[TCML] NST Measurements

[jimlux]

Godfrey Loudner wrote:
> It's too bad we can't get hold of the sophisticated software companies use
> to design transformers. The ones I saw advertised on the web cost about
> $80,000 or can be rented. They seem to use finite elements to deal with the
> horror of the differential equation involved. The programs can even deal
> with the thermodynamics of the transformers. I'm beginning to believe that
> the details of the mathematical methods are kept secret for financial
> reasons. The software Mathematica keeps some of their methods secret and
> that is a real horror. It would be much fun to enter the geometry of a mot
> or NST and run the calculations in particular cases. 
> 
> Godfrey Loudner  
> 

There's no secret math in finite element programs like Maxwell or HFSS

What you get for your tens of $K is a useful user interface, both for 
data entry and for looking at the results, lots of library support in 
terms of materials and prebuilt models, etc.  You also get good 
interfaces to CAD/CAM software (so you can model the part you're 
actually going to build, without having to re-enter the model)

They also have fairly efficient computational algorithms for doing the 
FEM, rather than brute forcing it with a tiny time step and hoping that 
numerical precision doesn't bite you.

All of this costs real money to write in the first place, and it saves 
engineers time to use it, so they can justify the price they charge for 
it.  If a senior engineer costs about $250k/yr (salary, benefits, office 
space, etc.) and you save 20% of their time, you've just justified an 
analysis program that costs $50k/yr.  Whether or not an engineer spends 
enough time doing analysis (as opposed to adjusting Gantt charts in MS 
Project and doing powerpoint presentations) is another factor, but I 
will say that modern design/analysis tools are wonderful.

There ARE open source/free analysis codes out there, but after using 
them, you'll see why folks are willing to pay the kilobucks, if they're 
doing it for work. The free codes are often limited in terms of model 
size, or require you to write software to describe what it is you are 
modeling (i.e. you get the FEM grid stuff, but YOU have to load the grid 
and constraints), or were done as a senior project/master's thesis, and 
are 90% complete and untested, or, are some toy built in BASIC back in 
the 1980s as demo on a TRS-80, etc.

They won't import IGES or DXF, etc.  They might have bugs. They might 
require doing many runs to understand numerical precision effects (you 
can't just chop things up into a really fine grid and sum them all).