A pure-mechanical, gantry-mounted "ladle / trough" for picking up loose powder from a bed and depositing it at a target location. The trough is suspended between two vertical arms that grip its two short end caps; a single longitudinal pivot pin runs through both end caps along the trough's long axis L. There are no actuators on the bucket itself — the gantry's existing X / Z motion plus a fixed wall-/post-mounted smooth inclined cam ramp does all the work.
Geometry note (post-review). An earlier revision of this design used a transverse pivot pin (running across the trough's width D) and tipped the trough end-over-end against a sawtooth ledge. Both choices were retired after a high-effort review by Edison Scientific identified (i) a kinematic impossibility — a fixed sawtooth tooth cannot engage a fixed-radius hook on a horizontally-translating pivot without the lip plunging through the tooth — and (ii) a severe trapped- volume / arching problem when cohesive powders are forced to bottleneck through the narrow ~13 mm spill point of an end-over-end half-cylinder. The current design uses a longitudinal pivot pin and sideways tilt so powder pours over the full 80 mm long edge, and replaces the sawtooth with a smooth cam ramp that the trough's chamfered bumper slides up. A follow-on hand sketch (
PXL_20260423_231729467.jpg) refines the cam ramp into a pin-defined-path (peg-in-routed-slot) variant in which a peg on a stem hanging from the gantry is captured in a slot routed into a fixed external board, so the slot path deterministically programs the trough's tilt schedule. The full review and the verbatim Edison answers live indocs/edison/; the literature-aware brainstorming doc has been ported to a LaTeX manuscript with a proper BibTeX bibliography (seedocs/manuscript/main.texanddocs/manuscript/refs.bib; build withcd docs/manuscript && make).
Candidate parts for the auger drive motor, vibration motor, external
solenoid, and the Raspberry Pi Zero 2 W driver boards needed to
control them are documented in
hardware/vibration-motor-and-solenoid.md.
The sketch above has been recreated as four labelled subpanels. Each is a
self-contained SVG with its own caption. All four are produced (along with
the mechanism animation below) by a single reproducible script,
scripts/generate_figures.py.
A follow-on hand sketch
(PXL_20260423_231729467.jpg) refines the
smooth cam ramp into a peg-in-routed-slot mechanism. A vertical stem hangs
from the gantry carriage on a pin pivot; a transverse peg at the top of the
stem rides in a slot routed into a fixed external board. The slot's shape
over X deterministically programs the trough's tilt schedule, with the peg
captive in the slot for the entire stroke (no approach-and-contact
ambiguity, and bidirectional return without relying on gravity). See
docs/manuscript/main.tex §"Pin-defined-path
actuation" for the full trade-off discussion.
The full literature-aware design discussion — framed roughly as the
introduction to a Digital Discovery manuscript on a new powder
dispenser, with references to recent SDL / powder-handling literature —
is now a LaTeX manuscript with a proper BibTeX bibliography:
docs/manuscript/main.tex,
docs/manuscript/refs.bib. Build with
cd docs/manuscript && make (requires pdflatex, bibtex, and
cairosvg).
Parametric CAD pipeline (cad/)
The same dimensions used in the figures and the manuscript also drive an
open, scriptable parametric CAD model built on
CadQuery, so the design can be
iterated as plain Python source and exported to STEP / STL for any
open-source slicer. A small cad/dfm.py runs design-for-manufacturing and
gantry-only kinematics checks (the explicit constraint from PR comment
4166621470 — "we have a gantry system and would like to avoid installing a
second axis"), so a parameter regression is caught the next time anyone
edits ExcavatorParams. See cad/README.md for the
rationale (why CadQuery over Rhino / Grasshopper / Fusion / nTop), the
target-hardware-matched toolchain for the Genmitsu 3018-Pro V2
(GRBL CNC) and Prusa MK3 / Ender (FDM) printers we plan to use, and
the full feedback-loop description.
pip install cadquery
python -m cad.build # writes STEP + STL + manifest into cad/build/
python -m cad.dfm # design-for-manufacturing + gantry-only kinematics checks
python -m unittest discover cad/tests -vThe bucket is an elongated half-cylinder trough (think: a long, narrow ladle with a semicircular cross-section and an open top). It is suspended between two parallel vertical arms that hang from the gantry carriage and grip the trough on its two short end caps. A single horizontal pivot pin runs from one arm, through a printed boss in the first end cap, lengthwise along L through the trough, out through a matching boss in the second end cap, and into the second arm (panels B and C). The pivot axis is therefore parallel to the trough's long axis L — the trough rolls about that axis like a spit-roast rotating about its skewer.
