Shield tunnels face significant structural health challenges attributed to the presence of random cracks. Due to the invisibility of underground structures, accurately quantifying crack morphology, particularly in depth, remains a formidable challenge. To address this, a cohesive-based multiscale modeling approach incorporating concrete macroscale and mesoscale information is employed to simulate the evolution of random cracks. Concrete is modeled as a three-phase composite of aggregates, mortar, and the interfacial transition zone (ITZ), with polyhedral aggregates generated according to Fuller’s grading curve and cohesive elements governed by a bilinear traction-separation law embedded in the mortar and along aggregate-mortar interfaces to capture mixed-mode cracking behavior. Red–green–blue (RGB) crack images obtained from numerical simulations are subsequently processed using a quantitative recognition frame-work based on the A* algorithm to measure crack depth. Results reveal that energy dissipation increases nonlinearly with the total crack length in depth, reflecting a progressive transition of fracture modes from pure tensile to tensile-shear, and ultimately to compression-shear. The framework enables precise identification of complex multi-crack patterns and quantification of crack depth, providing insights into fracture mode transitions and associated energy dissipation, and supporting reliable assessment of tunnel lining structural health.