set that up or if that would work for my problem. Any advice here?
  3. Would you recommend I consider switching over to a frequency dependent problem and trying a Boundary Element Method for the air domain? I figure I can run at every frequency needed, then do a Fourier Superposition and iFFT to get back to the time domain. I was a little confused on how I would adjust the force I apply to the plate for the frequency dependent study as it is a time dependent impulse excitation.
  4. Other suggestions?

Thank you so much in advance! I really appreciate it!

Best, Allison

4 Replies Last Post Mar 20, 2024, 9:31 a.m. EDT
Edgar J. Kaiser Certified Consultant
Send Private Message Flag post as spam

Please login with a confirmed email address before reporting spam

Posted: 8 months ago Mar 11, 2024, 4:17 p.m. EDT

Allison,

this is indeed a heavy one. Even if you had the memory it would probably run for ages.

May be I can provide a few hints for the option 3.

BEM methods can be an advantage in case the respective boundaries are relatively small. Not sure if this is the case in your project, but you can try.

In order to introduce the time dependent excitation into the frequency scan you need to calculate the Fourier transform of the impulse and use it as a window function either in the frequency scan or in the IFFT. It is important that all excitations are close to zero a t = 0 and the final time of the IFFT. So the frequency resolution of the frequency domain scan must be fine enough. Otherwise you will see aliasing in the IFFT.

I successfully implemented this approach some time ago. The motivation then was to run the frequency sweep in many parallel tasks, because frequency sweeps can be parallelized much better than transient studies.

Good luck, Edgar

-------------------
Edgar J. Kaiser
emPhys Physical Technology
www.emphys.com
Allison, this is indeed a heavy one. Even if you had the memory it would probably run for ages. May be I can provide a few hints for the option 3. BEM methods can be an advantage in case the respective boundaries are relatively small. Not sure if this is the case in your project, but you can try. In order to introduce the time dependent excitation into the frequency scan you need to calculate the Fourier transform of the impulse and use it as a window function either in the frequency scan or in the IFFT. It is important that all excitations are close to zero a t = 0 and the final time of the IFFT. So the frequency resolution of the frequency domain scan must be fine enough. Otherwise you will see aliasing in the IFFT. I successfully implemented this approach some time ago. The motivation then was to run the frequency sweep in many parallel tasks, because frequency sweeps can be parallelized much better than transient studies. Good luck, Edgar
Mark Cops COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 8 months ago Mar 12, 2024, 11:44 a.m. EDT
Updated: 8 months ago Mar 12, 2024, 11:44 a.m. EDT

Hi Allison,

This is a very large scale problem. The ratio of geometry scale to wavelength at your high frequency is close to ~50 and the model is full 3D.

The best technique (in terms of memory efficiency) for solving large transient acoustic problems in the Time Explicit physics. See here, for example: https://www.comsol.com/model/submarine-scattering-time-domain-simulation-and-fft-55921. This uses discontinuous galerkin FEM. It is more memory efficient, meaning one can solve larger models for a given amout of RAM - but it doesn't mean the run time will be fast.

It sounds like you are comparing the simulation to measured data from accelerometers and microphones. For simplicity, why not use a steady state noise in the experiment as the excitation? Then switch to the frequency domain study.

The frequency domain study itself is still large. You could consider that for stiff structures vibrating in air, the problem can be modeled as one way coupled - that is first solve for the surface vibration, then the acoustics. The assumption is that the air load is negligable on the vibration of the structure.

As a suggestion, start with solving lower frequency and work your way up, noting the scale of the problem, RAM use, and run times. Even BEM has its limitation at high frequency, since the surface mesh still needs to resolve the wavelength.

