Reconstruct laminated shale cores/specimens for numerical modeling by physics-informed image processing

This paper proposes end-to-end physics-informed image processing to construct the digital twin of laminated shale cores/specimens and extract the number, spacing, length, and frequency of bedding planes. Specifically, the image stitching technique was applied to eliminate any distortion that may arise from projecting 3D objects onto 2D images. Then, based on shale physics information (gradient difference between different pixels in the image due to various mineral components), an edge detection algorithm was used to extract one-pixel-wide bedding planes from the shale image. Furthermore, a depth-first search algorithm was used to collect the point coordinates of each line. The coordinates gathered from the 2D image were used to generate the model of the laminated shale specimens/cores in UDEC. A variable sensitivity analysis of the reconstructed shale was performed using the proposed method. This work will improve the reliability of laminated shale modeling to further investigate the influence of bedding planes on the macro mechanical properties of shale rockmass.