Sensor Design V2.1

Legend: ✅ Locked · 🔲 Decide (still open) · ⭐ Recommended default (use unless you object) · ✏️ Fill in (a value/coord/measurement) · ☐ checkboxes = build/test actions.

0 · What the pond test proves

The pond is a mechanism shakedown, not the convergence-zone study. Success = the pump-filter loop runs end to end and the data comes off clean.

• ✅ Primary objective: "The ASV reaches a set of pond waypoints, pumps a measured volume through the filter at each, and recovers a lab-countable sample — with flow, current, GPS and SD all logging together, no clog lockout, no electronics brownout."
• ⭐ Scope = mechanism + collection. Run the full pump/flow/clog/log loop and pull a real filter for lab counting, but don't gate success on a perfect particle count (that's the recovery trial, §8).
• Success criteria:
  • ☐ Reaches each waypoint and holds position long enough to pump
  • ☐ Pumps a repeatable target volume — ✏️ target (e.g. 5 L), ✏️ ± tolerance
  • ☐ Flow + INA219 + GPS + (turbidity) all log to SD in one synced CSV
  • ☐ No brownout/reset when the pump kicks on (see §5 power)
  • ☐ Clog-recovery routine fires and clears, or never trips
• Out of scope (write it down): FTIR/Raman confirmation, multi-day repeats, wave/chop handling, the real coastline sites.

1 · Architecture — ✅ LOCKED: on-boat pump + filter

• ✅ Boat goes to a waypoint, a 12 V diaphragm pump draws surface water through a staged filter, the YF-S201 logs the exact volume, the analytical mesh is pulled for lab counting. This is the mechanism for the pond and the coast.
• Why this over grab-and-sieve: it keeps the V2 hardware story, it's the method we actually want to field at convergence zones, and pumping a measured volume gives a clean concentration normalization (particles ÷ litres) instead of relying on grab-jar representativeness.
• What it requires (all covered below): pump + switch (§5), volume measurement (§5), a staged filter that won't blind instantly (§3), clog handling (§3/§5), and one removable cartridge per site/run (§4).
• Grab-and-sieve (Jeznach's river method) is set aside — noted here only so we remember we chose against it on purpose.

2 · Where to sample (pond + real deployment)

Jeznach's river site rules (wadeable/bridge access, streamflow gauges, up/down-stream of WWTPs, straight reaches not bends, mid-channel) don't transfer — those are river-science controls. We drop them.

Pond (July 18) — a mechanism grid, not a science design. Pick waypoints that test repeatability and navigation, not a pollution gradient:

• ⭐ 4 waypoints spread across open water + one near the inflow + one in a sheltered corner. Enough to test "go there, pump, log, move on" several times; few enough to fit one battery.
• ✏️ Fill in — the actual pond waypoints (name + GPS lat/long). Drop a pond map screenshot here.
• ⭐ 2 runs of the grid for repeatability (replaces "4 × 200 mL replicates" — we get repeatability from repeated pumped volumes, not from her jar count).
• ✏️ Fill in — sampling depth. Buoyant plastics ride the top → ⭐ surface intake (intake mouth 2–5 cm below waterline). Write down the choice.

Real deployment (coast, later). Site selection is driven by our hypothesis — current-convergence / accumulation zones — not river reaches. We'll redo §2 properly for the coast once the mechanism is proven.

3 · Filtration (pump-filter cascade) — Jeznach sizing applied

• ✅ Target band: 300 µm – 5 mm (Jeznach). Samples come back fiber-dominated, so the mesh has to retain fibers, not just fragments.
• ✅ Mesh material: 316 stainless or PET. Never nylon — nylon sheds polyamide fibers into a fiber-dominated count. This matters more than it did in V2.
• ✅ 50 µm is out. Going coarser is the point.
• Why we still need stages (even though we went coarser): we're pumping raw pond water full of algae and organics. A lone 300 µm mesh would still blind fast. A short cascade spreads the load:

• 🔲 Decide — analytical pore size. ⭐ 300 µm (matches Jeznach's floor, low clog/contamination risk). Going to 100–200 µm catches more but reintroduces blinding + fiber-contamination risk. Confirm 300 or pick a number.
• 🔲 Decide — 3-stage (above) vs 2-stage (2 mm → 300 µm). ⭐ Start 3-stage for raw pond water; drop stage 2 if it turns out unnecessary. The 500 µm from V2 is gone (too close to a 300 µm analytical to do real work).
• ✅ Keep the INA219 clog-recovery routine — coarse mesh clogs less, but pumping raw water still fouls; cheap insurance, and it's the headline "smart" feature for judging. (Routine detail in §5.)