- The two arms are rigidly bolted to the carriage and always stay vertical during operation.
- The trough is the only part that ever rotates, and only about the longitudinal pin — rolling sideways when its rim-mounted bumper catches the cam ramp. Powder pours over the full 80 mm long edge, not through a narrow end-cap point.
- The pin sits a few millimetres above the loaded trough's centre of mass, so gravity returns the trough to "open-up" once the dump cam is cleared (a stable pendulum / gondola-style pivot).
Picking shape matters:
- A half-cylinder maximises retained volume per unit "scoop depth" while presenting a long flat top edge that pours uniformly along its whole length.
- An elongated trough (length L ≈ 3 × diameter D) gives the spill edge a large area (≈ L × few-mm depth) so cohesive powders cannot arch across it, eliminating the bottleneck/bridging failure mode of an end-over-end tilt of the same trough (Edison v2 §3).
- No moving lid / hinge in the baseline design — powder is held in by gravity alone, which keeps the part count and the failure modes to a minimum.
- J-curve plunge into the bed. The gantry lowers the carriage (Z↓) so the trough enters the powder bed, then translates a few millimetres in X while continuing down — a shallow J-curve rather than a pure Z-plunge. A blunt flat half-cylinder driven straight down compresses the bed underneath it instead of flowing in (Edison v1 §2); the J-curve lets powder spill into the open top as the bucket sweeps forward.
- Lift past the strike-off bar. The carriage rises (Z↑) along an
X-coordinate that takes the rim of the trough under a fixed bed-
edge strike-off bar. The bar wipes the heaped powder back into the
bed and leaves a defined fill volume. Without this step, dose CV
sits in the 15–30 % range across powder classes (Edison v2 §4); with
it, ≈10 % CV is realistic on cohesive inorganics — comparable to the
positive-displacement-pipette baseline reported by Alsenz (see
docs/manuscript/main.tex). - Transport to the deposit X. The carriage translates in X over to the deposit station. Arms still vertical; trough still open-up, level under gravity.
- Sideways tilt against the cam → deposit. The gantry pushes the trough in X into a fixed, wall-/post-mounted smooth inclined cam ramp at end-cap-rim height. The chamfered bumper on the trough's long-side rim slides up the cam's hypotenuse; because the cam surface is a continuous incline rather than a fixed point, pure-X gantry travel is geometrically compatible with the resulting roll of the trough about its longitudinal pin (Edison v2 §1). Continued X-push rolls the trough sideways and powder pours over the full long edge. Backing the carriage off lets gravity right the trough under its pendulum action.
The "push against a cam to dump" trick is what makes this fully mechanical — no servo / solenoid is needed on the bucket itself.
A fixed, wall- or post-mounted inclined block at roughly trough- end-cap-rim height (see Panel D, Step 4, and the GIF above). Its top surface is a smooth ramp (a chamfered, ideally polished hypotenuse), not a comb of teeth. It is not on the floor of the powder bed and is not part of the moving assembly. The trough's rim bumper rides up the ramp as the gantry pushes the carriage in X; the height the bumper reaches up the ramp (set by how far in X the gantry pushes) determines the tilt angle and therefore how much powder is poured.
Why a smooth ramp and not a sawtooth?
- Kinematic compatibility. A fixed sawtooth tooth cannot engage a fixed-radius hook on a horizontally-translating pivot pin without the lip having to plunge through the tooth — the pin-to-tooth distance must change as the pivot moves in X, but the hook's radius is fixed. An inclined cam removes the constraint entirely: the contact point slides continuously along the cam surface as the gantry travels (Edison v2 §1).
- Continuous tilt control. With a smooth ramp, the tilt angle is a continuous function of gantry X position, so pour rate and final dose can be modulated in software without changing hardware.
- No catch-and-skip failure mode. A printed sawtooth tooth and a printed hook can mis-engage, snag, or skip; a chamfered bumper on a smooth incline cannot.
Because the bumper is in continuous frictional contact with the cam during the tilt, the gantry can be driven to rapidly oscillate ±2 mm in X while the bumper is engaged. This momentarily slams the trough against the cam and acts as a free pneumatic-knocker analogue to break up bridges of cohesive powder that would otherwise refuse to pour (Edison v1 §6). It is a software-only feature — the same hardware that performs the tilt also performs the knock.
- Manufacturing. Target is a 3D-printable trough + arms + end- cap pivot bosses (PETG / nylon for the prototype, with a glued-in brass sleeve for the pin holes if wear becomes an issue). The pivot pin itself is a stock dowel pin or shoulder bolt, ideally metal so it can also serve as the ground path for an optional conductive trough lining (e.g. interior copper tape) that mitigates triboelectric charging on fine dry inorganics (Edison v1 §2). The cam ramp can be printed too. A machined-aluminium revision would only be needed if the printed parts wear or charge problematically.
