Skip to content
noggin

Cams, automata and printed mechanisms for the next Mech

Researched 2026-11-02. Purpose: literature basis for the next Mech’s (#51) PETG skeleton and PLA shells: cams and servo-driven coupling, what automata teach, FDM fits, anisotropy, creep, fasteners, magnet pockets, snap fits and printed pivots. Builds on 2026-10-31-filament.md (PLA Pure colours, density), 2026-10-31-shafts.md (horn geometry) and model/beams.py (beam checks); does not repeat them.

Tags: [A] abstract only (full text not read), [I] my inference. No tag means the source’s full text or the cited page was read.

Sources:

Not reached: Cuellar et al. 2018 (Additive Manufacturing 21:150) and Lussenburg et al. 2021 (Additive Manufacturing 39:101846), the TU Delft reviews of non-assembly mechanisms. The repository refused connections. Chapuis & Droz, Automata (1958), is lending-only on archive.org.

  • The fundamental law of cam design: follower position, velocity and acceleration must be continuous and jerk finite. Discontinuities excite vibration [MR].
  • Profile laws [NO, guidelines 2–5]:
    • For rest-to-rest (“double-dwell”) moves, the 3-4-5 polynomial or modified sine is the “workhorse”.
    • The 4-5-6-7 polynomial vibrates least; the cycloidal law is an acceptable substitute if its higher peak acceleration is tolerable.
    • Motions with specified intermediate positions or velocities call for polynomials or B-splines.
  • Pressure angle [NO, guideline 6]: about 30° maximum for translating followers (35° on a low-friction ball slide), and about 35° for oscillating-arm followers.
  • Roller followers [NO, guideline 7]: the pitch curve’s radius of curvature must exceed the roller radius everywhere to avoid undercutting. Target ρ/Rf > 2; 1.5 has been used.
  • Flat-faced followers [NO, guideline 8]: the cam may have no concave (negative-radius) section.
  • Track cams [NO, guideline 12]: a single roller in a groove has crossover shock and slips at each load reversal. The cures are a spring holding the roller against one side, or two preloaded rollers.
  • A servo-driven cam is not a timing cam [I]. A classical cam turns at constant speed, so its profile is the motion law. A PG001 position servo sets the timing itself, so a cam on its horn only maps one angle to another. The smoothness law belongs in the firmware trajectory (motion control covers that); the cam profile only needs to be C2 and within the pressure-angle limits.
  • Driving a second feature from one servo [I]: all seven #51 degrees of freedom have their own servo. A lid that follows eye pitch is therefore a firmware sum (lid command + k·pitch) costing zero parts, and it keeps the wink. A mechanical cam or link from pitch to lids would fight the lid servos.
  • Printed cams [ZF]:
    • PLA FFF cams matched milled aluminium within 0.75% on dimensions in the print plane (XY), but erred up to 1.64% along Z (across layers).
    • The working face (the side wall) had Ra 10.41 µm, against 0.50 µm milled, at a 0.15 mm layer height. Metal cams want Ra < 0.4 µm.
    • The authors judge printed cams fit only for low-power transmission.
  • Print a cam flat [I from ZF]: profile in the XY plane, so the profile is traced by the nozzle and not stair-stepped by layers.

2. Automata: how natural motion was encoded

Section titled “2. Automata: how natural motion was encoded”
  • Maillardet’s automaton (c. 1800) [FI]:
    • Its “memory” is a stack of brass cam discs turned by a clockwork motor.
    • Three steel followers trace the cam edges. Through levers and rods they drive the hand side to side, front to back and up and down: one follower per axis, no follower shared between axes.
    • It holds four drawings and three poems.
    • Its restorer believes the head and eye motions were originally more nuanced.
  • Jaquet-Droz automata (Neuchâtel, 1768–1774) [LH]:
    • The Writer is programmed by setting input tabs that select the cams moving the pen, up to 40 characters. All gears, cams and followers sit inside the body.
    • The Musician’s eyes follow her fingers, and her breast rises and falls while she waits. The Draughtsman pauses to blow charcoal dust from the paper.
  • What carries over [I]:
    • The keyframe table is the cam. With one servo per axis, the Mech’s gesture tables play the role of the cam stack. The automata keep axes independent (one follower each) rather than coupling them mechanically, and #51’s seven independent servos already do the same.
    • Life comes from secondary motion: breathing, gaze following the action, small pauses. These are cheap in firmware and belong to the animation and motion-control notes (2026-11-01-toon-eyes.md, 2026-11-02-literature-motion-control.md).

All specimens were printed at 100% infill and annealed and dried for 8 h before testing [TDS-PB, TDS-PH, TDS-LB].

