IoT-based virtual sensing application for Bridge Static Deflection estimation via Data-Driven approaches at the Edge

This research investigates the enhancement of low-cost sensor performances using data-driven approaches by deploying models directly at the edge. An experimental setup, consisting of a bridge model and a vehicle equipped with a smart sensor, was used to scan the bridge deflection and reconstruct the mid-span displacement caused by the vehicle passage. Two data driven models have been compared to find the best performing to be deployed at the edge. The findings demonstrate that incorporating physics-related features into the training process improves both model accuracy and computational efficiency on microcontrollers, significantly reducing the required computational effort and processing time. Moreover deploying algorithms at the edge offers greater flexibility within an IoT infrastructure, facilitating efficient information exchange between smart sensor nodes, gateways, and the cloud through wireless technologies.