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.
basic help with parametric modeling
Posted 16 nov 2010, 12:13 GMT-5 5 Replies
Please login with a confirmed email address before reporting spam
I hope someone can give me a clue.
I have a large model (quarter geometry yields 2.3 million DOF, running on 12Gb 8 core system) that is having trouble converging (at four days and counting). It is a magnetostatics/electrostatics AC/DC module model with high external currents in the superconducting range for my set of coils surrounded by the non-linear 1010 steel yoke material. It is the non-linear bit that creates the problem.
It has been suggested to use solutions from smaller current densities in combination with parametric solver. Current density sources two orders of magnitude lower than my final value solve in about 1/2 hour.
1) It appears this can be done with a continuation parameter in the Stationary solver settings. As described in one of the forum topics, I have multiplied my source terms (external current densities) by a parameter, in this case "Jm". I then presumably ramp this value to unity in some number of steps.
How do I define this parameter in the Stationary study settings dialogue?? If I choose "add" there are no variables to select from the dialogue box that appears and there and no option to type one in. I have tried defining this "Jm" as a Parameter under Global definitions, but I still have no selection choices when I click "add" under the Stationary Study Settings.
I believe I am mission something simple and basic to define this continuation parameter.
I am also using "Initial Values" set to "solution" method from "solver 1" (lower current density solution) in my "Study 2" under "Dependent Variables 1" node. I have left scaling as "automatic", "Variables not solved for" as "initial expression" method and "zero" solution.
2) It would also seem I could approach this with a parametric sweep added to the study, but I don't have any instructions for implementing this option.
Any clues?
I have a large model (quarter geometry yields 2.3 million DOF, running on 12Gb 8 core system) that is having trouble converging (at four days and counting). It is a magnetostatics/electrostatics AC/DC module model with high external currents in the superconducting range for my set of coils surrounded by the non-linear 1010 steel yoke material. It is the non-linear bit that creates the problem.
It has been suggested to use solutions from smaller current densities in combination with parametric solver. Current density sources two orders of magnitude lower than my final value solve in about 1/2 hour.
1) It appears this can be done with a continuation parameter in the Stationary solver settings. As described in one of the forum topics, I have multiplied my source terms (external current densities) by a parameter, in this case "Jm". I then presumably ramp this value to unity in some number of steps.
How do I define this parameter in the Stationary study settings dialogue?? If I choose "add" there are no variables to select from the dialogue box that appears and there and no option to type one in. I have tried defining this "Jm" as a Parameter under Global definitions, but I still have no selection choices when I click "add" under the Stationary Study Settings.
I believe I am mission something simple and basic to define this continuation parameter.
I am also using "Initial Values" set to "solution" method from "solver 1" (lower current density solution) in my "Study 2" under "Dependent Variables 1" node. I have left scaling as "automatic", "Variables not solved for" as "initial expression" method and "zero" solution.
2) It would also seem I could approach this with a parametric sweep added to the study, but I don't have any instructions for implementing this option.
Any clues?
5 Replies Last Post 20 nov 2010, 15:31 GMT-5
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
for many non-linear problems ramping up some variables manually with a parametric sweep helps often on the convergence. To use a "continuation" stationary solver input, you must first define a Global Definitions Parameter with the name you want. Then you can go down in the node tree and "add" it to the continuation list, and give it some range values.
Note I mostly use a stadard range from 0-1 for my parameters and multiply the amplitude by this value, it's easier to find your way, rather than to have a fullrange that changes from case to case, then you can also reuse the same parameter for several variables.
Another point, if you have a non linear material (iron) I expect you have defined a new main material and physics node specifically for your iron: of the type "Ampere Law and Current Conservation" node for (mef)
--
Good luck
Ivar
for many non-linear problems ramping up some variables manually with a parametric sweep helps often on the convergence. To use a "continuation" stationary solver input, you must first define a Global Definitions Parameter with the name you want. Then you can go down in the node tree and "add" it to the continuation list, and give it some range values.
Note I mostly use a stadard range from 0-1 for my parameters and multiply the amplitude by this value, it's easier to find your way, rather than to have a fullrange that changes from case to case, then you can also reuse the same parameter for several variables.
Another point, if you have a non linear material (iron) I expect you have defined a new main material and physics node specifically for your iron: of the type "Ampere Law and Current Conservation" node for (mef)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thanks. I appear to be on the right track.
Yes, I have define a parameter "Jm" that is equal to unity in the global definitions. I have used that same variable as a multiplicative in front of all my external current densities. Although there are three sets of current densities, as you suggested, by using a single variable that ranges from 0-1, I can apply that same variable to all the coils.
I can solve the system when it is two orders of magnitude lower current density so I assume I should apply a range that starts in this neighborhood. That would mean a start value of 0.01. Next I need to define a step. I am uncertain what fidelity I require to produce a decent convergence. Should it be 0.025, 0.05 or something else. Does this generate a "step" parameter in the segregated solver for each step in the continuation parameter? The step values above would be something like either 40, or 20 steps respectively. I also made the mesh finer for the iron materials representing the non-linear portion of the problem. Any other suggestions?
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
the solver is normally set to do automatic stepping, but you can change that to strict or intermediate. With the automatic solver it starts generally gently and then increases the step if the solution looks linear, such that if you have a sudden short oscillation this might be jumped over. Try rather coarse first, and if it does not solve try to make it finer
--
Good luck
Ivar
the solver is normally set to do automatic stepping, but you can change that to strict or intermediate. With the automatic solver it starts generally gently and then increases the step if the solution looks linear, such that if you have a sudden short oscillation this might be jumped over. Try rather coarse first, and if it does not solve try to make it finer
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thanks Ivar,
I couldn't find any setting related to "strict" or "intermediate" in the solver associated with the stepping. There is a "Linearity" pull down under the general group within the Stationary Solver.
The solution was progressing farther and faster than previous attempts but seems to have bogged down at Step value 7 (sitting at 3.05 on the convergence after 1034 iterations). The Segregated group is 33% complete.
Any other nuances to tweak? I sense I could be close to solving this thing.
I couldn't find any setting related to "strict" or "intermediate" in the solver associated with the stepping. There is a "Linearity" pull down under the general group within the Stationary Solver.
The solution was progressing farther and faster than previous attempts but seems to have bogged down at Step value 7 (sitting at 3.05 on the convergence after 1034 iterations). The Segregated group is 33% complete.
Any other nuances to tweak? I sense I could be close to solving this thing.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
my way in these situations is to start reading the docs again (3.5 have a few chapter on solver tweaking) 4 too, but now with 4.1 I need to restart my reading, many new things in there so probably you will find something interesting there too. Then there are the courses of COMSOL, worth to take a few, or to look up the doc tere too, and finally the forum and support
--
Good luck
Ivar
my way in these situations is to start reading the docs again (3.5 have a few chapter on solver tweaking) 4 too, but now with 4.1 I need to restart my reading, many new things in there so probably you will find something interesting there too. Then there are the courses of COMSOL, worth to take a few, or to look up the doc tere too, and finally the forum and support
--
Good luck
Ivar