Page 1 of 1

Inverse heat tranfer Problem applied to a GTA/TIG welding

Posted: 07 Mar 2013, 18:35
by sebastien ROUQUETTE
here i come back.... so i took back my HEAT TRANSFER simulation of stationnary TIG welding case.
I modified slightly the gtaw1.sif file in order to parametrize some simulation's parameters (more easily).
I only have here 2 parameters to estimate: tig efficiency and its gaussian radius (as i modelled the heat flux with a gaussian function).
then i used python27 for writing a file that runs first the gtaw1.sif in order to get reference temperatures at 2 locations.
then i initialize the 2 parameters to different values and i started my inverse/optimization loop in order to get back the right values of tig efficiency and gaussian radius from the reference temperatures.
I used the Levenberg-Marquardt algorithm here to solvemy inverse problem. It requires to run 2 times the gtaw1.sif in order to get the sensitivity coefficients for each parameters to estimate.
So unzip the archive, use python27 (or above i guess) ... normally if you run the python file (if there is no problem of python path or elmersolver path stuffs) it should work. It takes less then 15min on my computer 2.3GHz and 4 processors but under win32bits...
I imagine easily that the optimization solver of Elmer can do this easily. I'm just trying to switch to commercial softwares to Elmer and python so i was only using Comsol(r) and matlab(r). I just started elmer 3weeks ago and python 1week ago... i will be grateful if you have any suggestion in order to simplify and speed it up...
See you soon and feel free to email me or pm me if you have any questions
Cheers

Re: Inverse heat tranfer Problem applied to a GTA/TIG welding

Posted: 29 Apr 2013, 14:49
by sebastien ROUQUETTE
additionnal document to understand the python file