[Getdp] Magnetostatic 3D Nonlinear case review

[janvrot at infolink.com.br]

Hello Patrick, Christophe, Johan, everybody

I'm running a Project connected to other projects at PUC-Rio University
that target to automate defect dimensioning in a steel plate (rough
estimate) based on field features. The approach requires hundreds or
even thousands of FE calculations varying defect dimensions,
magnetization level, etc.

My fellows at Physics Department are using a commercial FE software
(OPERA-3d) and MathLab, both available at the Department, in a related
research project with a simple model that I try to reproduce. In those
projects, this arrangement gave satisfactory results for around a
hundred cases for which experimental data is available, and my fellows
are really involved in supporting my project in this environment.

But not always I follow the easy way. I see the open source / free
software phenomenon as a major breakthrough in our way to face work and
cooperation these days. As a Linux user for many years I'm engaged in
doing all my job in this scenario, including software development.

In this context, I found Gmsh and GetDP, and realized that's the way to
go. A lot of automation programming is already completed and now I'm
more involved in validating the model and FE calculations, to run the
cases.

Attached you may find the files to run the problem, so I can eventually
get some extra help to complete my job.

The main topics I'm concerned about are:

FIRST - the geometry

The simple model (1/4 due to simetry) has a plate with a defect in one
side and a magnet in the other. There is a small region of air over and
under the plate. Field is forced to be normal in the magnet end surface
and in the opposing hole end surface. Field is forced to be parallel in
the simetry plane from magnet to defect and at the opposing boundary
surface. No constraints are made to the upper and lower air boundaries.

I was not able to fit an AirInf region to this case with the
corresponding formulation, so I used a big cylindrical section that adds
a lot of elements and probably slows down calculations.

So, is there a better alternative?

SECOND - comparison to experimental data

Flux in the region of interest compares quite well in some tested cases
in maximum values and in the X axis. However, in the Y axis flux
decreases much faster than observed in practice. This is a conceptual
issue, I guess, because the above mentioned tools can reproduce it.

THIRD - noise

As can be seen with gnuplot  results in the region of interest (about 4
mm from the plate) are quite noisy. File command.cmd has adequate lines
to cut&paste and get the figure. Flux at these points are much smaller
than in the plate, so maybe that's the problem. Another point is that
material is strongly nonlinear with the magnetization levels involved.
In the list I found mentions to higher order interpolations, Newton
method and another term in the formulation. I don't know how to deal
with these and some help would be greatly appreciated. 

FOURTH - convergence 

As the number of calculations is huge, any help in speeding up is
welcome. With free software I can always put a couple of machines to
work over the night, but anyway ...
 

Thanks in advance for any help.

-- 
Janvrot IVM
janvrot at infolink.com.br
janvrot at cetuc.puc-rio.br
-------------- next part --------------
A non-text attachment was scrubbed...
Name: case.tgz
Type: application/octet-stream
Size: 6790 bytes
Desc: not available
URL: <http://www.geuz.org/pipermail/getdp/attachments/20021017/f304fdbc/attachment.tgz>