Internal Crack Detection and Cross-Section Reconstruction of Reinforced Concrete Structure Based on Ultrasonic Tomography and Deep Learning

Reinforced concrete (RC) structures are susceptible to cracks. Surface character of cracks on concrete structures is not sufficient for structural condition assessment due to the anisotropy. Largely to this date, human-conducted periodic and visual inspection, along with prior studies are not competent for automatic and non-destructive detection of internal cracks in RC structures. In the present study, a novel deep learning-based framework for automatic and non-destructive ultrasonic tomography reconstruction internal crack detection of RC elements and was proposed. Concrete ultrasonic low-frequency tomograph device A1040 MIRA was utilized to acquire B-scans. UNet was leveraged as the underlying framework to detect crack(s) in B-scan and reconstruct geometrical morphology of a designated cross-section. Four RC beams were designed and made to collect dataset. Intersection-over-Union of validation, and testing set reported to be 44.80% and 43.44% respectively, indicating the efficiency of the proposed method.