Re: [femm] Metglas square loop BH curves

[David Meeker <dmeeker@xxxxxxxxxxxxxxxxx>]

Dave Squires wrote:

> I just wondered if FEMM is able to handle reliably square loop low
> hysteresis materials with relative permeabilities from 400,000 to 1,000,000?
> I did put in a curve of Metglas 2605SA1 material and it seems mostly OK
> showing saturation about where it is supposed to. I just wondered if there
> anything I should watch out for where FEMM might break down with these
> types of curves.

> I did notice that the plotting feature allows me to identify bad data points.
>
> It seems you must be using a cubic spline interpolation function or at
> least a polynomial one. I noticed that bad data points cause it to give the
> sinx/x type ringing behavior when it blows up the curve.

This is just the sort of thing to watch out for. The program does use cubic
splines to interpolate between data points. Unfortunately, the splines can do
"funny" things around sudden changes, like around the sharp "knee" in the B-H
curve for materials like you are interested in. The fix is to add lots of
points in the neighborhood of the knee, until things look OK when you plot the
BH curve.

I've been meaning to put some consistency-checking to make sure that it doesn't
let you enter in a B-H curve that is non-monotonic, but I haven't gotten around
to it yet--oh, well.


> Sometimes I see FEMM appear to hang in solving if the mesh size is too small.
>
> I double the mesh size and the solution is done in a few seconds.
> Just to let you know.

Every once in a while I have seen this. It often indicates a B-H curve that
isn't well-behaved. For example, since the B-H curve rises so suddenly on a
material with a really high permeability, it is hard to tell whether or not it
is well-behaved in the initial section.

Convergence problems also sometimes occur in materials with a really pronounced
"initial permeability" region, in which permeability starts out low, then
increases, then saturates out. This is one of those annoying "Newton's method
isn't _guaranteed_ to converge" problems. I need to work up some sort of
"brute force" solution for these cases, in which Newton's method determines the
direction in which the solution is modified during each iteration, but the step
length is modified so as to make sure that error is reduced during every
iteration. However, one would only want to use this option in a small subset
of problems, because explicitly evaluating the residuals for the purpose of
choosing a step length is pretty slow.

> Regards,
> Dave Squires

Thanks for the comments.

Dave Meeker



begin:vcard 
n:Meeker;David
tel;fax:781-890-3489
tel;work:781-684-4070
x-mozilla-html:TRUE
url:http://members.aol.com/dcm3c
org:Foster-Miller, Inc.;Electrical and Electronic Systems Group
version:2.1
email;internet:dmeeker@xxxxxxxxxxxxxxxxx
title:Senior Engineer
adr;quoted-printable:;;350 Second Avenue=0D=0A;Waltham;MA;02451-1196;USA
fn:David Meeker, Ph.D.
end:vcard