AI Smart Energy Monitor based on nRF9160

Client

An energy analytics company providing residential energy monitoring that breaks down electricity usage by individual appliance without per-appliance sensors.

Challenge

Sampling mains current at sufficient resolution (1MHz) to capture appliance electrical signatures, then classifying them on-device with consumer-grade privacy.

Product Concept & Architecture

A Rogowski coil clamps around the mains feed. A 24-bit ADS131M04 ADC samples at 1MHz. The Nordic nRF9160 processes 1-second current windows through a neural network for appliance identification.

Engineering Solution

The nRF9160's Cortex-M33 runs a 1D-CNN + LSTM hybrid model (150KB, INT8 quantized) that processes the 1MHz current waveform and extracts appliance-level signatures. The model identifies 12 appliance types with 93% accuracy. Only the appliance-level breakdown is transmitted via LTE-M.

Result

93% appliance disaggregation accuracy across 12 common appliances. Instant clip-on installation. Zero raw data leaves the home.