Aging bridge infrastructure in the United States presents critical safety risks and economic burdens. Early detection of structural deterioration can significantly reduce maintenance costs and prevent catastrophic failures. Post-earthquake safety evaluation of buildings also requires rapid assessments in the affected area. However, current structural health monitoring practices rely primarily on visual inspections and localized testing and are slow, labor-intensive, and insufficient. Ground Penetrating Radar (GPR) provides a promising alternative or complementary approach, enabling fast, non-destructive evaluation of large structures by revealing internal features such as reinforcing bars, voids, and delamination. This study aims to develop a lightweight, compact GPR system mountable on an unmanned aerial vehicle (UAV) enabling rapid scanning of hard-to-access or hazardous locations and potentially detecting issues in stability, stiffness, and internal cracks, while ensuring operator safety. What differentiates this work from others is the use of software-defined radio (SDR), enabling the operation of multiple GPR methods without hardware alterations, unlike conventional radar systems. Software implementation allows flexible switching between GPR modes (impulse, SFCW, and custom methods) with power and frequencies selected for a specific target.
Moreover, our laboratory has recently reconstructed various objects buried inside multilayer soils using deep learning-based signal processing. For the current study, a neural network is being trained on simulated datasets of real bridge and building structures to detect internal anomalies that are difficult to detect with conventional signal processing. Experimental validation will be performed in collaboration with the Earthquake Engineering Laboratory at the University of Nevada, Reno, which provides small-scale bridge structures for real measurements. The combination of multimodal SDR-based sensing and deep learning post-processing demonstrates the ability to accurately detect subsurface features and embedded defects for structural health monitoring. Though these findings reflect controlled experimental conditions, future work will extend this approach to more complex real-world structural environments.