[Prev][Next][Index][Thread]
MODELING AND SIMULATION
Jim, Malcolm, Mark, Richard, Richard H, and All,
I decided to try for a new thread since I think we need one for
modeling and simulation.
We are fortunate to have with us some excellent theorists and
some dedicated, meticulous experimentalists - all the ingredients
needed for improving our understanding of how TCs work. Perhaps a
separate thread will help bring some additional focus to this
important topic.
Here's an excerpt from some recent correspondence from Malcolm
Watts in which he responds to an earlier post of mine.
PG: Experimental and theoretical work are two sides of the same
coin. For me, the big high in engineering is carrying through a
theoretical development, building the widget, and seeing it work
as predicted. This almost always requires several iterations
refining and reconciling theory and practice, but eventually
things work and you understand why.
MW: Agree totally! If the numbers don't add up, then the model is
missing something IMO. Interestingly enough, the Corum's mention
this point at the start of their analysis and then go on to use
some quite unrealistic assumptions (e.g. the point about
operating at kc is clearly false as their own oscillograms
show). I am doing what I can to help with modelling a spark-
producing system which is what most if not all coilers are
concerned with.
==========
Richard Hull wrote:
[snip]
It will be hard for me to remain positive on this issue ( I have yet to
see one of these *#$%-at- "modeling wonders" do beans towards correctly
predicting any thing close for disruptive discharge systems.)
I feel that this particular modeling package (Pspice) was made up by
people who knew all the equations and had little feel for practical
application. Also, it is crucial to follow their logic in entering the
data or you can skew the results or get completely bogus returns. As I
am not a using the package regularly, I can't review the inputs you sent.
Everything seems to be in order, except the spark resistance. I feel it
is too low. This is the big bugaboo surrounding entering data for Tesla
coils. No one really knows the spark gap's resistance to even an order
of magnitude (sometimes), thus the output could be skewed or off plus or
minus an order of magnitude or two (never a happy though to us design
enegineers). Garbage in, garbage out.
Good luck on getting this "computing marvel" to spit out genuine data
descriptive of reality.
Remember, you said you'd welcome any comments.
==========
Spice and its derivatives do a fine job simulating routine circuits.
I often use Spice as an alternative to breadboarding and find it a real
timesaver. If the job is fairly routine, I don't hesitate to skip
breadboarding altogether.
But there's no substitute for benchwork when one doesn't understand
the problem well enough to model it. That's where we are with TCs
now. After enough iterations between careful experiment and simulation,
the numbers will start to come out right. Then we'll understand TCs
much better and will evolve better and more reliable designs.
Simulation is not a panacea. In competent hands, simulation is an
invaluable and practical tool for learning and design. GIGO? Of course.
But that stands for "Good stuff in, good stuff out" too.
==========
Richard wrote:
[snip]
I am trying to model a Tesla coil using Pspice. To start with, I
thought I'd try and model the lumped circuit as you would see in any
schematic and see what I got. While I seem to be getting vaguely
believable waveforms, the secondary voltage and current magnitudes
seem way off (approx 10kV and 200mA peak respectively).
Has anyone else tried this, or could anyone look at my source code
below and tell me what I might need to do to get more beleivable
results. Is Pspice capable of modeling this accurately?
==========
Lumped parameter models are useful but, as you know, how you model
things often depends on what you're after. Modeling a conventional primary
and secondary as simple tuned circuits with a few elements will allow
you to examine coupled resonant circuits in detail. You should be able
to see the "double peak" frequency response and how it varies with
coupling. A transient analysis should show the classical "coupled
pendulum" response with best energy transfer when the coupling is
such that the frequency response peaks occur at small integer ratios.
If you want to look at the operation of the secondary in some detail,
you will have to use lots of components if you stick with lumped
parameters. For example, each turn on the secondary has self inductance
and is capacitively and inductively coupled to every other turn and to
ground. If you model only the self inductance of each winding plus the
coupling between adjacent turns, you end up with five state variables
per turn or 5,000 state variables for a 1,000 turn secondary.
I'd start with maybe a hundred turns which ought to run pretty fast.
Model the secondary only and drive it with a current source. This
should be plenty to see transmission line effects clearly and to
examine the effects of, for example, top loading and turn spacing.
Put a resistive load between the top hat and ground as a rough
approximation of energy discharge and see how this effects things
like input impedance.
I hope you decide to give it a shot. I'd be *very* interested in the
results and so would others - including Richard Hull who seems to have
a burr in his blanket about Pspice just now. But you can always count
on him for objective, valuable feedback.
==========
Mark S. Rzeszotarski wrote:
[snip]
I have been doing a number of simulations here myself, using both
lumped circuit and helical resonator theory...
==========
I've been doing some simulations too but have just gotten started.
I'm presently concentrating on simulating secondaries and working
mostly with field simulations rather that lumped components.
There has been some good simulation work done in Japan on Normal Mode
Helical Antennas which resemble TC secondaries. Unfortunately, many
of the papers reporting details are only available in Japanese. That
lets me out.
The Japanese work I've seen considers NMHAs with of the order of ten
turns and compares well with measured data even though it neglects
transverse surface currents arising from interwinding capacitance.
Some of this work is reported in English in "Small Antennas" by K. Fujimoto,
et. al., published in the UK by Research Studies Press, ISBN 0 8630 048 3,
and in the US by John Wiley, ISBN 0 471 91413 4.
Conventional moment methods applied to close-wound helical resonators
with of the order of 1000 turns result in numerically ill-conditioned
equations. You can get what seem to be reliable results, but only
at the expense of using 75 to 100 significant digits. Thirty hour runs
tend to break my train of thought. And I have a rule of thumb that says
anytime you need more than 96-bit numbers you need another algorithm.
Right now, I'm looking at some sectional basis functions that may
give better numerical results.
There are claims of some recent work in the UK on long, close-wound
NMHAs using an as yet undisclosed method said to do away with numerical
difficulties. This work is said to show that interwinding displacement
currents become significant with more turns and closer winding. I expect
that is true. Unfortunately, these savants are hoarding information at
the moment. They will presumably publish eventually if they really have
something.
Regards,
Pike