2026-10-14T18:43:16Z https://www.scidb.cn/oai

10.57760/sciencedb.012gx 2026-09-14T11:16:14Z

2026-09-14 Discontinuity-Informed REconstruction and Continuous Tracking (DIRECT) for Speckle-Free Reconstruction of Rock Deformation doi:10.57760/sciencedb.012gx en This dataset contains uniaxial compression mechanical response data for 20 double-flawed sandstone specimens and raw images of both surfaces of Specimen 1 during the post-cracking stage. It supports analyses of macroscopic mechanical response, surface crack propagation, and failure characteristics. Following the experimental procedures described in the paper, sandstone was obtained from Zigong, Sichuan Province, China. The specimens were 75 mm high with a rectangular cross-section of 20 mm × 40 mm. Each specimen contained two prefabricated flaws, each 20 mm long and inclined at 30° to the horizontal. Before testing, the specimens were dried at 110 °C for 24 h. Displacement-controlled uniaxial compression tests were conducted using an Instron 5982 mechanical testing system, with axial deformation measured using an Instron 2630-series extensometer. An artificial speckle pattern was applied to one observation surface, while the opposite surface retained its natural texture. Two industrial cameras simultaneously recorded surface deformation and crack evolution under LED illumination.The mechanical data are stored in the Sheet1 worksheet of Dataset.xlsx. The 20 data groups are arranged horizontally from Specimen 1 to Specimen 20, with a blank column separating adjacent groups. The first row contains specimen identifiers, the second row contains variable labels, and data records begin in the third row. Each group comprises four columns: time t in seconds (s), testing-machine crosshead displacement U in millimetres (mm), axial load F in kilonewtons (kN), and dimensionless strain ɛ. Each row contains measurements corresponding to a single sampling instant for the relevant specimen, and the time series for each specimen starts at 0 s. The nominal sampling frequency is 50 Hz, corresponding to a regular sampling interval of 0.02 s. For some specimens, the interval between the final two records is shorter than 0.02 s; users should therefore refer to the actual time values provided in the worksheet. Dataset.xlsx uses the standard Excel workbook format and can be opened using Microsoft Excel or other software supporting XLSX files.The image data are organized by observation surface into two directories, Specimen1_Non-speckled and Specimen1_Speckled. The former contains raw images of the natural-texture surface of Specimen 1, while the latter contains raw images of its artificially speckled surface. The original images have a resolution of 2048 × 2048 pixels and were acquired at 20 Hz, corresponding to an interval of 0.05 s between consecutive original frames. This release includes images 6500 through 6972, inclusive, from the original sequence for each observation surface, comprising 473 image pairs and 946 images in total. The elapsed time between the first and last frames of this continuous segment is 23.60 s. These image numbers indicate positions within the original image sequences and do not correspond to row numbers in the mechanical data table. Alignment between the images and mechanical response data requires reference to the acquisition time bases.The complete original image dataset for Specimen 1 contains 6972 image pairs, totalling 13,944 images and approximately 54 GB. Given the large volume of the complete sequences, this release provides only a selected continuous segment from the post-cracking stage. Earlier images were deliberately excluded from the shared subset; their exclusion does not indicate that images were not acquired during those stages. The released images document surface crack propagation and subsequent failure but do not cover the entire loading process. In the analyses reported in the paper, natural-texture images underwent region-of-interest cropping, resizing to a consistent resolution, and spatial registration before AllTracker-based dense trajectory extraction, DeepLabv3+-based semantic crack segmentation, and subsequent deformation-field reconstruction. The image directories in this release contain the raw photographs.Potential sources of error include load and deformation measurement errors, loading misalignment, end-contact effects, and image-acquisition effects associated with illumination variations, perspective projection, and out-of-plane deformation. Rock fracture; Speckle-free deformation measurement; Dense point tracking Mr. PUBLIC https://opendatacommons.org/licenses/by/1-0/ dataset Science Data Bank