Edge-AI Enabled Structural Health Monitoring of Civil Infrastructure

Abstract

Ageing bridges, buildings, and other civil infrastructure require continuous condition assessment to prevent catastrophic failure, yet conventional structural health monitoring (SHM) systems stream raw vibration data to a central server, incurring prohibitive bandwidth, latency, and energy costs that limit scalability. This paper proposes an edge-artificial-intelligence (edge-AI) SHM framework in which lightweight one-dimensional convolutional neural networks execute damage detection directly on resource-constrained sensor nodes, with a fog layer performing damage localization and a cloud layer maintaining the long-term structural record. The hierarchical edge–fog–cloud design draws on architectural principles established for fog-computing-enabled intelligent transport systems. Evaluated on a benchmark bridge-damage dataset and a numerical girder model, the on-device network attained 95.5 percent damage detection accuracy at a 15 dB signal-to-noise ratio while reducing end-to-end detection latency from 1850 milliseconds for a cloud-only pipeline to 95 milliseconds and cutting uplink bandwidth by 93 percent. The framework localized damage on a thirty-metre girder to within one metre. These results demonstrate that pushing inference to the edge makes dense, real-time, and energy-efficient monitoring of large infrastructure portfolios technically and economically viable.