Damage Identification in Concrete Bridges Using Unmanned Aerial Vehicles and Neural Networks

Abstract

Bridge monitoring systems using cameras and unmanned aerial vehicles (UAV) are increasingly being used worldwide. Additionally, artificial intelligence techniques are being used to improve performance in the structural damage detection and processing stage. This article shows a non-destructive methodology for damage identification using neural networks in a real bridge on the coast of Peru. The 104 m long Villena Rey bridge is the case study inaugurated in 1960 to improve the conditions and vehicular resilience of the Malecon de la Reserva avenue crossing in Lima. As a first step, many images were taken using photogrammetry with a UAV and the noise was filtered for data preparation. The data is then prepared and labeled to train the neural network model in conjunction with flexible training tools and an optimal architecture using one of the most efficient systems known as YOLOv7. The results show an optimal calibration of the system with percentages that exceed 60% in the identification of structural damage in bridges. Finally, this research work has a great contribution since it would be the first time that these modern technologies are used in developing countries such as Peru in South America.