Towards a Solution for The Pre-Failure Concrete Element Crack Detection Problem

Abstract: Crack detection in concrete bridge elements is critical to the bridge’s durability and safety. The ability to link cracks with the typy of damaged element, location, and the moment of occurrence is critical for understanding the structure’s behaviour. This paper discusses a solution for segmenting structural elements on images and segmenting cracks using a deep learning network trained on a prepared dataset of pre-failure concrete cracks. While the detection of cracks in concrete representing the failure condition is currently a relatively straightforward task, the identification of narrow cracks representing the pre-failure state has not yet received a satisfactory solution.