Scanning with parametrized keywords (parameter sweep)

The graphical user interface of Elmer
Post Reply
CrocoDuck
Posts: 79
Joined: 12 May 2016, 13:15
Antispam: Yes

Scanning with parametrized keywords (parameter sweep)

Post by CrocoDuck »

Hi there!

I wanted to share on the forums a simple way to do scanning simulations (kinda like COMSOL parameter sweeps). I have two reasons for doing that:
  • People have been helping me, so I wanted to "balance the karma" sharing a solution for a problem which I hope will help someone else.
  • There are other threads on this, and also documentation covers the topic. However, solution in threads is not clearly stated and documentation on this topic is not very detailed. For a newbie like me it is not the best situation.
So, I think that just using what I did as an example will make it clear for other newbies.

Let's assume we have an acoustic simulation we solve with the Helmholtz solver. We define the frequency vector containing all the frequencies we want to test in the Header of the sif (Model > Setup > Header > Free Text):

Code: Select all

$ f = 500 1000 2200
We shift now to Simulation and we select Scanning. f is declared, with that syntax, as a MATC vector. It has clearly 3 components, so we set Timestep intervals to 3. Scanning produces a sort of time axis with integer values. So, time will look like this: time = 1, 2,3 ...

Then, when we edit the Helmholtz equation, we write this into Angular Frequency:

Code: Select all

Variable time; Real MATC "2 * pi * f(tx - 1)"
Which means that at each timestep a value of frequency is taken from the vector f and used to calculate the simulation Angular Frequency. You can do the same for boundary conditions. For example, this updates the flux of a constant velocity radiator:

Code: Select all

Boundary Condition 2
  Target Boundaries(1) = 10 
  Name = "Radiator"
  Wave Flux 1 = 0
  Wave Flux 2 = Variable time;Real MATC "2 * pi * f(tx - 1) * 1.205 * v"
End
tx - 1 is used because MATC vectors are 0 indexed: f(0) = 500, f(1) = 1000, f(2) = 2200. tx has the current value of time, which can be 1 ,2 or 3. Hence, you see why we subtract 1.

If you have any solver saving data for you, just set Exec Solver to After Timestep (Equation > Your-Saver-Equation > Edit Solver Settings > General). Here how my Save-My-Stuff setup looks like:

Code: Select all

Equation 2
  Name = "Result Output EQ"
  Active Solvers(1) = 3
End

Solver 3
  Equation = Result Output
  Save Geometry Ids = True
  Output Format = Vtu
  Procedure = "ResultOutputSolve" "ResultOutputSolver"
  Save Linear Elements = True
  Output File Name = case
  Exec Solver = After Timestep
End
This way, the solution after each simulation at a given frequency is saved.

Hope this helps somebody.

Related threads:
viewtopic.php?f=3&t=4239
viewtopic.php?f=3&t=3392

Related documentation:
http://www.nic.funet.fi/pub/sci/physics ... Manual.pdf
http://www.nic.funet.fi/pub/sci/physics ... ection.2.3

P.S.

Ehm... look at this equation:

Code: Select all

Real MATC "2 * pi * f(tx - 1) * 1.205 * v"
1.205 is the density of air, which is a parameter of the material I am using. Is there a way to get it from the material parameters? Couldn't find a way to do it... :oops:
Last edited by CrocoDuck on 22 May 2016, 18:33, edited 3 times in total.
annier
Posts: 1168
Joined: 27 Aug 2013, 13:51
Antispam: Yes

Re: Scaning with parametrized keywords (parameter sweep)

Post by annier »

Hi CrocoDuck
Thank you for your excellent contribution and illustration.

Yours Sincerely
Anil Kunwar
Anil Kunwar
Faculty of Mechanical Engineering, Silesian University of Technology, Gliwice
drone50
Posts: 4
Joined: 09 Jul 2015, 15:58
Antispam: Yes

Re: Scanning with parametrized keywords (parameter sweep)

Post by drone50 »

Dear CrocoDuck,

Thank you very much for your kind and detailed descriptions !!

Cheers,
Kee
Post Reply