Abstract: Addressing the low efficiency, poor collaboration, and inadequate geological adaptability in the conversion between 2D drawings and 3D models for coal mine roadway design, this study proposes an efficient conversion technology that integrates semantic parsing, multi-constraint parametric modeling, and a bidirectional dynamic linkage mechanism.By constructing a metasemantic knowledge graph for roadway, combined with geological pre-imagined section constraints and equipment operational parameters, a feature recognition-based 3D automatic reconstruction algorithm is established, which realizes that 2D design is modified as 3D model and bidirectional driving of tunneling equipment.Experimental results demonstrate that the method achieves a modeling efficiency of 32 m³/s in typical complex geological roadways, with a geometric error of ≤0.3%,a dynamic response latency of <300 ms, and supports real-time inverse driving tunneling equipment's posture adjustments.This provides a high-precision, strongly collaborative digital foundation for intelligent tunneling in coal mine.