A Scalable Full-Stack System for High-Precision Industrial Vibration Monitoring

Abstract

Industrial vibration monitoring is critical for predictive maintenance, but existing solutions face fundamental limitations in both sensing and inference. High-precision sensing often relies on power-hungry active hardware, while multi-source reasoning in structurally coupled environments remains challenging due to signal superposition and unknown source counts. These limitations hinder accurate and scalable deployment in real industrial settings. We present a full-stack vibration monitoring system that jointly addresses these challenges. On the sensing side, we design VibScatter, a battery-free vibration sensor that embeds mechanical vibrations into RFID backscatter, enabling high-resolution vibration measurement without wired power. On the reasoning side, we propose NeVOP, a physics-aware neural vibration operator that amortizes inverse wave propagation to enable scalable multi-source inference under sparse sensor deployments. We prototype the system and evaluate it through comprehensive experiments. In practical deployments under 30 dBm transmit power with a 12 dBi antenna (42 dBm EIRP), the full-stack system achieves up to 93.1% reasoning precision, with around 0.1 m source localization error, and 0.23 Hz frequency estimation error. Under the legal 36 dBm EIRP setting (24 dBm transmit power with a 12 dBi antenna), the system also maintains reliable communication at a 3 m reader-tag distance.