The lithology and composition type of an aquifuge in overburden play a crucial role in influencing the crack evolution and permeability changes of the aquifuge. This study utilized the high-temperature and high-pressure rock triaxial seepage test system to conduct triaxial compression tests on mudstone, sandstone, and their combined rock samples. The mechanical characteristics and permeability evolution of each lithology law during the failure were investigated. Furthermore, computed tomography (CT) scanning technology was utilized for the three-dimensional (3D) reconstruction and theoretical permeability calculation of single and combined rock samples. The results indicated that the stress–strain curves for single and combined rock samples exhibited similar patterns, which were divided into four stages: pore compaction, linear elasticity, yield deformation, and post-peak residual deformation. The peak strength of rock samples positively correlated with confining pressure. Permeability trends for mudstone and sandstone exhibited an “N”-type pattern characterized by “slow decrease–gradual stabilization–sudden increase–rebound decrease”, while the permeability of mudstone–sandstone combined rock followed a “U”-type pattern of “initial decrease–stabilization–subsequent increase”. Notably, the permeability of the combined rock samples was significantly lower compared to the single rock samples. The failure mode indicated that fractures in a single rock sample transversed the entire sample, whereas failures in the combined rock samples were confined to the mudstone component. This observation accounted for the differences in the permeability changes between the rock sample types. Additionally, the theoretical permeability results from the 3D reconstruction correlated with the experimental results.

To overcome the limitations of conventional numerical simulations of cemented gangue backfill (CGB), this study developed a refined PFC2D model that incorporates the actual particle size distributions of coal gangue and river sand. Randomly distributed steel fibers were generated using FISH programming. Based on uniaxial compression tests and scanning electron microscopy (SEM) observations, the influence of steel fibers on the mesoscopic damage mechanism of CGB is systematically investigated. The results indicate that: (1) the refined model significantly improves the reliability of numerical simulations, accurately reproducing stress concentration within coarse aggregates and the steel fiber “bridging effect”; (2) a steel fiber volume fraction of 0.8% optimizes force chain distribution and suppresses crack propagation, promoting a transition in failure mode from brittle shear failure to ductile compressive–extrusion failure mode, with the peak strength and residual strength increased by 23.7% and 40.2%, respectively, compared with the fiber-free specimen; (3) PFC simulations reveal that steel fibers markedly retard damage accumulation by modifying the force chain network and crack propagation paths; and (4) SEM analysis demonstrates that steel fibers enhance the toughening effect through the interfacial transition zone, whereas excessive fiber content (1.2%) leads to fiber agglomeration and a 62.5% increase in porosity, resulting in performance deterioration. This study provides a robust theoretical framework for gradation reconstruction and refined fiber modeling in the design of roadside backfill materials.