Segmentation of Spatial Crack Structures in Concrete by Deep Learning Enabling Image Based Characterization

Cracks in concrete structures are an integral part of this heterogeneous material. Characteristics of cracks induced by standardized tests yield valuable information about the tested concrete formulation and its mechanical properties. Observing cracks on the surface of the concrete structure leaves a wealth of structural information unused, since a crack in concrete is rarely a planar structure but rather spatial. Computed tomography enables looking into the sample without interfering with or destroying the microstructure. The reconstructed tomographic images are 3d images, consisting of voxels whose gray values represent local X-ray absorption. In order to identify voxels belonging to the crack, so to segment the crack structure in the images, appropriate algorithms need to be developed. Convolutional neural networks are known to solve this type of task very well given enough and consistent training data. However, fiber reinforced concrete has not been widely analyzed in the literature, since it is hard to obtain segmented cracks excluding the fibers. We overcome this problem by adapting a 3d version of the well-known U-Net and training it on semi-synthetic 3d images of real concrete samples equipped with simulated cracks. We thus lay the foundations for large experimental quantitative studies of 3d crack initiation and development in various types of concrete.