# Record 01 — a 70 µm membrane leaked **Status:** rescued into the repository 2026-09-11 from an untracked working note. --- ## What was attempted An OpenMFDA pneumatic micro-valve with a horizontal actuation diaphragm, roughly 2 mm across (91 px at 22 µm/px), separating a fluid void below from a pneumatic void above. Both voids approximately 440 µm tall. The membrane was drawn at **70 µm — 7 layers at 10 µm/layer.** Printer and resin as recorded in the source: a 22 µm pixel / 10 µm layer DLP machine running *generic clear resin*, explicitly "NOT a tuned custom high-absorber formulation". > Note on the pitch: the source says 22 µm/px. That figure has since been > corrected — the physical machine is a Phrozen Sonic Mighty 12K at 19 × 24 µm, > and 22 µm was the specification of a different machine. See > [`index.md`](index.md) and `/hardware/#printers`. The correction does not > change this record: the membrane thickness is set by **layer height**, not by > pixel pitch, and the layer height is unaffected. The "91 px / 2 mm" diameter > is the one figure here that was expressed in pixels. > > The source's "DLP" is inaccurate for that machine too. The Phrozen Sonic > Mighty 12K is an LCD printer: an LCD screen masks light from a UV source. It > does not project an image the way a DLP printer does. ## What happened — OBSERVED **The membrane leaked.** Fluid crossed into the pneumatic chamber. The CAD wall was intact; the printed wall was not. Verbatim from the source: > FAILURE: a CAD-intact 70µm (7-layer) membrane leaked fluid into the pneu > chamber. ## Why — INFERRED The explanation given in the source, not an independent measurement: > Root cause = cure-through from the layers above + sag over the 440µm void, so > the 7-layer wall doesn't fully/solidly form (NOT a CAD defect). This is the > same class Gong/Nordin (PMC10384158) controlled with a tuned resin + reduced > membrane exposure; OpenMFDA's stock resin has more bleed-through, so 70µm is > marginal. Two mechanisms are proposed — optical bleed-through from the layers exposed above the membrane, and mechanical sag over the 440 µm span before gel. Neither was isolated or measured. The attribution is reasoning from the failure, not evidence of which mechanism dominated. ## What was concluded — INFERRED The source's own recommendation, explicitly marked unverified in the original: > RECOMMENDATION (rule of thumb on stock resin, UNVERIFIED on this exact combo — > needs a calibration ladder): > - Min reliable horizontal separator on stock clear resin/default exposure: > ~120-150µm (12-15 layers). > - **Recommended membrane = 120µm (12 layers).** Stops the leak, still actuates. > - Stiffness ceiling … a 2mm disc at 10-50 kPa deflects usefully only up to > ~200-250µm. Past ~250µm (25 layers) the diaphragm is too stiff to pinch a > 440µm channel at these pressures. So usable window is roughly **12-22 layers > (120-220µm)**. ## What it became in the code `gen_centerfed_block.py` now enforces a **100 µm floor** on membrane thickness: ```python # Printability floor: membrane must be ≥ 100 µm (10 layers @ 10 µm) # for these scaled-up cells (bumped from 90 µm 2026-05-09). # --allow-thin-membrane lowers the floor to 50 µm (5 layers, absolute # hard minimum) and fires an explicit stderr warning so a thinner # diaphragm can be trialled when actuation has failed at 100 µm. ``` `openmfda_flow/flow/platforms/h.r.3.3/pdk/py_scripts/gen_centerfed_block.py` **That the leak caused this floor is INFERRED.** The code comment dates the 90 → 100 µm bump to 2026-05-09 and gives a rationale ("for these scaled-up cells"), but it does not cite the leak. The leak, the 120 µm recommendation and the 100 µm enforced floor are three different numbers, and no file connects them explicitly. ## Dates | what | when | how known | |---|---|---| | membrane floor bumped 90 → 100 µm | 2026-05-09 | dated in the `gen_centerfed_block.py` comment | | bump recorded in the progress narrative | 2026-05-11 | an untracked progress summary, "CURRENT STATE (as of 2026-05-11)" | | source note last written | 2026-06-05 | file mtime | The leak predates the 90 µm floor that the 100 µm floor replaced, so it happened before 2026-05-09. That is the tightest bound available. ## Source Recovered from an untracked working note in the lab's working copy of the design software, `openmfda_flow`, kept under the name `membrane_thickness_finding.md`. Related, kept beside it and also not a durable record: `cure_through_membrane_rule.md` — the optical argument behind the cure-through half of the attribution (penetration ≈ 1.3 × layer height on a strongly absorbing resin; more on a weakly absorbing one). A working note is not the project's record. That is why this document exists.