Microstructural crack segmentation of three-dimensional concrete images based on deep convolutional neural networks

Abstract As a nondestructive imaging technology, X-ray CT has become an effective tool for studying the microstructural damage of concrete. However, autonomous identification and segmentation of microstructural cracks remains a challenge due to the same greyscales of voids and cracks in CT images. To address this problem, this paper develops a new method for microstructural crack segmentation of three-dimensional concrete images based on the deep convolutional neural networks. The model architecture and training scheme of the proposed network are specifically designed to achieve the high accuracy in the segmentation of narrowly opened cracks. Meanwhile, the method can also be used to separate aggregates from mortar with high precision. The segmentation results are compared with manual segmentation to validate the performance of the proposed method, demonstrating that the proposed method is capable of successfully separating microcracks from voids through their shapes and the aggregates from the mortar matrix with high precision. Finally, the three-dimensional concrete microstructure is reconstructed with microcrack patterns dependent on freeze-thaw actions, further manifesting the capability of the proposed method in the internal damage analysis of concrete.