Application of machine learning technique for X-ray CT image segmentation of cement grouted rock

X-ray CT is a useful tool for observing internal structure of rock samples in non-destructive manner. It has been employed by many researchers in rock engineering and geological fields to see void structure, crack distribution, crack apertures, inhomogeneous material distribution and so on. This study deals automatic image segmentation using deep learning algorithm for X-ray CT imaging. 15 core samples were collected from check hole (depth = 10~20 m) after grouting in a gravity dam construction site. The samples were scanned with X-ray CT. Two cores are collected from mudstone layers to analyze, namely, Cores A and B. Core A is at the depth of 18.00 meter with a length of 130 mm, which has planar open joint with cement grout. Core B is at the depth of 27.5 meter with a length of 100 mm, which looks crushed and filled with cement grout. The CT images were automatically segmented into rock part, grout part and void part using a deep learning algorithm. The automatic segmentation method successfully identified rock part and grout part for the mudstone specimen with simple planar fracture, although the CT value range of rock part and grout part are overlapping each other. It also identified the difference of cement milk mixtures. However, for the specimen having intricate fracture geometry that was extracted from sheared stratum, narrow ungrouted cracks (less than 1.0 mm width) were sometimes misidentified. This indicates that automatic segmentation requires more training data for intricate fracture geometry.