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.
Acoustics - Large models convergence
Posted Aug 26, 2015, 2:32 p.m. EDT Acoustics & Vibrations, Geometry, Mesh, Modeling Tools & Definitions, Parameters, Variables, & Functions, Studies & Solvers Version 5.1 5 Replies
Please login with a confirmed email address before reporting spam
Hi,
I'm driving crazy about this subject. I think don't know many things about this model, but regardless how mesh, solver configuration (multigrid, iterative, direct, etc.) or any other thing I change doesn't matters, it can't reach the convergence.
I'm trying to simulate the noise generated by a solid source that emits noise in all frequencies, like a power transformer, in an electrical distribution substation. I'm using the acoustics module in the frequency domain.
I don't know very well how to set the sources pressure (or the pressure level) with the boundary conditions. I'm using the "pressure" boundary condition setting an arbitrary pressure in order to perform a test.
I'm using an iterative solver with multigrid method (2 or 3 levels), with Krylov preconditioning. Then I use SOR or Jacobi methods.
No matter if set or not a parametric sweep (for frequencies), the solver not converge.
There's any "trick" or something that guides me to reach the convergence in these type of models?
Please, see the attached images in order to understand the described above.
Thanks in advance.
I'm driving crazy about this subject. I think don't know many things about this model, but regardless how mesh, solver configuration (multigrid, iterative, direct, etc.) or any other thing I change doesn't matters, it can't reach the convergence.
I'm trying to simulate the noise generated by a solid source that emits noise in all frequencies, like a power transformer, in an electrical distribution substation. I'm using the acoustics module in the frequency domain.
I don't know very well how to set the sources pressure (or the pressure level) with the boundary conditions. I'm using the "pressure" boundary condition setting an arbitrary pressure in order to perform a test.
I'm using an iterative solver with multigrid method (2 or 3 levels), with Krylov preconditioning. Then I use SOR or Jacobi methods.
No matter if set or not a parametric sweep (for frequencies), the solver not converge.
There's any "trick" or something that guides me to reach the convergence in these type of models?
Please, see the attached images in order to understand the described above.
Thanks in advance.
5 Replies Last Post Apr 3, 2017, 6:50 a.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Any clue about this?
Thanks in advance.
Thanks in advance.
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hernan,
difficult to tell from the screen shots. A general advice is to start with a simpler geometry that can be tested faster and easier. Starting in 2D is also in general a good approach. Develop the physics, the mesh and the solver and go ahead to 3D and more complexity.
Cheers
Edgar
--
Edgar J. Kaiser
emPhys Physical Technology
www.emphys.com
difficult to tell from the screen shots. A general advice is to start with a simpler geometry that can be tested faster and easier. Starting in 2D is also in general a good approach. Develop the physics, the mesh and the solver and go ahead to 3D and more complexity.
Cheers
Edgar
--
Edgar J. Kaiser
emPhys Physical Technology
www.emphys.com
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi Hernan
To me it seems your mesh is way too coarse. The maximal element size should be lambda/6 (for 2nd order elements) here it is only lambda = c0/fi. Also fi is your lowest frequecny - if you intend to solve this up to fs = 2000 Hz, you should define the mesh (wavelength) in terms of this mximal frequecy.
I suggest you have a look at our documentation, more specifically at the Modelling section under Pressure Acoustics in the Acoustics Module User's Manual (version 5.1). Here meshing as well as solver suggestions are discussed.
For iterative solvers you can also have a look at the Bessel Pannel and the Test Bench Car Interior models in the Application Library.
Best regards
Mads
To me it seems your mesh is way too coarse. The maximal element size should be lambda/6 (for 2nd order elements) here it is only lambda = c0/fi. Also fi is your lowest frequecny - if you intend to solve this up to fs = 2000 Hz, you should define the mesh (wavelength) in terms of this mximal frequecy.
I suggest you have a look at our documentation, more specifically at the Modelling section under Pressure Acoustics in the Acoustics Module User's Manual (version 5.1). Here meshing as well as solver suggestions are discussed.
For iterative solvers you can also have a look at the Bessel Pannel and the Test Bench Car Interior models in the Application Library.
Best regards
Mads
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
Hello,
I am trying to solve an acoustic model with a piezoelectric as ultrasound transducer in the frequency domain and measure the pressure in the domain. Since I am in the near field, my mesh size is wave_length/20. This is the point that my result converges using direct solver.
Because 5 MHz frequency is used in the model, the number of meshes are really high. I want to use the iterative solver to solve this model. The setting that I used for iterative solver can be seen in Figure 1 (Attached). In the "Multigrid" node, I used "Geometric multigrid" as a solver and in "Krylov preconditioner", I used FGMRES as a solver.
The problem is it takes a great number of iteration to solve the model (Attached figure) and it takes a long time.
Any suggestion how the iterative solver setting can be set to converge the results sooner?
Thanks,
I am trying to solve an acoustic model with a piezoelectric as ultrasound transducer in the frequency domain and measure the pressure in the domain. Since I am in the near field, my mesh size is wave_length/20. This is the point that my result converges using direct solver.
Because 5 MHz frequency is used in the model, the number of meshes are really high. I want to use the iterative solver to solve this model. The setting that I used for iterative solver can be seen in Figure 1 (Attached). In the "Multigrid" node, I used "Geometric multigrid" as a solver and in "Krylov preconditioner", I used FGMRES as a solver.
The problem is it takes a great number of iteration to solve the model (Attached figure) and it takes a long time.
Any suggestion how the iterative solver setting can be set to converge the results sooner?
Thanks,
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
8 years ago
I cannot remember which solvers we had in 5.1 but you need to use different preconditioners on the different part of your model. I suggest:
* Use GMRES as iterative solver
* Use GMG (default settings) as preconditioner for the acoustic DOFs
* Use a direct preconditioner for the piezo+sold DOFs
You can add several preconditioners by using the Hybridization option and select Multiple preconditioners, then select which DOFs it should be applied to. I hope this already existed in 5.1.
* Use GMRES as iterative solver
* Use GMG (default settings) as preconditioner for the acoustic DOFs
* Use a direct preconditioner for the piezo+sold DOFs
You can add several preconditioners by using the Hybridization option and select Multiple preconditioners, then select which DOFs it should be applied to. I hope this already existed in 5.1.