An Attention-Based Detection Method of Fatigue Cracks on Steel

Abstract

Steel structures are susceptible to fatigue cracking under cyclic loading, which can lead to catastrophic structural failure. In the incipient phase of crack propagation, the width of fatigue cracks typically measures less than 0.1 mm, making them difficult to detect using standard imaging techniques. This study presents a novel approach to crack detection on steel structures by tracking the displacement field on the structural surface derived from visual data. Initially, video or sequential images of the target structure under loading are captured and processed using an enhanced dense feature-matching model. The surface displacement field is then computed from the coordinate difference of the numerous matched feature points. By extracting discontinuities within the displacement field, fatigue cracks can be localized. Two case studies were conducted to validate the methodology: one involving a with a pre-existing crack and another steel plate with fatigue crack propagation. The findings indicate that the proposed method can be used to detect minuscule cracks, with crack widths as small as 5 μm. Factors potentially influcencing the method, including the texture of the steel surface, the region of interest (ROI) area ratio, the density of matching, and the resolution of input images, were discussed. Compared to traditional image-based semantic segmentation techniques, this approach is more convenient and precise, offering a promising avenue for the nondestructive evaluation of steel structures in civil engineering.