E X-Y / Z (MPa) tensile X-Y / Z (MPa) elongation X-Y / Z (%) bend X-Y / Z (MPa) HDT 1.8 / 0.45 MPa (°C)
PETG Basic 1460 / 1120 48 / 39 11.2 / 5.7 81 / 64 65 / 69
PETG HF 1810 / 1540 34 / 23 8.6 / 5.1 64 / 48 62 / 69
PLA Basic 2680 / 2160 39 / 35 12.2 / 7.5 76 / 68 54 / 57
  • Z costs a lot: about 0.7–0.9 of the X-Y tensile strength, and about half the elongation.
  • These are best-case values [I]: 100% infill and annealing. Walls and infill on real parts do worse.
  • Transverse to the raster: with every raster in one direction and no walls, PETG gave 50.4 MPa along the raster and 22.5 MPa across it (E 2.33 vs 2.19 GPa) [ST].
  • Oil and grease: both TDSs rate PETG and PLA “resistant to most kinds of oil and grease” [TDS-PB, TDS-LB].
  • PETG creep [ST], measured at 21 °C:
    • Below 50% of strength, creep added up to 1% strain (X) or 0.6% (Y) in 5 h.
    • At 50% of strength, creep was about 5% of the elastic strain in 5 h and doubled by 20 h.
    • At 70%, it reached 25% of the elastic strain (X).
    • Plastic, unrecovered strain appears once creep exceeds 0.1% (X) or 0.05% (Y), which happened above 50% of strength.
    • One relaxation spectrum fitted both directions, so anisotropy enters through the moduli only.
  • Constant strain relaxes too [I from ST]: an interference fit in PETG is constant strain, so its grip relaxes over time. A press fit is not a permanent retainer.

