Abstract: In mining engineering, physical similarity model testing has been widely utilized as a critical research tool. However, traditional two-dimensional simulations face technical limitations in accurately replicating the coupling mechanisms between three-dimensional geological environments and dynamic mining processes. To address this challenge, a high-precision intelligent 3D physical simulation platform (5 m×5 m×3.2 m) based on source damage reduction and self-healing mining principles was developed. The platform employs modular programmable servo control for 280 square excavation units (±1 mm precision) and 280 load sensors, enabling 3D dynamic reconstruction of spatiotemporal stress evolution in mining fields. Validated through simulations of 8.8 m ultra-high mining height at the 12401 working face of Shangwan coal mine, the platform confirmed the "two major, one minor" periodic weighting pattern (strong weighting interval: 15 m; weak: 8–11 m), with key parameters including a support pressure coefficient of 2.5 and surface subsidence coefficient of 0.72. A multidimensional monitoring system was innovatively integrated, combining a 2 mm industrial endoscope (WS-K), ground-penetrating radar (ZTR11-Z), 3D laser scanning (FARO Focus Premium70, 2 million point clouds), and motion camera time-lapse imaging. This system captured the gradient attenuation process of overlying rock failure in a elliptical platform (trapezoidal rotation) shape, quantitatively revealing core indicators: subsidence basin proportion (34%), caving-mining ratio (3.8), and fracture-mining ratio (14), thereby elucidating the nonlinear diffusion characteristics of mining-induced disturbances. A sand-paste-ash low-strength, high-brittleness material system (uniaxial compressive strength ≤1 MPa, 2–3 cm layered structure) was developed, integrated with a "quantitative feeding–lateral pressure feedback–humidity monitoring" collaborative control strategy. This effectively resolved challenges in material uniformity and compaction control for 3D simulations. The platform’s modular design—combining 280 square and 40 strip excavation modules—enables refined simulations of strip mining, backfill mining, and deformation-controlled scenarios, ensuring standardized and repeatable testing. This advancement drives the transformation of coal mining similarity simulations from experience-driven practices to data-quantified paradigms.