A probabilistic approach for evaluating crack defects in reinforced concrete bridges

Cracks are one of the main evident symptoms related to structural health issues in reinforced concrete (RC) bridges. As such, assessing the position of cracks and the related features (e.g., orientation) can lead to perform a proper assessment of the structural health of bridges and then, to plan adequate interventions. As onsite surveys of these structures are the preferred method to perform structural health monitoring (SHM), automatic tools to support engineers are preferred, given the complexity and the burden posed on surveyors that, under specific circumstances, may lead to incorrect assessments. These issues could be amplified by the specificities of cracks, which represent a complex typology of defects that can be strongly related to the residual life of the structure. To address the above aspects, this study proposes an automated tool that, starting from a stack of images gathered for the structural elements in the RC bridge under investigation, performs a series of preprocessing steps to provide a comprehensive view of the structure and then uses a procedure based on deep neural networks to segment cracks automatically. Once detected, the tool can extract a complete pattern per each crack, thus providing the surveyor with a reliable and effective report containing crack’s location, orientation, and extension, thus defining an overall crack pattern for each element, which can be used as a base for planning future interventions. The method was tested on real-life case studies, showing its effectiveness and potentiality to improve traditional onsite surveys and defect detection techniques.