Energy theft and tamper
Flow signatures, magnetic interference and seal-break patterns flagged at the meter — no waiting on cloud aggregation windows.
Edge AI
TinyML on the meter. Anomalies caught before they become incidents.
Our Edge AI Module runs anomaly detection and predictive calibration on the device itself — sub-milliwatt power budget, no cloud round-trip, decisions in milliseconds.
What it catches
Calibration drift
On-device cross-reference against reference patterns predicts drift weeks before legal-for-trade tolerance is breached.
Leak signatures (water/gas)
Continuous low-flow patterns under expected zero-flow windows flagged as potential leaks — false-positive rate tuned per network.
Technical envelope
- Compute · ARM Cortex-M33 with CMSIS-NN, 256 KB SRAM, 1 MB Flash
- Power · sub-mW inference, 10-year coin-cell life for sensor-node variant
- Models · TFLite-Micro, INT8 quantised, customer-trainable via our Model Studio
- Integration · DLMS event push, MQTT publish, REST callback — same protocol surface as the rest of our hardware
Curious whether it fits your fleet?
We'll run a 30-day pilot with your reference patterns and report the false-positive / true-positive split before you commit to anything.
Request a pilot