Re: [femm] problem with convergence

[David Meeker <dmeeker@xxxxxxxx>]

giovanni.bortolan@xxxxxxxxxxxxx wrote:
>   Hi
> David,
>
> I have a problem with the attached file: no solution can be found.
> This model tries to solve a highly saturated lamination.
> Note that I have changed the material specifications adding some points
> up
> to B=3.0T (that should be never reached).
> As far as I know the points that I have added should give a smooth
> monotonic function (verified in femmplot and in excel diagram)
> In case the problem was spline interpolation, why not to add a lower degree
> interpolation as an option? I had the same problem with the function "fit"
> in gnucap and order=3 for interpolating functions. The problem disappeared
> with order 1 or 2.
Well, the program tests to see there are any extrema between node points.
 If there are, it smooths the data and tries again.  This forces the curve
to be monotonically increasing (although not exactly going through the originally
defined points).  However, it is still not impossible to give a curve that
it doesn't like.  I've seen this happen sometimes with curves with a strong
"initial permeability" region in which the curve is monotonically increasing,
but in which there is an inflection point.  Anyhow, I picked some different
points on the same curve to make a "new" bh curve.  I've attached the file
in which I did this.  It seems to converge OK now, with the modified curve.
 In the long run, there probably needs to be some "stiff" version of the
Newton iteration that can adjust the step size if things aren't converging.
> Bye
>
> P.S. did you see my message dated 29 January? I have seen no comment about
> it
Sorry--sometimes I get busy with other things, and I'm not very good about
responding.
>  
> 1- In the preprocessor in point insertion mode, it woud be nice that
the
> input line coul accept a formula instead of just a number (this should
be
> easy to do with the lua parser or with a parser generator like yacc
or
> bison)

This would be a nice feature, but it might be nontrivial to implement.  Maybe
this could be done without too much trouble by writing a subclass of CEditBox
or something  that calls Lua to parse the input.

> 2- Another thing in the preprocessor: When you edit the features of
more
> than one entity (selection + space-bar pop-up dialog, modify, save)
all
> selected entities will than have the same features. It is not possible
to
> change only one feature for a group of selected entities. Sometime
it
> causes that you have to select and edit entity by entity. It would
be nice
> that only changed features be saved in the group of selected entities
and
> that features that are not common to all the entities in the group
be not
> shown on the dialog (they should be blank or ghost, but you should
be able
> to force them to a value). This would be the same behavoiur of Word
or
> Excel .

I'd have to think about this a bit.  Sometimes you really do want to change
all the attributes together (although sometimes, you might not).

> 3-Mesher/solver: are you thinking about an adaptive mesh?

Triangle has some built-in capabilities to do mesh refinement, so again,
this would not be impossible to implement.  However, there are some other
things that I had wanted to get around to before mesh adaptation.

Dave.

--
David Meeker

Attachment: zip00027.zip
Description: Zip compressed data