Digital twin technology serves as the cornerstone for real-time monitoring and bidirectional mapping in intelligent coal mining. Current digital twin creation methods predominantly rely on commercial software such as Unity3D (U3D) and Unreal Engine (UE), which are entirely dependent on foreign technologies. These approaches face bottlenecks in handling complex coal mining scenarios, including excessively large model sizes and poor real-time interactivity on web platforms. To address these challenges, this paper proposes a lightweight method for constructing a WebGL-based digital twin model of three key machines in intelligent mining faces. By integrating glTF format conversion and a progressive mesh merging algorithm, the method preserves critical geometric features (achieving a compression rate of up to 95.3%) while employing octree-based ray picking, device-adaptive rendering strategies, and hardware-performance matching to create lightweight geometric models for complex equipment. Experimental results demonstrate that this approach significantly reduces model initialization and rendering time, enhances rendering frame rates, and exhibits excellent compatibility with mainstream browsers. This work provides a lightweight, compatible, and self-controllable technical pathway for building intelligent digital twins in coal mining, ensuring both efficiency and intellectual property autonomy.