TRIAL 2026-11-18
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
Travis Johnson df385e9594 Swap in annotated renders; retire the hand-drawn SVG P&ID
- hero is now the flow-annotated render (self-labelling, shows direction),
  replacing the plain render and its numbered callout overlay
- section 02 uses the pictorial P&ID in place of the inline SVG
- exploded assembly view added alongside the BOM
- captions call out four deviations from spec across the three renders:
  filter on the suction side, strainer downstream of the pump, yellow
  brass on the makeup fitting, and an AC pump that can't be PWM-throttled
2026-08-26 03:33:17 -05:00
docs Swap in annotated renders; retire the hand-drawn SVG P&ID 2026-08-26 03:33:17 -05:00
esphome Industrial dog waterer: spec, firmware, and alerting 2026-08-26 00:10:45 -05:00
prometheus Industrial dog waterer: spec, firmware, and alerting 2026-08-26 00:10:45 -05:00
.gitignore Industrial dog waterer: spec, firmware, and alerting 2026-08-26 00:10:45 -05:00
DESIGN.md Industrial dog waterer: spec, firmware, and alerting 2026-08-26 00:10:45 -05:00
LICENSE Industrial dog waterer: spec, firmware, and alerting 2026-08-26 00:10:45 -05:00
README.md Swap in annotated renders; retire the hand-drawn SVG P&ID 2026-08-26 03:33:17 -05:00

Dog waterer, industrialised

Take a $40 plastic dog fountain and build the version an engineer would build: copper pipe, real cleaning, a bigger tank, filters you don't change monthly, a surplus reservoir you refill occasionally, and something that tells you before it runs dry.

That was the brief. This repo is the answer, and the most useful thing in it is the finding that you probably shouldn't build the expensive one.

Annotated concept render of the tier 3 waterer

Concept render, not a built machine — nothing here has been fabricated. Faithful to the spec where it counts: gravity makeup through a 0 psi float valve, hard-piped copper recirculation, strainer ahead of the pump, blowdown at the low point, all on a drip tray over a drain. Three things to correct if you build from it: the cartridge filter is on the suction side (put it on the discharge — suction-side ΔP rises as the element loads, and that's how you cavitate a pump), the makeup fitting is yellow brass where the spec calls for lead-free NSF-61 or stainless, and an AC mag-drive pump can't be PWM-throttled — you need variable speed to get both a quiet 2 L/min serve and a 10 L/min scouring CIP.

P&ID Exploded assembly
Pictorial P&ID. Note two draughting errors: ST-301 is drawn downstream of P-101 (the strainer belongs upstream, protecting the impeller), and the return leg never closes back to the spout. Order of record: bowl → ST-301 → P-101 → F-301 → UV-401 → riser → spout → bowl. Exploded assembly. Rough build order is bottom-up: frame, drip tray and bowl, then hang the loop — solder every copper joint on the bench before anything is mounted.

📄 Read the full illustrated spec → — P&ID, sizing tables, the numbers. Open it in a browser; it's a single self-contained file.


Pick a tier and stop there

Each row is a complete, buildable system. Value does not increase linearly — tier 2 is where the curve bends.

Tier System Cost Consumables Refill Tells you how
0 Typical consumer fountain (~2 gal, ABS) $40 carbon pads, monthly ~2 d an LED
1 Livestock gravity waterer + a drain $70 none ~4 d you look at it
2 Tier 1 + load cells + stack light ← build this $130 none ~4 d blinking light, no cloud
3 Copper recirc loop, UV, auto-blowdown, 57 L reservoir $1,060 UV lamp + gaskets, yearly ~12 d Prometheus → email

A 5 gal livestock waterer plumbed to a drain wins on tank size, refillability, and filter changes by deleting components — there is no pump to disassemble and no filter to replace, ever. What it can't do is move the water or disinfect it, and that is the entire case for tier 3. Build tier 2 first, live with it a month, and let the data tell you whether you want the copper.

Contents

Path What
docs/spec.html The illustrated spec. P&ID, sizing, materials, cost. Start here.
DESIGN.md Full tier-3 engineering spec, rationale, BOM
docs/TIER2.md The cheap build, in detail
docs/HOSE-TIMER.md Automating a flush-on-fill livestock waterer: which timer, and the freeze/backflow/water-bill gotchas
docs/CIP.md Clean-in-place procedure and maintenance schedule
esphome/dogfountain.yaml Firmware — sensors, interlocks, blowdown, stack light
prometheus/ Scrape config, alert rules, Alertmanager email, compose stack

The findings worth stealing even if you build none of this

Delete the carbon. In a closed recirculating loop, activated carbon is exhausted within days for anything it actually adsorbs, it strips the chlorine that was keeping the water stable, and it then becomes a warm dark high-surface-area bacterial substrate. That is why those pads "need" replacing monthly — colonisation, not exhaustion. Mechanical strainer + washable pleated cartridge + UV-C gets you to one $25 consumable per year.

Blowdown, not filtration. The real problem is recirculating the same 2 L of saliva-water for a week, and no filter fixes it. Use the cooling-tower answer: with evaporation E = 0.5 L/day and a target of C = 1.5 cycles of concentration, a bleed of B = E/(C−1) = 1.0 L/day holds bowl TDS at ≤1.5× tap forever. Trigger it on conductivity, but keep a hard 24 h forced dump underneath — TDS is a weak proxy for saliva, and the control law should admit that.

Oversize the pump for cleaning, not for serving. 2 L/min through ½" line is 0.22 m/s — quiet, pleasant, and useless for cleaning. PWM to 10 L/min for CIP: 1.1 m/s, Re ≈ 15,000, fully turbulent. Turbulent shear is what removes biofilm; a slow rinse does not.

Quantify the copper before you commit to it. Water at the EPA action level of 1.3 mg/L adds ~1.5 mg/day against a ~2.6 mg/day NRC allowance for a 30 kg dog — not toxic, but ~60% on top of diet. Continuous recirculation kills the contact time that drives leaching. Check your utility's pH and alkalinity, use lead-free solder and NSF-61 brass, and actually test the water (target < 0.3 mg/L). Copper-associated hepatopathy is breed-linked; with a Bedlington, Labrador, or Doberman, go 316L wetted and use copper as cladding.

Verify actuators, not commands. "Relay closed" is not "lamp emitting." Current monitoring on the UV ballast and the pump turns a silent failure into an alert. Same logic makes the DIY hose timer worth building: a flow sensor proves the valve actually opened.

Don't let the alert path share fate with what it watches. If the server is down, the server cannot tell you the dogs have no water. Bowl-dry and leak fire from the device itself, and the red stack light needs no network at all.

predict_linear beats a fixed threshold. predict_linear(reservoir_liters[12h], 48*3600) < 0 gives constant lead time regardless of how fast they're drinking.

The reason to instrument rather than just enlarge. Reservoir mass on load cells, minus blowdown and modelled evaporation, is daily household water intake. Sustained polydipsia leads chronic kidney disease, diabetes mellitus, and hyperadrenocorticism by months. Route it at severity: info, never to anything that pages — it's a household aggregate, so it says someone changed, not who, and it's a prompt to book a vet visit, not a diagnosis.

Status

Prometheus rules validated with promtool (15 rules), Alertmanager config with amtool. The ESPHome config is YAML-valid but has not been compiled or run against hardware — component syntax drifts between releases, so run esphome config before flashing. Load-cell scale/tare, flow k-factor, and the TDS polynomial are placeholders that need calibrating against your build.

Nothing here has been through a plumbing inspection. Backflow prevention on any potable connection is not optional, and is the one part of this you should not improvise.

License

MIT. See LICENSE.