[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: ScanTesla program -> Optimization

Original poster: Terry Fritz <teslalist@xxxxxxxxxxxxxxxxxxxxxxx>

Hi Antonio, Marco, Jim, Malcolm, All,

Marco Wrote:

I have been doing a number of times with problems like this. This is
called "optimization", "linear programming" or "hill-climbing" in a
multidimensional space. You have:

.....

Neat!! Right now I am just trying to get the most simple program to "just work". Not fancy or optimized at all. The methods you mention will be very useful!!! When I get further along, maybe you could help out with some of that stuff. I could really use C-programming examples of anything like this. My problem is dealing with the language and syntax of C. I spend hours looking for the right commands and dealing with real simple C problems. I know what I want to do, but my programming skill level is poor. The only other C program I have dealt with in the last 15 years is E-Tesla, and others did most of it for me ;-))

Thank goodness for the web!! I can just type in say "complex.c" and find the basic info pretty fast at least...

I would take a C class, but I don't have time since I am doing so much C programming now :o)))


Malcolm Wrote: 
Hi Terry,
           I have one difficulty with your model: Rl isn't present
all the time and I don't think an average value cuts it when
calculating Vs. Any comment?

"I" sort of think it is ok. Right now it seems to work for everything. But that could change and there are folks that don't agree. That is why I am trying to make the program very easy to modify to handle things like you mention. Hopefully, the program can be easily changed to work the way anyone wants it to. Steve Conner's models work like you mention too, so that is a "must have" capability.


Antonio Wrote: 
Some comments:
I started to write a simulator in C. Not difficult, but I will not
finish it in the next weeks because I will be away, in a conference.
It's not necessary to change the model of the transformer. Work with
reciprocal inductances (invert the inductance matrix) and the
formulation of state equations or nodal equations becomes simple.
The T model just complicates the simulation.
A capacitor-discharge Tesla coil can be simulated with a step
function instead of the square-wave input. A square wave with zero
frequency.

Wow Cool!! You know far far more than me at this point!! I find I have forgotten most of the mesh equation and complex number stuff over the years. I will press on just to refresh my knowledge there in any case. Jim also strongly suggested this "matrix stuff" too so it is probably a "real good idea" ;-)) As far as I can tell, the C language has all the function built in and ready to play.

The step function and input square wave (Fourier series) should be about 10000X faster to compute than a pure say 10-nS stepwise number grinder. For the linear time-invariant case, that is the way to go. Program like MathCad can also reduce the equations to run faster. But the equation are not super bad.

If Lp, Cp, k, Ls, Cs (Rs and Rp probably are one time functions (of say Ls)) each have say 100 points (linear or log steps too...). That is 10^10 models to run. I think the methods Marco mentions could reduce that by a factor of at least 10^3. So we get 10^7 models. If a computer can do say 100/second. That is 10^5 seconds or say 28 hours!!

So in a ~~day, anyone with a newer PC could scan the entire reasonable Tesla coiling parameter space!!

From there, perhaps slower but more detailed versions could be used to study any interesting details. Perhaps a fast "big picture" program and a slow and versatile "micro detail" program are needed. What ever we find, it does not sound hard to adapt the programs to further the study as needed.

While you are at the conference (Cool!!), maybe we can better define the basic details and do a little testing to be sure things are going in the right direction. I don't think it will be "all done" when you get back :o)))

I use something like this in my Tesla coil and Magnifier optimizers. 

;-)))

What would be the objective? Maximum peak power at the output resistance? Maximum total energy there? Maximum efficiency? Maximum output voltage?

"I think" maximum power to Rl/bang. Maybe max peak power though?? I think efficiency will take care of itself. I think max output voltage will also just be a natural result of maximum power / bang. We won't worry about 6000BPS stuff "yet" ;-))

In the DRSSTC case, MicroSim modeling has shown that there can be high Rl powers at the expense of enormous Lp currents. So maybe we need a "reality check" there. If Lp is like 7 billion amps, we should probably stop ;-)) I think we need to watch Lp and VCp to be sure they do not go into crazy values. This is not such a problem in the normal fixed energy coil case. But we may get a Cp at 1pF at 1 trillion amps and stuff like that...

I have put my little program so far at:

http://hot-streamer.com/temp/scantesla.c

I will try to keep it up to date as I struggle on ;-) It is really just a text file that can be opened with say notepad. I note Lcc does do a few automatic fixes so let me know if something is not compatible with the world (off list). I think it took care of the complex number typedef struct thing... The 0.0 + 0.0i vs 0.0 + 0.0*I...

We may need a little testing to find out some stuff. Like what is Rp in a DRSSTC???? Paul has a ton of secondary info on his site for secondary stuff:

http://www.abelian.demon.co.uk/tssp/


Eventually, we might even have to "really build" a coil that the programs say is the best :o))))

It is an interesting thought that Tesla coiling programs and computer studies might die out. The time may come when they tell us all that we need to know... Perhaps if we want 5 foot arcs, the values of Lp, Cp, k, Ls, Cs.... will just be "known optimal values"...

Cheers,

        Terry