Auto pad/feature centre detection

[mharris]

I think detecting the centre of a pad made out of printed ink, or fiducial out of etched copper on the board would be relatively simple to probe for automatically, likewise user drilled vias.

I propose for edge finding of a positive feature to get the user to roughly align the probe with the centre, then:
* Probe the point for a height reference
* Move sufficiently to the side, based on the gerber data to hit bare board, and then get a height reference for the board.
* The probe is advertised as 20um accurate, so either raise up 20um and move inwards towards the pad until the probe activates - or move the probe till it touches the pad and then move in all directions until it "falls off" the pad.
* Find centroid of the pad from the touch or fall off data.

I prefer moving from outside in, as pads can have significant variation in height where the "kiss" is in the centre or when it has been printed over multiple times. The edge seems to be more reliable for height.

Likewise, this could be applied to vias, where you would touch off the board, and then drive towards the via until you "fall in", raise the probe and then move to the other sides and drive inwards. Determine centroid from this data.

This would give much more precise positioning of features on the board - be it fiducials or vias for ink deposition, or pads for paste deposition (especially when you have very small pads getting pasted!)

Even if you are pasting a pre-fabricated board with full ground pour/adjacent tracks, I feel the probe tip is fine enough that you would be able to detect the gap between copper layers even with 0.2mm isolation, assuming 32uM copper. I haven't measured the tip diameter to determine this though.

[troy]

I like this idea for alignment, where there are only 2 points. I'll make a ticket for this.

Automated via filling would be great, and a hole finding algorithm would be critical there. I think the algorithm would need to be smart enough to handle non-circular holes (since we could be dealing with hand drilling).

camera based hole/center finding is another option....i have concerns about handing all the environmental factors we could encounter in the field (reflections, sunlight, etc.).

[mharris]

I dont think non-circular holes would be such a big deal, as long as you know where the centre is you know the needle will be in the hole. Either way you need to approach the hole from at least 4 directions, if you're worried about circularity you could do 6 or 8 approaches to find it more accurately. You'd also be able to see how deep the hole is to determine volume for ink

OpenCV is pretty robust, take a look at what OpenPNP is doing with vision pipelines to get really good results. If ambient light is unreliable, just add LEDS!

[a2retro]

Wondering if the probe location could be shown on the screen in relation to the work surface?

[troy]

Hey @a2retro,

That would be cool, please explain your thinking.

I'd be worried about confusing people during the alignment step. I suspect that some would see the screen and try to move the physical board to match it...also moving the probe during that step would move the on screen probe and the on screen board.