-Mark

Hi Allison, This is a very large scale problem. The ratio of geometry scale to wavelength at your high frequency is close to ~50 and the model is full 3D. The best technique (in terms of memory efficiency) for solving large transient acoustic problems in the Time Explicit physics. See here, for example: https://www.comsol.com/model/submarine-scattering-time-domain-simulation-and-fft-55921. This uses discontinuous galerkin FEM. It is more memory efficient, meaning one can solve larger models for a given amout of RAM - but it doesn't mean the run time will be fast. It sounds like you are comparing the simulation to measured data from accelerometers and microphones. For simplicity, why not use a steady state noise in the experiment as the excitation? Then switch to the frequency domain study. The frequency domain study itself is still large. You could consider that for stiff structures vibrating in air, the problem can be modeled as one way coupled - that is first solve for the surface vibration, then the acoustics. The assumption is that the air load is negligable on the vibration of the structure. As a suggestion, start with solving lower frequency and work your way up, noting the scale of the problem, RAM use, and run times. Even BEM has its limitation at high frequency, since the surface mesh still needs to resolve the wavelength. -Mark
Allison King
Send Private Message Flag post as spam

Please login with a confirmed email address before reporting spam

Posted: 7 months ago Mar 19, 2024, 12:03 p.m. EDT
Updated: 7 months ago Mar 19, 2024, 12:03 p.m. EDT

Hi Edgar and Mark,

I really appreciate your insight into this problem. It is definitely a very large scale problem as my measurements contain very high frequencies. It sounds like trying to lower the frequencies in my problem or switching to a steady state excitation on the plate may be best.

Edgar, I am following your advice with running a frequency domain study. I am starting small and just trying to model the plate vibration before I extend this study to the air domain as well. I appreciate your input and it has been very helpful so far.

Mark, I am very interested in checking out the Submarine scattering problem that you linked, but cannot find any files associated with the example. Is there a PDF on it or an mph file that perhaps just needs to be attached to the site you linked? I am open to running the plate vibration as a separate study from the acoustics in the air domain if that will help things be more memory efficient. When you run them together, you have an acoustic-structure boundary. If I run them as two separate steps, how would you link the vibration of the plate to the acoustics of the air domain?

Thank you both!

  • Allison King
Hi Edgar and Mark, I really appreciate your insight into this problem. It is definitely a very large scale problem as my measurements contain very high frequencies. It sounds like trying to lower the frequencies in my problem or switching to a steady state excitation on the plate may be best. Edgar, I am following your advice with running a frequency domain study. I am starting small and just trying to model the plate vibration before I extend this study to the air domain as well. I appreciate your input and it has been very helpful so far. Mark, I am very interested in checking out the Submarine scattering problem that you linked, but cannot find any files associated with the example. Is there a PDF on it or an mph file that perhaps just needs to be attached to the site you linked? I am open to running the plate vibration as a separate study from the acoustics in the air domain if that will help things be more memory efficient. When you run them together, you have an acoustic-structure boundary. If I run them as two separate steps, how would you link the vibration of the plate to the acoustics of the air domain? Thank you both! - Allison King
Mark Cops COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 7 months ago Mar 20, 2024, 9:31 a.m. EDT
Updated: 7 months ago Mar 20, 2024, 9:24 a.m. EDT

Hi Allison, Thanks, we will check to get the files updated for the submarine model. In the meantime you can see other models with similar physics like here: https://www.comsol.com/model/wave-based-time-domain-room-acoustics-with-frequency-dependent-impedance-90551 .

To run acoustic-structure physics as one-way coupled, a clever way to do it is to setup a segregated solver and then set the number of iterations to 1. Then solve it in one study. This approach is done here: https://www.comsol.com/model/electric-motor-noise-permanent-magnet-synchronous-motor-90081 . Alternatively you can have two studies where you solve for the structural acceleration first, then use those acceleration values from the first study as the source condition in an acoustic acceleration boundary condition for the second study.

-Mark

Hi Allison, Thanks, we will check to get the files updated for the submarine model. In the meantime you can see other models with similar physics like here: https://www.comsol.com/model/wave-based-time-domain-room-acoustics-with-frequency-dependent-impedance-90551 . To run acoustic-structure physics as one-way coupled, a clever way to do it is to setup a segregated solver and then set the number of iterations to 1. Then solve it in one study. This approach is done here: https://www.comsol.com/model/electric-motor-noise-permanent-magnet-synchronous-motor-90081 . Alternatively you can have two studies where you solve for the structural acceleration first, then use those acceleration values from the first study as the source condition in an acoustic acceleration boundary condition for the second study. -Mark

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.

Suggested Content