Abstract: To address the problems of fragmented multi-source data, spatiotemporal misalignment, and low fault diagnosis efficiency in coal mine working face historical scene reconstruction and fault tracing, this study investigated a coal mine working face historical scene reconstruction and fault tracing technology based on digital twin. Six key technologies were used to realize complex scene reconstruction and fault tracing: ① a high-precision 3D model at the scale of equipment such as the "three machines" was constructed using Physically-Based Rendering (PBR). ② A hierarchical data preloading mechanism was proposed to establish a three-level loading system consisting of a dynamic data layer, a static data layer, and an event data layer. ③ Full-element state synchronization of equipment such as shearers and hydraulic supports was carried out, focusing on geometric position and orientation, physical parameters, behavior logic, and rule-based judgment. ④ A multi-host time unification algorithm based on shearer position data was proposed, which performed feature extraction and time alignment on position data recorded by each host, thereby realizing precise time synchronization of multi-source heterogeneous data. ⑤ A playback consistency verification mechanism was established, and process parameter similarity was used to quantify playback accuracy. ⑥ A support loss fault tracing algorithm was designed, and a multi-variable coupled support-shifting dynamics model was established. A three-level early warning mechanism was implemented based on the matching degree between theoretical and actual stroke. Experimental results showed that the hierarchical data preloading algorithm maintained stable running time, with an average data loading time of 35.5 s, improving efficiency by 29.3% compared with the full backup algorithm. The average similarity of the middle-section follow-up parameters was relatively low, at 93.73%, while the average similarity of the follow-up parameters in the triangular coal area was relatively high, at 98.56%. Through historical scene reconstruction of the working face, changes in the appearance color of the support with support loss could be observed, which verified the correctness of the support loss fault tracing algorithm.