Skip to content

HANDOVER.md

Latest commit

 

History

History
132 lines (111 loc) · 6.38 KB

HANDOVER.md

File metadata and controls

132 lines (111 loc) · 6.38 KB
Raw

Handover — Dripito Rev-B → Rev-C

State of the project at submission (2026-05-14) and what the next person working on this device needs to know to pick it up.

What Rev-B is

A single-board, single-AA-cell prototype IV drip flow-rate monitor with dual-beam optical drop detection on an STM32G071C8. The device counts drops at the drip-chamber orifice, times them across two horizontally-aligned IR beams (10.2 mm separation), and reports per-drop volume + cumulative flow on a 4×16 LCD. Submitted as an ETH GHE MSc semester project; not a certified medical device.

What's reproducible:

  • docker compose up regenerate-figures-sample in analysis/ cold-builds the headline figure on a clean machine.
  • The 2026-05-13 bench dataset (data/raw/2026-05-13_*) is committed, documented in data/README.md with explicit column schemas, and re-analysable from the included pipeline.
  • The hardware is fully spec'd: KiCad 8 project, BOM-CSV, gerbers, three custom footprints under CC BY-SA 4.0.

What works (within Rev-B scope)

  • Optical detection chain. Both beams reliably trigger on individual drops at the morning calibration mount position. Drop counting matches the gravimetric drop count to within a few percent over a 3-10 minute run at that position.
  • Sphere-model volume estimate after calibration. Post-V_CAL_K=1.27 per-run residual against gravimetric is ±30 % across the operating envelope on V_50_01..05.
  • Velocity (transit time) is position-agnostic. The two beams deliver reliable time-of-flight regardless of mount height.
  • Reproducible build + analysis. STM32CubeIDE GUI builds the firmware; the Docker pipeline regenerates results on any machine.

What does not work (Rev-B's three blockers)

  1. Position-dependent chord measurement (§17 of docs/limitations.md). The 2026-05-13 afternoon session measured the K required to make the LCD-summed drop volume match gravimetric at four mount positions on the same drip chamber. With the shipped mean-of-pulses algorithm, K_fit ranged 0.42–1.27 across positions (3× spread); across all eight algorithm variants trialled, K_fit spanned 0.42–1.98 (4.7× spread driven by mount geometry alone). Three contaminants isolated: umbilical at TOP (filament still attached when drop crosses TOP beam), backsplash at BOT (splash particles cross BOT beam after the drop body has cleared), threshold-vs-peak mismatch on the dual- threshold core pulse. No (algorithm, K) combination passed ±5 % at all positions.
  2. Drop oscillation between oblate and prolate shapes in the first 10–30 mm of free fall. The beam plane catches each drop at an unpredictable point in the oscillation. Mean averages out (hence V_CAL_K works on the run mean); per-drop CV stays inflated.
  3. Tilt sensitivity, un-quantified. Sub-mm chamber reposition doubled TOP pulse width on bench. The chord-time architecture is sensitive to off-vertical drop trajectories.

Rev-C path (named throughout docs/limitations.md)

The Rev-C optical front-end has to deliver position-invariant chord measurement, not just position-invariant counting:

  • Second beam pair, horizontally offset. Catches the horizontal extent of each drop. With vertical + horizontal extent, the firmware can infer the oscillation phase and apply an oscillation-aware volume formula instead of sphere-from-vertical-chord.
  • Wider TOP-to-BOT separation. Drops fully detach from the chamber tip before reaching the TOP beam — kills the umbilical contaminant.
  • TOP beam further from the pool surface or shielded against splash. Kills the backsplash contaminant at high mounts.
  • TOP/BOT gain equalisation. Removes the per-board dynamic-range scatter that breaks fixed-margin core thresholds.
  • Mechanical: verticality budget on the chamber holder. The improvised broom-clip Rev-B holder is bench-grade. Rev-C needs a printed clamp with a stated tolerance.
  • Power: validate the TPS610981 boost chain. Rev-B was tethered to 3.3 V via ST-Link; the single-AA loop was never validated end-to-end.

docs/decisions/firmware-build-and-test-rev-c.md is the ADR for the deferred firmware-CI work (Dockerfile + Makefile, host-buildable Unity harness for the drop-detection state machine). Both were skipped for Rev-B to avoid regression risk to the bench-validated firmware in the final 4-day window before submission.

Where things live

Area Path
Firmware (Rev-B) firmware/STM32CubeIDE/Dripito/
Hardware (KiCad 8) hardware/flow_monitor.kicad_sch, .kicad_pcb
Enclosure (ASA print) enclosure/
Validation dataset data/raw/2026-05-13_*, data/raw/2026-05-14_pm/
Analysis pipeline analysis/ (docker-composed)
Architecture Decision Records docs/decisions/
Known limits, ranked by impact docs/limitations.md
Headline results docs/results.md
Validation protocol docs/testing-and-validation.md
Bench tooling tools/ (UART/scale loggers, figure builders)

Open experiments (not run for Rev-B)

  • Rev-A baseline (count-only firmware on STM32G030, commit ed84ff5) was removed before bench day; a head-to-head against fixed-factor literature methods is the obvious paper-grade comparison.
  • Multi-board scatter. n=1 board for geometry. The print-tolerance MC error budget predicts 49 % of MAPE from per-board sensor-arm geometry; not refuted.
  • Multiple drip sets. Single macro-20 set on bench; macro-15, macro-10, micro-60 untested.
  • Multiple head heights. Single drip-chamber-to-TOP-beam height; clinical range 60–150 cm uncovered.
  • Multi-session generalisation. Single bench day; session-to-session drift un-quantified.

Reproducing the bench day

The full 2026-05-13 protocol-as-run is in docs/testing-and-validation.md. The morning calibration runs (V_50_01..04) fit V_CAL_K; V_50_05 is the independent verification with K baked into firmware. The afternoon position-drift runs are in a separate data/raw/2026-05-13_pm_position_drift/ folder with a per-run README documenting each mount position and the K required at that position.

Contact

Leandro Catarci · lcatarci@ethz.ch · ETH MSc, D-MAVT, Global Health Engineering. Bachelor thesis baseline: 10.5281/zenodo.16902366. Rev-B archive DOI: 10.5281/zenodo.20199232.