TinyML-enabled structural health monitoring for real-time anomaly detection in civil infrastructure

Abstract

SHM is an essential requirement to maintain civil infrastructure safety while extending its operational lifespan, including bridges, buildings, and dams. The conventional SHM systems need centralized data processing together with high-power sensors, even though they remain expensive, while needing significant amounts of energy and are not appropriate for areas lacking resources or infrastructure. The TinyML-based SHM (TSHM) system performs edge computing and machine learning-based computations which resulting in real-time structural integrity analysis with low power consumption. This work proposes a scalable TinyML-based SHM framework capable of real-time anomaly detection using low-power edge devices. The integrated system utilizes inexpensive accelerometers together with strain gauges and environmental sensors for the continuous acquisition of real-time data that includes vibration patterns, deformations, and environmental factors such as temperature and humidity. A resource-conserving anomaly detection model operates on edge devices to monitor and identify structural defects as well as damage in real-time. TSHMS achieves device-based critical decisions at minimal delay through the unification of structural dynamics principles with real-time sensor information and without needing cloud-based processing. The developed system performs structural anomaly detection with 92 % accuracy when compared to ordinary SHM systems while using 40 % less energy. The study illustrates how TinyML technology enables effective and sustainable structural health monitoring of civil infrastructure through AI-based decentralized operations with reduced energy needs.