LinuxCNC PnP - placement/vision repeatability test

Рет қаралды 4,511

5 жыл бұрын

DIY PnP machine using Mitsubishi MR-J3-B SSCNET III servos under LinuxCNC control.
placing a TQFP144 4 times with 90 deg rotation. a few vision changes to improve speed/accuracy. disabled OpenCV video buffering, recalibrated camera perspective distortion correction, removed delay - vision now completes after getting 3 samples with equal offsets.
machine/PC screen weren't shot at the same time, so some non-sync is present
yurtaev.com

Пікірлер: 20

@keithlegg7681 5 жыл бұрын

amazing!

@spsarolkar 2 жыл бұрын

Great job, what motion controller you are using?

@IceTurf 5 жыл бұрын

neat

@telelaci2 3 жыл бұрын

Probably you came from a CNC mill world or something like that. I've seen similar construction before, and this is not the proper design of a PnP machine, although it can work nicely. Your motors and drivers are all overkill (and I put that very kindly) and your rails and ballscrews are even more. And you solve it by "power". It can work, but you have to fix your very rigid table to a concrete base with huge bolts, (or you have to make it very heavy) if you don't want a continously moving shaking workplace. A PnP machine has very high accelerations and decelerations so you have to design it as light as possible. The most serious forces in a PnP machine are not chipping forces but inertia forces and moments. Just a thought.

@m3rkwurdigliebe 3 жыл бұрын

totally agree. i mostly designed it around leftover components i already had on hand - the base, x and y rails, ballscrew and servos. the flimsy shelf has been bolted to the wall since then :)

@telelaci2 3 жыл бұрын

@@m3rkwurdigliebe OK, I understand. You have the bolts. Maybe it's not a portable machine but works nicely, it impresses people. The power is power.

@trinhdinhhuu5079 4 жыл бұрын

Is it compatible with Openpnp? It is amazingly fast and accurate.

@m3rkwurdigliebe 4 жыл бұрын

my initial plan was to use openpnp, but the problem is it issues g-code moves one--by-one. but to keep the speed up i needed trajectory blending on sharp corners. linuxcnc does that with look-ahead

@AntonChromjak 4 жыл бұрын

@@m3rkwurdigliebe How much % does it take longer without trajectory blending for your average job? How much % longer it takes to setup new board design compared to using openPnp ? I work on corexy pnp machine with very light head and with posibly 90W servomotors so I'm curious if it's something worth digging. You mentioned somewhere that you don't use juki nozzle because they are big for your small machine. Could you explain more? They seem all the same to me. Thanks! And Great JOB!

@m3rkwurdigliebe 4 жыл бұрын

@@AntonChromjak honestly, currently i spend 90% of the time preparing cut tape feeders and 10% actually placing :) so it really doesn't matter, it just hurts me to watch a machine moving without traj blending :) i've spent some time designing motorized feeders but with this lockdown thing they are on hold now.. hope will resume the work on them soon. nozzles - i jus happened to have have a lathe and could turn my own to fit, so i didn't bother buying ones. i saw no need for spring-loaded tips so it was a pretty straight-forward job...

@AntonChromjak 4 жыл бұрын

@@m3rkwurdigliebe I'm designing my pnp around PushPullFeeder. You should check it out.

@coolfox999 5 жыл бұрын

Does it self correct placement? What is calibration routine?

@m3rkwurdigliebe 5 жыл бұрын

perspective calibration was manual by drawing a grid on the image and moving corners doing a perspective transform until the chip on the image fit the grid. the perspective distortion seems to be due to lens to image sensor misalignment in the ELP camera module, as i tried to keep all other things square. no self correction, top camera images are just to check the placement visually

@IceTurf 4 жыл бұрын

Is this openPnP? The vision looks different?

@m3rkwurdigliebe 4 жыл бұрын

no it is not. motion - linuxcnc, vision - a python script using opencv. everything else is written in g-code

@IceTurf 4 жыл бұрын

@@m3rkwurdigliebe Gotcha, I'm surprised you can get away with a leadscrew on only one side - must be very sturdy. How did you manage that? Any word on the parts per hour placement rate? Perhaps thats mentioned in another video and I didn't notice?

@m3rkwurdigliebe 4 жыл бұрын

carriages on the left side translate most of the bending force the gantry experiences into linear motion. far from ideal, but seems to work. those were parts i had on hand. pph - no idea, sorry. without motorized feeders its like an hour preparing cut tapes and 5 minutes placing :)

@IceTurf 4 жыл бұрын

@@m3rkwurdigliebe Haha. Figures that setup would take the most time.