Fracture evolution behavior of deep coal with bedding planes: An in-situ μCT study based on deep learning

Deep coalbed methane (CBM), a clean and low-carbon unconventional natural gas resource, plays a crucial role in energy security. Its production efficiency depends heavily on fracture behavior, which is intrinsically influenced by distinct natural fractures and bedding planes. In this study, in-situ compression experiments based on in-situ micro X-ray computed tomography (μCT) were conducted to investigate the fracture evolution behaviors of deep coal with different bedding angles, as well as the roof/floor rock. A deep learning-based segmentation method, called enhanced U-Net was proposed to precisely capture the fracture characteristics, demonstrating superior performance over traditional U-Net across all selected metrics. Based on the segmented results, internal fracture evolution was analyzed, and the influence of bedding angles was further explored. A model combining the weak plane criterion and Griffith criterion was employed to elaborate on these effects. The results indicate that the coal with a bedding angle of 70° undergoes significant shear fracture propagation along the bedding direction during the compaction stage, exhibiting mixed shear-tensile failure; while the coal with a bedding angle of 80° presents tensile failure. Then the relationship between the permeability and complexity of fracture networks was discussed. Permeability analysis reveals that although coal with a 70° bedding angle forms a more complex fracture network at failure, it has lower permeability than coal with an 80° bedding angle. This study provides a micro-scale insight into the fracture evolution mechanism of deep coal by revealing the influence of bedding angle, laying the foundation for optimizing hydraulic fracturing schemes and enhancing CBM extraction efficiency.