GaugeEye

• Single-meter AI nodes already sell for $96 → one 5 MP wide view covers up to 6 dials, bringing the price per dial to $67–133, with tap-to-calibrate setup
• Camera dev boards sleep at about 3 mA → custom board powers the whole ESP32-S3 + camera domain to 0 V between reads; sleep target ≤ 12 µA
• Li-SOCl₂ cells cannot supply camera and LED pulses → Tadiran HLC-1520A pulse buffer, LED pulses capped
• Long arms resonate → short serrated-rosette arm (69 or 118 mm), short exposure, 3-frame median and drift-tolerant registration
• Glare, sun and night → flared hood, cross-polarised 850 nm IR flash, and a quality gate that holds a bad read instead of sending it
• No independent accuracy figure exists → no accuracy claim until the bench test passes (95% of reads within ±2% FS on 8 gauges)

Utility operatorOperations superintendentGaugeEye (LTE-M)IEMI cloud

1. 01Trigger: every 15 minutes the camera wakes and reads the two pump discharge gauges and the wet-well level gauge
2. 02Detection: the cloud converts readings to psi and ft and sees pump 1 discharge below 8 psi for 2 reads while pump 2 is normal
3. 03Alert: the on-call operator gets a text flagging a possible air-lock or worn impeller, with the 24 h trend
4. 04Action: the operator decides whether to roll a truck; the superintendent exports the monthly CSV

Outcome: 96 readings a day per gauge instead of one visit, and trucks roll only on exceptions; about $28–43k/yr saved across 10 stations (est.)