Week 02 - system spec
group page // repo source files // objectives
Contents
meta >
preliminary system specification
body >
the danke system is a pnp machine for placing discrete, reversible electronics tiles in 3d space.
in keeping w/ feature parity present in COTS components, many features present in conventional pnp machines (like component feeders) are built into this system design. heavily inspired by machines like the lumenpnp from opulo.
major subsystems as follows:
- motion system
- feeders
- end effector
- build surface
- global error correction
- cv
- in-circuit electrical test jig
there are many directions in which to take the research arc, but the first priority (and primary purpose of this machine) are dedicated towards scaling rapidly and reliably.
eventually, the tiles themselves are intended to be self-correcting, to rapidly and reliably scale whilst relying less on global error correction.
motion system >
motion system is based on Jake Reed’s clank design; the first iteration will be specifically based on the lo-slung variant.
the bed has a modular grid system meant for easy fastening and repositioning of bed objects, such as the in-circuit electrical test jig.
holes have been specified to enable bottom-up computer vision, for recognizing and orienting components.
feeders >
automated feeders are hard systems to get reliably working.
luckily, lumenpnp and adjoining communities have developed both passive and active feeders.
however, an interesting challenge is repackaging DICE tiles for reel form factors.
the progression for feeders are as follows:
- tray-based feeding
- magazine-based feeding
- active reel feeder feeding
end effector >
there are 2 components of this system that need pnp, w/ variations in each category:
- tiles
- connectors
both can be picked in a couple of different ways. the two most popular ideas are mechanically, with a gripper, and pneumatically, w/ a vacuum pump.
initial experiments picking tiles with a gripper show that more iteration is necessary; the current thought is using pneumatics to pick instead.
for the connectors, grippers are still likely needed, given their odd geometry. if the connectors are re-engineered to have a flat surface, then pneumatic picking may work as well.
build surface >
the build surface itself is macro-tile. the current design sports 8x8 tiles, and is intended to separate from the build volume itself as part of the finished part.
the intention is to modularize blocks for a final build; this introduces hierarchical error correction for a much larger system and avoids the problem of “building it all in one go”
error correction >
computer vision >
leveraging existing platforms like openpnp and built-in computer vision capabilities, computer vision from the bottom and the top enable correcting the build process as it goes.
electronic verification >
the build surface itself is electronic and enables in-circuit testing. this is good for final verification after build has been completed. It also enables on-demand functional testing, to verify that design intent matches fabrication.