Combining DPM with Shape Semantics-Driven 3D Model Reconstruction to Improve Parametric Design Efficiency for Intelligent Products

Abstract

The rapid iteration of intelligent products places increasingly higher demands on parametric modeling, making it difficult for traditional modeling approaches to meet strict industrial design requirements for complex mechanical parts and editable functional structures. To address this, this paper proposes a 3D model reconstruction method that combines DPM with shape semantics. First, a high-fidelity point cloud is generated using a diffusion probability model. Semantic label parsing converts features into geometric parameters. Subsequently, constraints such as length and curvature are embedded in the sampling process to achieve controllable generation. The point cloud is then reconstructed into a mesh and optimized to ensure geometric integrity and parametric editability. Finally, the integration with Computer-Aided Design (CAD) systems enables rapid adjustment and redesign, thereby improving the efficiency of parameterized modeling of intelligent products. Experiments show that the proposed method achieves a Chamfer distance of 0.012 mm in the Electronics category, significantly outperforming the Generative Adversarial Network (GAN) method by 0.025 mm. When the point cloud complexity is 25,000 points, the generation time is only 3.1 s, which is highly efficient. The method achieves an accuracy rate of over 87% in semantic consistency. The research results significantly improve the efficiency of parametric modeling, making it possible to rapidly generate and adjust complex 3D models for industrial design. For antenna housings, radomes and other wave-related industrial structures, controllable 3D reconstruction can provide reusable geometric modeling support.