AI Body Composition Scale based on nRF52832

Client

A health tracking company developing a multi-user smart scale that automatically identifies who is stepping on it without manual profile selection.

Challenge

Achieving accurate body composition metrics across a wide range of weights (5-200kg) and distinguishing between household members who have similar BMIs.

Product Concept & Architecture

The Nordic nRF52832 reads four 3-wire load cells via a 24-bit ADS1256 ADC for weight. Bio-impedance is measured using a TI AFE4300 analog front-end. Four stainless steel electrodes are embedded in a tempered glass platform.

Engineering Solution

A two-stage recognition pipeline runs on the nRF52832: first, weight alone narrows candidates to 2-3 users. Second, a 1D-CNN classifier processes the BIA phase angle and impedance at 50kHz across 8 frequencies to generate a unique bio-impedance signature. The system learns new users over 3 weigh-ins. Data syncs to cloud via ESP32 companion module for long-term trend tracking.

Result

99% auto-identification accuracy across 6-person households. Medical-grade BIA accuracy validated against DEXA scans (r²=0.94 for body fat). User adherence increased 3x compared to manual-selection scales.