- Target powders are dozens-of-microns in diameter — catalysts, ceramics, salts. Many are cohesive, hygroscopic, and/or triboelectrically charged; some clump and resist removal from a scoop. This drives several of the open questions below.
- Trough geometry sweep. Pure semicircle vs. a slightly deeper "U" vs. a V-bottom — which retains powder best while still pouring cleanly when rolled? A 3D-printed bake-off across our worst-case cohesive powders is cheap.
- Cam-ramp angle sweep. Cam slope, length, and surface finish are the three knobs most likely to dominate dose CV. Steeper slopes give more aggressive tilts per unit gantry travel but risk bumper slip; shallower slopes give finer control but a longer dump stroke.
- Strike-off bar profile. Square cross-section vs. a chamfered or rounded leading edge — which leaves the cleanest fill at the rim without flicking powder away? Worth varying.
- Bumper profile. Chamfer angle and surface finish on the rim bumper directly affect cam engagement; too sharp and it digs in, too rounded and it slips. Print + sand a small set and characterise.
- Bed depletion. Repeated scoops from the same X, Y in a static bed will form a crater and the dose will decay (Edison v1 §3). The gantry must raster the plunge X, Y across the bed, or a separate "bed-stir" cycle must be added.
- Cleaning / cross-contamination. For multi-material campaigns the pivot pin is the natural quick-release point — pull the pin, swap the trough, re-insert. For very sticky powders a per-material consumable trough (snap-in printed liner) may be cheaper than cleaning.
- Use-environment caveat. A bulk-transfer scoop unavoidably exposes the powder to ambient air during transport and leaves an open crater in the stock bed. For hygroscopic salts this is a real limitation; the device's intended envelope is bulk transfer in ambient or globally-controlled environments (e.g. inside a glovebox or on a benchtop with a desiccated stock container), not as a direct loader for highly moisture-sensitive workflows.
- Reversible tilt — cam ramps on both sides of the work area let the same bucket dump left or right by which way it's pushed.
- Two troughs back-to-back — one fills while the other is being dumped, doubling throughput with no extra actuators.
- Auger / screw inside the trough — adds one rotary actuator but gives controlled metered dosing instead of "tilt and pour all".
- Passive flap lid for fine powders. A lightweight flap hinged on the trough's long upper edge that gravity holds closed over the mouth while the trough hangs open-up (so fluffy powder is not shed during X-travel). When the trough rolls against the cam ramp in step 4, a small projection on the flap strikes a separate fixed tang on the same cam assembly, swinging the flap clear so powder can pour out. Returns to closed under gravity once the trough returns upright. Still purely mechanical — no actuator on the bucket.
The design changes between PR #2 and the present revision were prompted
by a high-effort design review from Edison Scientific,
plus a literature search via the same service. Verbatim responses are
archived under docs/edison/ for traceability.
Presentation at POSE 2026
A written brainstorming pass over candidate dispenser architectures —
parallel-channel, cartridge carousel, X-Y gantry, vibratory, pneumatic,
and shared-auger-with-purge — with the benefits and pitfalls of each
under the autonomous-loop / 30-powder requirements is in
design/brainstorming.md. No CAD or sketches
are produced at this stage; that document is the output of the
brainstorming session called for in the corresponding tracking issue.
The set of metal powders we intend to dispense (high-purity Si,
AlSi10Mg, and the broader BYU VCL digital-alloy-lab element
palette), together with handling/storage constraints and the
custom-Al-crucible dispensing targets, is documented in
docs/candidate-powders.md.
A bare-bones LaTeX template in the Digital Discovery (Royal Society of
Chemistry) submission style is available in paper/. It is built on
the official RSC "Paper" article template (two-column, 9 pt Times,
natbib/rsc.bst bibliography, RSC running headers/footers) and uses
lipsum placeholder text. A pre-built PDF is committed alongside it at
paper/main.pdf.
To rebuild the PDF locally you need a TeX Live installation that includes
latexmk plus the standard packages used by the template (natbib,
mhchem, balance, caption, fancyhdr, lastpage, lipsum,
url, times/mathptmx, extsizes). On Debian/Ubuntu these are provided
by texlive-latex-recommended, texlive-latex-extra,
texlive-fonts-recommended, texlive-science, and latexmk. Only the
RSC-specific assets (rsc.bst, rsc.bib, headers/) are vendored in
paper/; all other packages are taken from the system TeX distribution.
cd paper
latexmk -pdf main.tex