updated v3.3a3

[David Meeker <dmeeker@xxxxxxxx>]

I've put an updated 3.3 development version on the FEMM website at:

http://femm.berlios.de/femm33bin.exe
http://femm.berlios.de/femm33src.zip

The most important change in this update is a set of new block integral 
used for calculating forces and torques. These new integrals are 
denoted: "Force via Weighted Stress Tensor" and "Torque via Weighted 
Stress Tensor". Instead of having to draw the "right" contour around the 
region of interest as with a the usual Stress Tensor line integral 
approach, the new integral allows one to just select the blocks upon 
which forces are desired and hit "integrate". This formuation converts 
the stress tensor line integral into a block integral in which the 
results of many possible different integration paths are averaged to 
give a result that is more accurate than any one line integral. 
Typically, more accurate force and torque results can be realized at 
lower mesh densities than with the stress tensor line integral.

A paper which describes the force integral approach can be downloaded at:
http://www.esat.kuleuven.ac.be/electa/publications/fulltexts/pub_942.pdf
However, the idea isn't really all that new. Also see:
S. McFee, J. P. Webb, and D. A. Lowther, A tunable volume integration 
formulation for force calculation in finite-element based computational 
magnetostatics, IEEE Transactions on Magnetics, 24(1):439-442, January 
1988.

FEMM finds the weighting function by solving a Laplace equation for a 
potential (that we could call W) over the same set of elements as used 
for the magnetics analysis. The boundaries and non-air objects are fixed 
at a W=0, and all nodes in selected blocks at W=1. Level contours of W 
can then be interpreted as stress tensor integration paths. In the code, 
there are a number of different schemes implemented for weighting the 
individual elements; the one that is actually turned on in the 
self-installing executable version is a scheme in which each element is 
weighted by the mesh size specification for the region in which it lies. 
This tends to make regions with a finer mesh (and therefore more 
accuracy) contribute more to the weighted stress tensor integrations 
than regions with a coarse mesh. At any rate, it typically takes the 
program a second or two to solve for the weighting function (because it 
is actually solving a finite element problem to get the weighting 
function). A new check box has been added to the contour plot dialog 
that turns on the display of the level contours of the weighting function.

I've also implemented a "Depth" parameter for 2D-planar so that one 
isn't constantly multiplying by a scaling factor to get forces, 
inductances, etc, for planar problems. I think that I've propagated the 
Depth parameter to everywhere it needs to go in the postprocessor--tell 
me if you notice any place that I missed. For v3.2 and older 3.3a 
problems, this latest 3.3a3 version assumes a depth of 1 meter, since 
v3.2 and the older 3.3a versions calculated all planar results per meter 
of depth.

The manual has been updated--not all v3.3 stuff (e.g. "nonlinear time 
harmonic" formulation) was documented in the manual before.

There is also a new selection View|BH Curves in the postprocessor that 
lets you plot out the B-H curves that the program actually used. This is 
especially valuable for nonlinear time harmonic problems, where the 
program computes an "effective" B-H curve based on the DC curve, 
hysteresis lag parameter, lamination dimensions, etc.

Dave.

--
David Meeker
dmeeker@xxxxxxxx
http://femm.berlios.de/dmeeker