4. Fits, clearances, fasteners, magnets, snap fits

Section titled “4. Fits, clearances, fasteners, magnets, snap fits”
  • Clearances in non-assembly revolute joints [AN]:
    • Literature values: 0.2 mm (PolyJet), 0.2 and 0.05 mm (FDM, the latter with shape changes), 0.4 mm (Stratasys F370 FDM).
    • The measured clearance is zero (fused) below a critical designed clearance, then follows it once large enough.
    • So every printer and material pair needs its own test coupon [I].
  • Hitting ISO 286 clearance fits in PLA (0.4 mm nozzle, 0.1 mm layers) needed a 0.13 mm hole horizontal expansion in the slicer [GR].
  • Hole accuracy depends most on layer height, then on the number of shells around the hole [PO, A].
  • Threaded inserts [KA, A]: pull-out depends most on infill in the joining zone, which should be ≥70%. Walls around the pre-hole need ≥2.4 mm for the insert they tested; thicker walls did not help.
  • Printed thread vs inserts [TH], M8 in SLS PA12: the printed thread failed at 3.41 kN, a heat-set insert at 3.83 kN (+12%), a Helicoil at 4.78 kN (+40%). Heat-set insertion bulges material at the top face. The authors limit printed threads to large sizes and low loads. Transfer to FDM PETG is [I].
  • Direct screwing [MH, A]: AM screw bosses built to injection-moulding guidelines differ significantly with build orientation and process.
  • Adhesives [OZ, A]: the adhesive type explains 91.6% of joint-strength variance and the substrate only 5.3%. PETG bonded best with a methyl methacrylate adhesive; PLA did best with PU, epoxy and acrylate.
  • Snap fits [KL, A]: the injection-moulding working principles hold for FDM. The guidelines must add FDM’s restrictions, notably the beam’s orientation to the build direction.
  • Snap-fit strain [I from the TDS]: Z elongation is about half of X-Y. A snap beam should lie in the X-Y plane, along its length, with a design strain well under the Z elongation (≤ ~2% for PETG).
  • PETG on steel, dry:
    • Unfilled FFF PETG on hardened C45 steel (Ra 0.06–0.08 µm, 0.64–1.91 MPa, 0.2 m/s) gave COF 0.57–0.64 [VA].
    • Vaught & Polycarpou measured a steady COF of 0.34–0.44, highest on vertically printed specimens, where layer interfaces are exposed [VP via VA].
  • Adding PTFE: press-fit PTFE plugs covering 30–40% of the face cut COF to 0.18–0.24 and mass loss by up to 76% [VA].
  • Print direction: sliding across the rasters gives a higher COF than sliding along them, for PLA and ABS; COF rises with load and speed [CH, A].
  • What this means for the Mech [I]:
    • Our pivots run far gentler than these tests. A ~0.5 N eye load on a 3 × 3 mm projected bearing is about 0.06 MPa. Even the 9.2 N stall force at the 12.75 mm horn pin (beams.py) on a 2 × 3 mm pin is only ~1.5 MPa, briefly.
    • Play, not wear, is the failure mode: wear is not the life limit at desk duty; the fit clearance the print needs is.
    • Use metal shafts on busy joints: a steel shank in a printed bore avoids the weak polymer-on-polymer and across-layer contact.
    • Orient the bore axis along Z: the shaft then slides along the perimeter lines, not across layers.
  1. Put any smoothing law (3-4-5 or 4-5-6-7 polynomial, or cycloidal, rest to rest) into the firmware trajectory, not into a printed cam. A cam on a position servo only maps angles [NO, I].
  2. Use no cam or link to couple lids to eye pitch. Add k·pitch to each lid command in firmware, which keeps the wink and adds no parts [I].
  3. If a cam is used anyway:
    • keep the pressure angle ≤30° (translating follower) or ≤35° (oscillating arm);
    • keep ρ/Rf > 2 for a roller follower, and no concave flank for a flat follower [NO];
    • print it flat, with the profile in X-Y [ZF].
  4. Size every skeleton beam with PETG Z-direction values: worst case PETG HF, E 1.54 GPa and 23 MPa tensile [TDS-PH]. Use X-Y values only where the bending stress runs along the layers, and confirm which PETG is on the spool.
  5. Keep sustained stress (parked loads, magnet pull, preloads) under ~1/3 of the Z strength: ≤ ~8 MPa for PETG HF, ≤ ~13 MPa for PETG Basic [I from ST: plastic creep above 50%].
  6. Retain magnets with adhesive, not by interference alone. PETG relaxes under constant strain [ST, I], and hole size varies with layer height and shell count [PO, GR]. Pick the adhesive by test: an MMA adhesive bonded PETG best [OZ, A].
  7. Print a clearance coupon on the Bambu (0.4 mm nozzle; PETG and PLA Pure) before fixing fits. Step pins in 0.05 mm from 0.1 to 0.4 mm per side, because below a critical clearance the joint fuses [AN]. Hold holes to the slicer’s hole compensation [GR].
  8. Busy pivots (eye gimbal axes, lid arms, jaw) get a steel shaft, such as a smooth screw shank, in a printed PETG bore with its axis along Z. Printed-on-printed bearings are only for rarely moving joints [VA, CH, I].
  9. Give screw bosses and insert holes ≥2.4 mm wall and ≥70% local infill [KA, A]. Self-tapping kit screws are enough at these loads; printed threads are the weakest option [TH, I].
  10. Lay snap-fit and flexure beams in X-Y with strain ≤ ~2% (PETG), never bending across layers [KL, TDS, I].
  11. Grease is permitted on PETG and PLA [TDS-PB, TDS-LB]. A drop of PTFE grease on the steel-in-PETG pivots attacks the 0.34–0.64 dry COF [VA, VP, I].
  1. The SF 2 checks assume PLA, but the skeleton is PETG.
    • #51 ranks candidates on servo torque at SF 2 and has the model check that every part holds.
    • model/beams.py uses E 3.5 GPa and yield 50 MPa (PLA). Bambu’s own PLA Basic TDS gives 2.68 GPa and 39 MPa X-Y. PETG HF gives 1.81/1.54 GPa and 34/23 MPa (X-Y/Z); PETG Basic gives 1.46/1.12 GPa and 48/39 MPa.
    • With PETG HF across layers, a beam showing SF < 4.35 today falls below SF 2, and deflections grow 1.9–3.1×.
    • Change: the #51 model reads skeleton constants from the PETG TDS (Z values by default), and the spool’s PETG variant is recorded.
  2. Eyeball caps “may grow up to touching”.
    • Printed parts err up to 0.75% in X-Y and 1.64% in Z [ZF]. On a ~40 mm cap that is ±0.3–0.65 mm, before magnet-seat play.
    • Touching PLA caps on linked eyes would rub, and PLA-on-PLA dry contact is the worst pairing of all those above.
    • Change: “up to a gap of ≥1 mm between caps at every pose” [I], checked in the model like the other collisions.
  3. “The one new purchase: magnets” (conditional).
    • Practice 8 wants steel shafts at the busy pivots. If the kit’s spare screws cannot serve as smooth shanks of the right length, that needs a second small purchase (steel pins or rod). This is within the fixed limits; the hardware list is not fixed.
    • Change: allow steel pins if the kit’s screws do not fit.
byYellow PineYellow Pine

noggin v0.2.26-6-g09bf9b7 · 09bf9b7