Meso-damage analysis of concrete based on X-ray CT in-situ compression and using deep learning method

Traditional concrete failure research mostly based on surface cracks to analyze its macro mechanical properties, which cannot obtain the internal structure and failure process on a mesoscopic level. The internal topography of concrete can be acquired by using X-ray CT to scan the entire concrete specimen, and the internal structure change caused by stress can be captured when combined with an in-situ CT loading device. Thus, in-situ X-ray computed tomography (X-ray CT) scanning was performed on concrete under uniaxial compression, and the real-time CT images of the meso-damage evolution process inside concrete were obtained. Deep Learning method was applied to extract two-dimensional (2D) and three-dimensional (3D) cracks for visual analysis and quantitative characterization. The damage evolution process, damage variables and Poisson's ratio of concrete were analyzed based on CT values. The results revealed that crack formation and propagation did not necessarily occur at the interface, the low-density area was prone to cause concrete damage. The crack morphology obtained by deep learning was highly consistent with the distribution of cracks in concrete. The law of crack extension, distribution and expansion could be represented through crack thickness, crack voxel number and crack area. The initial meso-damage distribution of concrete had an important impact on the occurrence of localized damage. The results of CT value showed that the meso-damage of concrete develops from inside-out. The damage evolution process was divided into compression stage Ⅰ, damage stage Ⅱ and destruction stage Ⅲ. The damage variable and Poisson's ratio were evaluated by CT value could further characterize the deterioration degree of concrete.