Minelab 8000

[Aziz]

Watch out, you might convince Ivica that AI is actually useful!

I have to use yet another AI or a better AI query. Or trying to solve the problem otherwise.
At the moment, the dilate/contract feature isn't solved good. The work-around does it at the moment.

[JoyJo]

Hi JoyJo,



I don't know. They don't contact and ask me.

It is also getting difficult for after market coil manufacturers. The new coils and detectors have a chip built-in (in the coil plug) to prevent after market coils.

No good.

The translator did not translate my message quite correctly: I meant how your diagrams can be useful for those amateurs who make coils for a metal detector themselves.

[Aziz]

Hi JoyJo,

The translator did not translate my message quite correctly: I meant how your diagrams can be useful for those amateurs who make coils for a metal detector themselves.

well, we are comparing different coil configurations against other coil configurations. And trying to understand why and improving it. Or simply testing new coil configurations and look how they perform. Just like the Minelabs new coil.

It is often a question of coil geometry comparison. This is also the reason, why we have a Geometric-Mean-Rule for coils. We have to know and understand the physics first. And the tool I am developing just simulates the physics behind the coils. According to the EM simulations, we get the answers.
Aziz

[rolling rock]

I approve... (uh huh, uh huh).


(Picture from he office of Gavin Newsom, Gov California) "Let them eat nuggets."
Well I do not know how to manage photos. Sorry 'bout that. Hope this won't be censored for "off topic" but that it is. Made me laugh. That's twice in only fifty years, maybe less.

[Aziz]

Hi all,

now lets try with a different AI approach to solve the dilate/contract issue. I am sure, it can solve the issue with ease this time.

[Aziz]

Hi all,

the dilate/contract feature works now perfekt. Even with z-coordinate variation.
The coil bundle is perfect now.

[Aziz]

It is even working correct for an open coil bundle.
Nice.

[Aziz]

What's next?
Upgrade from C64/CGA graphics output into Amiga 500 graphics output?

I have reverse engineered the code. No need to change anything. It's already implemented.

[ivconic]

Aziz, your enthusiasm is commendable and highly respected.
Unfortunately, there are fewer and fewer such people on the forum.
You remind me of me 15 years ago, with the difference that you are much more conversant and have greater formal knowledge than I did.

[Aziz]

Oh man!,

the AI on spline interpolation fails, if there is a collinear path. I had to fix it. And it deforms a rectangular path with round corners too.
But there is a work-around in the geometry description language, which fixes it.

//round rect 20x20 cm (mean), 2 cm radius on corners
ref 0, 0, 0
xyz
0.1    0 0
0.1    0.08
sec path
secspl 10

ref 0.08 0.08 0
pol 0 0.02
dpol 30 0.02
dpol 30 0.02
dpol 30 0.02
ref 0, 0, 0
sec st
xyz
0, 0.1
sec path
secspl 10

sec all
dups -1 1 1

sec all
dups 1 -1 1

sec all
secspl 100

dupd 0.0025
dupd 0.005
dupd -0.0025
dupd -0.005
sec path
dupm 0 0 0.0025
dupm 0 0 0.005
dupm 0 0 0.0075
end

This is the outcome:

[ivconic]

A little advice; Claude is known to work very well with code.
However... DeepSeek with a very clear prompt does the code much better and without errors (moderate code and not "novel").
Most importantly, if the free version is important to you; Claude will leave you stranded at the last moment, DeepSeek will give you much more space.
Claude handles C-oriented languages better (slightly stupid on my part, because they all more or less originated from C) and DeepSeek
is much much much better for fast Windows/Linux applications, especially with a lot of math.
At least that's my experience so far.
But a good, detailed and clear prompt is the mother of all successes.
And if you use the paid version of AI; then it doesn't matter which AI you choose from the two.
The latest versions of ChatGPT pleasantly surprised me, you can see considerable progress.

[Aziz]

Hi ivconic,

A little advice; Claude is known to work very well with code.
However... DeepSeek with a very clear prompt does the code much better and without errors (moderate code and not "novel").
Most importantly, if the free version is important to you; Claude will leave you stranded at the last moment, DeepSeek will give you much more space.
Claude handles C-oriented languages better (slightly stupid on my part, because they all more or less originated from C) and DeepSeek
is much much much better for fast Windows/Linux applications, especially with a lot of math.
At least that's my experience so far.
But a good, detailed and clear prompt is the mother of all successes.
And if you use the paid version of AI; then it doesn't matter which AI you choose from the two.
The latest versions of ChatGPT pleasantly surprised me, you can see considerable progress.

Thanks for your tips.
Claude is very limitted to me. And I have to wake up very early to use it.

Anyway, I have to find out all the "hidden" bugs with every new use case. This will make me busy for a while.
I will see, whenever it isn't working and start a better AI chat next time.

This code is using duplicate rotate (dupr) function instead of duplicate scale (dups). Spline interpolation isn't required. And it is efficient for calculations (less wires).

//round rect 20x20 cm (mean), 2 cm radius on corners
ref 0, 0, 0
​
xyz
0.1 0 0
0.1 0.08
ref 0.08 0.08 0
pol 0 0.02
rep 4
dpol 22.5 0.02
ref 0 0 0
xyz
0, 0.1

sec path
dupr 0 0 90
dupr 0 0 180
dupr 0 0 270

sec path
dupd 0.0025
dupd 0.005
dupd -0.0025
dupd -0.005

sec path
dupm 0 0 0.0025
dupm 0 0 0.005
dupm 0 0 0.0075
end

[Aziz]

Well ok,

3d spline interpolation issue solved now. This was a quick fix by the AI itself.

[Aziz]

What about optimizing features to a wire path to speed up the numerical calculations?
All collinear wire pieces could be compacted into a single wire piece. This would be a nice feature too I think.
Spline interpolation reduction may not be always applicable.

A new command:
sec comp (section compact command)

Every eliminated wire piece is a good wire piece I think.

[Aziz]

Hi all,

compacting collinear wires is also working fine now.
With the help of Google AI. I had to fix it by myself however. The AI couldn't do this on the special case (begin and end points handling).

Look, after making spline interpolation and having lot's of short wires, all collinear (straight) wires are now one long wire.
This feature is good.