4 · Filter handling — one cartridge per site, no cross-contamination

• ⭐ Swappable 300 µm cartridge, one per waypoint/run. Pump → pull cartridge → drop into a labeled sealed non-plastic container → fresh cartridge → next waypoint.
• ✏️ Fill in — cartridge housing (inline filter holder spec / 3D-printed body + the mesh disc size).
• ☐ Carry a field blank cartridge (exposed, never pumped) to catch airborne fiber contamination.
• ☐ Rinse the intake/pump line with filtered water between waypoints so site N doesn't bleed into N+1.

5 · Sensor + electronics suite — THE BUILD ⭐

Taken from V2 and made electrically coherent. The headline risks aren't the sensors individually — they're voltage levels (ESP32 is 3.3 V and not 5 V-tolerant), the I2C bus sharing, and power isolation so the pump doesn't brown out the logic.

Making them work together

I2C bus — ✅ no address conflicts. INA219 0x40, ADS1115 0x48, MPU-9250 0x68 (its AK8963 magnetometer at 0x0C) are all distinct, so they share one bus fine. Use ESP32 default SDA=GPIO21 / SCL=GPIO22 with 4.7 kΩ pull-ups to 3.3 V. (Even with the ADS1115 powered at 5 V, the bus is open-drain pulled to 3.3 V, so the ESP32 sees safe 3.3 V — fine.)

Flow sensor level-shift — the easiest thing to fry the ESP32 with. YF-S201 runs at 5 V and its pulse is ~5 V. Feed it through a level shifter or a simple divider (e.g. 10 kΩ / 20 kΩ → ~3.3 V) into an interrupt-capable GPIO (e.g. GPIO27). Calibrate volume gravimetrically (V2 already planned this). ✏️ Check the flow rate — YF-S201 is happy at 1–30 L/min; if our pump pushes <1 L/min through a fine mesh, swap to a YF-S401 (0.3–6 L/min) or it'll under-read.

Pump switching. ⭐ Drive the pump with a logic-level MOSFET (IRLZ44N or a MOSFET module) straight from a 3.3 V GPIO — quieter than a relay, and lets you PWM the flow if you want. Add a flyback diode (e.g. 1N5822) across the pump: it's an inductive load and the back-EMF will eventually kill an unprotected switch. If you keep the V2 relay instead, note it needs a 5 V coil and many relay boards don't trigger reliably from 3.3 V — use a "low-level-trigger" optocoupler board or a transistor.

INA219 placement + range. Wire it in series on the pump's 12 V feed (it reads bus voltage up to 26 V, fine). ✏️ Check clog current: the common breakout shunt (0.1 Ω) tops out near 3.2 A — if the pump spikes past that when blinded, the reading saturates and clog-detect goes blind exactly when you need it. Verify stall current, or use a board with a bigger shunt.

Turbidity + ADS1115 voltage. SEN0189 swings 0–4.5 V, which exceeds the ADS1115's input ceiling if the ADC is powered at 3.3 V. ⭐ Power the ADS1115 at 5 V (input ceiling rises to ~5 V, the 4.5 V signal fits) and keep the I2C pull-ups at 3.3 V as above. Mount the turbidity probe inline before the filter so it reads intake water, not post-filter. (Turbidity is optional for the mechanism test — 🔲 keep it or drop it to cut a variable on day one.)

IMU — the one sensor most likely to misbehave. The MPU-9250's magnetometer is a compass; mounted next to a 12 V pump, thruster motors, and a stainless mesh it'll read garbage (hard-iron). If you run autonomous heading: mount the IMU as far from the pump/motors/SS as possible, on a small mast if you can, and run a figure-8 hard/soft-iron calibration. ⭐ For the first pond run, you don't need it — drive by RC and let GPS course-over-ground give heading while moving. Re-add IMU heading for true autonomy. (Note: MPU-9250 is discontinued; an ICM-20948 is the modern drop-in if you ever rebuy.)

Power architecture — the thing that makes it "work in the field, not just the bench."

• One 3S LiPo (≈11.1–12.6 V) feeds the pump and thrusters directly, and a 12 V→5 V buck for the 5 V rail (SD, flow, turbidity, ADS1115). The ESP32's onboard 3.3 V regulator runs off that 5 V.
• ⚠️ The pump and thrusters are noisy inductive loads. If the logic buck shares the rail naively, pump inrush will sag the 12 V and reset the ESP32 mid-mission. Mitigate: dedicated buck for logic with fat input capacitance (≥470 µF bulk), keep pump/motor wiring away from signal wiring, and — cleanest — a separate small battery/BEC for electronics.
• Common ground everywhere (12 V, 5 V, 3.3 V domains tied) or the INA219 shunt reading, the MOSFET gate reference, and the flow-sensor signal won't make sense.

