Re: acmi accuracy...

["Tesla list" <tesla-at-pupman-dot-com>]

Original poster: "by way of Terry Fritz <twftesla-at-uswest-dot-net>" <paul-at-abelian.demon.co.uk>

Terry Fritz <twftesla-at-uswest-dot-net> wrote:

> I have finally gotten the program and have used it.  Very nice!
> Many thanks for the DOS version for us non-UNIX folks 0:-)

> I see that the error seems to increase when the primary and
> secondary are aligned at the base or when the coupling is greatest.
>  I think there are six possible causes for this:

> 1. Floor too close - In my measurements, the primary was only 4
> inches off the concrete floor. Perhaps this is causing the error.

acmi predictions of Lp are always high by a few percent.  Perhaps that
offset will reduce if the primary is raised.

> 2. Primary to secondary capacitance - Perhaps when the coils are
> close, the primary to secondary capacitive effects are a factor.

Are you at 60Hz or 1kHz for your measurements? Is one end of the
secondary grounded when you make the readings? I'll calculate the
total low frequency capacitance between your primary and secondary
and get back to you with the figure.

> 3. Winding direction - perhaps related to #2, the winding direction
> may tend to add or subtract from the coupling. This may explain why
> Bart and my measurements trend in opposite directions.

Agreed. Perhaps Bart will compare readings obtained when the
connections to one of his windings is reversed. The k trend is
significant in that a basic weakness of current filament summing might
be expected to manifest itself as a systematic trend in error. We have
yet to see that. The only systematic error appearing so far is the
tendency of Lp to be a few percent high.

> 4. Measurement error - Perhaps there is something in the technique
> of taking the measurement that is causing the error.  Also, the
> meters I used at the time were probably 1% accurate.  I now have
> two true RMS meters. One 4 digit and the 6 digit HP34401A. 
> I could probably stretch the accuracy out by two more digits now.

I don't think your 1% instruments are the limiting factor here.

> 5. The program - Coding causing the error but the measurements are
fine...

A possibility. A mistake in the table interpolation code might do it.
The tables values themselves have been proof-read by someone else.

> 6. Unused primary turns - I use 15 of 19 turns.  Perhaps the unused
> turns are having some auto transformer or capacitive effect.

Maybe. Use the full winding for future measurement runs.

> To check into these I could try the following tests:
> 
> 1. Measure the 1000Hz primary to secondary capacitance at the 15
> heights.

I'll calculate for height 0" - ought to the the worst case - and we'll
compare. Will this be for the same pri and sec arrangement as
described in sgap.html and amended by your recent posts, and which
appears to be the same secondary that I already have a model of?

> 2. Raise the coil off the floor by a foot or so and retake the
> measurements.
> 3. Use the better meters and big fixed ceramic resistors for current
> limiting to reduce pure measurement error.
> 4. I can't think of a way to change winding direction.  If you turn
> the secondary upside down it stays the same winding direction. 
> Turning the primary over is awkward at best in my case and I don't
> think I would trust that.
> 5. Try using all the primary turns to eliminate the effects of the
> unused ones.
> 6. ??? Can anyone else think of others or why one of these is not
> possible??

My hypothesis is that the weakness of the current filament
approximation of the thick primary conductor is exposed at close 
coupling, due to the angle subtended by the primary conductor as seen
from a nearby secondary conductor. We might consider averaging the
induction over a number of filaments spanning the cross-section of the
conductor, but the question arises as to how to distribute the current
appropriately, and we might have to deal with circulating current
loops within the primary conductor.

> I have a request:
> It would be nice if acmi could input and output the numbers to many
> more digits, at least while were trying to split hairs here. The
> 0.001 resolution for k seems rather course (wow, aren't we
> picky :-)).  Also, if it could do primary turns to the nearest
> 100th or 1000th, I could put in my 15.125 primary turns without
> worry.  With higher resolution, some of these errors would be easier
> to quantify and "err%" my be much better.  So I would ask that the
> outputs have more digits for the time being.

We'll I'm anxious to avoid the elementary error of reporting more
digits than are significant. I've added another digit to k - there
are now 4 after the decimal point, so that k is representable to 0.1%.
Any more would not be justifiable until reliable agreement to at least
3 digits is demonstrated.

> My free C compiler gave some error (LCC didn't like "strncasecmp")
> or I would do this myself.

Replace the strncasecmp with either strncmpi or stricmpn, whichever
your C library supports.

Cheers,
--
Paul Nicholson,
Manchester, UK.
--