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.
Out of Memory problem
Posted Jun 15, 2017, 3:52 a.m. EDT 3 Replies
Please login with a confirmed email address before reporting spam
Dear.
I am using COMSOL 5.2 to simulate "Acoustical characteristic of a 560m long tunnel". The requirement is the frequency applied should be up to 4kHz while the domain size is too big (560m in length, and appx 4x7 m in cross section). Therefore, because of the huge number of DOFs, the meshing step generates problem of "out of memory" every time I tried to meshing the domain size less than 0.5m (it is recommended the size should be 343/10/4000 = 0.0087m).
My PC is using Core i7 3.4GHz and 256GB of RAM. The amount of RAM is the maximum that can be installed.
So, is there any idea so solve the out of memory problem?
Thanks for help
I am using COMSOL 5.2 to simulate "Acoustical characteristic of a 560m long tunnel". The requirement is the frequency applied should be up to 4kHz while the domain size is too big (560m in length, and appx 4x7 m in cross section). Therefore, because of the huge number of DOFs, the meshing step generates problem of "out of memory" every time I tried to meshing the domain size less than 0.5m (it is recommended the size should be 343/10/4000 = 0.0087m).
My PC is using Core i7 3.4GHz and 256GB of RAM. The amount of RAM is the maximum that can be installed.
So, is there any idea so solve the out of memory problem?
Thanks for help
3 Replies Last Post Jun 25, 2017, 11:06 p.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Some problems are still too big to solve.
Things to try include: getting clever about meshing; invoking symmetries; finding a smaller problem that is close enough to the original.
Maybe if you explain more what question you are trying to answer people can make more specific suggestions. After all, the objective of simulation is not "simulate X" but rather "find attenuation," "find resonant frequency," "find forces," "optimize dimension," or whatever.
D. W. Greve
DWGreve Consulting
Things to try include: getting clever about meshing; invoking symmetries; finding a smaller problem that is close enough to the original.
Maybe if you explain more what question you are trying to answer people can make more specific suggestions. After all, the objective of simulation is not "simulate X" but rather "find attenuation," "find resonant frequency," "find forces," "optimize dimension," or whatever.
D. W. Greve
DWGreve Consulting
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Dear Nguyen
In version 5.3 we have made improvements on several fronts to model large acoustic problems. First of all, there are now predefined iterative solvers that you can enable if your problem is too large. Here you can save a lot of memory.
If you can work in the time domain I suggest you have a look at the convected wave equation interface, if you keep the background flow to be 0, then you basically solve pressure acoustics. The interface uses the DG method and is very memory efficient, it can solve for many million degrees of freedom. Moreover you only need to mesh with lambda/1.5 to lambda/2 so it will help on the mesh side also.
As also suggested make sure to use symmetries, take care of your mes etc.
Best regards
Mads
In version 5.3 we have made improvements on several fronts to model large acoustic problems. First of all, there are now predefined iterative solvers that you can enable if your problem is too large. Here you can save a lot of memory.
If you can work in the time domain I suggest you have a look at the convected wave equation interface, if you keep the background flow to be 0, then you basically solve pressure acoustics. The interface uses the DG method and is very memory efficient, it can solve for many million degrees of freedom. Moreover you only need to mesh with lambda/1.5 to lambda/2 so it will help on the mesh side also.
As also suggested make sure to use symmetries, take care of your mes etc.
Best regards
Mads
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Thank for your help.
I am trying another solution, could be named as "subsectional computation". I divide the whole domain (560m) into 56 sections (10m length each). The acoustic source is located at the first section, and I can obtain the pressure at the boundary surface between 1st and 2nd section. That data set of pressure is used as initial data of 2nd section and find the pressure at the boundary surface between 2nd and 3rd section, and keep going on until the end.
However, the frequency response of combined section is differed from frequency response of union section. I am trying to figure out why.
Is there any similar model using this computation method?
Thank you.
I am trying another solution, could be named as "subsectional computation". I divide the whole domain (560m) into 56 sections (10m length each). The acoustic source is located at the first section, and I can obtain the pressure at the boundary surface between 1st and 2nd section. That data set of pressure is used as initial data of 2nd section and find the pressure at the boundary surface between 2nd and 3rd section, and keep going on until the end.
However, the frequency response of combined section is differed from frequency response of union section. I am trying to figure out why.
Is there any similar model using this computation method?
Thank you.