Anisotropic non linear BH curve

[gforti]

Hello all,

Does anyone here have experience with simulation of anisotropic non linear BH curve? I
tried a simple simulation using an arbitrary transformer geometry and the results are unstable.
The same simulation with isotropic material gives good results in comparison with a commercial software.

When I change the step size, I get weird results:

With step = 1/30 * period

comp_30.png (64.57 KiB) Viewed 1692 times

With step = 1/50 * period

comp_50.png (64.43 KiB) Viewed 1692 times

With step = 1/100 * period

comp_100.png (73.49 KiB) Viewed 1692 times

The simulation files are here.
https://github.com/gforti84/ElmerFiles/ ... in/debug_2

[gforti]

Hello fellows,

I have some new and updated case files and results. I got better results using input voltage supply and messing a little with the solvers, also the reluctivity function.
I would like to know if it is possible to improve simulation time for the anisotropic case, it is significantly longer than isotropic one.

Please refer to the link for more details:
https://github.com/gforti84/ElmerFiles/ ... in/debug_3

[mika]

I would like to know if it is possible to improve simulation time for the anisotropic case, it is significantly longer than isotropic one.

It would be interesting to know whether the difference in the simulation times (Elmer vs. other software) seen is the case of anisotropy is due to different behavior of nonlinear iterations (so that more linear system solutions are needed) or worse behavior the linear solver (the count of nonlinear solutions about the same).

[gforti]

Thank you for your response.

Is there anything that I can do? I am not very good to go into the FEM formulation, I am more of a user.

[kevinarden]

I think the question is what is the ElmerSolver doing that takes so long, is it doing linear iterations and converging slowly, or is it doing non-linear iterations, and converging slowly.
if you change your max output level from 5 to 15 you will see more of what elmer is doing as it runs
Max Output Level = 15

Element quality can be an issue, commercial codes generally have increased element matrix development, which is proprietary.
There are some commercial codes that automatically upgrade higher order elements internally in the solver, which can help reduce mesh issues.
Slow convergence can be sometimes traced back to element mesh quality issues.

[gforti]

Thank you. I`ll do some tests with the mesh quality.

As per my solution for the anisotropic BH curve (the reluctivity function). Is there room for improvements?

[raback]

Hi,

My colleague did run this case and indicated that smaller "nonlinear system relaxation factor" (often values 0.5...0.7 are used) could speed up the solution. I guess the maximum number of iterations was used which takes some time.

-Peter

[kevinarden]

The isotropic case has a Linear System Convergence Tolerance of 1.E-8
The anisotropic case has a Linear System Convergence Tolerance of 1.E-9

The solver is spending a considerable amount of time nearly converged trying to get down to 1.E-9 from 1.E-8.

The isotropic case is much faster because it doesn't need these 100 of iterations per time step to achieve 1.E-9.

[gforti]

Thank you for the attention, Kevin. Using 'Linear System Convergence Tolerance' of 1.E-8 for anisotropic case was causing the results oscillation, as illustrated in the figures in a earlier post. This is why I used tighter convergence tolerance.
Also, thanks Peter for the tip. I'll try it.

[gforti]

Hello all, sorry for the delay. I was travelling on vacations.

Tried to increase mesh quality by using 2nd order elements but got the following errors (used ElmerGrid -increase parameter).

First, just with the quadratic element:

Code: Select all

ERROR:: GetEdgeBasis: Can't handle but linear elements, sorry.

Then, with "Quadratic Approximation = True"

Code: Select all

ERROR:: PElementBase::TetraNodalPBasis: Unknown node for tetrahedron

Is there anything missing?

Thanks again.

- forti