Starter pin map (ESP32 WROOM — adjust as needed)

• ✏️ Fill in — power budget. Measure pump A and thruster A at 12 V, sum, add margin, pick LiPo mAh for runtime across the grid.

6 · Navigation & control (pond)

• ⭐ RC-drive the first pond run. It isolates the one thing we're testing (the pump-filter loop) from autonomy bugs. GPS still logs position + timestamp at each waypoint. Add autonomous waypoint nav (GPS + IMU heading) on a later pond day.
• Stop-and-sample state machine: transit → arrive at waypoint → hold position → pump for ✏️ ___ s (or until ✏️ ___ L on the flow sensor) → confirm volume logged → pull cartridge → next waypoint → return.
• ✏️ Station-keeping while pumping: calm water makes this easy — ⭐ low-throttle hold or brief drift. Note which.

7 · Lab + data pipeline (Jeznach lab methods kept)

• Per-site concentration: Concentration = Particle Count ÷ (Pumped Volume × Recovery Efficiency) → particles/L at each waypoint.
• Lab method (Jeznach-aligned): KOH digest of organics → Nile Red stain (lab only — illegal to discharge in the field, fine in the lab) → microscope count → hot-needle test to confirm plastic. Expect mostly fibers. ✏️ confirm what we can actually run.
• ✏️ Fill in — microscope + lab access (school lab? magnification? who counts?).
• ☐ Per-site CSV: site ID, GPS, pumped volume, current trace, (turbidity), timestamp.

8 · Recovery efficiency (do this before the pond)

• ☐ Spike-and-recover in a tub: drop a known count of reference particles spanning 300 µm–5 mm — include fibers — into a measured volume, run the full pump-filter loop, count what the analytical mesh caught.
• ✏️ Fill in — reference particle source (cut/sieved known plastics, or a purchased microplastic reference set).
• Feeds the Recovery Efficiency term in §7 and the success criterion in §0.

9 · Timeline — backward from July 18

• ☐ By ~Jun 27 — design freeze. Confirm analytical mesh size (§3) + flow-sensor model (§5). Architecture's already locked.
• ☐ By ~Jun 30 — BOM ordered (§11). Mesh + buck + MOSFET + level shifter first; shipping is the long pole.
• ☐ Jul 1–9 — build + bench test. Wire the suite, prove I2C + flow + SD log together, prove the pump switches without browning out the ESP32 (the §5 power test).
• ☐ Jul 9–12 — recovery-efficiency trial (§8) in a tub.
• ☐ Jul 12–15 — full dry run (single waypoint, pond edge or tub).
• ☐ Jul 16–17 — buffer / fix.
• ☐ Jul 18 — pond deployment.

10 · Still open (everything 🔲, in one place)

Resolved this revision: ✅ architecture = pump + filter, ✅ not copying river methodology, ✅ INA219 clog-recovery kept, ⭐ RC-first nav.

1. Analytical mesh size: confirm ⭐ 300 µm, or pick another.
2. Filter stages: ⭐ 3-stage for raw pond water, or 2-stage?
3. Flow sensor: YF-S201, or YF-S401 if our pump flow is <1 L/min?
4. Pump switch: ⭐ MOSFET, or keep the V2 relay?
5. Turbidity sensor: keep for per-site data, or drop to cut a variable on day one?
6. Target pumped volume + pump-time per waypoint (§0/§6).
7. Pond waypoints + GPS (§2) and power budget numbers (§5).

11 · BOM / procurement

12 · Risks / watch-items

• Pump-inrush brownout — the #1 "worked on the bench, died on the pond" failure. Isolate logic power (§5).
• Frying the ESP32 — 5 V flow pulse / 4.5 V turbidity into 3.3 V pins. Level-shift / divide (§5).
• Magnetometer interference — IMU heading garbage next to the pump/motors/SS mesh (§5). RC-first sidesteps it.
• Clog-current saturation — INA219 default shunt blinds above ~3.2 A right when the filter blinds (§5).
• Filter blinding — raw pond water on a fine mesh; the cascade + clog-recovery exist for this (§3).
• Fiber contamination — fiber-dominated samples + any nylon/airborne fiber inflate counts. Nitrile gloves, cotton garments, field + procedural blanks (§4/§7).
• Cross-contamination between waypoints (§4).
• Time/power budget — waypoints × (transit + hold + pump) must fit one charge (§5).